Convenient photovoltaic string configuration method for new energy station system

Abstract

The invention relates to a convenient photovoltaic string configuration method for a new energy station system, in particular to the technical field of intelligent operation and maintenance of a photovoltaic power station, and realizes technical spanning from traditional artificial experience configuration to data-driven automatic optimization by constructing an intelligent photovoltaic string configuration closed-loop system. According to the method, automatic parameter rule extraction, historical data strategy mining, multi-objective optimization generation and online closed-loop verification are integrated, the configuration efficiency and accuracy are remarkably improved, the problem that manual operation is prone to errors is solved, the system can adapt to environment changes and equipment aging, the configuration template is continuously optimized, and the system performance is improved. Therefore, the power generation efficiency, the equipment reliability and the full-life-cycle economic benefit of the new energy station are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for photovoltaic power plants, and more specifically, to a convenient configuration method for photovoltaic strings in new energy power plant systems. Background Technology

[0002] With the rapid iteration and large-scale application of photovoltaic power generation technology, the models of string inverters and their supporting photovoltaic modules used in modern photovoltaic power plants are becoming increasingly diversified. Every year, numerous new models and specifications of photovoltaic string equipment are introduced to the market, differing in electrical parameters, operating characteristics, and communication protocols. During routine operation and maintenance, expansion, or equipment replacement at power plants, maintenance personnel face the challenge of quickly and accurately integrating new string equipment models into existing monitoring systems. Currently, complete equipment parameter information is typically only recorded in unstructured documents such as technical manuals and data sheets provided by manufacturers. Maintenance personnel need to manually consult, understand, and extract key parameters, a process that is not only inefficient but also highly susceptible to human error in parameter entry or misunderstanding, making it difficult to meet the real-time requirements of large-scale power plants for equipment integration.

[0003] Currently, the template library used in photovoltaic power plant monitoring systems to assist in configuring string equipment information relies heavily on manual operation for its creation and updating. Maintenance engineers must manually input various technical parameters, operating thresholds, alarm conditions, and other information into the system one by one, based on paper or electronic technical documents for new equipment, to create new configuration templates. This manual template update model suffers from significant lag, failing to keep pace with equipment updates, resulting in many newly commissioned devices lacking accurate model support in the monitoring system. The fundamental problem lies in the current technology's lack of ability to automatically extract, parse, and format equipment parameters from unstructured data sources, failing to build a self-evolving intelligent knowledge system. This technological deficiency directly leads to a configuration information update cycle in photovoltaic power plant monitoring systems that is far longer than the actual equipment commissioning cycle. This not only affects the integrity of monitoring data for new equipment but may also mask potential operational hazards due to untimely or inaccurate parameter configuration, ultimately posing a potential risk to the overall power generation efficiency and safe operation of the power plant. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a convenient configuration method for photovoltaic strings in new energy power station systems, thereby solving the problems mentioned in the background art.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: a convenient configuration method for photovoltaic strings in a new energy power station system, comprising the following steps: Step S1: The server receives the technical documents of the photovoltaic string equipment, uses natural language processing technology to parse the parameter description text in the technical documents, extracts the equipment parameter rules, and generates a structured parameter rule library containing parameter names, parameter value ranges, and parameter relationships. Step S2: Based on the structured parameter rule base generated in step S1, and combined with historical configuration data and corresponding runtime performance data, the server uses a deep learning model to mine parameter optimization strategies and generate an optimization strategy library containing parameter configuration priorities and parameter optimization directions. Step S3: Based on the optimization strategy library generated in step S2, the server uses a multi-objective optimization algorithm to generate an adaptive configuration template that adapts to different environmental conditions, performs simulation verification on the adaptive configuration template, and outputs the final configuration template that passes the verification. Step S4: The server monitors the actual running data of the deployed configuration template. When the deviation between the actual running data and the simulation data exceeds a preset threshold, the deep learning model in step S2 and the multi-objective optimization algorithm in step S3 are updated using an online learning algorithm to achieve closed-loop optimization of the configuration template.

[0006] In a preferred embodiment, the specific operation of using natural language processing technology to parse the parameter description text in the technical document and extract the device parameter rules in step S1 is as follows: First, optical character recognition (OCR) is performed on the technical documents to convert the document content into machine-readable text data. Then, named entity recognition technology is used to locate parameter name entities and parameter value entities in the text; Finally, relation extraction technology is used to identify the correspondence between parameter name entities and parameter value entities, and complete equipment parameter rules are extracted based on the identified correspondence.

[0007] In a preferred embodiment, step S1, specifically the operation of generating a structured parameter rule base containing parameter names, parameter value ranges, and parameter relationships, is as follows: The extracted device parameter rules are organized and stored according to a preset rule pattern. The rule pattern assigns a specific data type to the parameter name, defines a standard representation format for the parameter value range, and specifies a logical expression form for the parameter relationship. By systematically mapping and storing all parameter rule instances, a machine-readable structured parameter rule library is constructed.

[0008] In a preferred embodiment, the specific operation of using a deep learning model to mine parameter optimization strategies in step S2 is as follows: The server performs multi-source data fusion processing on the structured parameter rule base generated in step S1 and the historical configuration data. The multi-source data fusion processing includes feature alignment processing and time sequence alignment processing to form a standardized training sample set. A deep reinforcement learning algorithm is used to train on a standardized training sample set to generate parameter configuration priority scores and parameter optimization direction vectors; An optimization strategy library is constructed based on parameter configuration priority scoring and parameter optimization direction vector.

[0009] In a preferred embodiment, the specific operation of constructing the optimization strategy library in step S2 is as follows: The parameter configuration priority scores are normalized to obtain standardized priority weights; Cluster analysis is performed on the parameter optimization direction vectors to generate a set of typical optimization patterns; By associating standardized priority weights with a set of typical optimization patterns, an optimization strategy library with weight allocation and pattern recognition functions is formed.

[0010] In a preferred embodiment, step S3, which involves generating an adaptive configuration template using a multi-objective optimization algorithm, specifically involves the following steps: Based on the optimization strategy library generated in step S2, the server establishes a multi-objective optimization model with three optimization objectives: maximizing power generation efficiency, maximizing equipment lifespan, and maximizing voltage stability, and with equipment safety operation boundaries and on-site physical constraints as limiting conditions. By solving this multi-objective optimization model, an adaptive configuration template that can maintain performance balance under various environmental conditions is obtained.

[0011] In a preferred embodiment, the specific operation of simulating and verifying the adaptive configuration template in step S3 is as follows: A photovoltaic system simulator was used to simulate the varying irradiance, ambient temperature, and module temperature conditions throughout the year, and to verify the power generation efficiency, equipment heat loss, and voltage fluctuation rate of the adaptive configuration template under the simulated conditions. When all verification metrics meet the preset standards, the adaptive configuration template is output as the final configuration template.

[0012] In a preferred embodiment, step S4 involves monitoring the actual operational data of the deployed configuration template as follows: The server continuously collects power generation efficiency data, equipment temperature data, and voltage fluctuation data of the photovoltaic strings as actual operating data. The actual operating data is compared with the simulation data corresponding to the simulation verification stage in real time, and the deviation value of each data item is calculated through the deviation calculation model. When the deviation of any data item exceeds its corresponding preset threshold, the model update process is triggered.

[0013] In a preferred embodiment, step S4, which involves updating the deep learning model and the multi-objective optimization algorithm using an online learning algorithm, specifically involves the following steps: An incremental learning algorithm is used to fine-tune the parameters of the deep learning model in step S2 using actual running data as new training samples. At the same time, an adaptive optimization algorithm is adopted to adjust the weight allocation strategy of the multi-objective optimization algorithm in step S3 according to the actual running effect; The updated deep learning model and multi-objective optimization algorithm are used to generate a new round of adaptive configuration templates to achieve closed-loop optimization.

[0014] In a preferred embodiment, step S4 involves the following specific operations to achieve closed-loop optimization of the configuration template: The updated deep learning model and multi-objective optimization algorithm are then reapplied to the configuration template generation process in steps S2 and S3. The regenerated adaptive configuration template is subjected to real-time performance verification. When the verification indicators reach the preset standards, the closed-loop optimization process is completed. For configuration templates that do not meet the standards, a new round of model update and optimization iteration process will be triggered.

[0015] A convenient configuration system for photovoltaic strings in a new energy power plant system includes a structured parameter module, a parameter optimization module, a simulation verification module, and a configuration module. The structured parameter module involves the server receiving technical documents for photovoltaic string equipment, using natural language processing technology to parse the parameter description text in the technical documents, extracting equipment parameter rules, and generating a structured parameter rule library containing parameter names, parameter value ranges, and parameter relationships. The parameter optimization module uses a deep learning model to mine parameter optimization strategies based on the generated structured parameter rule library and combined with historical configuration data and corresponding runtime performance data. This generates an optimization strategy library that includes parameter configuration priorities and parameter optimization directions. The simulation verification module uses a multi-objective optimization algorithm based on the generated optimization strategy library to generate adaptive configuration templates that adapt to different environmental conditions, and performs simulation verification on the adaptive configuration templates, outputting the final configuration template that passes the verification. The configuration module monitors the actual running data of the deployed configuration template. When the deviation between the actual running data and the simulation data exceeds a preset threshold, the deep learning model and multi-objective optimization algorithm are updated using an online learning algorithm to achieve closed-loop optimization of the configuration template.

[0016] The beneficial effects of this invention are as follows: By constructing an intelligent closed-loop configuration system for photovoltaic strings, a technological leap from traditional manual experience-based configuration to data-driven automatic optimization is achieved. This method integrates automatic parameter rule extraction, historical data strategy mining, multi-objective optimization generation, and online closed-loop verification, significantly improving configuration efficiency and accuracy while avoiding the errors inherent in manual operation. The system can adapt to environmental changes and equipment aging, continuously optimizing configuration templates to ensure that photovoltaic strings are in optimal operating condition for extended periods, thereby effectively improving the power generation efficiency, equipment reliability, and overall lifecycle economic benefits of new energy power plants. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] This embodiment provides, for example Figure 1 The present invention discloses a convenient configuration method for photovoltaic strings in a new energy power station system, comprising the following steps: Step S1: The server receives the technical documents of the photovoltaic string equipment, uses natural language processing technology to parse the parameter description text in the technical documents, extracts the equipment parameter rules, and generates a structured parameter rule library containing parameter names, parameter value ranges, and parameter relationships. Step S2: Based on the structured parameter rule base generated in step S1, and combined with historical configuration data and corresponding runtime performance data, the server uses a deep learning model to mine parameter optimization strategies and generate an optimization strategy library containing parameter configuration priorities and parameter optimization directions. Step S3: Based on the optimization strategy library generated in step S2, the server uses a multi-objective optimization algorithm to generate an adaptive configuration template that adapts to different environmental conditions, performs simulation verification on the adaptive configuration template, and outputs the final configuration template that passes the verification. Step S4: The server monitors the actual running data of the deployed configuration template. When the deviation between the actual running data and the simulation data exceeds a preset threshold, the deep learning model in step S2 and the multi-objective optimization algorithm in step S3 are updated using an online learning algorithm to achieve closed-loop optimization of the configuration template.

[0022] In this embodiment, it is specifically necessary to explain the following steps in step S1: parsing the parameter description text in the technical document using natural language processing technology to extract the device parameter rules. First, the technical document (TD) undergoes Optical Character Recognition (OCR) processing to convert the document content into machine-readable text data. Specifically, a deep learning-based OCR engine is used to recognize the text in scanned PDF or image-format technical documents. The accuracy of the text recognition is guaranteed by the following formula:

[0023] This formula is used to calculate the accuracy of optical character recognition (OCR). It compares the recognized characters... With real characters Consistency, using indicator functions Count the number of correctly recognized characters, then divide by the total number of characters. The average accuracy rate is obtained and then used to evaluate the quality of OCR processing, ensuring the reliability of converting technical documents into machine-readable text. High accuracy is the foundation for subsequent natural language processing steps, preventing parameter extraction failures due to OCR errors.

[0024] in, Indicates OCR recognition accuracy. Total number of characters This is an indicator function that takes a value of 1 when the recognized character matches the actual character. This process ensures the quality of the input data for subsequent Natural Language Processing (NLP).

[0025] Then, Named Entity Recognition (NER) technology is used to locate parameter name entities (PNEs) and parameter value entities (PVEs) in the text. Specifically, a sequence labeling model based on BERT-BiLSTM-CRF is used, where the labeling accuracy of Named Entity Recognition (NER) is calculated using the following formula:

[0026] This formula calculates the F1 score for Named Entity Recognition (NER), which is the harmonic mean of precision and recall. Precision represents the proportion of correctly identified entities, while recall represents the proportion of correctly identified real entities. This formula is used to comprehensively evaluate the performance of NER models, balancing precision and recall to avoid focusing on only a single metric. It ensures that the recognition of both parameter names and value entities is accurate and comprehensive.

[0027] in, The F1 score of NER is used to comprehensively evaluate the model performance; For accuracy, the proportion of correctly identified entities out of all identified entities; Recall rate is the proportion of correctly identified entities out of all real entities.

[0028] Finally, relation extraction (RE) technology is used to identify the correspondence between parameter name entities (PNEs) and parameter value entities (PVEs), and complete device parameter rules (DPRs) are extracted based on the identified correspondences. Relation extraction employs a neural network model based on an attention mechanism, and its relation classification confidence calculation formula is as follows:

[0029] This formula calculates the confidence score of relation extraction (RE) using the Sigmoid function. The linearly transformed feature vector is mapped to the interval [0,1]. The feature vector consists of parameter name entities (PNE), parameter value entities (PVE), and context (…). It is composed of features concatenated together and weighted by a weight matrix. ) and bias terms ( A linear transformation is performed. This is then used to evaluate the reliability of the classification of relationships between parameter names and value entities; a higher confidence level indicates a more reliable relationship. This value can be used to filter low-quality relationships and improve the accuracy of parameter rule extraction.

[0030] in, For the Sigmoid function, , These represent the feature vectors of the parameter name entity and the parameter value entity, respectively. This represents the context feature vector.

[0031] In step S1, the specific operations for generating a structured parameter rule base (SPRL) containing parameter names, parameter value ranges, and parameter relationships are as follows: The extracted Device Parameter Rules (DPRs) are organized and stored according to a preset rule schema (RM). The rule schema (RM) assigns a specific data type (DT) to parameter names, defines a standard representation format (RF) for parameter value ranges, and specifies a logical expression form (LE) for parameter relationships. In practice, an ontology-based knowledge representation method is used to construct the rule schema (RM), and its schema integrity is evaluated using the following formula:

[0032] This formula calculates the completeness of the rule pattern (RM), that is, the number of parameter rules that are successfully mapped. Percentage of total parameter rules ( The proportion of the rule pattern is used to evaluate the degree to which the rule pattern covers the parameter rules. Higher completeness indicates a more reasonable rule pattern design that can effectively organize most parameter rules.

[0033] in, Indicators representing the integrity of the rule schema (RM) This indicates the number of parameter rules that were successfully mapped. This indicates the total number of parameter rules.

[0034] By systematically mapping and storing all parameter rule instances, a machine-readable structured parameter rule base (SPRL) is constructed. In practice, a graph database is used to store the relationships between parameter rules, and its storage efficiency is optimized using the following formula:

[0035] in, To optimize rule base efficiency and comprehensively evaluate query performance, there is no fixed value; the higher the value, the better. This represents the number of successful queries. This represents the total number of queries. Average query time, in seconds.

[0036] This formula calculates the efficiency of the Structured Parameter Rule Base (SPRL), taking into account the query success rate. / ) and query time ( The efficiency value is the reciprocal of the rule base's query performance. A higher efficiency value indicates better query performance of the rule base. This is then used in the formula to calculate the efficiency of the Structured Parameter Rule Base (SPRL), taking into account the query success rate. / ) and query time ( The reciprocal of (). A higher efficiency value indicates better query performance of the rule base.

[0037] The structured parameter rule base generated in this step will serve as the input for step S2, providing an accurate parameter constraint basis for optimization strategy mining. The parameter relationships in the rule base will guide the feature selection process in step S2, ensuring the directionality and physical rationality of the optimization.

[0038] In this embodiment, it is specifically necessary to explain the specific operation of using a deep learning model to mine parameter optimization strategies in step S2: The server (SVR) performs multi-source data fusion processing on the Structured Parameter Rule Base (SPRL) generated in step S1 and the Historical Configuration Data (HCD). During this multi-source data fusion process, feature alignment is first performed, using an attention-based feature projection algorithm to map features from different sources to the same vector space.

[0039] This formula is the core computational step of the attention mechanism, used to process the query. ,key Sum The matrix is ​​then weighted by attention to obtain the aligned feature representation. The function ensures that the attention weights sum to 1, and the scaling factor... It is used to prevent the gradient from vanishing due to excessively large dot products; and further used in multi-source data fusion to achieve vector space alignment of features from different sources (such as structured parameter rule bases and historical configuration data), ensuring that features are comparable and providing standardized input for subsequent model training.

[0040] in, This represents the aligned feature matrix. Represents the query matrix. Represents the key matrix. Represents a value matrix, This is the feature dimension. Next, time series alignment is performed, using a dynamic time warping algorithm to address the issue of inconsistent lengths between different time series.

[0041] This formula calculates two time series. and The minimum normalized distance between them is found by finding the optimal path. This formula minimizes the cumulative distance after sequence alignment. It solves the alignment problem caused by inconsistent time series lengths. In time series data fusion, this formula aligns the time axes of historical configuration data and operational performance data, ensuring data consistency across the time dimension and providing a foundation for time series analysis.

[0042] in, Represents a sequence and The minimum normalization cost between them To standardize the path, the above processing forms a standardized training sample set (STS).

[0043] A deep reinforcement learning (DRL) algorithm is used to train the standardized training sample set (STS). Specifically, the proximal policy optimization (PPO) algorithm is used for policy learning, with the objective function being:

[0044] This formula is the objective function of the Proximal Policy Optimization (PPO) algorithm, which is determined by the pruning probability ratio. This limits the step size of policy updates, ensuring training stability. It is then used to optimize policy parameters during deep reinforcement learning training. This is used to generate parameter configuration priority scores and optimize direction vectors, avoiding drastic fluctuations during the training process.

[0045] in, To prune the objective function, This represents the probability ratio between the old and new strategies. For the estimation of the advantage function, For parameter pruning, a parameter configuration priority score (PPS) and a parameter optimization direction vector (PDV) are generated during training.

[0046] An optimization strategy library (OSL) is constructed based on parameter configuration priority score (PPS) and parameter optimization direction vector (PDV). The optimization strategy library is stored in a graph database, where nodes represent configuration parameters and edges represent optimization relationships between parameters.

[0047] In step S2, the specific operations for building the optimization strategy library are as follows: The parameter configuration priority score (PPS) is normalized, and the standardized priority weight (PW) is obtained using the min-max scaling method:

[0048] This formula linearly scales the parameter priority score (PPS) to the [0,1] interval, achieving normalization. Wherein, Hehe These are the minimum and maximum values ​​of PPS, respectively. These are then used to standardize the priority scores output by the model into weights, eliminating the influence of unit weights and facilitating subsequent weight allocation and comparison.

[0049] in, This represents the standardized weight of the i-th parameter. For the original priority score, and These are the minimum and maximum scores, respectively.

[0050] Cluster analysis was performed on the parameter optimization direction vector (PDV), and the K-means++ clustering algorithm was used to generate a set of typical optimization patterns (POS). The clustering objective function is:

[0051] This formula calculates the sum of squared errors within each cluster in K-means clustering, i.e., the sum of squared errors for all data points. To its cluster center The algorithm minimizes the sum of squared distances. To optimize cluster center Cluster allocation Furthermore, the parameter optimization direction vectors are clustered to generate a set of typical optimization patterns, thereby summarizing and simplifying optimization strategies.

[0052] in, For clustering error, For the number of clusters, For the i-th cluster, As the cluster center, To optimize the direction vector.

[0053] Standardized priority weights (PW) are associated with a set of typical optimization patterns (POS), forming an optimization policy library (OSL) that includes weight allocation and pattern recognition functions. The association mapping employs a nearest neighbor search algorithm based on cosine similarity.

[0054] in, Scoring the cosine similarity; This is the priority weight vector; Let be the vector of the j-th optimization mode.

[0055] This formula calculates the cosine of the angle between two vectors (priority weight PW and optimization mode POS_j), with a range of [-1, 1]. A larger value indicates greater similarity in direction. In the association mapping, it measures the degree of matching between the priority weight and the optimization mode, selecting the most similar optimization mode for association.

[0056] The optimization strategy library generated in this step will serve as input for step S3, providing strategy guidance for adaptive template generation. The priority scores and optimization directions in the strategy library will directly affect the decision-making process of the multi-objective optimization algorithm in step S3.

[0057] In this embodiment, the specific operation of generating the adaptive configuration template using a multi-objective optimization algorithm in step S3 is as follows: Based on the Optimization Strategy Library (OSL) generated in step S2, the Server (SVR) establishes a multi-objective optimization model (MOM) with three optimization objectives: maximizing generation efficiency (MEE), maximizing equipment lifetime (MEL), and maximizing voltage stability (MSV). The Model uses the Safe Operating Boundary (DSB) and Field Physical Constraints (FPC) as constraints. The MOM is solved using a decomposition-based multi-objective evolutionary algorithm (MOEA / D), and its objective function is expressed as follows:

[0058] This formula combines three conflicting objective functions (power generation efficiency, equipment lifespan, and voltage stability) into a vector optimization problem. By minimizing this vector function, it seeks the Pareto optimal solution set. This is then used to find the best balance point among multiple conflicting objectives, generating a configuration scheme that maintains balanced performance under different environmental conditions.

[0059] in, This represents a negative value for power generation efficiency (maximizing efficiency is equivalent to minimizing negative efficiency). This represents the negative value of the reciprocal of the device's lifespan (maximizing the lifespan is equivalent to minimizing the negative value of the reciprocal of the lifespan). This represents voltage fluctuation rate (minimizing fluctuation rate). The constraint condition is handled using the penalty function method:

[0060] This formula is used to handle constraints in optimization problems. When a constraint is violated, the penalty function value increases, guiding the search towards the feasible region. It transforms a constrained optimization problem into an unconstrained one, ensuring that the final solution satisfies all equipment safety boundaries and on-site physical constraints.

[0061] in, For inequality constraints, The equality constraints are used. By solving this multi-objective optimization model (MOM), an adaptive configuration template (ACT) that can maintain performance balance under various environmental conditions is obtained.

[0062] In step S3, the specific operation for simulating and verifying the adaptive configuration template is as follows: A photovoltaic system simulator (PSS) was used to simulate the annual variation in irradiance. Ambient temperature and component temperature Conditions were established to verify the power generation efficiency of the Adaptive Configuration Template (ACT) under simulated conditions. Equipment heat loss and voltage fluctuation rate Simulation verification employs a statistical evaluation framework based on the Monte Carlo method:

[0063] This formula is used to calculate the simulation results relative to the required confidence level. Essentially, it's a p-value calculation for a one-sample Z-test. It quantifies the statistical significance of the simulation results, ensuring that the performance evaluation of the configuration template is statistically significant.

[0064] in, The average of the simulation results. To find the required value, The standard deviation of the simulation results. For the sample size, This represents the standard normal distribution function. When all validation metrics meet the preset criteria, i.e.:

[0065] The formula consists of three independent inequalities that set minimum acceptable standards for power generation efficiency, equipment losses, and voltage fluctuation rate. These standards then serve as criteria for determining the suitability of a configuration template; only templates that simultaneously meet all three conditions will be adopted as the final configuration.

[0066] Output the adaptive configuration template (ACT) as the final configuration template (FCT).

[0067] The final configuration template generated in this step will be directly used for the actual deployment in step S4. The parameter settings and operating strategies included in the template will provide precise guidance for on-site configuration, ensuring the optimality and reliability of the configuration results.

[0068] In this embodiment, the specific operation of monitoring the actual running data of the deployed configuration template in step S4 is as follows: The server (SVR) continuously collects power generation efficiency (ED), equipment temperature (TD), and voltage fluctuation (VD) data of the photovoltaic (PV) strings as actual operating data (ARD). In practice, data acquisition employs multi-sensor fusion technology, using sensor arrays installed at key nodes of the PV strings to collect operating parameters in real time. The sampling frequency is set to 1-5 minutes per sampling, depending on system requirements, to ensure the real-time nature and accuracy of the data.

[0069] The actual operating data (ARD) is compared in real time with the simulation data (SD) corresponding to the simulation verification phase, and the deviation value (DV) of each data item is calculated using the deviation calculation model (DCM). The deviation calculation model (DCM) adopts a time-weighted dynamic deviation algorithm.

[0070] This formula is a time-weighted dynamic deviation calculation model used to quantify the difference between actual operating data and simulation data. The formula weights historical data using time-weighted coefficients and employs a relative error calculation method to eliminate the influence of dimensions. It is used to monitor the system's operating status; when the deviation exceeds a threshold, it triggers a model update, ensuring the system can detect performance degradation in a timely manner.

[0071] in, This represents the deviation value of the i-th data item. The size of the time window. For time weighting coefficients, The actual running data at time t, To correspond to the simulation data, To prevent division by zero errors by using extremely small constants.

[0072] When the deviation value (DV) of any data item exceeds its corresponding preset threshold (TH), the Model Update Process (MUP) is triggered. The threshold setting adopts an adaptive mechanism, which is dynamically adjusted based on historical operating data and environmental conditions.

[0073] In step S4, the specific operations for updating the deep learning model and the multi-objective optimization algorithm using the online learning algorithm are as follows: Incremental Learning Algorithm (ILA) is employed, using real-world training data (ARD) as new training samples to fine-tune the parameters of the deep learning model (DLM) in step S2. Specifically, an elastic weight consolidation strategy is used to retain important weight information while updating model parameters, preventing catastrophic forgetting. The update formula is:

[0074] This formula is based on the Elastic Weight Consolidation (EWC) incremental learning algorithm. It adds a regularization term to the gradient descent update to prevent the forgetting of old knowledge when learning on new data. This enables online updates of model parameters, balancing new knowledge learning and old knowledge retention, and avoiding catastrophic forgetting.

[0075] in, The updated model parameters include new knowledge information; The original model parameters serve as the baseline for updates; The learning rate controls the update step size, typically 0.001-0.01. The loss function measures the model's prediction error. These are the diagonal elements of the Fisher information matrix, representing the importance of the parameters; This is the regularization coefficient, used to balance the weights of new and old knowledge.

[0076] Simultaneously, an adaptive optimization algorithm (AOA) is employed to adjust the weight allocation strategy (WDS) of the multi-objective optimization algorithm (MOOA) in step S3 based on actual operational results. Through multi-objective game analysis, the weight ratios of the three objectives—power generation efficiency, equipment lifespan, and voltage stability—are dynamically adjusted to better align the optimization direction with actual operational needs.

[0077] The updated deep learning model (DLM) and multi-objective optimization algorithm (MOOA) are used to generate a new round of adaptive configuration templates (ACT), achieving closed-loop optimization (CLO). Model updates are managed through a version control mechanism to ensure a smooth system transition.

[0078] In step S4, the specific operations for implementing closed-loop optimization of the configuration template are as follows: The updated deep learning model (DLM) and multi-objective optimization algorithm (MOOA) are then reapplied to the configuration template generation process in steps S2 and S3. In practice, a model version management system is established to ensure the traceability and rollback capability of the update process.

[0079] Real-time performance (PV) verification is performed on the regenerated adaptive configuration template (ACT). Verification metrics include power generation efficiency verification (EEV), equipment safety verification (ESV), and system stability verification (SSV). A statistical test method based on confidence intervals is employed.

[0080] in, Confidence level indicates the reliability of the result. This represents the average of actual performance indicators; To meet the required performance index values; The standard deviation of the performance index; The number of samples; The standard normal distribution function is used; this formula, based on the confidence level calculation formula of the one-sample Z-test, is used to assess the statistical significance difference between actual performance and required performance. It is then used to quantify the reliability of the validation results, providing a basis for closed-loop optimization decisions.

[0081] When the verification indicators reach the preset standards (PS), the closed-loop optimization process (CLOP) is completed. The preset standards are set according to actual operating requirements and industry standards, such as an increase in power generation efficiency of no less than 3% and an extension of equipment life of no less than 5%.

[0082] For configuration templates that do not meet the standards, a new round of model update and optimization iteration (OIP) is triggered. The iteration termination conditions include: performance meeting the standards, reaching the maximum number of iterations, or performance improvement stabilizing.

[0083] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0087] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A convenient configuration method for photovoltaic strings in a new energy power station system, characterized in that, Includes the following steps: Step S1: The server receives the technical documents of the photovoltaic string equipment, uses natural language processing technology to parse the parameter description text in the technical documents, extracts the equipment parameter rules, and generates a structured parameter rule library containing parameter names, parameter value ranges, and parameter relationships. Step S2: Based on the structured parameter rule base generated in step S1, and combined with historical configuration data and corresponding runtime performance data, the server uses a deep learning model to mine parameter optimization strategies and generate an optimization strategy library containing parameter configuration priorities and parameter optimization directions. Step S3: Based on the optimization strategy library generated in step S2, the server uses a multi-objective optimization algorithm to generate an adaptive configuration template that adapts to different environmental conditions, performs simulation verification on the adaptive configuration template, and outputs the final configuration template that passes the verification. Step S4: The server monitors the actual running data of the deployed configuration template. When the deviation between the actual running data and the simulation data exceeds a preset threshold, the deep learning model in step S2 and the multi-objective optimization algorithm in step S3 are updated using an online learning algorithm to achieve closed-loop optimization of the configuration template.

2. The convenient configuration method for photovoltaic strings in a new energy power station system according to claim 1, characterized in that: In step S1, the specific operation of using natural language processing technology to parse the parameter description text in the technical document and extract the device parameter rules is as follows: First, optical character recognition (OCR) is performed on the technical documents to convert the document content into machine-readable text data. Then, named entity recognition technology is used to locate parameter name entities and parameter value entities in the text; Finally, relation extraction technology is used to identify the correspondence between parameter name entities and parameter value entities, and complete equipment parameter rules are extracted based on the identified correspondence.

3. The convenient configuration method for photovoltaic strings in a new energy power station system according to claim 2, characterized in that: In step S1, the specific operations for generating a structured parameter rule base containing parameter names, parameter value ranges, and parameter relationships are as follows: The extracted device parameter rules are organized and stored according to a preset rule pattern. The rule pattern assigns a specific data type to the parameter name, defines a standard representation format for the parameter value range, and specifies a logical expression form for the parameter relationship. By systematically mapping and storing all parameter rule instances, a machine-readable structured parameter rule library is constructed.

4. The convenient configuration method for photovoltaic strings in a new energy power station system according to claim 3, characterized in that: In step S2, the specific operation of using a deep learning model to mine parameter optimization strategies is as follows: The server performs multi-source data fusion processing on the structured parameter rule base generated in step S1 and the historical configuration data. The multi-source data fusion processing includes feature alignment processing and time sequence alignment processing to form a standardized training sample set. A deep reinforcement learning algorithm is used to train on a standardized training sample set to generate parameter configuration priority scores and parameter optimization direction vectors; An optimization strategy library is constructed based on parameter configuration priority scoring and parameter optimization direction vector.

5. A convenient configuration method for photovoltaic strings in a new energy power station system according to claim 4, characterized in that: In step S2, the specific operations for building the optimization strategy library are as follows: The parameter configuration priority scores are normalized to obtain standardized priority weights; Cluster analysis is performed on the parameter optimization direction vectors to generate a set of typical optimization patterns; By associating standardized priority weights with a set of typical optimization patterns, an optimization strategy library with weight allocation and pattern recognition functions is formed.

6. A convenient configuration method for photovoltaic strings in a new energy power station system according to claim 5, characterized in that: In step S3, the specific operation of generating the adaptive configuration template using a multi-objective optimization algorithm is as follows: Based on the optimization strategy library generated in step S2, the server establishes a multi-objective optimization model with three optimization objectives: maximizing power generation efficiency, maximizing equipment lifespan, and maximizing voltage stability, and with equipment safety operation boundaries and on-site physical constraints as limiting conditions. By solving this multi-objective optimization model, an adaptive configuration template that can maintain performance balance under various environmental conditions is obtained.

7. A convenient configuration method for photovoltaic strings in a new energy power station system according to claim 6, characterized in that: In step S3, the specific operation for simulating and verifying the adaptive configuration template is as follows: A photovoltaic system simulator was used to simulate the varying irradiance, ambient temperature, and module temperature conditions throughout the year, and to verify the power generation efficiency, equipment heat loss, and voltage fluctuation rate of the adaptive configuration template under the simulated conditions. When all verification metrics meet the preset standards, the adaptive configuration template is output as the final configuration template.

8. A convenient configuration method for photovoltaic strings in a new energy power station system according to claim 7, characterized in that: In step S4, the specific operation for monitoring the actual running data of the deployed configuration template is as follows: The server continuously collects power generation efficiency data, equipment temperature data, and voltage fluctuation data of the photovoltaic strings as actual operating data. The actual operating data is compared with the simulation data corresponding to the simulation verification stage in real time, and the deviation value of each data item is calculated through the deviation calculation model. When the deviation of any data item exceeds its corresponding preset threshold, the model update process is triggered.

9. A convenient configuration method for photovoltaic strings in a new energy power station system according to claim 8, characterized in that: In step S4, the specific operations for updating the deep learning model and the multi-objective optimization algorithm using the online learning algorithm are as follows: An incremental learning algorithm is used to fine-tune the parameters of the deep learning model in step S2 using actual running data as new training samples. At the same time, an adaptive optimization algorithm is adopted to adjust the weight allocation strategy of the multi-objective optimization algorithm in step S3 according to the actual running effect; The updated deep learning model and multi-objective optimization algorithm are used to generate a new round of adaptive configuration templates to achieve closed-loop optimization.

10. A convenient configuration system for photovoltaic strings in a new energy power station system using the method of claim 1, characterized in that, It includes a structured parameter module, a parameter optimization module, a simulation verification module, and a configuration module: The structured parameter module involves the server receiving technical documents for photovoltaic string equipment, using natural language processing technology to parse the parameter description text in the technical documents, extracting equipment parameter rules, and generating a structured parameter rule library containing parameter names, parameter value ranges, and parameter relationships. The parameter optimization module uses a deep learning model to mine parameter optimization strategies based on the generated structured parameter rule library and combined with historical configuration data and corresponding runtime performance data. This generates an optimization strategy library that includes parameter configuration priorities and parameter optimization directions. The simulation verification module uses a multi-objective optimization algorithm based on the generated optimization strategy library to generate adaptive configuration templates that adapt to different environmental conditions, and performs simulation verification on the adaptive configuration templates, outputting the final configuration template that passes the verification. The configuration module monitors the actual running data of the deployed configuration template. When the deviation between the actual running data and the simulation data exceeds a preset threshold, the deep learning model and multi-objective optimization algorithm are updated using an online learning algorithm to achieve closed-loop optimization of the configuration template.

Images