An Artificial Intelligence-Based Photovoltaic Array Fault Diagnosis System and Method

Abstract

This invention relates to the field of photovoltaic power generation fault diagnosis technology, and in particular to a photovoltaic array fault diagnosis system and method based on artificial intelligence. By encrypting and preprocessing multi-source data, a multimodal dataset is constructed, providing a comprehensive and collaborative data foundation for subsequent analysis. The collaborative diagnosis model can analyze the battery and inverter status in parallel. By combining circuit equivalent models, spatial adjacency relationships, and theoretical power models for calculation, the diagnosis results have physical interpretability. A physical consistency verification rule based on power flow analysis is introduced to cross-validate and logically fuse the preliminary diagnosis results, generating reliable comprehensive fault diagnosis results. This approach solves the technical problem in existing technologies that focus on independent analysis of single components such as photovoltaic cells or inverters, lacking effective fusion and collaborative utilization of multi-source heterogeneous data, which limits the accuracy of diagnosis results.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation fault diagnosis technology, and in particular to a photovoltaic array fault diagnosis system and method based on artificial intelligence. Background Technology

[0002] As a crucial component of clean energy, photovoltaic (PV) power generation plays an increasingly vital role in the global energy structure transformation. With the continuous expansion of PV power plant scale and the increasing complexity of deployment environments, PV arrays, exposed to the natural environment for extended periods, are susceptible to multiple factors such as aging, shading, hot spots, connection failures, and inverter performance degradation, leading to decreased power generation efficiency or even system failure. Therefore, timely and accurate fault diagnosis of PV arrays is a core element in ensuring the safe and stable operation of power plants, improving power generation efficiency, and reducing operation and maintenance costs.

[0003] However, existing fault diagnosis methods focus on independent analysis of single components such as photovoltaic cells or inverters, lacking effective integration and collaborative utilization of multi-source heterogeneous data, which limits the accuracy of diagnostic results. Summary of the Invention

[0004] The purpose of this invention is to provide a photovoltaic array fault diagnosis system and method based on artificial intelligence, which solves the technical problem that the existing technology focuses on independent analysis of single components such as photovoltaic cells or inverters, lacks effective fusion and collaborative utilization of multi-source heterogeneous data, and thus limits the accuracy of the diagnosis results.

[0005] To achieve the above objectives, the present invention provides an artificial intelligence-based photovoltaic array fault diagnosis method, comprising the following: After acquiring multi-source data from the photovoltaic array and transmitting it in encrypted form, the multi-source data is fused and preprocessed to obtain a preprocessed multimodal dataset. The multimodal dataset is input into the collaborative diagnostic model for analysis, generating preliminary battery anomaly diagnostic results and inverter anomaly diagnostic results. Based on the battery anomaly diagnosis results, combined with the circuit equivalent model and spatial adjacency relationship of the photovoltaic array, the location-related anomaly level of the abnormal photovoltaic cell is calculated; based on the inverter anomaly diagnosis results and the theoretical output power model, the inverter power correction deviation is calculated. Based on the location of abnormal photovoltaic cells, the location-related anomaly level, and the inverter power correction deviation, a physical consistency verification rule based on power flow analysis is introduced to cross-validate and logically fuse the cell anomaly diagnosis results and the inverter anomaly diagnosis results to generate the final comprehensive fault diagnosis results and fault types. Based on historical diagnostic result sequences and comprehensive fault diagnosis results, a time series prediction model is used to predict the future health status of the photovoltaic array and update the array health record.

[0006] The methods for encrypting the transmission of multi-source data are as follows: When the transport layer receives a transport task, the management layer requests the generation of an encryption key and a decryption key, which are then transmitted to the encryption layer and the decryption layer, respectively. The encryption layer uses the encryption key to encrypt the original data to generate ciphertext, which is then transmitted to the preprocessing module via the transport layer. The preprocessing module calls the decryption layer to decrypt the ciphertext using the decryption key to obtain the original data.

[0007] The collaborative diagnostic model consists of a dual AI model collaborative diagnostic architecture composed of a photovoltaic cell IV curve diagnostic model and an inverter IV curve diagnostic model. Furthermore, a meta-heuristic-deep learning hybrid evolutionary optimization strategy guided by array physical topology is adopted to synchronously and automatically optimize the model type selection, model structure, and hyperparameters in the dual AI model collaborative diagnostic architecture, resulting in an optimized collaborative diagnostic model.

[0008] The methods for preprocessing multi-source data fusion are as follows: Real-time acquisition of photovoltaic cell IV characteristic curve data, inverter IV characteristic curve and power data, environmental irradiance and temperature data, as well as photovoltaic cell location and service information; Based on the location information of photovoltaic cells and the relative motion law of the sun, the effective data time period length for feature analysis is adaptively determined; the IV curve data within the selected time period is sampled by a sliding window, and transient distortion points caused by environmental noise are filtered out. For the sampled IV curve data, nonlinear feature dimensionality reduction based on kernel principal component analysis is first performed to extract the dominant feature components. Then, for IV curve segments with missing data or local anomalies, a similarity measurement method based on k-means clustering is used to find the most similar segments from historical normal curves in adjacent locations or similar working conditions for data repair and filling. The processed IV curve feature data, environmental data, location and service information are spatiotemporally aligned and feature-stitched to construct the multimodal dataset.

[0009] Among them, combining the circuit equivalent model of the photovoltaic array with spatial adjacency relationships, the location-related anomaly levels of abnormal photovoltaic cells are calculated, including: Based on the battery anomaly diagnosis results, locate the abnormal photovoltaic cell and its labeled area in the array; Calculate the first deviation: Based on environmental data and photovoltaic cell characteristics, generate the theoretical predicted IV curve of the abnormal cell under the current conditions, and calculate the characteristic deviation between it and the actual IV curve; Calculate the second deviation: Based on the electrical wiring diagram of the photovoltaic array, identify the adjacent cells that are in the same series branch or parallel group as the abnormal cell; calculate the spatial deviation between the IV curve characteristics of the abnormal cell and the average IV curve characteristics of these electrically adjacent cells. The location-related anomaly level is calculated by weighting and fusing the first deviation and the second deviation, and then introducing an aging degradation factor related to battery service time for correction to obtain a quantified location-related anomaly level; wherein, the weight of the second deviation is dynamically adjusted according to the closeness of the electrical adjacency relationship.

[0010] When cross-validating and logically fusing battery anomaly diagnosis results with inverter anomaly diagnosis results: Cross-validation includes: based on the labeling region of the abnormal photovoltaic cell and the location-related abnormality level, using a simplified circuit model of the photovoltaic array, simulating and calculating the theoretical impact range of the abnormality on the current, voltage, and even the DC-side input power of the inverter in the branch, generating a "predictive verification interval" for the inverter state; comparing the key features extracted from the inverter abnormality diagnosis results with the "predictive verification interval", if they fall within the interval, the verification passes; otherwise, they are marked as inconsistent. Logical fusion includes a logical fusion module that combines a rule-based decision tree with an attention-based feature fusion network.

[0011] This invention also provides an artificial intelligence-based photovoltaic array fault diagnosis system for performing the artificial intelligence-based photovoltaic array fault diagnosis method described above. It includes a multi-source data fusion preprocessing module, a dual-model collaborative diagnostic architecture management module, a hybrid evolutionary optimization engine, a collaborative diagnostic execution module, an intelligent analysis and fusion decision-making module, and a health prediction and record management module; The multi-source data fusion preprocessing module is used to acquire and fuse preprocessed multi-source data from the photovoltaic array to obtain a multimodal dataset; the dual-model collaborative diagnostic architecture management module is used to construct, store, and manage a collaborative diagnostic architecture composed of a photovoltaic cell IV curve diagnostic model and an inverter IV curve diagnostic model; the hybrid evolutionary optimization engine is used to execute a metaheuristic-deep learning hybrid evolutionary optimization strategy to automatically optimize the collaborative diagnostic architecture; the collaborative diagnostic execution module is used to call the optimized model to diagnose the input data and generate preliminary diagnostic results; the intelligent analysis and fusion decision module is used to calculate the location-related anomaly level and power correction deviation, and perform cross-validation and logical fusion to output a comprehensive fault diagnosis result; the health prediction and file management module is used to predict future health status and update and manage the digital twin health file of the photovoltaic array.

[0012] This invention discloses an artificial intelligence-based photovoltaic array fault diagnosis system and method. By encrypting and preprocessing multi-source data through transmission and fusion, a multimodal dataset is constructed, solving the problem of information silos and providing a comprehensive and collaborative data foundation for subsequent analysis. The collaborative diagnosis model can analyze the status of batteries and inverters in parallel, avoiding the one-sidedness of single-component diagnosis. Furthermore, by combining circuit equivalent models, spatial adjacency relationships, and theoretical power models for calculation, the diagnostic results are made physically interpretable. At the same time, physical consistency verification rules based on power flow analysis are introduced to cross-validate and logically fuse the preliminary diagnostic results, greatly reducing the false alarm and false negative rates and generating reliable comprehensive fault diagnosis results. This approach solves the technical problem in the prior art that focuses on independent analysis of single components such as photovoltaic cells or inverters, lacking effective fusion and collaborative utilization of multi-source heterogeneous data, which limits the accuracy of diagnostic results. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0014] Figure 1 This is a flowchart of the artificial intelligence-based photovoltaic array fault diagnosis method of the present invention. Detailed Implementation

[0015] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0016] Please see Figure 1 , Figure 1 This is a flowchart of the artificial intelligence-based photovoltaic array fault diagnosis method of the present invention.

[0017] This invention provides an artificial intelligence-based method for diagnosing photovoltaic array faults, comprising the following: S1. Acquire multi-source data from the photovoltaic array, and after encrypted transmission, perform fusion preprocessing on the multi-source data to obtain the preprocessed multimodal dataset. In this specific implementation, the method for encrypting and transmitting multi-source data is as follows: When the transport layer receives a transport task, the management layer requests the generation of an encryption key and a decryption key, which are then transmitted to the encryption layer and the decryption layer, respectively. The encryption layer uses the encryption key to encrypt the original data to generate ciphertext, which is then transmitted to the preprocessing module via the transport layer. The preprocessing module calls the decryption layer to decrypt the ciphertext using the decryption key to obtain the original data.

[0018] The following are methods for preprocessing multi-source data fusion: Real-time acquisition of photovoltaic cell IV characteristic curve data, inverter IV characteristic curve and power data, environmental irradiance and temperature data, as well as photovoltaic cell location and service information; Based on the location information of photovoltaic cells and the relative motion law of the sun, the effective data time period length for feature analysis is adaptively determined; the IV curve data within the selected time period is sampled by a sliding window, and transient distortion points caused by environmental noise are filtered out. For the sampled IV curve data, nonlinear feature dimensionality reduction based on kernel principal component analysis is first performed to extract the dominant feature components. Then, for IV curve segments with missing data or local anomalies, a similarity measurement method based on k-means clustering is used to find the most similar segments from historical normal curves in adjacent locations or similar working conditions for data repair and filling. The processed IV curve feature data, environmental data, location and service information are spatiotemporally aligned and feature-stitched to construct the multimodal dataset.

[0019] S2. Input the multimodal dataset into the collaborative diagnostic model for analysis to generate preliminary battery anomaly diagnostic results and inverter anomaly diagnostic results; In this specific implementation, the collaborative diagnostic model is a dual AI model collaborative diagnostic architecture consisting of a photovoltaic cell IV curve diagnostic model and an inverter IV curve diagnostic model. Furthermore, a meta-heuristic-deep learning hybrid evolutionary optimization strategy guided by array physical topology is adopted to synchronously and automatically optimize the model type selection, model structure, and hyperparameters in the dual AI model collaborative diagnostic architecture, thereby obtaining an optimized collaborative diagnostic model.

[0020] Define a hybrid search space: the search space contains both discrete and continuous variables; wherein, the discrete variables are used to select the base model type from the pre-selected model pool for each sub-diagnostic model in the dual AI model collaborative diagnostic architecture, the pre-selected model pool including at least one-dimensional convolutional neural networks, long short-term memory networks and support vector machines; the continuous variables are used to define the structural parameters and hyperparameters of the selected base model; The design of the physical information-guided initialization and optimization process is as follows: Before optimization, based on the electrical connection topology (series / parallel structure) of the photovoltaic array, prior guidance is provided for the generation of the initial population by the metaheuristic algorithm, so that the initial model combination is more inclined to learn features with circuit correlation. Design a two-stage adaptive optimization process: The first stage is coarse screening and exploration: using an improved metaheuristic optimization algorithm, a large-scale, large-step global search is performed in the hybrid search space; the improvement is that the fitness function of the optimization algorithm simultaneously considers the diagnostic accuracy of a single model on the validation set and the logical consistency score of the preliminary collaborative diagnosis of the two models. The second stage is fine-tuning: After locking in the best combination of model types obtained in the first stage, within the narrowed parameter range, the improved metaheuristic optimization algorithm and the gradient-based deep learning optimizer are alternately iterated. The metaheuristic algorithm is responsible for escaping local optima, and the deep learning optimizer is responsible for fast local convergence, thereby working together to complete the fine-tuning of the model structure and hyperparameters. During the optimization process, a dynamic population management mechanism is introduced to adaptively adjust the population size based on the optimization progress and population diversity, and to eliminate individual models with poor synergistic effects while retaining individuals with high accuracy and high synergistic potential.

[0021] S3. Based on the battery anomaly diagnosis results, combined with the circuit equivalent model and spatial adjacency relationship of the photovoltaic array, calculate the location-related anomaly level of the abnormal photovoltaic cell; based on the inverter anomaly diagnosis results and the theoretical output power model, calculate the inverter power correction deviation. For this specific implementation, combining the circuit equivalent model and spatial adjacency relationship of the photovoltaic array, the location-related anomaly level of abnormal photovoltaic cells is calculated, including: Based on the battery anomaly diagnosis results, locate the abnormal photovoltaic cell and its labeled area in the array; Calculate the first deviation: Based on environmental data and photovoltaic cell characteristics, generate the theoretical predicted IV curve of the abnormal cell under the current conditions, and calculate the characteristic deviation between it and the actual IV curve; Calculate the second deviation: Based on the electrical wiring diagram of the photovoltaic array, identify the adjacent cells that are in the same series branch or parallel group as the abnormal cell; calculate the spatial deviation between the IV curve characteristics of the abnormal cell and the average IV curve characteristics of these electrically adjacent cells. The location-related anomaly level is calculated by weighting and fusing the first deviation and the second deviation, and then introducing an aging degradation factor related to battery service time for correction to obtain a quantified location-related anomaly level; wherein, the weight of the second deviation is dynamically adjusted according to the closeness of the electrical adjacency relationship.

[0022] S4. Based on the location of abnormal photovoltaic cells, the location-related abnormality level, and the inverter power correction deviation, a physical consistency verification rule based on power flow analysis is introduced to cross-validate and logically fuse the battery abnormality diagnosis results and the inverter abnormality diagnosis results to generate the final comprehensive fault diagnosis results and fault types. For this specific implementation, cross-validation includes: based on the labeling area of ​​the abnormal photovoltaic cell and the location-related abnormality level, using a simplified circuit model of the photovoltaic array, simulating and calculating the theoretical impact range of the abnormality on the current, voltage, and even the DC-side input power of the inverter in the branch, generating a "predictive verification interval" for the inverter state; comparing the key features extracted from the inverter abnormality diagnosis results with the "predictive verification interval," if they fall within the interval, the verification passes; otherwise, they are marked as inconsistent. Logical fusion includes: a logical fusion module that combines a rule-based decision tree with an attention-based feature fusion network; If cross-validation passes, the diagnostic results for the battery and inverter are assigned high confidence weights, and feature-weighted fusion is performed directly to output a comprehensive fault diagnosis result.

[0023] If cross-validation passes, the diagnostic results for the battery and inverter are assigned high confidence weights, and feature-weighted fusion is performed directly to output a comprehensive fault diagnosis result.

[0024] If the cross-validation is inconsistent, deep analysis is triggered: the IV curve features of the battery side, the location-related anomaly level, the IV curve features of the inverter side, the power correction deviation, and the environmental data are input into the feature fusion network. The network learns the importance of features from different sources in contradictory situations through the attention mechanism, performs comprehensive reasoning again, and outputs the final diagnostic result. Based on the comprehensive fault diagnosis results, the fault types are mapped to a preset fault type library, which includes at least normal, battery aging, partial shading, battery short circuit / open circuit, inverter MPPT fault, inverter conversion efficiency abnormality, and combined faults. S5. Based on the historical diagnostic result sequence and comprehensive fault diagnosis results, use a time series prediction model to predict the future health status of the photovoltaic array and update the array health record.

[0025] In this specific implementation, the AI-based photovoltaic array fault diagnosis method further includes: monitoring whether the data stream is interrupted, whether the model diagnosis confidence level is consistently low, and whether the frequency of verification rule conflicts is too high; when the system itself experiences an "abnormality," a self-check alarm is issued to ensure the reliability of the diagnostic system.

[0026] In the diagnosis and verification process, external environmental data (such as irradiance, temperature, and humidity data from meteorological stations, or satellite cloud imagery data) are proactively introduced and integrated. These data are used to dynamically correct the theoretical output power model or to distinguish between power changes caused by environmental fluctuations and anomalies caused by actual faults, thereby reducing the false alarm rate.

[0027] A confidence score or uncertainty range is provided for both the initial results output by the collaborative diagnostic model and the final results after physical verification. For example, the variance of a Bayesian neural network or an ensemble model is used to quantify the uncertainty. When the confidence score falls below a preset threshold, a manual review process is triggered or the system is marked as "to be observed".

[0028] In addition to encrypted data transmission, the system adds auditing for data integrity (such as hash verification) and system operation logs. Version management of diagnostic results is also implemented. If subsequent data or human feedback proves that a diagnosis may have been incorrect, the system supports rolling back the health record to a previous state and recording the reason for the correction for model learning.

[0029] This invention also provides an artificial intelligence-based photovoltaic array fault diagnosis system for performing the artificial intelligence-based photovoltaic array fault diagnosis method described above. It includes a multi-source data fusion preprocessing module, a dual-model collaborative diagnostic architecture management module, a hybrid evolutionary optimization engine, a collaborative diagnostic execution module, an intelligent analysis and fusion decision-making module, and a health prediction and record management module; The multi-source data fusion preprocessing module is used to acquire and fuse preprocessed multi-source data from the photovoltaic array to obtain a multimodal dataset; the dual-model collaborative diagnostic architecture management module is used to construct, store, and manage a collaborative diagnostic architecture composed of a photovoltaic cell IV curve diagnostic model and an inverter IV curve diagnostic model; the hybrid evolutionary optimization engine is used to execute a metaheuristic-deep learning hybrid evolutionary optimization strategy to automatically optimize the collaborative diagnostic architecture; the collaborative diagnostic execution module is used to call the optimized model to diagnose the input data and generate preliminary diagnostic results; the intelligent analysis and fusion decision module is used to calculate the location-related anomaly level and power correction deviation, and perform cross-validation and logical fusion to output a comprehensive fault diagnosis result; the health prediction and file management module is used to predict future health status and update and manage the digital twin health file of the photovoltaic array.

[0030] The present invention provides an artificial intelligence-based photovoltaic array fault diagnosis system and method. By encrypting and preprocessing multi-source data, a multimodal dataset is constructed, solving the problem of information silos and providing a comprehensive and collaborative data foundation for subsequent analysis. The collaborative diagnosis model can analyze the status of batteries and inverters in parallel, avoiding the one-sidedness of single-component diagnosis. Furthermore, by combining circuit equivalent models, spatial adjacency relationships, and theoretical power models for calculation, the diagnosis results are physically interpretable. At the same time, the introduction of physical consistency verification rules based on power flow analysis allows for cross-validation and logical fusion of preliminary diagnosis results, greatly reducing false alarms and false negatives and generating reliable comprehensive fault diagnosis results. This approach solves the technical problem in the prior art that focuses on independent analysis of single components such as photovoltaic cells or inverters, lacking effective fusion and collaborative utilization of multi-source heterogeneous data, which limits the accuracy of diagnosis results.

[0031] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A photovoltaic array fault diagnosis method based on artificial intelligence, characterized in that, Including the following: After acquiring multi-source data from the photovoltaic array and transmitting it in encrypted form, the multi-source data is fused and preprocessed to obtain a preprocessed multimodal dataset. The multimodal dataset is input into the collaborative diagnostic model for analysis, generating preliminary battery anomaly diagnostic results and inverter anomaly diagnostic results. Based on the battery anomaly diagnosis results, combined with the circuit equivalent model and spatial adjacency relationship of the photovoltaic array, the location-related anomaly level of the abnormal photovoltaic cell is calculated. Based on the inverter anomaly diagnosis results and the theoretical output power model, the inverter power correction deviation is calculated. Based on the location of abnormal photovoltaic cells, the location-related anomaly level, and the inverter power correction deviation, a physical consistency verification rule based on power flow analysis is introduced to cross-validate and logically fuse the cell anomaly diagnosis results and the inverter anomaly diagnosis results to generate the final comprehensive fault diagnosis results and fault types. Based on historical diagnostic result sequences and comprehensive fault diagnosis results, a time series prediction model is used to predict the future health status of the photovoltaic array and update the array health record.

2. The photovoltaic array fault diagnosis method based on artificial intelligence as described in claim 1, characterized in that, The following are methods for encrypting the transmission of multi-source data: When the transport layer receives a transport task, the management layer requests the generation of an encryption key and a decryption key, which are then transmitted to the encryption layer and the decryption layer, respectively. The encryption layer uses the encryption key to encrypt the original data to generate ciphertext, which is then transmitted to the preprocessing module via the transport layer. The preprocessing module calls the decryption layer to decrypt the ciphertext using the decryption key to obtain the original data.

3. The photovoltaic array fault diagnosis method based on artificial intelligence as described in claim 2, characterized in that, The collaborative diagnostic model is a dual-AI model collaborative diagnostic architecture consisting of a photovoltaic cell IV curve diagnostic model and an inverter IV curve diagnostic model. Furthermore, it adopts a meta-heuristic-deep learning hybrid evolutionary optimization strategy guided by array physical topology to simultaneously and automatically optimize the model type selection, model structure, and hyperparameters in the dual-AI model collaborative diagnostic architecture, thereby obtaining the optimized collaborative diagnostic model.

4. The photovoltaic array fault diagnosis method based on artificial intelligence as described in claim 3, characterized in that, The following are methods for preprocessing multi-source data fusion: Real-time acquisition of photovoltaic cell IV characteristic curve data, inverter IV characteristic curve and power data, environmental irradiance and temperature data, as well as photovoltaic cell location and service information; Based on the location information of photovoltaic cells and the relative motion law of the sun, the effective data time period length for feature analysis is adaptively determined; the IV curve data within the selected time period is sampled by a sliding window, and transient distortion points caused by environmental noise are filtered out. For the sampled IV curve data, nonlinear feature dimensionality reduction based on kernel principal component analysis is first performed to extract the dominant feature components; Subsequently, for IV curve segments with missing data or local anomalies, a similarity measurement method based on k-means clustering was used to find the most similar segments from historical normal curves of adjacent locations or similar working conditions for data repair and filling. The processed IV curve feature data, environmental data, location and service information are spatiotemporally aligned and feature-stitched to construct the multimodal dataset.

5. The photovoltaic array fault diagnosis method based on artificial intelligence as described in claim 4, characterized in that, Combining the circuit equivalent model and spatial adjacency relationship of the photovoltaic array, the location-related anomaly level of abnormal photovoltaic cells is calculated, including: Based on the battery anomaly diagnosis results, locate the abnormal photovoltaic cell and its labeled area in the array; Calculate the first deviation: Based on environmental data and photovoltaic cell characteristics, generate the theoretical predicted IV curve of the abnormal cell under the current conditions, and calculate the characteristic deviation between it and the actual IV curve; Calculate the second deviation: Based on the electrical wiring diagram of the photovoltaic array, identify the adjacent cells that are in the same series branch or parallel group as the abnormal cell; calculate the spatial deviation between the IV curve characteristics of the abnormal cell and the average IV curve characteristics of these electrically adjacent cells. The location-related anomaly level is calculated by weighting and fusing the first deviation and the second deviation, and then introducing an aging degradation factor related to battery service time for correction to obtain a quantified location-related anomaly level; wherein, the weight of the second deviation is dynamically adjusted according to the closeness of the electrical adjacency relationship.

6. The photovoltaic array fault diagnosis method based on artificial intelligence as described in claim 5, characterized in that, When cross-validating and logically fusing the battery anomaly diagnosis results and the inverter anomaly diagnosis results: Cross-validation includes: based on the labeling area of ​​the abnormal photovoltaic cell and the location-related abnormality level, using a simplified circuit model of the photovoltaic array, simulating and calculating the theoretical impact range of the abnormality on the current, voltage, and even the DC-side input power of the inverter in the branch, generating a "predictive verification interval" for the inverter state; comparing the key features extracted from the inverter abnormality diagnosis results with the "predictive verification interval", if they fall within the interval, the verification passes; otherwise, they are marked as inconsistent. Logical fusion includes a logical fusion module that combines a rule-based decision tree with an attention-based feature fusion network.

7. An artificial intelligence-based photovoltaic array fault diagnosis system, used to execute the artificial intelligence-based photovoltaic array fault diagnosis method as described in claim 6, characterized in that, It includes a multi-source data fusion preprocessing module, a dual-model collaborative diagnostic architecture management module, a hybrid evolutionary optimization engine, a collaborative diagnostic execution module, an intelligent analysis and fusion decision-making module, and a health prediction and record management module; The multi-source data fusion preprocessing module is used to acquire and fuse preprocessed multi-source data of photovoltaic arrays to obtain a multimodal dataset; the dual-model collaborative diagnostic architecture management module is used to construct, store and manage a collaborative diagnostic architecture composed of photovoltaic cell IV curve diagnostic model and inverter IV curve diagnostic model; the hybrid evolutionary optimization engine is used to execute metaheuristic-deep learning hybrid evolutionary optimization strategy to automatically optimize the collaborative diagnostic architecture. The collaborative diagnosis execution module is used to call the optimized model to diagnose the input data and generate preliminary diagnosis results; the intelligent analysis and fusion decision module is used to calculate the location-related anomaly level and power correction deviation, and perform cross-validation and logical fusion to output comprehensive fault diagnosis results; the health prediction and record management module is used to predict future health status and update and manage the digital twin health records of the photovoltaic array.

Images