CN109145430B - Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network - Google Patents
Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network Download PDFInfo
- Publication number
- CN109145430B CN109145430B CN201810926844.7A CN201810926844A CN109145430B CN 109145430 B CN109145430 B CN 109145430B CN 201810926844 A CN201810926844 A CN 201810926844A CN 109145430 B CN109145430 B CN 109145430B
- Authority
- CN
- China
- Prior art keywords
- double
- neural network
- output power
- layer
- hidden
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention discloses a method for evaluating the output power of a double-sided double-glass photovoltaic assembly based on a neural network, which comprises the steps of taking four parameters of air temperature, air speed, front irradiance and back irradiance of the double-sided double-glass photovoltaic assembly measured at the same time in an experimental place as independent variables, taking the output power of the assembly measured at the same time as a dependent variable, inputting the obtained data into the neural network subjected to structural optimization for nonlinear fitting analysis, and finally inputting any four parameters into the neural network to obtain a fitting calculation result of the output power of the double-sided double-glass photovoltaic assembly.
Description
Technical Field
The invention relates to a method for evaluating output power of a double-sided double-glass photovoltaic assembly based on a neural network, and belongs to the technical field of performance analysis of photovoltaic assemblies.
Background
With the development of the photovoltaic industry becoming perfect, the research gravity center is expanded from the improvement of the conversion efficiency of a photovoltaic module to the improvement of the comprehensive performance of a photovoltaic system. In recent years, double-sided double-glass photovoltaic modules attract extensive attention due to the outstanding characteristics of durability, reliability, safety and high power generation capacity. However, due to the difference in power generation performance between the double-sided double-glass photovoltaic module and the conventional single-sided photovoltaic module, the evaluation method for the output power of the conventional single-sided photovoltaic module cannot be completely applied to the double-sided double-glass photovoltaic module, but the research on the aspect at home and abroad at the present stage is not completely developed, if the evaluation method is the same as the conventional single-sided photovoltaic module, the output power is evaluated only by taking the parameters of irradiance, air temperature, wind speed and wind direction on the front side of the module in a fitting manner, the characteristics of the double-sided double-glass photovoltaic module cannot be embodied, and the fitting accuracy is poor.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method for evaluating the output power of a double-sided double-glass photovoltaic assembly based on a neural network, which can be used for prejudging the output power condition according to air temperature, wind speed, front irradiance and back irradiance.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
the method for evaluating the output power of the double-sided double-glass photovoltaic module based on the neural network comprises the following steps:
1) measuring four parameters of air temperature, wind speed, front irradiance and back irradiance of the double-sided double-glass photovoltaic assembly and output power of the double-sided double-glass photovoltaic assembly on different days and different time periods in an experimental place;
2) processing the measured data and inputting the processed data into a neural network;
3) the number of hidden nodes of the neural network and the type of a network structure transfer function of the neural network are kept unchanged, and error minimization selection is carried out on the number of hidden nodes of the neural network;
4) under the condition of the number of hidden layers of the neural network determined in the step 3), the type of the network structure transfer function is kept unchanged, and the number of hidden layer nodes of the neural network is subjected to error minimization selection;
5) under the conditions of the number of hidden layers of the neural network determined in the step 3) and the number of hidden layer nodes of the neural network determined in the step 4), performing error minimization selection on the type of the transfer function of the network structure;
6) constructing a neural network by using the number of hidden layers of the neural network determined in the step 3), the number of hidden layer nodes of the neural network determined in the step 4) and the type of the network structure transfer function determined in the step 5), inputting all data measured in the step 1) into the neural network, and inputting a group of values of four parameters at will after fitting to obtain the value of corresponding output power.
In the step 1), different days of the experiment comprise 3 weather conditions of sunny days, cloudy days and cloudy days, and each weather type is subjected to the experiment for 4 days; the different time periods of the experiment included 8 to 10 am, 11 to 1 pm, and 2 to 4 pm, with the time steps measured being 1 minute.
In the step 1), the air temperature and air speed measuring devices arranged in the experiment place are positioned in the surrounding area of the tested double-sided double-glass photovoltaic module, and no obvious barrier exists;
the double-sided double-glass photovoltaic module front irradiance measuring device arranged in the experimental place is arranged in the middle of the side edge of the front side of the module; the irradiance measuring devices on the back of the double-sided double-glass photovoltaic assembly are two and are respectively arranged in the middle position of the bottom edge and the middle position of the top edge of the back of the assembly; the installation angles of all the irradiance measuring devices are consistent with the inclination angle of the component, and shielding is not formed on the power generation surface of the component;
the output power recording device is connected between the component power line and the inverter.
In the step 2), the processing of the measurement data means that two irradiance values at the middle positions of the bottom edge and the top edge of the back of the component measured at the same time are averaged to be used as the irradiance of the back of the component; and then arranging the air temperature, the wind speed, the front irradiance of the double-sided double-glass photovoltaic module, the back irradiance of the module and the output power of the module according to a time sequence, and enabling four parameter values and 1 output power value at the same moment to be 1 group.
In the step 3), the number of hidden nodes of the neural network and the transfer function of the network structure are preliminarily set as: the number of hidden layer nodes of the neural network is 4, the transfer function from the input layer to the first hidden layer is tan sig, if the number of hidden layers is more than or equal to 2, the transfer functions between the hidden layers are logsig, and the transfer function from the last hidden layer to the output layer is purelin.
In the step 3), the error minimization selection of the number of hidden layers of the neural network refers to adjusting the number of hidden layers of the neural network from 1 layer to 2, 3, 4 and 5, calculating errors of the fitting output power value and the actual power value under the condition of different numbers of hidden layers respectively, and taking the number of hidden layers with the minimum error.
In the step 4), the number of hidden nodes of the neural network is subjected to error minimization selection, specifically, the number of hidden nodes of the neural network is adjusted from 4 to 5, 6, 7 and 8, errors of the fitting output power value and the actual power value under the condition of different numbers of hidden nodes are respectively calculated, and the number of hidden nodes of the neural network with the minimum error is taken.
In the step 5), the error minimization selection is performed on the type of the network structure transfer function, specifically, the hidden layer transfer function and the output layer transfer function are combined, errors of the fitting output power value and the actual power value under different transfer function combination conditions are respectively calculated, and the transfer function combination condition with the smallest error is taken.
The hidden layer transfer functions include tansig and logsig, and the output layer transfer functions include tansig and purelin.
The foregoing process of fitting the output power value is as follows:
91) normalizing the input air temperature, wind speed, front irradiance of the assembly, back irradiance of the assembly and output power data;
92) multiplying the processing results of the input four parameters by weights from the corresponding input layer to the hidden layer respectively and adding the weights and the threshold, wherein the initial weights and the threshold are selected as uniformly distributed decimal numbers, arbitrarily selecting the decimal numbers between (-1, 1), and substituting the obtained results into the transfer function of the layer;
93) repeating the previous layer operation on the obtained result to finish the calculation among all the hidden layers;
94) respectively multiplying the calculated results by weights from the hidden layer to the output layer and adding the calculated results to a threshold value, substituting the obtained results into a transfer function from the hidden layer to the output layer, wherein the obtained results are the fitted output power signal values obtained for the first time;
95) calculating the error between the primarily fitted output power signal value and the input output power signal value, and performing back propagation on the error to adjust the weight and the threshold in the neural network;
96) after the adjustment is finished, the calculation process from the input layer to the hidden layer and from the hidden layer to the output layer is carried out again until the calculation error meets the specified requirement, and the calculation is stopped;
97) and performing inverse normalization on the final fitting output power signal value, and outputting the final fitting output power value.
The invention has the following beneficial effects: through the adjustment of parameters of the evaluation method for the output power of the conventional single-sided photovoltaic assembly, the power generation characteristic of the double-sided double-glass assembly is highlighted, the defects in the evaluation method for the output power of the double-sided double-glass photovoltaic assembly are made up, and the fitting accuracy of the output power is relatively improved.
Drawings
FIG. 1 is a schematic diagram of a BP neural network structure;
FIG. 2 is a flow chart of the method of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
The method for evaluating the output power of the double-sided double-glass photovoltaic module based on the neural network, disclosed by the invention, as shown in figure 2, comprises the following steps of:
(1) measuring air temperature, air speed, front irradiance and back irradiance of the double-sided double-glass photovoltaic module and output power of the double-sided double-glass photovoltaic module on different days and different time periods of an experimental place. In particular, the method comprises the following steps of,
the experiment is carried out on 3 weather conditions including sunny days, cloudy days and cloudy days, and each weather type is carried out for 4 days. The different time periods of experiment included 8 am ~10, 11 noon to 1 pm to and 2 pm ~4 pm, and the measuring time of day is 6 hours, and the time step of measurement is 1 minute.
The air temperature and air speed measuring devices arranged in the experiment place are all located in the area around the tested double-sided double-glass photovoltaic assembly, no obvious barriers exist, the air temperature unit is in centigrade degree (DEG C), and the air speed unit is in meters per second (m/s).
The front side of the double-sided double-glass photovoltaic module for experimental measurement,The unit of back irradiance is Watt per square meter (W/m)2) And the arranged double-sided double-glass photovoltaic module front irradiance measuring device is arranged in the middle position of the front side edge of the module. The irradiance measuring device on the back of the double-sided double-glass photovoltaic assembly is arranged at the middle position on the bottom edge of the back of the assembly and the middle position on the top edge of the assembly respectively.
The installation angles of the front and back irradiance measuring devices of the double-sided double-glass photovoltaic module arranged in the experiment place are consistent with the inclination angle of the module, and the module is not shielded on the power generation surface.
The unit of the output power of the double-sided double-glass photovoltaic module is watt (W), and the output power recording device is connected between the module power line and the inverter.
(2) Processing the measured data and inputting the processed data into a neural network; the processing of the measured data comprises averaging two irradiance values of the middle positions of the bottom edge and the top edge of the back of the assembly measured at the same moment to obtain the irradiance of the back of the assembly; the air temperature, the wind speed, the front irradiance and the back irradiance of the double-sided double-glass photovoltaic module and the module output power data are arranged according to a time sequence, the parameter values at the same moment are classified into 1 group, and the number of data groups in one day is 360 groups. Inputting all the data of different weather types measured in the step (1) into a neural network.
(3) Performing error minimization selection on the hidden layer number of the neural network; the concrete mode is as follows:
the number of hidden nodes of the neural network and the type of a network structure transfer function are kept unchanged, and the specific selection mode is as follows: the number of hidden layer nodes of the neural network is 4, a transfer function from an input layer to a first hidden layer is tan sig, if the number of hidden layers is greater than or equal to 2, transfer functions between the hidden layers are logsig, a transfer function from the last hidden layer to an output layer is purelin, the number of hidden layers is adjusted from 1 layer to 2 layers, 3 layers, 4 layers and 5 layers, errors of a fitting output power value and an actual power value under the condition of different numbers of hidden layers are calculated respectively, and the number of hidden layers with the smallest error is taken.
Taking the hidden layer as 1 layer as an example, the calculation process of the fitting output power value is as follows: firstly, normalizing input air temperature, wind speed, front irradiance, back irradiance and output power data, respectively multiplying processing results (signal values) of the input four parameters by weights from a corresponding input layer to a hidden layer and adding the processing results to a threshold value (the initial weights and the threshold value are selected as uniformly distributed decimal numbers and are arbitrarily selected between (-1 and 1)), substituting the obtained results into a transfer function of the layer, repeating the operation of the previous layer on the obtained results, respectively multiplying the weights from the hidden layer to the output layer and adding the weights to the threshold value, substituting the obtained results into the transfer function from the hidden layer to the output layer, and obtaining a result which is a fitted output power signal value obtained for the first time. And calculating the error between the primarily fitted output power signal value and the input output power signal value (target output power signal value), and performing back propagation on the error to adjust the weight value and the threshold value in the neural network. The error back propagation process is as follows: taking the weight of a certain parameter in a certain layer as an example, calculating a partial derivative value of an error to the weight, and subtracting the partial derivative value multiplied by a learning rate (a decimal within a range of 0-1) from the weight to obtain an adjusted weight, wherein the weight of each parameter in each layer and the threshold value in each layer are similar to the adjustment method. And after the adjustment is finished, the calculation process from the input layer to the hidden layer and from the hidden layer to the output layer is carried out again, the calculation is circulated until the calculation error meets the specified requirement, the calculation is stopped, the final fitting output power signal value is subjected to inverse normalization, and the final fitting output power value is output.
(4) Performing error minimization selection on the number of hidden nodes of the neural network; the concrete mode is as follows:
and (4) under the condition of the number of hidden layers determined in the step (3), keeping the type of a network structure transfer function unchanged, adjusting the number of hidden layers from 4 to 5, 6, 7 and 8, respectively calculating the errors of the fitting output power value and the actual power value under the condition of different numbers of hidden layers, and taking the condition when the errors are minimum. And (4) the calculation process of the fitting output power value is the same as the step (3).
(5) Performing error minimization selection on the type of the network structure transfer function; the concrete mode is as follows:
combining and matching network structure transfer functions under the conditions of the number of hidden layers determined in the step (3) and the number of hidden layer nodes determined in the step (4), wherein common hidden layer transfer functions include tansig and logsig, and output layer transfer functions include tansig and purelin, for example, under the condition that the hidden layers are 2 layers, the combination of the transfer functions is 8: the transfer function from the input layer to the hidden layer 1 is tansig, the transfer function from the hidden layer 1 to the hidden layer 2 is logsig, and the transfer function from the hidden layer 2 to the output layer is purelin; the transfer function from the input layer to the hidden layer 1 is tan sig, the transfer function from the hidden layer 1 to the hidden layer 2 is logsig, and the transfer function from the hidden layer 2 to the output layer is tan sig; the transfer function from the input layer to the hidden layer 1 is tansig, the transfer function from the hidden layer 1 to the hidden layer 2 is tansig, and the transfer function from the hidden layer 2 to the output layer is purelin; the transfer function from the input layer to hidden layer 1 is tansig, the transfer function from hidden layer 1 to hidden layer 2 is tansig, and the transfer function from hidden layer 2 to the output layer is tansig; the transfer function from the input layer to the hidden layer 1 is logsig, the transfer function from the hidden layer 1 to the hidden layer 2 is logsig, and the transfer function from the hidden layer 2 to the output layer is purelin; the transfer function from the input layer to the hidden layer 1 is logsig, the transfer function from the hidden layer 1 to the hidden layer 2 is logsig, and the transfer function from the hidden layer 2 to the output layer is tansig; the transfer function from the input layer to the hidden layer 1 is logsig, the transfer function from the hidden layer 1 to the hidden layer 2 is tansig, and the transfer function from the hidden layer 2 to the output layer is purelin; the transfer function from the input layer to the hidden layer 1 is logsig, the transfer function from the hidden layer 1 to the hidden layer 2 is tansig, and the transfer function from the hidden layer 2 to the output layer is tansig; and respectively calculating the errors of the fitting output power value and the actual power value under the conditions of different transfer function combinations, and taking the condition when the error is minimum. And (4) calculating the fitting output power value in the same step (3).
(6) Forming a neural network system by utilizing the number of hidden layers determined in the step (3), the number of hidden layer nodes determined in the step (4) and the network structure transfer function determined in the step (5), and inputting any group of four parameter values to obtain a corresponding output power evaluation result; the concrete mode is as follows: all data obtained by experimental measurement are input into the neural network which is subjected to network structure parameter error minimization selection, and after fitting, values of any group of four parameters are input at the moment, so that an evaluation result of a corresponding output power value can be obtained.
In the illustrated embodiment of the present invention, the experiment is performed under a clear day, the different time periods of the experiment include 8 to 10 am, 11 to 1 pm and 2 to 4 pm, the measurement time of one day is 6 hours, and the measurement time steps are all 1 minute.
Air temperature and wind speed data are measured by an automatic meteorological station near the tested double-sided double-glass photovoltaic module, the air temperature unit is in centigrade degree (DEG C), and the wind speed unit is in meters per second (m/s). The unit of irradiance of the front surface and the back surface of the double-sided double-glass photovoltaic module is measured to be watt per square meter (W/m)2) The irradiance on the front side of the double-sided double-glass photovoltaic module is measured by a thermopile radiometer, and the radiometer is arranged in the middle of the side edge of the front side of the module. The irradiance on the back side of the double-sided double-glass photovoltaic assembly is obtained by averaging data measured by two thermopile irradiation meters, and the two irradiation meters are respectively arranged in the middle positions of the bottom edge and the top edge of the back side of the assembly. The installation angles of the irradiators for measuring the irradiance on the front and back of the double-sided double-glass photovoltaic module are consistent with the inclination angle of the module, and the irradiators do not form shielding on the power generation surface of the module. The unit of the output power of the double-sided double-glass photovoltaic module is watt (W), and the output power recording device is connected between the module power line and the inverter.
After the experiment, the measured data needs to be processed, and the two irradiance values of the middle positions of the bottom edge and the top edge of the back of the component measured at the same moment are averaged to obtain the irradiance of the back of the component; the temperature, the wind speed, the irradiance on the front side and the back side of the double-sided double-glass photovoltaic module and the output power data of the module are arranged according to a time sequence, the parameter values at the same time are classified into 1 group, and the number of the data groups is 360.
Fitting calculation of a neural network is carried out by Matlab software, after the obtained data is input into a BP neural network established by the Matlab software, the BP neural network is structurally optimized in three aspects of hidden layer number, hidden layer node number and network structure transfer function, the BP neural network is schematically shown in figure 1, and W (ih) and W (ho) are weights transmitted among layers.
Error minimization processing is carried out on the hidden layer number, the number of hidden layer nodes of the neural network and the type of a network structure transfer function are adjusted from 1 layer to 2, 3 and 4, errors of a fitting output power value and an actual power value under the condition of different hidden layer numbers are calculated respectively, the fitting output power value can be obtained by inputting data of four parameters and output power and calculating the hidden layer and output layer transfer functions, the calculating method is the same as that in the step (3), and the method is realized by utilizing a train function and a sim function in software in Matlab. The calculation result shows that the error is gradually increased along with the increase of the number of hidden layers, so that the error is minimum when the number of hidden layers is 1.
The error minimization processing on the number of hidden nodes of the neural network needs to keep the type of a network structure transfer function unchanged on the basis of one hidden layer, adjust the number of the hidden nodes from 4 to 5, 6, 7 and 8, and respectively calculate the errors of a fitting output power value and an actual power value under the condition of different numbers of the hidden nodes. The calculation result shows that the error is continuously reduced and then gradually increased along with the increase of the number of the hidden nodes, and when the number of the hidden nodes is 11, the error is minimum.
The error minimization processing for the network structure transfer function type needs to combine and match the transfer functions on the basis of one hidden layer with hidden layer nodes of 11, common hidden layer transfer functions include tansig and logsig, and output layer transfer functions include tansig and purelin, and the result shows that the transfer functions from the input layer to the hidden layer and from the hidden layer to the output layer are combined: the combined error of (tansig + purelin), (logsig + tansig), and (tansig + tansig) is the smallest, and the average absolute percentage error is the smallest at this time, 5.4%.
Therefore, based on the number of hidden layers 1, the number of hidden layer nodes 11, and the network structure transfer function: in Matlab, a function with a data reading function, such as xlsread, is utilized, all data of four parameters and output power obtained through experimental measurement and four parameter values to be evaluated are read into the neural network, fitting calculation is carried out through the neural network, and then a function with a function is derived through data, such as xlsrite, so that an evaluation result of the output power corresponding to the four parameter values can be obtained.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, it is possible to make various improvements and modifications without departing from the technical principle of the present invention, and those improvements and modifications should be considered as the protection scope of the present invention.
Claims (9)
1. The method for evaluating the output power of the double-sided double-glass photovoltaic module based on the neural network is characterized by comprising the following steps of:
step 1), measuring four parameters of air temperature, wind speed, front irradiance and back irradiance of the double-sided double-glass photovoltaic component and output power of the double-sided double-glass photovoltaic component on different days and different time periods in an experimental place; the different days comprise 3 weather conditions of sunny days, cloudy days and cloudy days, and each weather type is tested in 4 days; the different time periods comprise 8 to 10 am, 11 to 1 pm and 2 to 4 pm, and the measured time steps are all 1 minute;
step 2), processing the measured data and inputting the processed data into a neural network;
step 3), keeping the number of hidden layer nodes of the neural network and the type of a network structure transfer function unchanged, and performing error minimization selection on the number of hidden layers of the neural network;
step 4), under the condition of the number of hidden nodes of the neural network determined in the step 3), the type of a network structure transfer function is kept unchanged, and error minimization selection is carried out on the number of the hidden nodes of the neural network;
step 5), carrying out error minimization selection on the type of the network structure transfer function under the conditions of the number of the hidden layers of the neural network determined in the step 3) and the number of the hidden layer nodes of the neural network determined in the step 4);
and 6) constructing a neural network by using the number of hidden layers of the neural network determined in the step 3), the number of nodes of the hidden layers of the neural network determined in the step 4) and the type of the network structure transfer function determined in the step 5), inputting all data measured in the step 1) into the neural network, and after fitting, arbitrarily inputting a group of values of four parameters, namely air temperature, wind speed, front irradiance and back irradiance of the double-sided double-glass photovoltaic module to obtain a value of corresponding output power.
2. The method for evaluating the output power of the double-sided double-glass photovoltaic module based on the neural network as claimed in claim 1, wherein in the step 1), air temperature and air speed measuring devices arranged at an experimental site are both positioned in the area around the double-sided double-glass photovoltaic module to be tested, and no obvious barrier exists;
the double-sided double-glass photovoltaic module front irradiance measuring device arranged in the experimental place is arranged in the middle of the side edge of the front side of the module; two double-sided double-glass photovoltaic module back irradiance measuring devices are arranged and are respectively arranged at the middle position of the bottom edge and the middle position of the top edge of the back of the module; the installation angles of all the irradiance measuring devices are consistent with the inclination angle of the component, and no shielding is formed on the power generation surface of the component;
the output power recording device is connected between the assembly power line and the inverter.
3. The method for estimating the output power of the double-sided double-glass photovoltaic assembly based on the neural network as claimed in claim 2, wherein in the step 2), the measurement data is processed by averaging two irradiance values measured at the middle position of the bottom edge and the top edge of the back of the assembly at the same moment to obtain an average value as the irradiance of the back of the assembly; and then arranging the air temperature, the wind speed, the front irradiance of the double-sided double-glass photovoltaic module, the back irradiance of the module and the output power of the module according to a time sequence, and enabling four parameter values and 1 output power value at the same moment to be 1 group.
4. The method for evaluating the output power of the double-sided double-glass photovoltaic assembly based on the neural network as claimed in claim 1, wherein in the step 3), the number of hidden nodes of the neural network and the transfer function of the network structure are preliminarily set as: the number of hidden layer nodes of the neural network is 4, a transfer function from the input layer to the first hidden layer is tansig, if the number of hidden layers is more than or equal to 2, transfer functions between the hidden layers are logsig, and a transfer function from the last hidden layer to the output layer is purelin.
5. The method for estimating the output power of the double-sided double-glass photovoltaic module based on the neural network as claimed in claim 4, wherein in the step 3), the selection of the number of the hidden layers of the neural network for error minimization refers to adjusting the number of the hidden layers of the neural network from 1 layer to 2, 3, 4 and 5.
6. The method for estimating the output power of the double-sided double-glass photovoltaic module based on the neural network as claimed in claim 5, wherein in the step 4), the number of hidden nodes of the neural network is selected to minimize the error, specifically, the number of hidden nodes of the neural network is adjusted from 4 to 5, 6, 7 and 8.
7. The method for estimating the output power of the double-sided double-glass photovoltaic module based on the neural network as claimed in claim 6, wherein in the step 5), the type of the network structure transfer function is selected for error minimization, specifically, a hidden layer transfer function and an output layer transfer function are combined, errors of a fitting output power value and an actual power value under different transfer function combination conditions are respectively calculated, and the transfer function combination condition with the minimum error is taken.
8. The method for evaluating the output power of the double-sided double-glass photovoltaic assembly based on the neural network as claimed in claim 7, wherein the hidden layer transfer function comprises tansig and logsig, and the output layer transfer function comprises tansig and purelin.
9. The method for estimating the output power of the double-glass photovoltaic assembly based on the neural network as claimed in claim 5, 6 or 7, wherein the process of fitting the output power value is as follows:
91) normalizing the input air temperature, wind speed, irradiance on the front surface of the assembly, irradiance on the back surface of the assembly and output power data;
92) multiplying the processing results of the four input parameters by weights from the corresponding input layer to the hidden layer respectively and adding the weights and the thresholds, wherein the initial weights and the thresholds are selected as uniformly distributed decimal numbers and arbitrarily selected between (-1, 1), and the obtained results are substituted into the transfer function of the layer;
93) repeating the previous layer operation on the obtained result to finish the calculation among all the hidden layers;
94) respectively multiplying the calculated results by weights from the hidden layer to the output layer and adding the calculated results to a threshold value, substituting the obtained results into a transfer function from the hidden layer to the output layer, wherein the obtained results are the fitted output power signal values obtained for the first time;
95) calculating the error between the primarily fitted output power signal value and the input output power signal value, and performing back propagation on the error to adjust the weight and the threshold in the neural network;
96) after the adjustment is finished, the calculation process from the input layer to the hidden layer and from the hidden layer to the output layer is carried out again until the calculation error reaches the specified requirement, and the calculation is stopped;
97) and performing inverse normalization on the final fitting output power signal value, and outputting the final fitting output power value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810926844.7A CN109145430B (en) | 2018-08-15 | 2018-08-15 | Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810926844.7A CN109145430B (en) | 2018-08-15 | 2018-08-15 | Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145430A CN109145430A (en) | 2019-01-04 |
| CN109145430B true CN109145430B (en) | 2022-07-22 |
Family
ID=64793296
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810926844.7A Active CN109145430B (en) | 2018-08-15 | 2018-08-15 | Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145430B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529095A (en) * | 2016-12-12 | 2017-03-22 | 广州市扬新技术研究有限责任公司 | Photovoltaic power generation prediction research system based on Matlab |
-
2018
- 2018-08-15 CN CN201810926844.7A patent/CN109145430B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529095A (en) * | 2016-12-12 | 2017-03-22 | 广州市扬新技术研究有限责任公司 | Photovoltaic power generation prediction research system based on Matlab |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145430A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020215405A1 (en) | Photovoltaic power station power prediction method and system based on grid-tie inverter operating data | |
| Zhong et al. | Prediction of photovoltaic power generation based on general regression and back propagation neural network | |
| CN111833202B (en) | Farmland evapotranspiration short-term prediction method considering crop coefficient dynamic change and rainfall | |
| CN103678872A (en) | Photovoltaic power generation system performance evaluation method and device | |
| Ventura et al. | Development of models for on-line diagnostic and energy assessment analysis of PV power plants: The study case of 1 MW Sicilian PV plant | |
| KR102230548B1 (en) | Power generation prediction and efficiency diagnosis system of solar power generation facilities using FRBFNN model | |
| CN113987823B (en) | Mesoscale numerical simulation method for land centralized photovoltaic power station climate effect evaluation | |
| Veldhuis et al. | An empirical model for rack-mounted PV module temperatures for Southeast Asian locations evaluated for minute time scales | |
| CN108320050B (en) | Method for improving photovoltaic short-term power prediction accuracy based on wind speed prediction | |
| Ramaswamy et al. | Forecasting PV power from solar irradiance and temperature using neural networks | |
| CN109599899B (en) | Setting method for new energy operation simulation boundary conditions | |
| Cabezon et al. | Comparison of methods for power curve modelling | |
| CN109145430B (en) | Method for evaluating output power of double-sided double-glass photovoltaic module based on neural network | |
| Castro et al. | Detailed effects of wind on the field performance of a 50 kW CPV demonstration plant | |
| CN106227981A (en) | A kind of double glass photovoltaic module crack preventing method | |
| Khenar et al. | Particle swarm optimisation‐based model and analysis of photovoltaic module characteristics in snowy conditions | |
| CN111967660B (en) | Ultra-short-term photovoltaic prediction residual error correction method based on SVR | |
| Tapakis et al. | Performance evaluation of a photovoltaic park in Cyprus using irradiance sensors. | |
| Liao et al. | Evaluation of temperature dependent models for PV yield prediction | |
| Koumpli | Impact of data quality on photovoltaic (PV) performance assessment | |
| CN109829572B (en) | Photovoltaic power generation power prediction method under thunder and lightning weather | |
| CN115034078B (en) | Modeling method for photovoltaic string fusion model of mountain power station | |
| CN104361399A (en) | Solar irradiation intensity minute-scale predication method | |
| CN115293190B (en) | Sand storm dynamic inversion method based on microwave signal blind source separation and SCA | |
| Magare et al. | Estimation of module temperature effects on seasonal performance variation of different photovoltaic technology modules |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |