Photovoltaic grid-connected practical training system and method

Abstract

The invention discloses a photovoltaic grid-connected practical training system and method, and relates to the technical field of photovoltaic grid-connected practical training, and the system comprises an operation state analysis unit, a configuration parameter determination unit, a model parameter output unit, a model detection unit, and a practical training score settlement unit. According to the invention, the operation state analysis unit is utilized to construct photovoltaic array models in different operation states, the operation state judged by a user is compared with the actual operation state, and the project score of the user is set according to the comparison result. A user can conveniently analyze the running state of a current photovoltaic array model according to a photovoltaic module model, a characteristic curve and recorded data, so that the user is helped to know related data of a photovoltaic module, and meanwhile, a configuration parameter determination unit selects configuration parameters in a long-short-term memory network model in a data set setting mode; and the analysis performance of the network model can be optimized by the configuration parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic grid-connected training, and specifically provides a photovoltaic grid-connected training system and method. Background Art

[0002] Photovoltaic grid-connected training systems are mainly designed to help trainees understand and master the basic principles, design, debugging, and maintenance skills of photovoltaic power generation. Such systems typically include key components such as photovoltaic modules, inverters, energy storage devices, and monitoring systems, and can simulate a real photovoltaic grid-connected power generation environment. In the patent with the application number 202420174831.X, "A photovoltaic off-grid and grid-connected integrated training system is disclosed, which includes a photovoltaic support installation platform and a control cabinet. Two photovoltaic supports are arranged side by side on the photovoltaic support installation platform. A photovoltaic module and a light irradiance monitoring module are provided at the middle position of each photovoltaic support. A simulated solar light source module is also connected to each photovoltaic support through a connecting rod. An environmental meteorological monitoring module is provided at the top of one of the photovoltaic supports. Two drawers are arranged side by side below the photovoltaic support installation platform. A photovoltaic string converging module is provided in one of the drawers, and a component temperature and meteorological data acquisition module is provided in the other drawer. The photovoltaic string converging module is connected to the photovoltaic module, and the environmental meteorological monitoring module and the light irradiance monitoring module are connected to the component temperature and meteorological data acquisition module. The photovoltaic off-grid and grid-connected integrated training system is suitable for professional teaching practice."

[0003] The above-mentioned prior art solves problems such as the large differences in currently highly integrated photovoltaic systems, which prevent students from effectively mastering photovoltaic application-related knowledge. However, during system operation, since a photovoltaic array model under different operating states is not constructed, users cannot effectively understand the corresponding characteristic curves and data differences under different operating states. At the same time, the system cannot automatically identify the operating state of the photovoltaic module model built by the user, making it difficult for users to detect and adjust in a timely manner even when incorrect operations occur, and it also cannot reflect the user's mastery of photovoltaic modules based on the connection situation. Summary of the Invention

[0004] The purpose of the present invention is to provide a photovoltaic grid-connected training system and method to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A photovoltaic grid-connected training system includes a model detection unit.

[0006] An operating status analysis unit. The operating status analysis unit uses Matlab software to build a photovoltaic power station model, extracts photovoltaic module models corresponding to different operating states under the same environmental parameters, calculates the similarity between the characteristic curves and recorded data of different module models, and then transmits the current photovoltaic module model, characteristic curves, and recorded data to the user interface, waiting for the user to return the operating state. It compares this operating state with the actual operating state and calculates the user score for this round according to the similarity between the two operating states.

[0007] A configuration parameter determination unit. After setting the number of data groups, spatial dimension, and number of loops, the configuration parameter determination unit initializes the values at each position in the data group according to the configuration parameter range in the long short-term memory network model, determines the accuracy corresponding to each data group, selects the two data groups with the highest accuracy and unmarked ones, swaps the values at the specified positions in the current data group to obtain the next-round data group, marks the original data group, counts the next-round data group and accuracy of each original data group, repeats the operation until the loop ends, and extracts the values at each position in the data group with the maximum accuracy to determine the optimal configuration parameters of the long short-term memory network model.

[0008] A model parameter output unit. After extracting the operation record data of the photovoltaic array under different light intensities, the model parameter output unit generates corresponding sample data according to the operation record data, constructs a set matrix using the sample data, determines the eigenvalues and eigenvectors of the current matrix, analyzes the main eigenvectors using the mapping function and eigenvectors, projects each sample data according to the main eigenvectors to obtain new sample data, and transmits the sample data to the long short-term memory network model for training to determine the network model parameters.

[0009] A training score settlement unit. After counting the weight values of different rounds and the user score for each round, the training score settlement unit uses deep learning technology to calculate the user's comprehensive training score according to the score of each round and the weight values of different rounds, and then returns it to the corresponding user interface.

[0010] Preferably, the operation status analysis unit includes an information storage module, a model construction module, and a status classification module. After the information storage module obtains all the equipment installation steps and introduction information involved in photovoltaic grid connection, it stores them in the database. The model construction module uses Matlab software to construct a photovoltaic module model, combines multiple photovoltaic modules in a series-parallel manner to obtain a photovoltaic array model, connects this model with other photovoltaic grid-connected equipment models to determine a photovoltaic power station model. The status classification module divides the operation status of the photovoltaic array into seven types, namely normal status, short-circuit status, over-attenuation status, occlusion status, aging status, dust accumulation status, and PID phenomenon status.

[0011] Preferably, the operation status analysis unit further includes a similarity calculation module and a status comparison module. The similarity calculation module extracts the photovoltaic module models corresponding to different operation statuses under the same environmental parameters, determines the characteristic curves and operation record data according to the model simulation results, and calculates the similarity between the curves and record data corresponding to different operation statuses. The status comparison module transmits the current photovoltaic module model, characteristic curves, and record data to the user interface, waits for the user to return the operation status, compares this operation status with the actual operation status, and calculates the current round of user scores according to the similarity between the two operation statuses.

[0012] Preferably, the configuration parameter determination unit includes an accuracy calculation module and a data group marking module. After the accuracy calculation module sets the number of data groups, spatial dimensions, and number of loops, it initializes the values at each position in the data group according to the configuration parameter range in the long short-term memory network model, calculates the accuracy when the network model runs using the values included in each data group as model parameters. The data group marking module determines the accuracy corresponding to each data group. If the accuracy is less than the threshold, the current data group is deleted; otherwise, the current data group is retained. After selecting the two data groups with the highest accuracy and not being marked, the values at the specified positions in the current data group are swapped to obtain the next round of data groups, and then the original data groups are marked.

[0013] Preferably, the configuration parameter determination unit further includes a difference value calculation module and an optimal parameter output module. After the difference value calculation module counts the next-round data groups of each original data group, it processes the next-round data groups, calculates the model accuracy corresponding to the data groups. If the current accuracy is higher than the accuracy of the original data group, the next-round data group is retained; otherwise, the original data group is used as the next-round data group. The accuracy analysis algorithm is used to analyze the accuracies of the original data group and the next-round data group to obtain a difference value. If the difference value is within a preset range, the next round of loop is entered; otherwise, the next round of loop is not performed. The optimal parameter output module extracts the data group with the maximum accuracy and determines the optimal configuration parameters in the long short-term memory network model according to the values at each position in the data group. The accuracy analysis algorithm is specifically as follows:

[0014]

[0015] Among them, represents the difference value between the accuracies of all data groups in the m-th round and the accuracies of all data groups in the (m - 1)-th round, represents the accuracy of the l-th data group in the m-th round, represents the accuracy of the l-th data group in the (m - 1)-th round, where m and l are parameters, and μ represents the data group.

[0016] Preferably, the model parameter output unit includes a sample storage module, a covariance calculation module, a feature vector analysis module, and a model training module. After the sample storage module extracts the operation record data of the photovoltaic array under different light intensities, it constructs corresponding sample data according to the operation record data and stores it in a set. After the covariance calculation module counts all the sample data in the set, it constructs a set matrix using the sample data, converts the data included in the set matrix through a mapping function, and calculates the corresponding covariance value according to the converted matrix data using a matrix analysis algorithm. The feature vector analysis module determines the eigenvalues and eigenvectors of the current matrix using the covariance value corresponding to the matrix, analyzes the number of main eigenvectors according to the mapping function and the eigenvectors, projects each sample data according to the main eigenvectors to obtain new sample data. The model training module adds corresponding label information to each sample data, where the label information is specifically the actual operation state of the photovoltaic array, and transmits the sample data and the corresponding label information to the long short-term memory network model for training to determine the network model parameters.

[0017] Preferably, the model detection unit includes a recorded data analysis module, a model layout determination module, and a prompt information display module. After receiving the photovoltaic component model constructed by the user, the recorded data analysis module determines the operation record data of the current photovoltaic component according to the model simulation result. After projecting the current operation record data according to the main feature vector, the projection result is transmitted to the long short-term memory network model for analysis, so as to obtain the operation state of the current photovoltaic component. If the operation state of the current photovoltaic component is in a normal state, the model layout determination module determines that the photovoltaic component model constructed by the user is reasonably laid out and determines the user score for this round. If the operation state of the current photovoltaic component is not in a normal state, the model layout determination module determines that the photovoltaic component model constructed by the user is not reasonably laid out and starts the timer. If the photovoltaic component model is still not reasonably laid out before the end of the timing, the prompt information display module outputs the operation state of the photovoltaic component analyzed by the long short-term memory network model through the visualization interface and determines that the user score for this round is zero. If the photovoltaic component model is determined to be reasonably laid out before the end of the timing, the prompt information display module returns a prompt message indicating the successful construction of the photovoltaic component model to the user interface and determines the score for this round according to the layout time.

[0018] The photovoltaic grid-connected training method includes the following steps:

[0019] S1. Analyze the operation state: Use Matlab software to construct a photovoltaic power station model, extract the photovoltaic component models corresponding to different operation states under the same environmental parameters, calculate the similarity between the characteristic curves and the recorded data of different component models, and then transmit the current photovoltaic component model, characteristic curve, and recorded data to the user interface, waiting for the user to return the operation state. Compare this operation state with the actual operation state, and calculate the user score for this round according to the similarity between the two operation states;

[0020] S2. Determine model parameters: After setting the number of data groups, spatial dimensions, and number of loops, initialize the values at each position in the data groups according to the configuration parameter range in the long short-term memory network model, determine the accuracy corresponding to each data group, select the two data groups with the highest accuracy and unmarked ones, swap the values at the specified positions in the current data group to obtain the next-round data group, mark the original data group, count the next-round data group and accuracy of each original data group, repeat the operation until the loop ends, extract the values at each position in the data group with the maximum accuracy to determine the optimal configuration parameters of the long short-term memory network model. After extracting the operation record data of the photovoltaic array under different light intensities, generate corresponding sample data according to the operation record data, construct a set matrix using the sample data, determine the eigenvalues and eigenvectors of the current matrix, analyze the main eigenvectors using the mapping function and eigenvectors, project each sample data according to the main eigenvectors to obtain new sample data, and transmit the sample data to the long short-term memory network model for training to determine the network model parameters;

[0021] S3. Deploy the photovoltaic module model: After receiving the photovoltaic module model constructed by the user, determine the operation record data of the current photovoltaic module according to the model simulation result, analyze the record data using the long short-term memory network model to obtain the operation state of the current photovoltaic module, and determine the user score for this round according to the operation state and deployment time;

[0022] S4. Output the training score: After counting the weight values of different rounds and the score of the user in each round, use deep learning technology to calculate the comprehensive training score of the user according to the score of each round and the weight values of different rounds, and return it to the corresponding user interface.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] The present invention uses the operation state analysis unit to construct a photovoltaic array model under different operation states, compares the operation state judged by the user with the actual operation state, and sets the project score of the user according to the comparison result, which is convenient for the user to analyze the operation state of the current photovoltaic array model based on the photovoltaic module model, characteristic curve, and record data, thereby helping the user understand the relevant data of the photovoltaic module. At the same time, the configuration parameter determination unit selects the configuration parameters in the long short-term memory network model by setting data groups to ensure that the configuration parameters can optimize the analysis performance of the network model. The model detection unit analyzes the operation state of the photovoltaic module model built by the user, and determines the user's mastery of component connection according to the operation state, thereby improving the user's training ability, and the training score settlement unit can more accurately reflect the user's familiarity with the relevant information of the photovoltaic module. Description of the Drawings

[0025] Figure 1 Schematic diagram of the overall system flow provided by the embodiment of the present invention;

[0026] Figure 2 Internal module block diagram of the operating state analysis unit provided by the embodiment of the present invention;

[0027] Figure 3 Internal module block diagram of the configuration parameter determination unit provided by the embodiment of the present invention;

[0028] Figure 4 Internal module block diagram of the model parameter output unit provided by the embodiment of the present invention;

[0029] Figure 5 Internal module block diagram of the model detection unit provided by the embodiment of the present invention.

[0030] In the figure: 1. Operating state analysis unit; 101. Information storage module; 102. Model construction module; 103. State division module; 104. Similarity calculation module; 105. State comparison module; 2. Configuration parameter determination unit; 201. Accuracy calculation module; 202. Data group marking module; 203. Gap value calculation module; 204. Optimal parameter output module; 3. Model parameter output unit; 301. Sample storage module; 302. Covariance calculation module; 303. Eigenvector analysis module; 304. Model training module; 4. Model detection unit; 401. Record data analysis module; 402. Model layout determination module; 403. Prompt information display module; 5. Training score settlement unit. Detailed implementation manners

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0032] Please refer to Figures 1 - 5 , the present invention provides a technical solution: a photovoltaic grid-connected training system, including a model detection unit 4;

[0033] The operating status analysis unit 1 uses Matlab software to construct a photovoltaic power station model, extracts photovoltaic module models corresponding to different operating states under the same environmental parameters, calculates the similarity between the characteristic curves and recorded data of different module models, and then transmits the current photovoltaic module model, characteristic curve, and recorded data to the user interface. It waits for the user to return the operating state, compares this operating state with the actual operating state, and calculates the user score for this round according to the similarity between the two operating states.

[0034] The configuration parameter determination unit 2 sets the number of data groups, spatial dimension, and number of loops, and then initializes the values at each position in the data group according to the configuration parameter range in the long short-term memory network model to determine the accuracy corresponding to each data group. After selecting the two data groups with the highest accuracy and not being marked, it swaps the values at the specified positions in the current data group to obtain the next-round data group, marks the original data group, and counts the next-round data group and accuracy of each original data group. It repeats the operation until the loop ends, and extracts the values at each position in the data group with the maximum accuracy to determine the optimal configuration parameters of the long short-term memory network model.

[0035] The model parameter output unit 3 extracts the operating record data of the photovoltaic array under different light intensities, generates corresponding sample data according to the operating record data, constructs a set matrix using the sample data, determines the eigenvalues and eigenvectors of the current matrix, analyzes the main eigenvectors using the mapping function and eigenvectors, projects each sample data according to the main eigenvectors to obtain new sample data, and transmits the sample data to the long short-term memory network model for training to determine the network model parameters.

[0036] The training score settlement unit 5 counts the weight values of different rounds and the user score for each round, and then uses deep learning technology to calculate the user's comprehensive training score according to the score of each round and the weight values of different rounds, and returns it to the corresponding user interface.

[0037] The operation status analysis unit 1 includes an information storage module 101, a model construction module 102, and a status division module 103. After the information storage module 101 obtains all the equipment installation steps and introduction information involved in photovoltaic grid connection, it stores them in the database. The model construction module 102 uses Matlab software to construct a photovoltaic module model, combines multiple photovoltaic modules in a series-parallel manner to obtain a photovoltaic array model, connects this model with other photovoltaic grid-connected equipment models to determine a photovoltaic power station model. The status division module 103 divides the operation status of the photovoltaic array into seven types, namely normal status, short-circuit status, over-attenuation status, occlusion status, aging status, dust accumulation status, and PID phenomenon status;

[0038] The operation status analysis unit 1 further includes a similarity calculation module 104 and a status comparison module 105. The similarity calculation module 104 extracts the photovoltaic module models corresponding to different operation statuses under the same environmental parameters, determines the characteristic curves and operation record data according to the model simulation results, and calculates the similarity between the curves and record data corresponding to different operation statuses. The status comparison module 105 transmits the current photovoltaic module model, characteristic curves, and record data to the user interface, waits for the user to return the operation status, compares this operation status with the actual operation status, and calculates the current round of user scores according to the similarity between the two operation statuses;

[0039] The configuration parameter determination unit 2 includes an accuracy calculation module 201 and a data group marking module 202. After the accuracy calculation module 201 sets the number of data groups, spatial dimensions, and number of loops, it initializes the values at each position in the data group according to the configuration parameter range in the long short-term memory network model, calculates the accuracy when the network model runs using the values included in each data group as model parameters. The data group marking module 202 determines the accuracy corresponding to each data group. If the accuracy is less than the threshold, the current data group is deleted; otherwise, the current data group is retained. After selecting the two data groups with the highest accuracy and not being marked, the values at the specified positions in the current data group are swapped to obtain the next round of data groups, and then the original data groups are marked;

[0040] The configuration parameter determination unit 2 further includes a gap value calculation module 203 and an optimal parameter output module 204. After the gap value calculation module 203 counts the next-round data groups of each original data group, it processes the next-round data groups, calculates the model accuracy corresponding to the data groups. If the current accuracy is higher than the accuracy of the original data group, the next-round data group is retained; otherwise, the original data group is used as the next-round data group. The accuracy analysis algorithm is used to analyze the accuracies of the original data group and the next-round data group to obtain the gap value. If the gap value is within the preset range, the next-round loop is entered; otherwise, the next-round loop is not performed. The optimal parameter output module 204 extracts the data group with the maximum accuracy and determines the optimal configuration parameters in the long short-term memory network model according to the values at each position in the data group. The accuracy analysis algorithm is specifically as follows:

[0041]

[0042] Among them, represents the gap value between the accuracies of all data groups in the m-th round and the accuracies of all data groups in the (m - 1)-th round, represents the accuracy of the l-th data group in the m-th round, represents the accuracy of the l-th data group in the (m - 1)-th round, where m and l are parameters, and μ represents the data group;

[0043] The model parameter output unit 3 includes a sample storage module 301, a covariance calculation module 302, a feature vector analysis module 303, and a model training module 304. After the sample storage module 301 extracts the operation record data of the photovoltaic array under different light intensities, it constructs the corresponding sample data according to the operation record data and stores it in the set. After the covariance calculation module 302 counts all the sample data in the set, it constructs a set matrix using the sample data, converts the data contained in the set matrix through a mapping function, and calculates the corresponding covariance value according to the converted matrix data using the matrix analysis algorithm. The feature vector analysis module 303 determines the eigenvalues and eigenvectors of the current matrix using the covariance value corresponding to the matrix, analyzes the number of main eigenvectors according to the mapping function and the eigenvectors, projects each sample data according to the main eigenvectors to obtain new sample data. The model training module 304 adds the corresponding label information to each sample data, where the label information is specifically the actual operation state of the photovoltaic array, and transmits the sample data and the corresponding label information to the long short-term memory network model for training to determine the network model parameters. The matrix analysis algorithm is specifically as follows:

[0044]

[0045] Among them, P represents the covariance value, N represents the total number of sample data, f(α k) represents matrix data, f(α k ) T represents the matrix data after transposition, k represents a parameter, and α k represents the value in the k-th matrix;

[0046] The model detection unit 4 includes a recorded data analysis module 401, a model layout determination module 402, and a prompt information display module 403. After receiving the photovoltaic module model constructed by the user, the recorded data analysis module 401 determines the operation record data of the current photovoltaic module according to the model simulation result. After projecting the current operation record data according to the main eigenvector, the projection result is transmitted to the long short-term memory network model for analysis, so as to obtain the operation state of the current photovoltaic module. If the operation state of the current photovoltaic module is in a normal state, the model layout determination module 402 determines that the photovoltaic module model constructed by the user is reasonably laid out and determines the current round of user score. If the operation state of the current photovoltaic module is not in a normal state, the model layout determination module 402 determines that the photovoltaic module model constructed by the user is unreasonably laid out and starts a timer to start timing. If the photovoltaic module model is still unreasonably laid out before the end of the timing, the prompt information display module 403 outputs the operation state of the photovoltaic module analyzed by the long short-term memory network model through the visualization interface and determines that the current round of user score is zero. If the photovoltaic module model is determined to be reasonably laid out before the end of the timing, the prompt information display module 403 returns a prompt message indicating the successful construction of the photovoltaic module model to the user interface and determines the score of this round according to the layout time;

[0047] A photovoltaic grid-connected training method includes the following steps:

[0048] S1. Analyze the operation state: Use Matlab software to construct a photovoltaic power station model, extract the photovoltaic module models corresponding to different operation states under the same environmental parameters, calculate the similarity between the characteristic curves and the record data of different module models, and then transmit the current photovoltaic module model, the characteristic curve, and the record data to the user interface, wait for the user to return the operation state, compare the operation state with the actual operation state, and calculate the current round of user score according to the similarity between the two operation states;

[0049] S2. Determine model parameters: After setting the number of data groups, spatial dimensions, and the number of loops, initialize the values at each position in the data groups according to the configuration parameter range in the long short-term memory network model, determine the accuracy corresponding to each data group, select the two data groups with the highest accuracy and unmarked, swap the values at the specified positions in the current data group to obtain the next-round data group, mark the original data group, count the next-round data group and accuracy of each original data group, repeat the operation until the loop ends, extract the values at each position in the data group with the maximum accuracy to determine the optimal configuration parameters of the long short-term memory network model, extract the operation record data of the photovoltaic array under different light intensities, generate corresponding sample data according to the operation record data, construct a set matrix using the sample data, determine the eigenvalues and eigenvectors of the current matrix, analyze the main eigenvectors using the mapping function and eigenvectors, project each sample data according to the main eigenvectors to obtain new sample data, and transmit the sample data to the long short-term memory network model for training to determine the network model parameters;

[0050] S3. Deploy the photovoltaic module model: After receiving the photovoltaic module model constructed by the user, determine the operation record data of the current photovoltaic module according to the model simulation results, analyze the record data using the long short-term memory network model to obtain the operation status of the current photovoltaic module, and determine the user score for this round according to the operation status and deployment time;

[0051] S4. Output the training score: After counting the weight values of different rounds and the score of the user in each round, use deep learning technology to calculate the comprehensive training score of the user according to the score of each round and the weight values of different rounds, and return it to the corresponding user interface.

[0052] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0053] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic grid-connected training system, comprising a model detection unit (4), characterized in that: An operating state analysis unit (1), wherein the operating state analysis unit (1) constructs a photovoltaic power station model using Matlab software, extracts photovoltaic component models corresponding to different operating states under the same environmental parameters, calculates the similarity between characteristic curves and recorded data of different component models, transmits the current photovoltaic component model, characteristic curve and recorded data to a user interface, waits for the user to return to the operating state, compares the operating state with the actual operating state, and calculates the user score for this round according to the similarity between the two operating states; A configuration parameter determination unit (2), wherein after setting the number of data groups, the spatial dimension and the number of cycles, the configuration parameter determination unit (2) performs an initialization operation on the values ​​of each position in the data group according to the configuration parameter range in the long short-term memory network model, determines the accuracy corresponding to each data group, selects two data groups with the highest accuracy and are not marked, swaps the values ​​at the specified positions in the current data group, obtains the next round of data groups, marks the original data groups, counts the next round of data groups and the accuracy of each original data group, repeats the operation until the cycle ends, and extracts the values ​​of each position in the data group with the highest accuracy to determine the optimal configuration parameters of the long short-term memory network model; A model parameter output unit (3), wherein the model parameter output unit (3) extracts operation record data of the photovoltaic array under different light intensities, generates corresponding sample data according to the operation record data, constructs a set matrix using the sample data, determines the eigenvalue and eigenvector of the current matrix, analyzes the main eigenvector using a mapping function and the eigenvector, projects each sample data according to the main eigenvector to obtain new sample data, and transmits the sample data to a long short-term memory network model for training, thereby determining network model parameters; A training score settlement unit (5) calculates the user's comprehensive training score by using deep learning technology based on the scores of each round and the weight values ​​of different rounds, and returns the score to the corresponding user interface after counting the weight values ​​of different rounds and the scores of each round of the user.

2. A photovoltaic grid-connected training system according to claim 1, characterized in that: The operation status analysis unit (1) comprises an information storage module (101), a model construction module (102) and a state classification module (103). The information storage module (101) acquires all equipment installation steps and introduction information involved in photovoltaic grid connection and stores them in a database. The model construction module (102) uses Matlab software to construct a photovoltaic component model, combines multiple photovoltaic components in a series-parallel combination manner, thereby obtaining a photovoltaic array model, and connects the model with other photovoltaic grid-connected equipment models to determine a photovoltaic power station model. The state classification module (103) divides the operation status of the photovoltaic array into seven types, namely normal state, short circuit state, over-attenuation state, shielding state, aging state, dust accumulation state and PID phenomenon state.

3. A photovoltaic grid-connected training system according to claim 2, characterized in that: The operating state analysis unit (1) further comprises a similarity calculation module (104) and a state comparison module (105). The similarity calculation module (104) extracts photovoltaic component models corresponding to different operating states under the same environmental parameters, determines characteristic curves and operating record data according to model simulation results, and calculates the similarity between the curves and record data corresponding to different operating states. The state comparison module (105) transmits the current photovoltaic component model, characteristic curve and record data to the user interface, waits for the user to return to the operating state, compares the operating state with the actual operating state, and calculates the user score for this round according to the similarity between the two operating states.

4. A photovoltaic grid-connected training system according to claim 1, characterized in that: The configuration parameter determination unit (2) comprises an accuracy calculation module (201) and a data group marking module (202). After the accuracy calculation module (201) sets the number of data groups, the spatial dimension and the number of cycles, it performs an initialization operation on the values ​​at each position in the data group according to the configuration parameter range in the long short-term memory network model, and calculates the accuracy of the network model when it is running after taking the values ​​contained in each data group as model parameters. The data group marking module (202) determines the accuracy corresponding to each data group. If the accuracy is less than a threshold, the current data group is deleted, otherwise the current data group is retained. After selecting the two data groups with the highest accuracy and not marked, the values ​​at the specified positions in the current data group are exchanged. After obtaining the next round of data groups, the original data group is marked.

5. A photovoltaic grid-connected training system according to claim 4, characterized in that: The configuration parameter determination unit (2) also includes a gap value calculation module (203) and an optimal parameter output module (204). The gap value calculation module (203) counts the next round of data groups of each original data group, processes the next round of data groups, and calculates the model accuracy corresponding to the data group. If the current accuracy is higher than the accuracy of the original data group, the next round of data groups are retained. Otherwise, the original data group is used as the next round of data groups. The accuracy of the original data group and the next round of data groups are analyzed using an accuracy analysis algorithm to obtain a gap value. If the gap value is within a preset range, the next round of circulation is entered. Otherwise, the next round of circulation is not performed. The optimal parameter output module (204) extracts the data group with the highest accuracy and determines the optimal configuration parameters in the long short-term memory network model according to the values ​​of each position in the data group.

6. A photovoltaic grid-connected training system according to claim 1, characterized in that: The model parameter output unit (3) comprises a sample storage module (301), a covariance calculation module (302), a feature vector analysis module (303) and a model training module (304); after the sample storage module (301) extracts the operation record data of the photovoltaic array under different light intensities, corresponding sample data is constructed according to the operation record data and stored in a set; after the covariance calculation module (302) counts all the sample data in the set, a set matrix is ​​constructed using the sample data; the data contained in the set matrix is ​​converted by a mapping function; and a matrix analysis algorithm is used to calculate the data according to the converted data. The corresponding covariance value is calculated from the matrix data, the eigenvector analysis module (303) determines the eigenvalue and eigenvector of the current matrix using the covariance value corresponding to the matrix, analyzes the number of main eigenvectors according to the mapping function and the eigenvector, projects each sample data according to the main eigenvector to obtain new sample data, the model training module (304) adds corresponding label information to each sample data, wherein the label information is specifically the actual operating status of the photovoltaic array, and transmits the sample data and the corresponding label information to the long short-term memory network model for training, thereby determining the network model parameters.

7. A photovoltaic grid-connected training system according to claim 1, characterized in that: The model detection unit (4) comprises a record data analysis module (401), a model layout determination module (402) and a prompt information display module (403). After receiving the photovoltaic module model constructed by the user, the record data analysis module (401) determines the operation record data of the current photovoltaic module according to the model simulation result, and after projecting the current operation record data according to the main feature vector, transmits the projection result to the long short-term memory network model for analysis, thereby obtaining the operation state of the current photovoltaic module. If the operation state of the current photovoltaic module is normal, the model layout determination module (402) determines that the photovoltaic module constructed by the user is normal. The photovoltaic module model constructed is reasonably arranged, and the user score of this round is determined. If the current operating state of the photovoltaic module is not normal, the photovoltaic module model constructed by the user is determined to be unreasonable, and the timer is started to start timing. If the photovoltaic module model is still unreasonable before the timing ends, the prompt information display module (403) outputs the photovoltaic module operating state analyzed by the long short-term memory network model through a visual interface, and determines that the user score of this round is zero. If the photovoltaic module model is determined to be reasonably arranged before the timing ends, a prompt information indicating that the photovoltaic module model is successfully constructed is returned to the user interface, and the score of this round is determined according to the arrangement time.

8. A photovoltaic grid-connected training method, characterized in that: The photovoltaic grid-connected training method is applicable to a photovoltaic grid-connected training system according to any one of claims 1 to 7, comprising the following steps: S1. Analyze the operation status: Use Matlab software to build a photovoltaic power station model, extract the photovoltaic module models corresponding to different operation states under the same environmental parameters, calculate the similarity between the characteristic curves and recorded data of different module models, and then transmit the current photovoltaic module model, characteristic curve and recorded data to the user interface, wait for the user to return to the operation status, compare the operation status with the actual operation status, and calculate the user score for this round according to the similarity between the two operation states; S2. Determine model parameters: after setting the number of data groups, spatial dimensions and number of loops, initialize the values ​​of each position in the data group according to the configuration parameter range in the long short-term memory network model, determine the accuracy corresponding to each data group, select the two data groups with the highest accuracy and are not marked, swap the values ​​at the specified positions in the current data group, and after obtaining the next round of data groups, mark the original data groups, count the next round of data groups and the accuracy of each original data group, repeat the operation until the end of the loop, extract the values ​​of each position in the data group with the maximum accuracy to determine the optimal configuration parameters of the long short-term memory network model, extract the operation record data of the photovoltaic array under different light intensities, generate the corresponding sample data according to the operation record data, use the sample data to construct a set matrix, determine the eigenvalues ​​and eigenvectors of the current matrix, use the mapping function and the eigenvector to analyze the main eigenvectors, project each sample data according to the main eigenvectors to obtain new sample data, and transfer the sample data to the long short-term memory network model for training, so as to determine the network model parameters; S3. Layout of photovoltaic module model: After receiving the photovoltaic module model constructed by the user, the operation record data of the current photovoltaic module is determined according to the model simulation results, and the record data is analyzed using the long short-term memory network model to obtain the current operation status of the photovoltaic module, and the user score of this round is determined according to the operation status and layout time; S4. Output training scores: After counting the weight values ​​of different rounds and the scores of each round of the user, use deep learning technology to calculate the user's comprehensive training score based on the scores of each round and the weight values ​​of different rounds, and return it to the corresponding user interface.

Images