1. A method for improving comprehensive utilization efficiency of a smart grid area is characterized by comprising the following steps:

the method comprises the steps of obtaining the distance between each spare load and a photovoltaic power station and the power consumption information of each spare load, and calculating the consumption capacity corresponding to each spare load according to the distance and the power consumption information;

clustering each spare load according to the consumption capacity corresponding to each spare load to obtain the spare load corresponding to each cluster set;

inputting the absorption capacity of the cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity to be absorbed by the photovoltaic power station into a constructed absorption capacity prediction network for reasoning, and extracting a characteristic tensor corresponding to the absorption capacity prediction network after the reasoning is finished;

calculating a combination condition corresponding to the redundant electric quantity which needs to be consumed by the photovoltaic power station according to the feature tensor and the number of the standby loads included by each cluster set, wherein the combination condition comprises the number of the standby loads which need to be input by each cluster set;

and calculating the total distance value corresponding to each combination mode meeting the combination condition, and taking the combination mode with the minimum total distance value as a target combination mode.

2. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 1, wherein the constructed consumption capability prediction network has the following inputs: the consumption capacity of a cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity to be consumed by the photovoltaic power station; the output of the constructed consumption capability prediction network is as follows: a consumption capacity prediction value corresponding to the surplus electric quantity required to be consumed by the photovoltaic power station; the loss of the constructed absorption capacity prediction network comprises the following steps: the regression loss, the characteristic tensor loss and the parameter loss of the full connection layer of the consumption capacity predicted value and the redundant electric quantity needing to be consumed.

3. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 2, wherein the regression loss of the consumption capacity predicted value and the surplus electric quantity to be consumed is calculated by adopting the following formula:

wherein the content of the first and second substances,L _r in order to accommodate the regression loss between the predicted consumption capacity and the extra power to be consumed,in order to accommodate the predicted value of the digestion capability,E _t in order to take up the excess amount of power,is composed ofL2A paradigm.

4. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 2, wherein the characteristic tensor loss is calculated by adopting the following formula:

wherein the content of the first and second substances,L _c1 in order to be a loss of the characteristic tensor,p _k is the first in the feature tensorkThe value of each of the elements is,Qis a constant greater than a set threshold value,Kis the total number of elements in the feature tensor.

5. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 2, wherein the parameter loss of the full connection layer is calculated by adopting the following formula:

wherein the content of the first and second substances,L _c2 in order to achieve a full link layer parameter loss,w _k1 is as followsk1The weight of the individual neuron or neurons is,K1is the total number of the neurons,Q1is a constant greater than a set threshold.

6. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 1, wherein the power consumption information of the backup load comprises a power consumption average value of the backup load in a set time length.

7. The method for improving the comprehensive utilization efficiency of the smart grid area according to claim 1, wherein the total distance value is a superposition of distances of the backup loads from the photovoltaic power plant in the combined mode.

Background

The photovoltaic power generation can provide clean power for the smart grid, and in recent years, large grid-connected photovoltaic power stations develop rapidly; however, the electric energy supplied by the photovoltaic power stations in some areas far exceeds the area requirements, that is, the power generation amount of the photovoltaic power stations is larger than the sum of the maximum transmission power amount and the load consumption power amount of the power system, and the light abandoning phenomenon exists.

In order to solve the light abandoning phenomenon, the existing control method is to add a plurality of backup loads on a micro-grid bus to consume the redundant electric quantity by putting in the plurality of backup loads, as shown in fig. 1, the backup loads are all flexible loads, that is, adjustable loads or transferable loads which can be freely switched in different time periods are allowed.

Although the existing control method can consume the surplus electric quantity by inputting a plurality of standby loads, the consumption capacities of the standby loads are different, the consumption capacities corresponding to different input combination modes are different, how to achieve large-degree consumption of the surplus electric quantity by inputting the standby loads is a problem to be solved urgently for improving the utilization rate of the photovoltaic power generation quantity.

Disclosure of Invention

The invention aims to provide a method for improving the comprehensive utilization efficiency of a smart grid area, which is used for solving the problem that the existing control method is low in photovoltaic power generation utilization rate.

In order to solve the above problems, the technical solution of the method for improving the comprehensive utilization efficiency of the smart grid area of the present invention includes the following steps:

the method comprises the steps of obtaining the distance between each spare load and a photovoltaic power station and the power consumption information of each spare load, and calculating the consumption capacity corresponding to each spare load according to the distance and the power consumption information;

clustering each spare load according to the consumption capacity corresponding to each spare load to obtain the spare load corresponding to each cluster set;

inputting the absorption capacity of the cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity to be absorbed by the photovoltaic power station into a constructed absorption capacity prediction network for reasoning, and extracting a characteristic tensor corresponding to the absorption capacity prediction network after the reasoning is finished;

calculating a combination condition corresponding to the redundant electric quantity which needs to be consumed by the photovoltaic power station according to the feature tensor and the number of the standby loads included by each cluster set, wherein the combination condition comprises the number of the standby loads which need to be input by each cluster set;

and calculating the total distance value corresponding to each combination mode meeting the combination condition, and taking the combination mode with the minimum total distance value as a target combination mode.

The method has the beneficial effects that: according to the invention, the consumption capacity corresponding to each backup load is calculated, the backup loads are clustered according to the consumption capacity corresponding to each backup load, each combination mode corresponding to the redundant electric quantity required to be consumed by the photovoltaic power station can be obtained by inputting the consumption capacity of the clustering center corresponding to each cluster set, the number of the backup loads included in each cluster set and the redundant electric quantity required to be consumed by the photovoltaic power station into the constructed consumption capacity prediction network, the total distance value corresponding to each combination mode is used as a reference index for further screening the optimal combination mode, the redundant electric quantity required to be consumed by the photovoltaic power station can be greatly consumed by the backup loads, and the problem of low photovoltaic power generation utilization rate of the existing control method is solved.

Further, the inputs of the constructed consumption capability prediction network are as follows: the consumption capacity of a cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity to be consumed by the photovoltaic power station; the output of the constructed consumption capability prediction network is as follows: a consumption capacity prediction value corresponding to the surplus electric quantity required to be consumed by the photovoltaic power station; the loss of the constructed absorption capacity prediction network comprises the following steps: the regression loss, the characteristic tensor loss and the parameter loss of the full connection layer of the consumption capacity predicted value and the redundant electric quantity needing to be consumed.

Further, the regression loss of the consumption capacity predicted value and the redundant electric quantity to be consumed is calculated by adopting the following formula:

wherein the content of the first and second substances,L _r in order to accommodate the regression loss between the predicted consumption capacity and the extra power to be consumed,in order to accommodate the predicted value of the digestion capability,E _t in order to take up the excess amount of power,is composed ofL2A paradigm.

Further, the feature tensor loss is calculated using the following formula:

wherein the content of the first and second substances,L _c1 in order to be a loss of the characteristic tensor,p _k is the first in the feature tensorkThe value of each of the elements is,Qis a constant greater than a set threshold value,Kis the total number of elements in the feature tensor.

Further, the parameter loss of the full connection layer is calculated by adopting the following formula:

wherein the content of the first and second substances,L _c2 in order to achieve a full link layer parameter loss,w _k1 is as followsk1The weight of the individual neuron or neurons is,K1is the total number of the neurons,Q1is a constant greater than a set threshold.

Further, the power consumption information of the backup load comprises a power consumption average value of the backup load under a set time length.

Further, the total distance value is the superposition of the distances from the backup loads to the photovoltaic power station in the combined mode.

Drawings

FIG. 1 is a schematic diagram of a prior art microgrid system architecture;

FIG. 2 is a flowchart of a method for improving comprehensive utilization efficiency of a smart grid area according to the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention.

As shown in the problem 1, the existing micro-grid system comprises a photovoltaic power station, a micro-grid energy storage unit, an original load and a standby load, wherein the standby load comprises the standby load1…, spare loadm. When the generated energy of the photovoltaic power station is larger than the sum of the maximum transmission electric quantity of the power system and the consumed electric quantity of the load (including the original load and the micro-grid energy storage unit), the redundant electric quantity needs to be consumed by putting in a standby load. The absorption capacities of the various backup loads are different, the absorption capacities corresponding to different input combination modes are different, how to realize the large absorption of the redundant electric quantity by inputting the backup loads is to improve the lightThe utilization rate of the photovoltaic power generation capacity is an urgent problem to be solved.

In order to solve the above problems, as shown in the figure2As shown, the method for improving the comprehensive utilization efficiency of the smart grid area of the embodiment includes the following steps:

(1) the method comprises the steps of obtaining the distance between each spare load and a photovoltaic power station and the power consumption information of each spare load, and calculating the consumption capacity corresponding to each spare load according to the distance and the power consumption information;

the absorption capacity of each backup load is related to the distance between the backup load and the photovoltaic power station and the power consumption of the backup load, and the absorption capacity of the backup load is in positive correlation with the distance between the backup load and the photovoltaic power station and the power consumption of the backup load: the distance between the standby load and the photovoltaic power station reflects the electric energy loss when the photovoltaic power station supplies power to the standby load, and the farther the distance between the standby load and the photovoltaic power station is, the more the electric energy loss when the photovoltaic power station supplies power to the standby load is; in this embodiment, the power consumption of the backup load has less fluctuation with time, and the power consumption of the backup load can be set for a set time length (for example, the power consumption of the backup load can be set by the backup load24h) The lower power consumption mean value represents that the power of the standby load is reflected, and the larger the power consumption of the standby load is, the stronger the power consumption of the standby power supply is. In this embodiment, the relationship between the absorption capacity of each backup load, the distance from the photovoltaic power station, and the power consumption is obtained by fitting based on a mathematical modeling method as follows:

wherein the content of the first and second substances,for the spare load to be able to absorb the capacity per unit time,lthe distance of the backup load from the photovoltaic power plant,for the power consumption of the spare load per unit time,kandbfor parameters derived based on least-squares fitting。

(2) Clustering each spare load according to the consumption capacity corresponding to each spare load to obtain the spare load corresponding to each cluster set;

after obtaining the absorption capability of each spare load, the embodiment is based onK-meansThe clustering algorithm clusters the spare loads to classify the spare loads with similar consumption capacities into a class. The number of cluster sets is related to the total number of the standby loads and the difference of the absorption capacity between the standby loads, and the more the cluster sets are, the more finely the standby loads are divided, and the smaller the difference of the absorption capacity of the standby loads in each cluster set is.

The number of the cluster sets is used as5The total number of the standby loads is20For example, the backup load absorption capacity is differentK-meansThe clustering algorithm will do this20Spare load division5Classes are ranked according to the consumption capacity of the corresponding standby load of each class center from small to large, and the classes are endowed with1-5The number of the standby loads in each set is counted, the corresponding set identification is distributed to the standby loads in the corresponding set, and each standby load is regarded as a node.

(3) Inputting the absorption capacity of the cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity to be absorbed by the photovoltaic power station into a constructed absorption capacity prediction network for reasoning, and extracting a characteristic tensor corresponding to the absorption capacity prediction network after the reasoning is finished;

the redundant electric quantity to be consumed by the photovoltaic power station refers to the total electric quantity to be consumed by the standby load, and the value of the total electric quantity to be consumed by the standby load is equal to the sum of the generated energy of the photovoltaic power station minus the maximum transmission electric quantity of the power system and the consumed electric quantity of the load (including the original load and the micro-grid energy storage unit). In the embodiment, in order to obtain the corresponding relationship between the excess electric quantity required to be consumed by the photovoltaic power station and the spare load investment, a consumption capacity prediction network is constructed, so that the corresponding relationship between the excess electric quantity required to be consumed by the photovoltaic power station and the spare load investment is obtained through the consumption capacity prediction network, that is, how to input the load can greatly consume the excess electric quantity required to be consumed by the photovoltaic power station.

The input of the consumption capability prediction network is the consumption capability of the cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity required to be consumed by the photovoltaic power station; the output of the consumption capability prediction network is: a consumption capacity prediction value corresponding to the surplus electric quantity required to be consumed by the photovoltaic power station; the architecture of the digestion capability prediction network is as follows:

constructing a TCN network having inputs of1*(K+1)Size, in this exampleK=5Enter before inKEach of the elements includes the number of nodes of the corresponding cluster set and the absorption capability of the cluster center, the firstK+1Each element is the surplus electric quantity which needs to be absorbed by the photovoltaic power station. The input of the network in this embodiment is1*(5+1)Size of sliding windowLSetting the void rate of the void convolution to beriThe input of the network is subjected to cavity convolution and then the characteristic tensor is output, and the size of the characteristic tensor is1*5Size, maintenance of size byPaddingThe operation is realized, and the characteristic tensor is input into the full connection layer and output as the combined absorption capacity;

the network specific training process comprises the following steps: the photovoltaic power station of one area corresponds to a group of the inputs, the corresponding inputs of the photovoltaic power stations of a plurality of areas are used as training data sets, and the loss function of the network is designed to be

Wherein the content of the first and second substances,L _r in order to accommodate the regression loss between the predicted consumption capacity and the extra power to be consumed,L _c1 in order to be a loss of the characteristic tensor,L _c2 in order to achieve a full link layer parameter loss,in order to accommodate the predicted value of the digestion capability,E _t in order to take up the excess amount of power,in the form of the L2 model,p _k is the first in the feature tensorkThe value of each of the elements is,Qis a constant greater than a set threshold value,Kis the total number of elements in the feature tensor,w _k1 is as followsk1The weight of the individual neuron or neurons is,K1is the total number of the neurons,Q1is a constant greater than a set threshold; it should be noted that: each element in the feature tensor corresponds to a neuron of the fully connected layer, so that the total number of elements in the feature tensor is equal to the total number of neurons, i.e. each element in the feature tensor corresponds to a neuron of the fully connected layerK= K1。

Loss of feature tensorL _c1 The purpose of (2) constraining the characteristic tensor element not greater than1Wherein a functionTo ensure that it is less than1The loss is smaller than1The loss is extremely large and is continuously derivable, so that the back propagation of the gradient is convenient to carry out, and other functions meeting the requirements can be selected as the loss in practical application; loss of feature tensorL _c1 Middle functionThe purpose of (2) constraining the feature tensor element not to be smaller than0Otherwise, the loss is extremely large, and in order to realize the constraint effect, the method is setQIs greater than a set threshold value so thatQAre very large constants. To realize the full connection layer parameterw _k1 Constraint effect of setting full link layer parameter lossL _c2 InQ1Is greater than a set threshold value so thatQ1Is a very large constant, e.g.Q1Is composed of1*10 ⁶ 。

Size of sliding windowLVoid rate of convolution with voidsriWhen the determination is made, a corresponding trained predictive network of the digestion ability can be obtained through the training process; in order to obtain an optimal predictive network of absorption capacity, the sliding window size is used in this embodimentLVoid rate of convolution with voidsriAs a hyper-parameter, the network is subjected to hyper-parameter traversal optimization, and the default network can be converged, namelyL _c1 、L _c2 Terms can be satisfied, since both provide a greater penalty, in the case of both being satisfied, forL _r Selecting optimal hyper-parametersL _r And taking the minimum corresponding hyper-parameter as the optimal hyper-parameter. Will optimize the size of the sliding windowLVoid rate of convolution with voidsriAnd the corresponding trained predictive network of the digestion ability is used as an inference network.

After the trained predictive network of the absorption capacity (namely the reasoning network) is obtained, the absorption capacity of the cluster center corresponding to each cluster set, the number of the standby loads included in each cluster set and the redundant electric quantity which needs to be absorbed by the photovoltaic power station are input into the constructed predictive network of the absorption capacity for reasoning, and then the characteristic tensor corresponding to the predictive network of the absorption capacity after reasoning is extracted.

(4) Calculating a combination condition corresponding to the redundant electric quantity which needs to be consumed by the photovoltaic power station according to the feature tensor and the number of the standby loads included by each cluster set, wherein the combination condition comprises the number of the standby loads which need to be input by each cluster set;

the absorption capacity prediction network obtained based on the training method has the characteristics tensor which is a sequence formed by the proportion of the standby loads needing to be input by each cluster set, the proportion of the standby loads needing to be input by each cluster set can be obtained based on the extracted characteristics tensor, the number of the standby loads needing to be input by each cluster set can be obtained by multiplying the number of the standby loads included in the corresponding cluster set, and the number of the standby loads needing to be input by each cluster set forms a combination condition corresponding to the redundant electric quantity needing to be absorbed by the photovoltaic power station.

E.g. when the optimum sliding window sizeL=4Void ratio of void convolutionriIn case of =2, the extracted feature tensor is: [b1、b2、0、b4、b5]Reference numeral is1-5The number of the standby load nodes corresponding to the cluster set is respectively as follows:S1、S2、S3、S4、S5then the input label is required1The number of the spare loads in the cluster set isb1*S1Reference numerals to be input2The number of the spare loads in the cluster set isb2*S2Reference numerals to be input3The number of the spare loads in the cluster set is0Reference numerals to be input4The number of the spare loads in the cluster set isb4*S4Reference numerals to be input5The number of the spare loads in the cluster set isb5*S5. If the number of the obtained backup loads of a certain cluster set to be input is not an integer, in order to avoid excessive consumption of the photovoltaic power generation amount due to excessive backup loads, the embodiment adopts a downward rounding mode to obtain the number of the backup loads to be input in the cluster set.

(5) And calculating the distance total value corresponding to each combination mode meeting the combination condition, and taking the combination mode with the minimum corresponding distance total value as a target combination mode.

The combination modes meeting the combination conditions are not unique, and there are many combination modes, because the larger the distance between the spare load and the photovoltaic power station is, the more uncontrollable factors are in the electric energy transmission process, for example, the local fault of the power transmission line, in order to obtain a more reliable combination mode, in this embodiment, the distances between the spare load and the photovoltaic power station in each combination mode are superposed to obtain a total distance value corresponding to each combination mode, and the combination mode with the minimum total corresponding distance value is used as a target combination mode, that is, a preferred combination mode.

The embodiment calculates the consumption capacity corresponding to each backup load, clusters the backup loads according to the consumption capacity corresponding to each backup load, and inputs the consumption capacity of the cluster center corresponding to each cluster set, the number of the backup loads included in each cluster set and the redundant electric quantity required to be consumed by the photovoltaic power station into the constructed consumption capacity prediction network to obtain each combination mode corresponding to the redundant electric quantity required to be consumed by the photovoltaic power station.

It should be noted that while the preferred embodiments of the present invention have been described, additional variations and modifications to these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.