1. An indoor natural illuminance modeling method is characterized by comprising the following steps:

s1, collecting indoor natural illuminance information, recording the position coordinates of the measuring points, and making a training data set of the radial basis function neural network by taking the position coordinates as input and the illuminance information as output;

s2, training a radial basis function neural network by adopting a training data set to obtain a reference model of indoor natural illuminance distribution;

s3, arranging a plurality of sensors indoors to monitor the real-time illumination of key points;

and S4, correcting the reference model by the real-time illumination value of the monitoring key point to obtain an indoor real-time illumination distribution condition model of the natural light.

2. The modeling method of indoor natural illuminance according to claim 1, wherein the radial basis function neural network is as follows:

wherein, C_uIs the center of neuron u, X is the input, b1 is a given constant offset affecting the shape of the radial basis function, ω_uB2 is the hidden layer to output layer weight and the hidden layer to output layer bias.

3. The modeling method of indoor natural illuminance according to claim 1, wherein in S2, the specific method for training the radial basis function neural network by using the training data set is as follows "

S21, initializing the network;

s22, determining a sample, and when the radial basis function taking the sample as the center is taken as a hidden node, minimizing the error;

s23, selecting the radial basis function with the sample as the center as a new hidden node;

s24, calculating the weight from the hidden layer to the output layer by adopting a least square method;

s25, calculating a new error of the new radial basis function neural network, and finishing training if the error meets the requirement; if the error is not satisfactory, S22 is executed.

4. The indoor natural illuminance modeling method for the indoor environment as claimed in claim 1, wherein the intrinsic factors affecting the indoor natural illuminance distribution include: the geographical location of the building, the orientation of the building, the location, size and number of windows, the reflective properties of the indoor walls and object materials, and the depth from the window into the room.

5. The method of claim 1, wherein the extrinsic factors affecting the distribution of the natural indoor illuminance comprise: variations in the radiant intensity of natural light, variations in peak areas of indoor illumination, variations in seasonal illumination intensity in the northern hemisphere, and daily weather effects on illumination.

6. The modeling method of indoor natural illuminance according to claim 1, wherein the model of indoor natural illuminance distribution is as follows:

wherein, w_rFor the correction parameters between the target position and the sensor, the calculation method is as follows:

wherein E' is the output of the reference model, E is the real-time natural illuminance value obtained after correction, R is the real-time illuminance value measured by setting R illuminance sensors, x_r,y_r,z_rIs the position of the sensor r.

Background

In an indoor lighting control system, rational utilization of natural light is an effective way to achieve energy-saving lighting. The indoor natural illuminance is effectively measured or estimated, and the artificial supplementary lighting is carried out on the indoor lighting environment through a proper control strategy to meet the lighting requirement. Therefore, obtaining real-time indoor natural light illuminance values is a prerequisite for lighting control systems. In the case where only the illuminance of a very small number of locations needs to be acquired, it is possible to measure the natural illuminance by arranging the illuminance sensor indoors. However, in the situation of needing accurate illumination measurement and light control, the control system needs to obtain natural illumination values of a large number of measurement points, and the measurement by using the sensor obviously has the problems of high arrangement cost, difficult maintenance, great difficulty in realizing engineering and the like. On the other hand, calculating the illuminance distribution of natural light generated indoors by modeling is also a "soft measurement" method. However, the indoor natural illuminance distribution is influenced by various factors such as indoor layout, outdoor environment and the like, the existing modeling method needs to acquire a large number of input parameters, the modeling is complex, and some conditions are idealized in the modeling process, so that the calculation result is not accurate.

Disclosure of Invention

The invention aims to overcome the defects and provide an indoor natural illuminance modeling method, which is characterized in that an illuminance sensor is used for acquiring illuminance and coordinates in advance to form a data set, and a radial basis function neural network is trained through the data set to obtain an illuminance reference model. A few of illumination sensors are arranged at key positions in the room to measure real-time natural illumination, real-time natural illumination data are used for correcting the illumination value predicted by the reference model, and finally a real-time natural illumination distribution estimation can be obtained.

In order to achieve the above object, the present invention comprises the steps of:

s1, collecting indoor natural illuminance information, recording the position coordinates of the measuring points, and making a training data set of the radial basis function neural network by taking the position coordinates as input and the illuminance information as output;

s2, training a radial basis function neural network by adopting a training data set to obtain a reference model of indoor natural illuminance distribution;

s3, arranging a plurality of sensors indoors to monitor the real-time illumination of key points;

and S4, correcting the reference model by the real-time illumination value of the monitoring key point to obtain an indoor real-time illumination distribution condition model of the natural light.

The radial basis function neural network is as follows:

wherein, C_uIs the center of neuron u, X is the input, b1 is a given constant offset affecting the shape of the radial basis function, ω_uWeight from hidden layer to output layer, b2 bias from hidden layer to output

In S2, the specific method for training the radial basis function neural network by using the training data set is as follows "

S21, initializing the network;

s22, determining a sample, and when the radial basis function taking the sample as the center is taken as a hidden node, minimizing the error;

s23, selecting the radial basis function with the sample as the center as a new hidden node;

s24, calculating the weight from the hidden layer to the output layer by adopting a least square method;

s25, calculating a new error of the new radial basis function neural network, and finishing training if the error meets the requirement; if the error is not satisfactory, S22 is executed.

Intrinsic factors affecting the natural illuminance distribution in a room include: the geographical location of the building, the orientation of the building, the location, size and number of windows, the reflective properties of the indoor walls and object materials, and the depth from the window into the room.

Extrinsic factors affecting the natural illuminance distribution in the room include: variations in the radiant intensity of natural light, variations in peak areas of indoor illumination, variations in seasonal illumination intensity in the northern hemisphere, and daily weather effects on illumination.

The model of the indoor natural illuminance distribution is as follows:

wherein, w_rFor the correction parameters between the target position and the sensor, the calculation method is as follows:

wherein E' is the output of the reference model, E is the real-time natural illuminance value obtained after correction, R is the real-time illuminance value measured by setting R illuminance sensors, x_r,y_r,z_rIs the position of the sensor r.

Compared with the prior art, the method has the advantages that indoor natural illuminance data are collected, and the radial basis function neural network is trained to obtain a reference model of illuminance distribution; and correcting the output of the model by monitoring the real-time illumination of key points by adopting a few sensors arranged indoors, so as to obtain real-time indoor natural illumination distribution estimation. The illumination control method acquires the illumination reference model by acquiring natural illumination data of the actual scene in a training mode, and corrects the illumination output of the reference model by using the real-time illumination change monitored by a few sensors, so that the rapid illumination distribution estimation is realized, and a basis is provided for comfortable illumination control. According to the method, the natural illuminance model constructed by the radial basis function neural network only needs to arrange a few illuminance sensors in an application scene to detect the change of the natural illuminance, so that the indoor illuminance distribution is quickly estimated, and the engineering problem caused by arrangement of a large number of sensors is solved.

Drawings

FIG. 1 is a diagram of a radial basis function neural network of the present invention;

FIG. 2 is a block diagram of a model of natural illumination according to the present invention;

FIG. 3 is a three-dimensional view of the interior of the embodiment;

FIG. 4 is a radial basis function neural network structure in an embodiment.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

For the indoor environment with good natural lighting, the reasonable utilization of the natural lighting can not only reduce the loss of lighting energy, but also benefit the physical and mental health of people. The invention establishes an accurate and effective natural illuminance model and provides support for an indoor lighting control system.

Step one, data acquisition;

arranging sensors indoors, wherein the sensors are arranged to cover all working surfaces concerned as far as possible, collecting indoor natural illumination information at a certain determined time (under natural light conditions) in a day, recording position information of a measuring point, and making a training data set of the radial basis function neural network by taking the position information as input and the illumination information as output.

Step two, training a neural network;

the reference model of the illumination distribution is obtained by training a radial basis function neural network:

the radial basis function neural network is a three-layer neural network consisting of an input layer, a hidden layer and an output layer. The hidden layer is also called radial basic layer, and the activation function of each neuron is a radial basic function, and the structure is shown in fig. 1.

C_uIs the center of the neuron u, having the same dimension as the input X, and b1 is a given constant bias that affects the shape of the radial basis function. Each neuron calculates the distance of the input vector to the neuron center and then multiplies the deviation b 1; the result is then obtained byAnd (5) transferring. Thus, a radial basis function neural network can be generally expressed as:

the learning method of the radial basis function neural network adopted by the invention is characterized in that the center is selected from sample input: initializing a radial basis function neural network without any hidden node, adding the hidden nodes into the network one by one, and stopping adding the nodes when the error meets the requirement. In order to minimize hidden nodes, the node which can reduce the error most is selected as a new node each time. The specific algorithm is as follows:

1) the network is initialized.

2) The error is reduced most by predicting which sample is centered radial basis function is added as the hidden node.

3) The radial basis function centered on the sample is selected as the new hidden node.

4) And calculating the weight from the hidden layer to the output layer by using a least square method.

5) A new error for the new network is calculated.

6) Whether the error meets the requirement, and otherwise, returning to 2).

Thirdly, arranging a small number of sensors to monitor real-time illumination;

when the model is deployed, a few sensors are arranged indoors to monitor the real-time illumination of key points. The selection principle of the key point position is a working surface capable of reflecting the obvious change of the natural light along with the change of the illumination value of time.

And step four, correcting the reference model by using the monitored real-time illumination value to obtain the indoor real-time natural light illumination distribution, wherein the correction formula is calculated according to the formulas (2) and (3).

Take an indoor office environment as an example. Generally, people pay attention to the illumination on a working surface, so that natural illumination modeling mainly aims at the illumination distribution on the working surface, enough illumination sensors are arranged on the working surface (the specific number is determined by practical application scenes and layout), a certain determined time period (under the condition of natural light) in one day is selected, the position coordinates of each sensor and the corresponding measured illumination value in the time period are obtained, and a training data set is obtained;

the reference model is trained with a radial basis function neural network with a sensor data set. The learning method adopted in the process of training the radial basis function neural network is 'center selection from sample input': initializing a radial basis function neural network without any hidden node, adding the hidden nodes into the network one by one, and stopping adding the nodes when the error meets the requirement. In order to minimize hidden nodes, the node which can reduce the error most is selected as a new node each time. The specific algorithm is as follows:

1) the network is initialized.

2) The error is reduced most by predicting which sample is centered radial basis function is added as the hidden node.

3) The radial basis function centered on the sample is selected as the new hidden node.

4) And calculating the weight from the hidden layer to the output layer by using a least square method.

5) A new error for the new network is calculated.

6) If the error meets the requirement, otherwise, returning to 2.

The indoor illumination distribution of the natural light is a result of common influence of various factors, and the illumination obtained by estimating the external environment change through the reference model inevitably generates large deviation, so that the real-time accurate estimation of the indoor natural illumination distribution cannot be obtained. Specifically, the following aspects are provided:

1) intrinsic factors affecting the natural illuminance distribution in a room include: the geographical location of the building, the orientation of the building, the position, size and number of windows, the reflective properties of the indoor walls and object materials, the greater the depth from the window into the room, the weaker the natural light intensity, etc.

2) The external factors influencing the natural indoor illuminance distribution mainly include: along with the rising and falling of the sun in one day, the radiation intensity of natural light is changed from weak to strong and then becomes weak, and the indoor illumination peak area is also shifted from east to west; for the northern hemisphere, the illumination in summer is strong, and the illumination in winter is weak; in cloudy and rainy days, direct sunlight is shielded by a cloud layer, so that indoor illumination is insufficient, and indoor illumination is sufficient in sunny days.

In view of the above, the indoor illumination distribution is affected by various factors and has a wide variation range. Therefore, the reference model cannot reflect dynamic changes in the natural illuminance distribution.

In a preferred embodiment of the present invention, a reference model of the illuminance distribution is obtained, the illuminance measurement is performed by the key point arrangement sensor, and the output of the model is corrected, so as to obtain the real-time natural indoor illuminance distribution under the dynamic condition. The selection principle of the key point position is that the working surface with obvious change of natural illumination along with time can be reflected;

in a preferred embodiment of the invention, the calibration based on the reference model is based on the following knowledge. The factors inherent in affecting the natural indoor light distribution for a given indoor environment are also determined. It is assumed that the natural indoor illuminance distribution at this time will be uniquely determined without considering the external factors. That is, the natural illumination of each indoor location has a certain correlation, and the reference model is obtained by fitting the correlation using the measured training data.

However, it is a fact that various external factors act to change the natural illuminance distribution in the room. Therefore, the output of the reference model is calibrated according to the real-time data of the key points, and an effective method for quickly estimating the illumination when the external factors change is realized.

The indoor natural illuminance distribution has continuity, so the smaller the distance is, the greater the illuminance value correlation between two points is. Selecting a few key points to arrange an illumination sensor, and measuring the illumination E in real time_rIlluminance E 'from model output'_rThe difference reflects the change of the natural illumination of two time periods at the same position.

The indoor illumination distribution is corrected according to the characteristics of the continuity of the illumination distribution by utilizing the illumination change of the key position to obtain the dynamic change characteristics of the illumination generated by the indoor natural light, and then the model output is corrected to obtain the real-time indoor natural illumination distribution. The specific correction formula is as follows:

in the formula, w_rTo correct the parameters, it can be calculated as:

(2) wherein E' represents the output of the reference model, E represents the real-time natural illuminance value obtained after correction, R represents that R illuminance sensors are provided to measure the real-time illuminance value, [ x ]_r,y_r,z_r]The position of the sensor r is defined. Formula (3) represents w_rDefined by the distance between the target location and the sensor.

Example (b):

1. experimental Environment

The laboratory was located in a laboratory (34.35 ° north latitude, 108.92 ° east longitude) which was a 4.5 × 7.0 × 3.2m space.

6 sets of tables and chairs are arranged indoors, the lower right corner of the building is the origin of coordinates, the upward direction is the x axis, the leftward direction is the y axis, and a rectangular coordinate system in the horizontal direction (hereinafter, the horizontal coordinate system is used). The horizontal coordinates of 6 work tables, based on the center of the table top, are shown in the following table:

TABLE 1 Table coordinates

Numbering	1	2	3	4	5	6
							Coordinate (m)	(0.75,1.70)	(3.75,1.70)	(0.75,4.13)	(3.75,4.13)	(0.75,6.56)	(3.75,6.56)

The indoor window faces south, the height of the windowsill is 0.8m, and the whole size of the window is 5.8m multiplied by 1.35 m.

2. Establishing a natural illuminance distribution reference model

The acquisition time of the training data of the natural illuminance distribution reference model is 11/26/11/2020: 00-11: 20, the weather condition is sunny day. As shown in FIG. 3, an indoor top plan view is obtained by collecting 30 points of illuminance on a surface with an indoor horizontal height of 0.75m by using a Guarda FX-101 LUX METER illuminometer, wherein red marks are illuminance sensors with a point for collecting the illuminance, blue marks are key positions, the horizontal heights are unified to be 0.75m and are respectively marked as s₁,s₂,s₃。

The collected illuminance values and horizontal coordinates are shown in the following table:

TABLE 2 training data

Coordinate (cm)	(84,175)	(141,175)	(197,175)	(253,175)	(309,175)	(366,175)
							Illuminance (lx)	641	698	756	851	995	1420
Coordinate (cm)	(84,292)	(141,292)	(197,292)	(253,292)	(309,292)	(366,292)
							Illuminance (lx)	646	722	783	882	1015	1444
Coordinate (cm)	(84,408)	(141,408)	(197,408)	(253,408)	(309,408)	(366,408)
							Illuminance (lx)	623	687	762	862	1037	1523
Coordinate (cm)	(84,525)	(141,525)	(197,525)	(253,525)	(309,525)	(366,525)
							Illuminance (lx)	516	572	641	750	926	1401
Coordinate (cm)	(84,642)	(141,642)	(197,642)	(253,642)	(309,642)	(366,642)
							Illuminance (lx)	351	403	442	501	581	716

The horizontal coordinates of the illuminance sensor are as follows:

TABLE 3 illuminance sensor coordinates

Illuminance sensor	s1	s2	s3
				Coordinate (cm)	(422,642)	(422,408)	(422,175)

And training the radial basis function neural network by taking the coordinates of the training data as input and the illumination as output to obtain a trained illumination distribution reference model. The trained neural network is shown in fig. 4, and the number of finally determined hidden layer neurons is 23.

3. Analysis of predicted results

After obtaining the reference model, the center points of the desktop numbers 3, 4 and 5 are selected as test points, and the coordinates are #3(75cm,413cm), #4(375cm,413cm) and #5(75cm,656cm), respectively. The illuminometer was used to measure the illuminance at different times of day 27, 11/2020, and the weather was cloudy. Real-time inputting the value of the illumination sensor at the key point position at the same moment into a natural illumination modelCalculating real-time natural illuminance, comparing the calculated real-time natural illuminance with the measured real-time natural illuminance, and analyzing as follows, wherein the illuminance values are lx including s₁,s₂,s₃An acquired real-time luminance value.

Table 4 statistical table of illuminance prediction results

Table E' is the illumination estimate predicted by the reference model,and calculating a value obtained by correcting the estimated value output by the reference model for the illumination distribution model, wherein the relative error delta is calculated as:

therefore, from the statistical results in the table above, most of the relative errors of the model-predicted illumination values can be controlled within 7%, the maximum error is not more than 10%, and the model can perform relatively accurate real-time natural illumination estimation.

The foregoing shows and describes the general principles and broad features of the present invention and advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.