Real-time dynamic evaluation system of solar cell performance based on deep learning

Abstract

The present invention relates to the field of renewable energy technology, and more specifically, to a real-time dynamic evaluation system for solar cell performance based on deep learning. The system comprises: a data acquisition module: used for real-time acquisition of internal key parameters and external key parameters of solar cells; a data processing module: using a deep learning algorithm to process and analyze the collected data, and establishing a solar cell performance evaluation model; a data visualization module: making a visual graph of the performance score of each time step; an abnormality prediction and diagnosis module: comparing the visual graph with the visual graph of the standard working state in the historical data, judging whether an abnormality occurs in the current and future certain time steps; the present invention uses an RNN model to consider the dependency relationship of each time step in a time series, and better learn and extract features.

Description

Technical Field

[0001] The present invention relates to the field of renewable energy technology, and more specifically, to a real-time dynamic evaluation system for solar cell performance based on deep learning. Background Art

[0002] Solar cells are a key component that converts solar energy into electrical energy, and their performance evaluation is crucial to the efficiency and reliability of solar power generation systems. Traditional solar cell performance evaluation methods mainly rely on static testing, which cannot dynamically monitor the performance changes of solar cells in real time, limiting the accurate evaluation of their performance in actual operation.

[0003] In recent years, the rapid development of deep learning technology has brought new opportunities for solar cell performance evaluation. Deep learning can build complex models by learning a large amount of data, thereby realizing the prediction and evaluation of solar cell performance. By analyzing the real-time data of the cell, the deep learning model can accurately determine the key indicators of the cell such as efficiency and power output, and promptly detect abnormal conditions and reasons for performance degradation.

[0004] Prior art: A Chinese patent with publication number CN104241160A discloses a method and device for testing the local optical properties of a solar cell assembly, comprising a hollow mesh structure for fixing the solar cell assembly to be tested, a hollow area, a shielding plate, a support member and a handrail for internally supporting the shielding plate arranged in the hollow mesh structure, wherein the hollow mesh structure is customized according to the solar cell assembly to be tested so that each solar cell assembly unit has a one-to-one corresponding hollow area; the invention can be used to simply, reliably and effectively measure any one or more (including all) solar cell assembly units of the solar cell assembly to be tested; by detecting the optical properties of the solar cell assembly units, the design defects of the solar cell process can be effectively evaluated.

[0005] However, there are still some challenges in the current deep learning model in predicting the performance of solar cells. Complex environmental conditions may make it difficult for existing models to accurately predict the performance of solar cells. To address these issues, it is necessary to further optimize the deep learning model and solve the challenges of data reliability and real-time performance.

[0006] Therefore, developing a real-time dynamic evaluation system for solar cell performance based on deep learning has important research significance and application value. The system will combine deep learning algorithms and real-time data acquisition technology to monitor and evaluate the performance of solar cells in real time, providing accurate reference and guidance for the operation and maintenance of solar power generation systems. At the same time, the system can also process data automatically and intelligently, improve the accuracy and stability of the evaluation, and realize the comprehensive evaluation and optimization of solar cell performance. Summary of the invention

[0007] To achieve the above objectives, the present invention proposes a real-time dynamic evaluation system for solar cell performance based on deep learning. The system takes into account the impact of sunlight on the performance evaluation of solar cells, divides the performance influencing factors into internal and external, and determines internal and external abnormalities by comparing visual graphics. It also provides a reference for maintenance personnel.

[0008] A real-time dynamic evaluation system for solar cell performance based on deep learning, the system includes:

[0009] Data acquisition module: collects the internal key parameters of solar cells in real time by installing sensors; collects the external key parameters of solar cells in real time by using photo sensors;

[0010] Data processing module: Use deep learning algorithms to process and analyze the collected internal and external key parameters to establish a solar cell performance evaluation model;

[0011] Data visualization module: visualize the performance scores of each time step;

[0012] Abnormal prediction and diagnosis module: By comparing the visualized graphics with the visualized graphics of the standard working status in the historical data, it is determined whether abnormalities occur in the current and future time steps. If abnormalities occur, the type of abnormality is determined based on the curve changes of the visualized graphics. The abnormality types include internal abnormalities and external abnormalities. Abnormal screening is performed, and the results of abnormal screening and the visualized graphics are imported into the database. At the same time, early warning information is sent to the background server.

[0013] Furthermore, the internal key parameters include operating current, operating voltage and operating temperature.

[0014] Furthermore, the external key parameters include light intensity, spectral distribution and light angle.

[0015] Furthermore, the solar cell performance evaluation model is an RNN model, and the construction method includes:

[0016] Define a time window, the length of the time window, i.e., the length of the time step is n, where n is a preset value, and the time step includes the corresponding internal key parameters and external key parameters in the time window;

[0017] Construct the initial framework of the solar cell performance evaluation model. The initial framework includes an input layer, an encoder LSTM layer, a bidirectional LSTM layer, an average pooling layer, a fully connected layer and a Softmax output layer. The input size of the input layer is the time step length and the feature dimension is 6, that is, the input size is (n, 6). The encoder LSTM layer uses LSTM units. The LSTM unit state includes the context state and the hidden state. The hidden state size is 128, the input to the hidden state matrix size is (128, 3), the context state to the hidden state matrix size is (128, 128), and the LSTM unit includes forget The gate and output gate, the forget gate coefficient matrix and the output gate coefficient matrix are of the same size and the size is (128, 128); the bidirectional LSTM layer includes the forward LSTM and the backward LSTM. The structure of the forward LSTM is the same as that of the encoder LSTM, and the structure of the backward LSTM is symmetrical. The hidden state of the backward LSTM is used to merge the output size of the forward LSTM and the backward LSTM, which is (30, 256); the average pooling layer has a pooling window of 1, a step size of 1, and an output size of (30, 256); the fully connected layer inputs 256-dimensional features, outputs 1-dimensional scores, and the weight matrix is ​​(1, 256); the Softmax output layer has no activation function;

[0018] Input the sequence, calculate and output the performance score through the initial framework; use cross entropy as the loss function for training; optimize the initial framework by comparing the performance score with the actual performance score, and finally complete the construction of the overall model.

[0019] Furthermore, the specific data transmission method of the solar cell performance evaluation model includes: the output of the input layer is sent to the encoder LSTM layer, each time step is input, and the hidden state is output as the input of the LSTM layer; the output of the encoder LSTM layer is sent to the bidirectional LSTM layer, the forward LSTM uses the encoder hidden state as input, and the backward LSTM uses the encoder hidden state for back propagation; the output of the bidirectional LSTM layer is sent to the average pooling layer, and the outputs of the last time steps of the front and back LSTMs are superimposed and averaged; the output of the pooling layer is sent to the fully connected layer, the feature vector is used as the input of the fully connected layer, and the output data of the fully connected layer is sent to the softmax output layer.

[0020] Furthermore, the solar cell performance evaluation model training method includes:

[0021] Obtain n groups of historical data of solar cells, the historical data include internal key parameters and external key parameters in the same time series and the performance scores corresponding to internal key parameters and external key parameters in the same time series; randomly separate the n groups of historical data into a training set F and a validation set G, wherein the training set F has m groups of historical data, and the ratio of the number of historical data in the training set F to the number of historical data in the validation set G is 4 / 1; import the training set F into the solar cell performance evaluation model, and calculate the output and loss function through the RNN loop structure; calculate the gradient of the current loss function with respect to the weight, and use the gradient descent method to update the parameters of each layer; repeat the forward calculation and then back-propagate to update the weight, and obtain the solar cell performance evaluation model through multiple rounds of iterative training;

[0022] Among them, the loss function L formula is:

[0023]

[0024] Where, X i is the i-th group of historical data, F(X i ) is the performance score of the i-th group of historical data prediction, Y i Score the actual performance of the i-th group of historical data;

[0025] The BPTT algorithm is used to calculate the gradient of the current loss function to the weight. Let the network parameter set be W, the input sequence be X, the output sequence be Y, and the loss function be L; then the gradient of the loss to the weight W in a single time step is calculated as:

[0026]

[0027] In the formula, is the gradient of the output Y with respect to the loss, which can be directly calculated using the chain rule, It is the gradient of output Y with respect to weight W, which is directly calculated using the kernel function formula;

[0028] Therefore, the total gradient B for the entire input sequence is calculated as:

[0029]

[0030] Where Yt is the RNN output at time step t; Represents the partial derivative of the output Yt with respect to the weight W of the solar cell performance evaluation model.

[0031] Furthermore, the abnormality judgment and screening method includes:

[0032] The performance scores of each time step corresponding to the historical normal data are combined into a normal reference curve graph. The historical normal data refers to the internal key parameters and external key parameters without abnormalities within a time step of one day. All normal reference curve graphs of solar cells are input into the normal reference curve graphic library; the performance scores of the current real-time monitoring are visualized as real-time curve graphs;

[0033] Compare the topological structures of the real-time curve in the real-time curve graph with those in the normal curve graphs in the normal reference curve graphic library, and calculate the curve similarity; if the similarity with all normal curves is lower than the preset threshold, it is determined to be abnormal; classify internal abnormalities and external abnormalities according to the curve characteristics, assuming that the real-time curve graph has a total of u time steps, the normal reference curve graph has a total of u time steps, and the time steps are the same;

[0034] If the performance score continues to decrease within u time steps, it is an internal anomaly;

[0035] If the slope of the real-time curve is greater than z in g consecutive time steps, it is an external anomaly, z is the preset value, g is greater than or equal to 1 and less than u;

[0036] If the performance score continues to decrease within u time steps, and the slope of the real-time curve is greater than z within g consecutive time steps, both internal and external anomalies exist.

[0037] Furthermore, the curve similarity calculation method includes:

[0038] Assume that the real-time curve is C1, the control point set is P1 = {p1, p2, ..., pm}, the normal reference curve is C2, and the control point set is P2 = {q1, q2, ..., qm};

[0039] Compute the control point distance matrix:

[0040] D1 is the distance matrix of C1 control points, D1[r, j] = ||pr-pj||, ||pr-pj|| represents the Euclidean distance between pr and pj, D2 is the distance matrix of C2 control points D2[r, j] = ||qr-qj||; ||qr-qj|| represents the Euclidean distance between qr and qj, r and j are both ∈[1, m];

[0041] Compute the control point order matrix:

[0042] Calculate the topological distance LD(D1, D2):

[0043]

[0044] Curve similarity Similarity(C1, C2):

[0045]

[0046] Furthermore, the threshold calculation method includes: initializing a threshold T0; updating the threshold each time an abnormality is detected; the threshold calculation formula is:

[0047] T i+1 =α×DTW(xi,yi)+(1-α)×T i ;

[0048] Where α is the learning rate, DTW(xi,yi) is the accumulated distance on the path, and T i is the ith threshold.

[0049] Beneficial effects: The present invention provides a real-time dynamic evaluation system for solar cell performance based on deep learning. The effects and advantages are as follows:

[0050] Effect:

[0051] It can collect and analyze multi-source data inside and outside the solar cell in real time, give the battery score in each time window, and realize continuous monitoring of battery performance; by visually displaying the score results of each time step, it can intuitively reflect the working status of the battery under different conditions; it can identify battery abnormalities and preliminarily locate the causes of abnormalities, providing a reference for subsequent problem diagnosis and processing;

[0052] advantage:

[0053] The use of RNN model can consider the dependency of each time step in the time series, better learn and extract features, and the complementary effect of multi-source monitoring data is stronger. The battery status is evaluated from multiple angles, and the results are more accurate and reliable. The system adopts end-to-end learning and automatic modeling. The operation and maintenance cost after training is low, the prediction speed is fast, and the evaluation results are real-time. It is conducive to real-time monitoring and management of batteries, identification of abnormal situations and initial identification of abnormal causes, and providing important decision-making support for battery operation management and maintenance; the results are presented in visual graphics, and the operation is more intuitive and the management is more efficient. The algorithm and system have strong scalability and can adapt to the evaluation needs of batteries of different models and sizes. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a data transmission schematic diagram of the real-time dynamic evaluation system of solar cell performance based on deep learning of the present invention;

[0055] Figure 2 It is a connection diagram of the real-time dynamic evaluation system of solar cell performance based on deep learning of the present invention;

[0056] Figure 3 This is a flow chart of a method for real-time dynamic evaluation of solar cell performance based on deep learning of the present invention;

[0057] Figure 4 is a schematic diagram of an electronic device of the present invention;

[0058] Figure 5 Schematic diagram of the storage medium of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Example 1

[0061] like Figure 1 and Figure 2 As shown, this embodiment provides a real-time dynamic evaluation system for solar cell performance based on deep learning, and the system includes:

[0062] Data acquisition module: collects the internal key parameters of solar cells in real time by installing sensors; collects the external key parameters of solar cells in real time by using photo sensors;

[0063] The key internal parameters include: operating current, operating voltage and operating temperature; the operating current and operating voltage determine the output of solar cells and are important indicators for judging the performance of solar cells. The operating temperature of solar cells affects the efficiency of the battery. The higher the operating temperature, the lower the electrical parameters will be. High temperature can easily lead to battery aging. For example, the battery efficiency is highest at 25-30℃.

[0064] Key external parameters include: light intensity, spectral distribution and light angle; light intensity affects the current and voltage generated by solar cells. The stronger the light intensity, the more photons can be provided to the cells, and the higher the current and voltage generated by the cells; the spectral distribution of sunlight affects the cell's absorption rate of light of different wavelengths, and light that is suitable for the cell's spectral response will be more efficiently converted into electrical energy; the angle of light affects the amount of photons received by the cell's electronic layer, which affects the battery's power output;

[0065] Data processing module: Use deep learning algorithms to process and analyze the collected internal and external key parameters to establish a solar cell performance evaluation model;

[0066] The solar cell performance evaluation model is an RNN model, and the construction method includes:

[0067] Define a time window, the length of the time window, that is, the length of the time step is n, n is a preset value, and the time step includes the corresponding internal key parameters and external key parameters in the time window, for example, time window 1 (8:00-8:15): working current value 1, working voltage value 1, working temperature value 1, light intensity 1, spectrum average value 1, spectrum energy 1, and illumination angle 1; time window 2 (8:15-8:30): working current value 2, working voltage value 2, working temperature value 2, light intensity 2, spectrum curve average value 1, spectrum curve energy 1, and illumination angle 1;

[0068] It is particularly important to note that since the spectral distribution in the external key parameters is multi-dimensional pixel data, the average value and energy in the spectral curve characteristics are directly extracted as the basis for spectral distribution judgment and data input;

[0069] Construct the initial framework of the solar cell performance evaluation model. The initial framework includes an input layer, an encoder LSTM layer, a bidirectional LSTM layer, an average pooling layer, a fully connected layer and a Softmax output layer. The input size of the input layer is the time step length and the feature dimension is 6, that is, the input size is (n, 6). The encoder LSTM layer uses LSTM units. The LSTM unit state includes the context state and the hidden state. The hidden state size is 128, the input to the hidden state matrix size is (128, 3), the context state to the hidden state matrix size is (128, 128), and the LSTM unit includes forget The gate and output gate, the forget gate coefficient matrix and the output gate coefficient matrix are of the same size and the size is (128, 128); the bidirectional LSTM layer includes the forward LSTM and the backward LSTM. The structure of the forward LSTM is the same as that of the encoder LSTM, and the structure of the backward LSTM is symmetrical. The hidden state of the backward LSTM is used to merge the output size of the forward LSTM and the backward LSTM, which is (30, 256); the average pooling layer has a pooling window of 1, a step size of 1, and an output size of (30, 256); the fully connected layer inputs 256-dimensional features, outputs 1-dimensional scores, and the weight matrix is ​​(1, 256); the Softmax output layer has no activation function;

[0070] Input the sequence, calculate and output the performance score through the initial framework; use cross entropy as the loss function for training; optimize the initial framework by comparing the performance score with the actual performance score, and finally complete the construction of the overall model.

[0071] The specific data transmission method of the solar cell performance evaluation model includes: the output of the input layer is sent to the encoder LSTM layer, each time step is input, and the hidden state is output as the input of the LSTM layer; the output of the encoder LSTM layer is sent to the bidirectional LSTM layer, the forward LSTM uses the encoder hidden state as input, and the backward LSTM uses the encoder hidden state for back propagation; the output of the bidirectional LSTM layer is sent to the average pooling layer, and the outputs of the last time steps of the front and rear LSTMs are superimposed and averaged; the output of the pooling layer is sent to the fully connected layer, the feature vector is used as the input of the fully connected layer, and the output data of the fully connected layer is sent to the softmax output layer;

[0072] The solar cell performance evaluation model training method comprises:

[0073] Obtain n groups of historical data of solar cells, the historical data include internal key parameters and external key parameters in the same time series and the performance scores corresponding to internal key parameters and external key parameters in the same time series; randomly separate the n groups of historical data into a training set F and a validation set G, wherein the training set F has m groups of historical data, and the ratio of the number of historical data in the training set F to the number of historical data in the validation set G is 4 / 1; import the training set F into the solar cell performance evaluation model, and calculate the output and loss function through the RNN loop structure; calculate the gradient of the current loss function with respect to the weight, and use the gradient descent method to update the parameters of each layer; repeat the forward calculation and then back-propagate to update the weight, and obtain the solar cell performance evaluation model through multiple rounds of iterative training;

[0074] Among them, the loss function L formula is:

[0075]

[0076] Where, X i is the i-th group of historical data, F(X i ) is the performance score of the i-th group of historical data prediction, Y i Score the actual performance of the i-th group of historical data;

[0077] The BPTT algorithm is used to calculate the gradient of the current loss function to the weight. Let the network parameter set be W, the input sequence be X, the output sequence be Y, and the loss function be L; then the gradient of the loss to the weight W in a single time step is calculated as:

[0078]

[0079] In the formula, is the gradient of the output Y with respect to the loss, which can be directly calculated using the chain rule, It is the gradient of output Y with respect to weight W, which is directly calculated using the kernel function formula;

[0080] Therefore, the total gradient B for the entire input sequence is calculated as:

[0081]

[0082] Where Yt is the RNN output at time step t; It represents the partial derivative of the output Yt with respect to the weight W of the solar cell performance evaluation model, reflecting the influence of the weight on the output of the current time step;

[0083] Data visualization module: visualize the performance scores of each time step;

[0084] Abnormal prediction and diagnosis module: By comparing the visualized graph with the visualized graph of the standard working status in the historical data, it is determined whether abnormalities occur in the current and future time steps. If abnormalities occur, the abnormality type is determined according to the curve change of the visualized graph. The abnormality types include internal abnormalities and external abnormalities. The abnormality screening is performed, and the results of abnormal screening and the visualized graph are imported into the database. At the same time, the warning information is sent to the background server;

[0085] Among them, the abnormal judgment and screening methods include:

[0086] The performance scores of each time step corresponding to the historical normal data are combined into a normal reference curve graph. The historical normal data refers to the internal key parameters and external key parameters without abnormalities within a time step of one day. All normal reference curve graphs of solar cells are input into the normal reference curve graphic library; the performance scores of the current real-time monitoring are visualized as real-time curve graphs;

[0087] Compare the topological structures of the real-time curve in the real-time curve graph with those in the normal curve graphs in the normal reference curve graphic library, and calculate the curve similarity; if the similarity with all normal curves is lower than the preset threshold, it is determined to be abnormal; classify internal abnormalities and external abnormalities according to the curve characteristics, assuming that the real-time curve graph has a total of u time steps, the normal reference curve graph has a total of u time steps, and the time steps are the same;

[0088] If the performance score continues to decrease within u time steps, it is an internal anomaly;

[0089] If the slope of the real-time curve is greater than z in g consecutive time steps, it is an external anomaly, z is the preset value, g is greater than or equal to 1 and less than u;

[0090] If the performance score continues to decrease within u time steps, and the slope of the real-time curve is greater than z within g consecutive time steps, both internal and external anomalies exist;

[0091] The curve similarity calculation method comprises:

[0092] Assume that the real-time curve is C1, the control point set is P1 = {p1, p2, ..., pm}, the normal reference curve is C2, and the control point set is P2 = {q1, q2, ..., qm};

[0093] Compute the control point distance matrix:

[0094] D1 is the distance matrix of C1 control points, D1[r, j] = ||pr-pj||, ||pr-pj|| represents the Euclidean distance between pr and pj, D2 is the distance matrix of C2 control points D2[r, j] = ||qr-qj||; ||qr-qj|| represents the Euclidean distance between qr and qj, r and j are both ∈[1, m];

[0095] Compute the control point order matrix:

[0096] Calculate the topological distance LD(D1, D2):

[0097]

[0098] Curve similarity Similarity(C1, C2):

[0099]

[0100] The threshold calculation method includes: initializing a threshold T0; updating the threshold each time an abnormality is detected; the threshold calculation formula is:

[0101] T i+1 =α×DTW(Xi,yi)+(1-α)×T i ;

[0102] Where α is the learning rate, DTW(xi,yi) is the accumulated distance on the path, and T i is the i-th threshold;

[0103] It is important to note that the calculation of curve similarity uses topological structure comparison, so DTW is a common mathematical symbol used in this method, which stands for dynamic time lag distance.

[0104] Example 2

[0105] See also Figure 3 As shown, this embodiment provides a real-time dynamic evaluation method for solar cell performance based on deep learning, the method comprising:

[0106] The internal key parameters of the solar cell are collected in real time by installing sensors; the external key parameters of the solar cell are collected in real time by using light sensors;

[0107] Use deep learning algorithms to process and analyze the collected internal and external key parameters to establish a solar cell performance evaluation model;

[0108] Visualize the performance scores at each time step;

[0109] By comparing the visualization graph with the visualization graph of the standard working status in the historical data, it is determined whether an abnormality occurs in the current and future time steps. If an abnormality occurs, the type of abnormality is determined based on the curve change of the visualization graph. The abnormality types include internal abnormalities and external abnormalities. Abnormal screening is performed, and the results of abnormal screening and the visualization graph are imported into the database. At the same time, early warning information is sent to the background server.

[0110] Example 3

[0111] See also Figure 4 As shown, according to another aspect of the present application, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are executed by one or more processors, a real-time dynamic evaluation method for solar cell performance based on deep learning as described above may be executed.

[0112] The method or system according to the embodiment of the present application can also be Figure 4 The electronic device architecture shown in FIG. Figure 4 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, ROM 503, RAM 504, a communication port 505 connected to a network, an input / output 506, a hard disk 507, etc. The storage device in the electronic device 500, such as ROM 503 or hard disk 507, may store the real-time dynamic evaluation method of solar cell performance based on deep learning provided in the present application. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 4 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 4 One or more components of an electronic device are shown.

[0113] Example 4

[0114] See also Figure 5As shown, it is a computer storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer storage medium. When the computer-readable instructions are executed by the processor, a real-time dynamic evaluation method for solar cell performance based on deep learning according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0115] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0116] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time dynamic evaluation system for solar cell performance based on deep learning, characterized in that the system include: Data acquisition module: used to collect the internal and external key parameters of solar cells in real time; Data processing module: Use deep learning algorithms to process and analyze the collected internal and external key parameters to establish a solar cell performance evaluation model; Data visualization module: visualize the performance scores of each time step; Abnormal prediction and diagnosis module: By comparing the visualized graph with the visualized graph of the standard working status in the historical data, it is determined whether abnormalities occur in the current and future time steps. If abnormalities occur, the abnormality type is determined according to the curve change of the visualized graph. The abnormality types include internal abnormalities and external abnormalities. The abnormality screening is performed, and the results of abnormal screening and the visualized graph are imported into the database. At the same time, the warning information is sent to the background server; Preliminary abnormality judgment and screening methods include: The performance scores of each time step corresponding to the historical normal data are combined into a normal reference curve graph. The historical normal data refers to the internal key parameters and external key parameters without abnormalities within a time step of one day. All normal reference curve graphs of solar cells are input into the normal reference curve graphic library; the performance scores of the current real-time monitoring are visualized as real-time curve graphs; Compare the topological structures of the real-time curve in the real-time curve graph with those in the normal curve graphs in the normal reference curve graphic library, and calculate the curve similarity; if the similarity with all normal curves is lower than the preset threshold, it is determined to be abnormal; classify internal abnormalities and external abnormalities according to the curve characteristics, assuming that the real-time curve graph has a total of u time steps, the normal reference curve graph has a total of u time steps, and the time steps are the same; If the performance score continues to decrease within u time steps, it is an internal anomaly; If the slope of the real-time curve is greater than z in g consecutive time steps, it is an external anomaly, z is the preset value, g is greater than or equal to 1 and less than u; If the performance score continues to decrease within u time steps, and the slope of the real-time curve is greater than z within g consecutive time steps, both internal and external anomalies exist; The curve similarity calculation method comprises: Assume that the real-time curve is C1, the control point set is P1 = {p1, p2, ..., pm}, the normal reference curve is C2, and the control point set is P2 = {q1, q2, ..., qm}; Calculate the control point distance matrix: D1 is the distance matrix of C1 control points, D1[r,j]=pr-pj, pr-pj represents the Euclidean distance between pr and pj, D2 is the distance matrix of C2 control points D2[r,j]=qr-qj; qr-qj represents the Euclidean distance between qr and qj, r and j are both ∈[1, m]; Calculate the control point order matrix: Calculate the topological distance LD(D1,D2): Curve similarity (C1, C2):

2. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 1 is characterized in that: The key internal parameters include operating current, operating voltage and operating temperature.

3. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 1 is characterized in that: The external key parameters include light intensity, spectral distribution and light angle.

4. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 1 is characterized in that: The solar cell performance evaluation model is an RNN model, and the construction method includes: Define a time window, the length of the time window, i.e., the length of the time step is n, where n is a preset value, and the time step includes the corresponding internal key parameters and external key parameters in the time window; Construct the initial framework of the solar cell performance evaluation model. The initial framework includes an input layer, an encoder LSTM layer, a bidirectional LSTM layer, an average pooling layer, a fully connected layer and a Softmax output layer. The input size of the input layer is the time step length and the feature dimension is 6, that is, the input size is (n, 6). The encoder LSTM layer uses LSTM units. The LSTM unit state includes the context state and the hidden state. The hidden state size is 128, the input to the hidden state matrix size is (128, 3), the context state to the hidden state matrix size is (128, 128), and the LSTM unit includes forget The gate and output gate, the forget gate coefficient matrix and the output gate coefficient matrix are of the same size and the size is (128, 128); the bidirectional LSTM layer includes the forward LSTM and the backward LSTM. The structure of the forward LSTM is the same as that of the encoder LSTM, and the structure of the backward LSTM is symmetrical. The hidden state of the backward LSTM is used to merge the output size of the forward LSTM and the backward LSTM, which is (30, 256); the average pooling layer has a pooling window of 1, a step size of 1, and an output size of (30, 256); the fully connected layer inputs 256-dimensional features, outputs 1-dimensional scores, and the weight matrix is ​​(1, 256); the Softmax output layer has no activation function; Input sequence, calculate and output performance score through the initial framework; train with cross entropy as loss function; The initial framework is optimized by comparing the performance score with the actual performance score, and finally the overall model is built.

5. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 4 is characterized in that: The specific data transmission method of the solar cell performance evaluation model includes: The output of the input layer is sent to the encoder LSTM layer, each time step is input, and the output Hiddenstate is used as the input of the LSTM layer; the output of the encoder LSTM layer is sent to the bidirectional LSTM layer, the forward LSTM uses the encoder hidden state as input, and the backward LSTM uses the encoder hidden state for back propagation; the output of the bidirectional LSTM layer is sent to the average pooling layer, and the output of the last time step of the front and back LSTM is superimposed and averaged; the output of the pooling layer is sent to the fully connected layer, and the feature vector is used as the input of the fully connected layer. The output data of the fully connected layer is sent to the softmax output layer.

6. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 5 is characterized in that: The solar cell performance evaluation model training method comprises: Obtain n groups of historical data of solar cells, the historical data include internal key parameters and external key parameters in the same time series and the performance scores corresponding to internal key parameters and external key parameters in the same time series; randomly separate the n groups of historical data into a training set F and a validation set G, wherein the training set F has m groups of historical data, and the ratio of the number of historical data in the training set F to the number of historical data in the validation set G is 4 / 1; import the training set F into the solar cell performance evaluation model, and calculate the output and loss function through the RNN loop structure; calculate the gradient of the current loss function with respect to the weight, and use the gradient descent method to update the parameters of each layer; repeat the forward calculation and then back-propagate to update the weight, and obtain the solar cell performance evaluation model through multiple rounds of iterative training; Among them, the loss function L formula is: Where, X i is the i-th group of historical data, F(X i ) is the performance score of the i-th group of historical data prediction, Y i Score the actual performance of the i-th group of historical data; The BPTT algorithm is used to calculate the gradient of the current loss function to the weight. Let the network parameter set be W, the input sequence be X, the output sequence be Y, and the loss function be L; then the gradient of the loss to the weight W in a single time step is calculated as: In the formula, is the gradient of the output Y with respect to the loss, which can be directly calculated using the chain rule, It is the gradient of output Y with respect to weight W, which is directly calculated using the kernel function formula; Therefore, the total gradient B for the entire input sequence is calculated as: Where Yt is the RNN output at time step t; Represents the partial derivative of the output Yt with respect to the weight W of the solar cell performance evaluation model.

7. The real-time dynamic evaluation system for solar cell performance based on deep learning according to claim 1, characterized in that: The threshold calculation method includes: initializing a threshold T0; updating the threshold each time an abnormality is detected; the threshold calculation formula is: T i+1 =α×DTW(xi,yi)+(1-α)×T i ; Where α is the learning rate, DTW(xi,yi) is the accumulated distance on the path, and T i is the ith threshold.

Images