1. A method for estimating enteromorpha disaster spatial distribution under dynamic time-space correlation of microblog public sentiments is characterized by comprising the following steps: the method comprises the steps of extracting corresponding enteromorpha disaster characteristics and microblog social public opinion characteristics to construct a bidirectional association neural network by using enteromorpha disasters and corresponding microblog social public opinion data of the past year, correcting results in a stable state of the bidirectional association neural network by using a residual error network in the process of predicting the enteromorpha disasters by using the bidirectional association neural network, correcting results of operation of the neural network by using an enteromorpha and microblog time-space association rule extracted based on an APRIORI algorithm, and finally obtaining estimated spatial distribution and development situation of the current enteromorpha disaster, so that the purpose of estimating the spatial distribution and development situation of the enteromorpha in the process of obtaining the microblog social public opinion data is realized; the method specifically comprises the following steps:

extracting enteromorpha disaster characteristics, wherein the enteromorpha disaster characteristics to be extracted comprise the coverage area of an enteromorpha disaster, the drift gravity center of the enteromorpha disaster and the length, width and rotation angle of a minimum enteromorpha disaster distribution circumscribed rectangle;

extracting microblog social public opinion characteristics, wherein the microblog social public opinion characteristics to be extracted comprise a hot spot radiation range, a hot spot radiation intensity, a standard deviation ellipse distribution range and a standard deviation ellipse rotation angle;

step three, constructing a bidirectional association neural network based on enteromorpha disaster characteristics and microblog social public opinion characteristics;

taking enteromorpha disaster characteristics and microblog social public opinion characteristics of different years as test sets, taking the rest years as training sets, carrying out bidirectional association network training of different years, and constructing a residual error network based on the test sets and result sets of different years;

step five, extracting enteromorpha and microblog social public opinion spatiotemporal rules based on enteromorpha disasters and microblog social public opinion characteristics in the past year;

and step six, inputting microblog social public opinion data corresponding to the enteromorpha disaster to be predicted, and performing enteromorpha distribution prediction based on the microblog social public opinion.

2. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the specific implementation manner of the step one is as follows,

when the enteromorpha disaster in history occurs, a group of N enteromorpha disasters exist_dThe MODIS image for enteromorpha disaster interpretation and analysis is Year ═ Year₁，year₂，...，year_NFor any Year in the set Yeast, the MODIS date for interpreting Enteromorpha disasters isWherein N is the number of total years, N_yearDays of year;

step 11, calculating the Coverage Area (CA) of the enteromorpha disaster extracted based on the MODIS image according to the formula (1);

CA＝n×cellsize×cellsize (1)

in the formula (1), n is the number of pixels actually covered by the enteromorpha disaster in the image, and cellsize is the resolution of the image;

step 12, calculating the drift gravity center of the enteromorpha disaster according to the formulas (2) and (3)

In the formulae (2) and (3), x_iAnd y_iRespectively representing the abscissa and the ordinate of the occurrence position of the enteromorpha, wherein n is the number of pixels actually covered by an enteromorpha disaster in an image;

step 13, any point (x) in the monitoring range is checked_i,y_i) The enteromorpha disaster is applied with a counterclockwise rotation of an angle theta to obtain a new coordinate position (x'_i(θ)，，y′_i(θ),) which are calculated by the formula (4) and the formula (5); theta is in the range of [0 DEG, 180 DEG ]]Step length is 1 degree, and initial value is 0 degree; calculating the abscissa x 'of the theta values'_i(theta) and ordinate y'_i(θ); all x'_i(theta) and y'_i(theta) range of maximum difference as the length L of the outer rectangle of the Enteromorpha distribution₁(theta) and width L₂(θ) having the calculation formulas of formula (7) and formula (8), respectively; calculating the minimum circumscribed rectangle area S (theta) when the theta takes different values, as shown in a formula (8); selecting the theta corresponding to the minimum area of the circumscribed rectangle₀As the angle characteristic of the distribution of the enteromorpha, the characteristic is shown in a formula (9); take theta₀Corresponding length L₁(θ₀) And width L₁(θ₀) As the characteristics of the length and width of the distribution of the enteromorpha, the characteristics are shown in a formula (10) and a formula (11);

L₁(θ)＝R(X′)，X＝{x′₀(θ)，x′₁(θ)，...，x′_n-1(θ)} (6)

L₂(θ)＝R(Y′)，Y＝{y′₀(θ)，y′₁(θ)，...，y′_n-1(θ)} (7)

S(θ)＝L₁(θ)L₂(θ) (8)

minS(θ)→θ₀ (9)

L₁＝L₁(θ₀) (10)

L₂＝L₁(θ₀) (11)

in the formula (6) and the formula (7), R (·) represents a range function; in the formula (9), min represents the minimum value.

3. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the specific implementation of the second step is as follows,

acquiring microblog data with keywords of enteromorpha and green tide by using a Singler microblog open interface, wherein the release time is every day in a Year set Year, and the acquired microblog social public opinion data comprises microblog text content, microblog text sending positions and microblog release time; setting that an MODIS image capable of interpreting enteromorpha disaster exists in Day, and the Year of the MODIS image is Year, the corresponding social public opinion data is all data before the date Day of Year, and extracting microblog social public opinion characteristics for each Day in the date Day corresponding to Year in Yeast;

step 2.1, discovering the hotspot position B of social public sentiment with date (year, day) according to the local Moran index, wherein B is { B ═ B₁，B₂，...，B_MM is the number of the discovered hotspot positions; for any social public opinion point, calculating the linear distance between the social public opinion point and all the hot point positions according to the spatial position of the social public opinion point, selecting the hot point position with the closest distance as the hot point position, and selecting the hot point position B_mThe corresponding social public opinion point set is P_m，P_mEach piece of data in and the hot spot position B_mThe set of abscissa distances therebetween is DX_mThe set of ordinate distances being DY_m；

Step 2.2, calculating any hot spot position B according to formulas (12) and (13)_mRadiation range ofAndB_m∈B；

in the equations (12) and (13),as a set P_mThe number of the elements of (a) is,andrespectively representing sets DX_mAndthe ith element;

step 2.3, calculating any hot spot position B according to the formula (14) and the formula (15)_mRadiation intensity p (B) of_m)；

In the formulae (14) and (15), λ (B)_m，t_i) Represents in unit time, B_mMean of public sentiment data points in the radiation range, t₀Represents unit time, E (-) represents mean, N (-) represents count;

step 2.4, setting the number of all social public opinion point sets P as np, and calculating the central coordinates (SDE) of the microblog social public opinion points according to a formula (16) and a formula (17)_x,SDE_y) (ii) a Calculating a standard deviation ellipse rotation angle alpha of the microblog social public opinion point according to a formula (18); calculating the distribution range sigma of the standard deviation ellipse of the microblog social public opinion point on the X axis and the Y axis according to the formula (19) and the formula (20)_xAnd σ_y；

In the equations (18), (19) and (20),is the abscissa x of the ith public opinion point_iTo the center abscissa SDE_xThe distance of (a) to (b),is the ordinate y of the ith public opinion point_iTo the center ordinate SDE_yThe distance of (c).

4. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the concrete implementation manner of the third step is as follows,

step 31, regarding enteromorpha disaster with date of (year, day), using the social public opinion characteristics of microblog as

Taking the characteristics of enteromorpha disaster asMixing X_year，dayAnd Y_year，dayEach feature in (1) is processed from the decimal data into binary form and 0 is replaced with-1; performing the step on enteromorpha disaster data and microblog social public opinion data of each date (Day) in the Year set and the corresponding Day set to form an X set and a Y set, wherein X ═ { X ═ X_year，day|year∈Year，day∈Day_year}，Y＝{Y_year，day|year∈Year，day∈Day_yearAll of X and Y contain N_dEach element of X and Y has a sequence length of N_XAnd N_Y；

Step 32, calculating a weight matrix W according to the formula (21) to obtain a bidirectional association neural network BAM;

in the formula (21), k is the sequence number recorded in X and Y, and W is a size N_X×N_YOf the matrix of (a).

5. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the specific implementation process in step four is as follows,

step 41, circularly executing (1) - (3) for any element Year in the Year set Year;

(1) recording that year is matched between X and Y, and recording the result meeting the requirement as X_yearAnd Y_yearAnd the result of the non-compliance is recorded as X_year' and Y_year', the number thereof is N_year′；

(2) According to equation (22), a weight matrix W is calculated_yearThe corresponding bidirectional associative neural network is BAM_year；

(3) Inputting X according to the formula (23), the formula (24) and the formula (25)_yearAnd the loop iterates the calculation until X_yearAnd Y_yearStable, i.e. no change, to obtain the predicted Y_year′；

In equations (23), (24) and (25), f (a, b) is the activation function;

step 42, summarize all Y_year', replacing-1 therein with 0 and restoring to decimal system to form newPredicting a sequence set Y' of characteristics of enteromorpha disasters; the residual network for Y and Y' can be described as equation (26); fitting formula (26) based on the principle of minimum root Mean Square (MSE) error to obtain parameter omega₁And b₁A value of (d);

Y-Y′＝ω₁Y′+b₁ (26)

in the formula (26), ω₁As a weight template, b₁Is an offset.

6. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the concrete implementation manner of the step five is as follows,

step 51, setting A and B as the states of some two features in X or Y, D as the set of all records in X or Y, and D as the set of all records in which these two feature states are A and B, respectively, and calculating the confidence (A → B) that B also occurs when A occurs according to the formula (27);

step 52, based on APRIORI algorithm, with the maximum confidence degree as the target, the rule with 100% confidence degree in the set X and Y is found and is marked as rule_XAnd rule_Y。

7. The method for estimating enteromorpha disaster spatial distribution under the dynamic space-time correlation of microblog public sentiments according to claim 1, which is characterized in that: the concrete implementation process of the step six is as follows:

step 61, inputting year 'and date day' of enteromorpha disaster distribution to be predicted, and acquiring microblog social public opinion data with year ', date before day' and keywords of enteromorpha and green tide by using a microblog open interface;

step 62, executing the step two, and obtaining the current public opinion information characteristics

Processing x' into binary form, and replacing 0 in the binary form with-1;

step 63, inputting x ' into the BAM according to the formulas (28), (29) and (25), performing cycle calculation, and ending the cycle when x ' and y ' are stable, namely no longer changed;

in the formulas (28) and (29), p and q are the serial numbers of the elements in x 'and y', respectively;

step 64, mixing y'_qIn the enteromorpha prolifera disaster distribution characteristics, the-1 is replaced by 0 and the enteromorpha prolifera disaster distribution characteristics are recovered

Step 65, according to the formula (30), correcting the BAM result by using a residual error network to obtain a corrected result y';

y′′＝ω₁(y′+I)+b₁ (30)

for each type of feature in y', rule is used, step 66_YJudging all the rules, if the rules are met, taking y 'as a final result, and if the rules are not met, replacing the part which is not met in the y' with the part which meets rule_YAny value of the medium conditions as the final result.

Background

In the process of enteromorpha disaster occurrence in the past year, business departments need daily emergency monitoring work of the enteromorpha disaster. In the daily monitoring process, all or part of direct enteromorpha monitoring data are often lost, such as: unmanned aerial vehicle and aircraft remote sensing are taken photo by plane and are restricted by weather greatly, the monitoring range of satellite is big but receive influences such as cloud, fog easily, ocean monitoring ship monitoring range is less and efficiency is lower etc.. In the process of occurrence, development, outbreak, extinction and disappearance of enteromorpha disasters in the past year, relevant information and individual viewpoints related to the enteromorpha disasters can be published on the enteromorpha microblogs by the users of the green microblogs in different places, so that a certain amount of social public opinion data exists on the green microblogs when the enteromorpha disasters occur, and the time and space distribution information of the occurrence of the enteromorpha disasters can be determined in an auxiliary manner. However, the microblog social public opinion information has wide directivity and low density and value, for example, the microblog social public opinion information can only provide the geographical position, microblog text content, social attention and the like of a microblog publisher, and less information directly pointing to enteromorpha disaster space distribution and disaster situation is provided. How to estimate the development situation and spatial distribution of enteromorpha prolifera on the same day based on microblog social public opinions under the condition of directly monitoring data loss or poor data quality on the same day so as to meet the requirement of enteromorpha prolifera disaster emergency rescue is a key difficulty.

Disclosure of Invention

In order to solve the key difficult point problems, the enteromorpha prolifera development situation on the same day is dynamically estimated according to the microblog public sentiments, the research fully considers the space-time correlation relationship between the characteristics of the microblog social public sentiments such as space-time distribution and quantity and the characteristics corresponding to the enteromorpha development, and a method for estimating the enteromorpha prolifera disaster space distribution under the dynamic space-time correlation of the microblog public sentiments is provided;

according to the technical scheme, firstly, the disaster monitoring result monitored based on the MODIS image is subjected to feature extraction, and the features comprise the Coverage Area (CA) and the length (L) of a minimum area circumscribed rectangle for enteromorpha distribution₁) Width (L)₂) Angle (theta)₀) Center of gravity of enteromorpha due to disaster driftExtracting features of microblog data from the current year to the expiration date by taking the date of the enteromorpha monitoring result as the expiration date, wherein the features comprise central coordinates (SDE) of Standard Deviation Ellipses (SDE) of microblog distribution_x，SDE_y) Angle (alpha), major axis (sigma)_x) Minor axis σ_yDistribution of hot spot areas (B)_m) Intensity of radiation ofAnd the range of radiationAnd then, training a bidirectional associative memory neural network (BAM) by using the characteristics of the enteromorpha disaster and the characteristics corresponding to the microblog social public sentiment. Dividing the enteromorpha disaster characteristics and social public opinion characteristics for training into N subsets Y by taking year as unit_i(i ═ 1, 2, …, N). Taking only one year as a test set each time, and taking the rest as a training set, and training out the corresponding bidirectional association neural network BAM_yearAnd result set Y thereof_year'. Merging all result sets Y_year', forming a new result set Y ', and carrying out residual error network training on Y ' and the test set Y; extracting association rules of microblog distribution features to obtain a rule (rule) of microblog features_X) (ii) a Extracting the characteristics of the enteromorpha disaster according to the association rule to obtain the enteromorpha characteristic rule (rule)_Y) (ii) a Finally, extracting features of microblog social public opinions needing to estimate the situation of the enteromorpha, inputting the extracted features into the BAM, and using a residual error network and Rule in the operation process of the BAM_XAnd Rule_YConstraining the result to obtain corresponding enteromorpha disaster characteristics to estimate the enteromorpha situation, and the specific steps are as follows:

the method comprises the following steps: and extracting enteromorpha disaster characteristics, wherein the enteromorpha disaster characteristics to be extracted comprise the coverage area of the enteromorpha disaster, the drifting gravity center of the enteromorpha disaster, and the length, width and rotation angle of a minimum enteromorpha disaster distribution circumscribed rectangle. If the enteromorpha disaster occurs in history, the enteromorpha disaster existsA group of number N_dThe MODIS image for enteromorpha disaster interpretation and analysis is Year ═ Year₁，year₂，...，year_NFor any Year in the set Yeast, the MODIS date for interpreting Enteromorpha disasters isWherein N is the number of total years, N_yearDays of year;

step 11: calculating the Coverage Area (CA) of the enteromorpha disaster extracted based on the MODIS image according to a formula (1);

CA＝n×cellsize×cellsize (1)

in the formula (1), n is the number of pixels actually covered by the enteromorpha disaster in the image, and cellsize is the resolution of the image, and is usually 250 m;

step 12: calculating the drift gravity center of the enteromorpha disaster according to the formulas (2) and (3)

In the formulae (2) and (3), x_iAnd y_iRespectively representing the abscissa and the ordinate of the occurrence position of the enteromorpha, wherein n is the number of pixels actually covered by an enteromorpha disaster in an image;

step 13: for any point (x) in the monitoring range_i，y_i) The enteromorpha disaster is applied with a counterclockwise rotation of an angle theta to obtain a new coordinate position (x'_i(θ)，，y′_i(θ),) which are calculated by the formula (4) and the formula (5); theta is in the range of [0 DEG, 180 DEG ]]Step length is 1 degree, and initial value is 0 degree; calculating the abscissa x 'of the theta values'_i(theta) and ordinate y'_i(θ); all x'_i(theta) and y'_i(theta) range of maximum difference as the length L of the outer rectangle of the Enteromorpha distribution₁(theta) and width L₂(θ) having the calculation formulas of formula (7) and formula (8), respectively; calculating the minimum circumscribed rectangle area S (theta) when the theta takes different values, as shown in a formula (8); selecting the theta corresponding to the minimum area of the circumscribed rectangle₀As the angle characteristic of the distribution of the enteromorpha, the characteristic is shown in a formula (9); take theta₀Corresponding length L₁(θ₀) And width L₁(θ₀) As the characteristics of the length and width of the distribution of the enteromorpha, the characteristics are shown in a formula (10) and a formula (11);

L₁(θ)＝R(X′)，X＝{x′₀(θ)，x′₁(θ)，...，x′_n-1(θ)} (6)

L₂(θ)＝R(Y′)，Y＝{y′₀(θ)，y′₁(θ)，...，y′_n-1(θ)} (7)

S(θ)＝L₁(θ)L₂(θ) (8)

minS(θ)→θ₀ (9)

L₁＝L₁(θ₀) (10)

L₂＝L₁(θ₀) (11)

in the formula (6) and the formula (7), R (·) represents a range function; in the formula (9), min represents the minimum value;

step two: and extracting microblog social public opinion characteristics, wherein the microblog social public opinion characteristics to be extracted comprise a hot spot radiation range, a hot spot radiation intensity, a standard deviation ellipse distribution range and a standard deviation ellipse rotation angle. Microblog data with keywords of enteromorpha and green tide are acquired by utilizing the open interface of the Sina microblog, and the release time is every day in the Year set Year. The acquired microblog social public opinion data comprise microblog text content, microblog text sending positions and microblog sending time; if an MODIS image capable of interpreting enteromorpha disaster exists in day and year is year, the corresponding social public opinion data is all data before day of day in year. Extracting microblog social public opinion characteristics for each Day in the date Day corresponding to each Year in Year;

step 21: according to the local Moran index, discovering the hotspot position B of social public opinion with date (year, day), wherein B is { B ═ B₁，B₂，...，B_MAnd M is the position number of the found hot spots. For any social public opinion point, calculating the linear distance between the social public opinion point and all the hot point positions according to the spatial position of the social public opinion point, and selecting the hot point position with the closest distance as the hot point position. Hotspot location B_mThe corresponding social public opinion point set is P_m，P_mEach piece of data in and the hot spot position B_mThe set of abscissa distances therebetween is DX_mThe set of ordinate distances being DY_m；

Step 22: calculating the position B of any hot spot according to the formulas (12) and (13)_m(B_mE.g. B) radiation rangeAnd

in the equations (12) and (13),as a set P_mThe number of the elements of (a) is,andrespectively representing sets DX_mAnd DY_mThe ith element;

step 23: calculating the position B of any hot spot according to the formula (14) and the formula (15)_m(B_mE.g. B) radiation intensity rho (B)_m)；

In the formulae (14) and (15), λ (B)_m，t_i) Represents in unit time, B_mMean of public sentiment data points in the radiation range, t₀Represents unit time, E (-) represents mean, N (-) represents count;

step 24: and (3) calculating the central coordinates (SDE) of the microblog social public opinion points according to a formula (16) and a formula (17) by setting the number of all the social public opinion point sets P as np_x，SDE_y) (ii) a Calculating a standard deviation ellipse rotation angle alpha of the microblog social public opinion point according to a formula (18); calculating the distribution range sigma of the standard deviation ellipse of the microblog social public opinion point on the X axis and the Y axis according to the formula (19) and the formula (20)_xAnd σ_y；

In the equations (18), (19) and (20),is the abscissa x of the ith public opinion point_iTo the center abscissa SDE_xThe distance of (a) to (b),is the ordinate y of the ith public opinion point_iTo the center ordinate SDE_yThe distance of (d);

step three: constructing a bidirectional association neural network based on enteromorpha disaster characteristics and microblog social public opinion characteristics;

step 31: for enteromorpha disaster with date (year, day), the social public opinion characteristics of microblog are taken as

Taking the characteristics of enteromorpha disaster asMixing X_year，dayAnd Y_year，dayEach feature in (1) is processed from the decimal data into binary form and 0 is replaced with-1; performing the step on enteromorpha disaster data and microblog social public opinion data of each date (Day) in the Year set and the corresponding Day set to form an X set and a Y set, wherein X ═ { X ═ X_year，day|year∈Year，day∈Day_year}，Y＝{Y_year，day|year∈Year，day∈Day_yearAll of X and Y contain N_dEach of X and YRespectively has a sequence length of N_XAnd N_Y；

Step 32: calculating a weight matrix W according to a formula (21) to obtain a bidirectional association neural network BAM;

in the formula (21), k is the sequence number recorded in X and Y, and W is a size N_X×N_YA matrix of (a);

step four: taking enteromorpha disaster characteristics and microblog social public opinion characteristics in different years as a test set, and taking the rest years as a training set, and performing bidirectional association network training in different years. Constructing a residual error network based on test sets and result sets of different years;

step 41: circularly executing (1) - (3) for any element Year in the Year set Yeast;

(1) recording that year is matched between X and Y, and recording the result meeting the requirement as X_yearAnd Y_yearAnd the result of the non-compliance is recorded as X_year' and Y_year', the number thereof is N_year′；

(2) According to equation (22), a weight matrix W is calculated_yearThe corresponding bidirectional associative neural network is BAM_year；

(3) Inputting X according to the formula (23), the formula (24) and the formula (25)_yearAnd the loop iterates the calculation until X_yearAnd Y_yearStable, i.e. no change, to obtain the predicted Y_year′；

In equations (23), (24) and (25), f (a, b) is the activation function;

step 42: all Y s are collected_year'replacing-1 in the enteromorpha prolifera into 0 and recovering to decimal to form a new sequence set Y' for predicting characteristics of enteromorpha prolifera disasters; the residual network for Y and Y' can be described as equation (26); fitting formula (26) based on the principle of minimum root Mean Square (MSE) error to obtain parameter omega₁And b₁A value of (d);

Y-Y′＝ω₁Y′+b₁ (26)

in the formula (26), ω₁As a weight template, b₁Is an offset;

step five: extracting enteromorpha and microblog social public opinion spatiotemporal rules based on enteromorpha disasters and microblog social public opinion characteristics in the past year;

step 51: setting A and B as the state of some two features in X or Y, D as the set of all records in X or Y, and D as the set of all records in A and B, respectively, according to the formula (27), the confidence (A → B) that B also occurs when A occurs can be calculated;

step 52: based on APRIORI algorithm, with the maximum confidence degree as the target, the rule with 100% confidence degree in the set X and Y is found and is marked as rule_XAnd rule_Y；

Step six: inputting microblog social public opinion data corresponding to enteromorpha disasters to be predicted, and performing enteromorpha distribution prediction based on microblog social public opinions;

step 61: inputting year 'and date day' of enteromorpha disaster distribution to be predicted, and acquiring microblog social public opinion data with year ', date before day' and keywords of enteromorpha and green tide by using a microblog open interface;

step 62: executing the step two, and obtaining the current public opinion information characteristics

Processing x' into binary form, and replacing 0 in the binary form with-1;

and step 63: inputting x' in BAM according to the formulas (28), (29) and (25), and performing a loop

Calculating, and ending the cycle when x 'and y' are stable, namely no change occurs;

in the formulas (28) and (29), p and q are the serial numbers of the elements in x 'and y', respectively;

step 64: will y'_qIn the enteromorpha prolifera disaster distribution characteristics, the-1 is replaced by 0 and the enteromorpha prolifera disaster distribution characteristics are recovered

Step 65: according to a formula (30), correcting the BAM result by using a residual error network to obtain a corrected result y';

y″＝ω₁(y′+I)+b₁ (30)

and step 66: for each type of feature in y', rule is used_YAll the rules are judged, if the rules are met, y' is taken as a final result, and if the rules are not metIf the rule is satisfied, the part which is not satisfied in y' is replaced by the part which satisfies rule_YAny value of the medium conditions as the final result.

According to the method, enteromorpha disaster in the past year and corresponding microblog social public opinion data are used, a bidirectional associative neural network is constructed according to the coverage area, the distribution length and width, the rotation angle and the drift gravity center of the enteromorpha disaster, the hot spot radiation range, the hot spot radiation intensity, the standard deviation ellipse angle and the standard deviation ellipse range of microblog social public opinion, and the purpose that the spatial distribution situation of enteromorpha can be estimated when the microblog social public opinion data are obtained is achieved. In the process of predicting enteromorpha disaster by using the bidirectional associative neural network, correcting a result of the bidirectional associative neural network in a stable state by using a residual error network, correcting a result of the operation of the neural network by using an enteromorpha and microblog time-space association rule extracted based on an APRIORI algorithm, and finally obtaining the estimated current enteromorpha disaster space distribution and development situation. Compared with the traditional manual interpretation method using remote sensing images, the method has the characteristics of high economy, high efficiency, high automation degree and the like;

drawings

FIG. 1 is a flow chart of the present invention.

Fig. 2 shows the microblog social public opinion data and the discovered hot spot area used in the present invention.

FIG. 3 shows the results of the experiment of the present invention.

Detailed Description

In the example, the enteromorpha disasters in 2016, 2017, 2018 and 2019 and the corresponding microblog social public opinion data are selected as training data for experiments, and the enteromorpha disasters in 2019 and the corresponding microblog social public opinion data are selected as test data in the experiments. The microblog social public opinion data format used in the experiment is a shape format, and four hot spot areas of Beijing, Qingdao, Jinan and Nantong are jointly explored and are shown in ArcMap as shown in figure 2. The method is characterized in that one embodiment of the method is provided for the whole process of enteromorpha disaster feature extraction, microblog social public opinion feature extraction, bidirectional association neural network construction, residual error network construction, enteromorpha and microblog social public opinion space-time feature extraction and enteromorpha distribution prediction based on microblog social public opinions;

enteromorpha disaster characteristic extraction

45 groups of data in 2016-2019 have good quality, and can be used for MODIS images for enteromorpha disaster monitoring, wherein the specific Year Year and the corresponding date Day are shown in Table 1, wherein the data for training are 2016, 2017 and 2018, and the data for testing are 2019;

data date distribution in the example of table 1.

Step 11: calculating the Coverage Area (CA) of the enteromorpha disaster extracted based on the MODIS image according to the formula (1), wherein the Coverage Area (CA) is 615km, for example, 2016, 5, 16 days in 2016 year and the distribution area of the enteromorpha disaster²；

Step 12: calculating drift barycenter of enteromorpha disaster according to formulas (2) and (3), such as 2016, 5, 16 days and distribution barycenter coordinate of enteromorpha disasterIs (373.86km, 3731.48 km);

step 13: calculating the length, width and rotation angle of the distribution of the enteromorpha disaster according to the formulas (4) to (11), such as 2016 year, 5 month and 16 days, and L₁500.58km, width L₂217.19km, rotation angle theta₀Is 141 degrees;

(II) extracting social public opinion characteristics of microblog

Step 21: according to local Molan indexes, 4 public sentiment distribution hot spot areas are dug out in 2016-2018 public sentiment data, as shown in figure 2. B ═ beijing, jiannan, Qingdao and Nantong };

step 22: calculating the radiation range of each hot spot position according to the formulas (12) and (13)Andfor example, 2016, 5, 16, Beijing has a radiation range of Beijing

Step 23: calculating the radiation intensity rho (B) of each hot spot position according to the formula (14) and the formula (15)_m). For example, 2016, 5, 16 th, in microblog social public opinion, the radiation intensity of Beijing is rho (B)₀)＝4；

Step 24: calculating the central coordinate (SDE) of the microblog social public opinion point according to the formula (16) and the formula (17)_x，SDE_y). For example, 2016 (5/16/2016), central coordinates of social public opinion points in microblog_x，SDE_y) Was (191.87km, 4024.20 km). And (4) calculating the standard deviation ellipse rotation direction alpha of the microblog social public opinion point according to a formula (18). For example, 2016, 5, 16, with an elliptical direction of rotation α of 102.83 ° standard deviation. Calculating the distribution range sigma of the standard deviation ellipse of the microblog social public opinion point on the X axis and the Y axis according to the formula (19) and the formula (20)_xAnd σ_y. For example, 2016, 5, 16, month and a distribution σ_x189.08km and σ_y＝208.82km。

(III) constructing enteromorpha disaster with date (year, day) by performing bidirectional association neural network construction based on enteromorpha disaster characteristics and microblog social public opinion characteristics, and taking microblog social public opinion characteristics as the enteromorpha disaster Taking the characteristics of enteromorpha disaster asX constructed by taking Enteromorpha disaster of 2016, 5, 16 days_2016，5.16＝(150.58km，38.10km，4，113.24km，224.99km，4，39.33km，401.70km，3，737.43km，430.69km，0，191.87km，4024.20km，102°，189.03km，20.88km)，Y_2016，5.16＝(615km²373.86km, 3731.48km, 141 °, 500.58km, 217.19 km). Mixing X_year，dayAnd Y_year，dayEach feature in (a) is processed from decimal data into binary data and replaces 0 with-1. For example, mixing X_2016，5.16The coverage area CA in the set is characterized by 1-1-111-1-1111;

step 32: calculating a weight matrix W according to a formula (21) to obtain a bidirectional association neural network BAM;

and (IV) taking the enteromorpha disaster characteristics and microblog social public opinion characteristics of different years as a test set, and taking the rest years as a training set to carry out bidirectional association network training of different years. Constructing a residual error network based on test sets and result sets of different years;

step 41: circularly executing (1) - (3) for any element Year in the Year set Yeast;

(1) taking 2016 as an example, the matching year of X and Y is 2016, and the result meeting the requirement is recorded as X₂₀₁₆And Y₂₀₁₆And the result of the non-compliance is recorded as X₂₀₁₆' and Y₂₀₁₆', the number thereof is 37.

(2) According to equation (22), a weight matrix W is calculated₂₀₁₆The corresponding network structure code is BAM₂₀₁₆

(3) Inputting X according to the formula (23), the formula (24) and the formula (25)₂₀₁₆And the loop iterates the calculation until X₂₀₁₆And Y₂₀₁₆Stable, i.e. no change, to obtain the predicted Y₂₀₁₆′；

Step 42: summary Y₂₀₁₆'replacing-1 in the enteromorpha prolifera into 0 and recovering to decimal to form a new sequence set Y' for predicting characteristics of enteromorpha prolifera disasters; the residual network for Y and Y' can be described as equation (26); fitting formula (26) based on the principle of minimum root Mean Square (MSE) error to obtain parameter matrix omega₁And b₁A value of (d);

extracting enteromorpha and microblog social public opinion spatio-temporal rules based on enteromorpha disaster and microblog social public opinion characteristics in the past year;

based on APRIORI algorithm, with the maximum confidence coefficient as the targetRule for confidence of 100% in digging set X and Y is recorded as rule_XAnd rule_Y. For example rule_XIn, L₂When the CA is less than or equal to 81.18km, the CA is less than or equal to 647km²；

Inputting microblog social public opinion data corresponding to the enteromorpha disaster to be predicted, and performing enteromorpha distribution prediction based on microblog social public opinion;

step 61: inputting 2019 years and 23 days of 6 months of the date of the distribution of the enteromorpha to be predicted, and acquiring microblog social public opinion data with 2020 years, the date of 6 months and 23 days before, and keywords of enteromorpha and green tide by using a microblog open interface;

step 62: executing the step two, and acquiring the current public opinion information characteristics;

x '═ 78.61km, 544.66km, 10,500.41km, 196.08km, 8, 152.55km, 198.48km, 5, 167.18km, 224.14km, 4, 15.91km, 4029.59km, 171 °, 695.94km, 14.08km), treat x' to binary form and replace all 0 s with-1;

and step 63: inputting x 'into the BAM according to the formulas (28), (29) and (25), circularly calculating, and ending the circulation when x' and y 'are stable, namely no longer changed, so as to obtain a final result y';

step 64: replacing-1 in y 'with 0, and recovering to obtain characteristic y' of enteromorpha disaster distribution (2299.17 km)²，336.09km，3946.37km，6.12°，512.05km，180.29km)；

Step 65: the result of BAM is corrected using the residual error network according to equation (30) to obtain a corrected result y ″ (2295.38 km)²，331.31km，3895.31km，11.62°，288.52km，233.26km)；

And step 66: for each type of feature in y', rule is used_YJudging all the rules, if the rules are met, taking y 'as a final result, and if the rules are not met, replacing the part which is not met in the y' with the part which meets rule_YAny value of the medium conditions as the final result. The prediction result of 23 days 6 months in 2019 meets rule_YTherefore, no correction is made as a direct result.

And (3) testing and analyzing: comparing the enteromorpha distribution result predicted by the method with the actual enteromorpha distribution result (as shown in fig. 3), it can be seen that the two have higher consistency, and the result is superior to the result obtained by predicting by using a bidirectional association neural network alone. The false alarm probability of the result of the method is 31%, the probability of missing judgment is 25%, and the accuracy is 79%, thus showing the effectiveness of the method. In addition, compared with the traditional manual interpretation method using remote sensing images, the method has the characteristics of high economy, high efficiency, high automation degree and the like.