A method for implementing life cycle management of large model photovoltaic energy management

Abstract

This invention belongs to the field of artificial intelligence and communication technology, and discloses a lifecycle management implementation method for large-scale photovoltaic energy management. It focuses on three new technological perspectives: intelligent perception of photovoltaic full-link data, collaborative optimization of power generation, grid, load, and storage, and intelligent fault diagnosis and self-healing. This covers the entire process of photovoltaic energy management, from data acquisition and perception, collaborative scheduling to fault diagnosis and maintenance. Through core algorithms such as spatiotemporal fusion and accuracy compensation of photovoltaic multi-source data, multi-objective collaborative optimization scheduling of power generation, grid, load, and storage, and fault feature extraction and self-healing decision-making for photovoltaic equipment, it overcomes the industry bottlenecks of traditional photovoltaic energy management, including "inaccurate data perception, lack of global optimization in collaborative scheduling, and passive and delayed fault handling." This achieves high-precision perception of photovoltaic data across all dimensions, global collaborative optimization of energy scheduling, and proactive diagnosis and self-healing of equipment faults, significantly improving photovoltaic energy consumption efficiency, power plant operational stability, and the level of full lifecycle maintenance management.

Description

Technical Field

[0001] This invention belongs to the field of information technology, specifically relating to a lifecycle management implementation method for large-scale photovoltaic energy management. Background Technology

[0002] In the current field of automated large-scale models for photovoltaic energy management, existing technologies have achieved basic intelligence in "basic data acquisition, simple scheduling execution, and manual-assisted fault diagnosis." However, in practical engineering applications, for the three core sub-scenarios of photovoltaic multi-source data fusion sensing, source-grid-load-storage coordinated scheduling, and photovoltaic equipment fault diagnosis and self-healing, there are still three specific and unresolved technical problems. These are all scenario-based practical problems, not macro-level industry challenges, as follows: The fusion and sensing of multi-source photovoltaic data lacks precision, and there is no dedicated algorithm for spatiotemporal fusion and accuracy compensation. Photovoltaic data sources cover four major ends: power plants (PV panels, inverters, combiner boxes), meteorology (irradiance, temperature, wind speed), power grid (load, voltage, frequency), and energy storage (SOC, charging and discharging power). Significant heterogeneity exists (heterogeneous format: structured, semi-structured; spatiotemporal heterogeneity: different acquisition frequencies, inconsistent spatiotemporal benchmarks; heterogeneous accuracy: sensor-level precise data, prediction-level approximate data). Existing technologies only adopt a crude fusion mode of "simple splicing + deduplication," lacking targeted spatiotemporal fusion and accuracy compensation algorithms. This fails to address issues such as data spatiotemporal offset, accuracy deviation, and feature weakening, resulting in fused data with problems such as "spatiotemporal misalignment, insufficient accuracy, and missing effective features." The data sensing accuracy cannot meet the refined management needs of real-time photovoltaic scheduling and accurate fault diagnosis, and the effective data utilization rate is less than 60%.

[0003] The current generation-grid-load-storage coordinated dispatch lacks global optimization and multi-objective coordination and dynamic adjustment algorithms. Photovoltaic energy dispatch requires comprehensive coordination among power plants, the grid, loads, and energy storage. Dispatch objectives involve multiple dimensions, including energy absorption, grid stability, energy storage utilization, and operation and maintenance costs. Furthermore, photovoltaic output is highly random due to weather conditions, grid load is dynamic, and energy storage capacity is constrained. Existing technologies employ a "single-objective fixed threshold dispatch" model, lacking multi-objective coordinated optimization algorithms. Focusing solely on the photovoltaic absorption rate, this easily leads to grid voltage fluctuations, overcharging and discharging of energy storage, and increased operation and maintenance costs. Simultaneously, the lack of accurate load forecasting and dynamic adjustment algorithms prevents incremental optimization of the dispatch scheme based on real-time operational data, resulting in problems such as "dispatch lag, global imbalance, and resource waste," with a generation-grid-load-storage coordination efficiency below 70%.

[0004] Photovoltaic equipment fault handling is passive and delayed, lacking implicit feature extraction and self-healing decision-making algorithms: Photovoltaic equipment (photovoltaic panels, inverters, combiner boxes, transformers) suffers from diverse fault types and complex fault characteristics. Some faults exhibit "implicit features and gradual development," and rapid response is required after a fault occurs to minimize power generation loss. Current technologies rely on a passive fault handling mode of "manual inspection + threshold alarm," lacking implicit feature extraction algorithms. They can only identify obvious threshold-type faults and cannot detect gradual or implicit faults. Furthermore, the lack of fault classification and self-healing decision-making algorithms necessitates manual on-site handling after each fault occurs, resulting in problems such as "late fault detection, low diagnostic accuracy, untimely handling, and lack of self-healing capabilities." The average fault handling time exceeds 2 hours, and the power generation loss rate of photovoltaic power plants due to faults exceeds 8%.

[0005] Existing photovoltaic (PV) energy management methods lack core algorithmic innovation to address the three specific problems mentioned above. Significant technological gaps exist, particularly in the spatiotemporal fusion and accuracy compensation of multi-source data, multi-objective collaborative optimization of source-grid-load-storage systems, and modeling and solving for hidden fault features and self-healing decisions. These shortcomings prevent the realization of end-to-end automation and intelligence in PV energy management. There is an urgent need for a method to achieve automated large-scale PV energy management models based on algorithmic innovation, focusing on three entirely new perspectives: accurate data perception, global collaborative scheduling, and proactive fault self-healing. This approach aims to overcome existing technological bottlenecks and fill the technological gaps in end-to-end intelligence for PV energy management. Summary of the Invention

[0006] The purpose of this invention is to provide a lifecycle management implementation method for large-scale photovoltaic energy management, which solves the problems of inaccurate perception of photovoltaic multi-source data fusion, lack of global optimization in source-grid-load-storage collaborative scheduling, and passive and lagging handling of photovoltaic equipment faults.

[0007] The specific technical solution of the present invention is as follows: (I) Photovoltaic Multi-Source Heterogeneous Data Fusion Sensing Module This module addresses the issue of inaccurate perception of multi-source photovoltaic data fusion. It enables spatiotemporal calibration, accuracy compensation, feature enhancement, and redundancy filtering of multi-source heterogeneous data, constructs an accurate fusion perception model for multi-source heterogeneous data, and outputs fusion perception data with full dimensions, high precision, and high utilization, providing a data foundation for subsequent collaborative scheduling and fault diagnosis.

[0008] Modeling Approach: Abandoning the traditional crude fusion modeling approach of "simple splicing + deduplication", we construct an integrated modeling logic of "feature extraction - spatiotemporal calibration - accuracy compensation - feature enhancement - redundancy filtering". Combining the heterogeneous characteristics (spatiotemporal, format, and accuracy) of photovoltaic multi-source data, we establish a multi-source data spatiotemporal fusion model, accuracy compensation model, feature enhancement model, and redundancy filtering model, and design corresponding four core sub-algorithms to achieve accurate data fusion and efficient perception, breaking through the technical bottlenecks of data spatiotemporal misalignment, insufficient accuracy, and missing features.

[0009] Solution Process: First, deploy multi-source data acquisition terminals to aggregate heterogeneous data from four major sources: power plants, meteorology, power grids, and energy storage, constructing a photovoltaic multi-source sensing data resource pool. Then, extract the core spatiotemporal features (collection timestamps, monitoring point coordinates) and attribute features (output, irradiance, load, SOC, etc.) of the data, identifying the spatiotemporal offset type, accuracy deviation level, and feature weakening dimension. Convert data from different spatiotemporal references into a unified photovoltaic power plant standard spatiotemporal system (using the power plant control room as the time reference and the power plant's local coordinate system as the spatial reference). Calculate the comprehensive fusion accuracy using a fusion accuracy calculation formula, and employ an iterative correction strategy to correct spatiotemporal offsets and accuracy deviations until the fusion accuracy reaches the scenario-defined threshold. Establish a heterogeneous data feature association mapping relationship, and use a multi-dimensional feature fusion strategy to enhance core features such as faults and output, improving the effective information density of the data. Based on a data correlation and redundancy calculation model, identify and eliminate duplicate and irrelevant redundant data to improve data utilization. Finally, construct a fusion sensing effect verification model to quantify fusion accuracy, feature recognition, and data utilization, dynamically optimizing algorithm parameters to ensure that the data meets the needs of refined photovoltaic management.

[0010] 1: Spatiotemporal fusion and accuracy compensation algorithm for photovoltaic multi-source data Modeling approach: To address the problems of "spatiotemporal misalignment and insufficient accuracy" in existing technologies, an integrated modeling logic of "spatiotemporal feature extraction - spatiotemporal calibration modeling - accuracy compensation modeling - fusion accuracy quantification - iterative correction" is constructed. A multi-source data spatiotemporal fusion model and an accuracy compensation model are established, and spatiotemporal calibration and accuracy compensation algorithms are designed to achieve accurate spatiotemporal fusion and accuracy compensation of data, thereby solving the problems of spatiotemporal offset and accuracy deviation.

[0011] Solution Process: Core spatiotemporal features such as timestamps and monitoring point coordinates from multi-source heterogeneous data are extracted to establish a spatiotemporal feature point set. A spatiotemporal coordinate transformation and timestamp synchronization algorithm is used to unify all data into the standard spatiotemporal system of a photovoltaic power station, completing preliminary spatiotemporal calibration. For the accuracy deviations of data in different monitoring dimensions, an accuracy compensation coefficient model is established, and preliminary compensation for deviation data is performed by combining equipment performance parameters and environmental impact coefficients. The accuracy of each dimension and the overall fusion accuracy are calculated using the fusion accuracy calculation formula to determine whether a set threshold has been reached. For data that does not reach the threshold, a "feature point matching-deviation tracing-iterative correction" strategy is adopted to trace the causes of spatiotemporal offset and accuracy deviation, dynamically adjusting the compensation coefficients until the overall fusion accuracy meets the scenario requirements. An accuracy verification model is constructed, and standard real values ​​are used to verify the fused data to ensure the effectiveness of accuracy compensation.

[0012] 2: Heterogeneous Data Feature Enhancement and Redundancy Filtering Algorithms Modeling approach: To address the issues of "weakened data features and high redundancy" in existing technologies, we construct an integrated modeling logic of "feature association modeling - multi-dimensional feature fusion - redundancy measurement - adaptive filtering", establish a heterogeneous data feature enhancement model and a redundancy filtering model, and design feature enhancement and redundancy filtering algorithms to improve the feature recognition and effective utilization of data.

[0013] Solution Process: The process involves analyzing the feature relationships of heterogeneous data from different sources and establishing feature mapping relationships such as "output-irradiance-temperature" and "SOC-charging / discharging power-grid load." A multi-dimensional strategy combining weighted fusion and deep learning feature extraction is employed to enhance core features such as photovoltaic output, equipment operation, and fault precursors, improving feature identification and exploitability. A data correlation and redundancy measurement model is constructed to calculate the correlation coefficient and redundancy between data, identifying duplicate, irrelevant, and low-value data. Adaptive filtering strategies are used to eliminate redundant data and retain core feature data, while ensuring data integrity, based on the needs of photovoltaic energy management scenarios (scheduling, diagnosis, and operation and maintenance). Feature identification and data utilization efficiency are quantified, and feature fusion weights and redundancy filtering thresholds are dynamically optimized to achieve data "quality and efficiency improvement."

[0014] (II) Photovoltaic Energy Coordinated Dispatch Optimization Module This module addresses the lack of global optimization in the coordinated scheduling of power generation, grid, load, and energy storage. It enables multi-objective coordination, accurate load forecasting, dynamic adjustment, and global coordination in photovoltaic energy scheduling. It constructs a multi-objective coordinated optimization scheduling model for power generation, grid, load, and energy storage, and outputs the globally optimal dynamic scheduling scheme, thereby improving the coordinated efficiency of power generation, grid, load, and energy storage and the photovoltaic energy absorption rate.

[0015] Modeling Approach: Abandoning the traditional extensive modeling approach of "single-objective fixed threshold scheduling", we construct an integrated modeling logic of "scheduling objective modeling - source-grid-load-storage constraint modeling - multi-objective optimization solution - accurate load prediction - dynamic incremental adjustment". Combining the dynamic characteristics of photovoltaic output, grid load and energy storage status, we establish a multi-objective collaborative optimization scheduling model, an accurate load prediction model and a dynamic adjustment model, and design three corresponding core sub-algorithms to achieve global collaborative scheduling of source-grid-load-storage, breaking through the technical bottlenecks of scheduling lag and global imbalance.

[0016] Solution Process: First, the multiple objectives of photovoltaic energy dispatch (energy absorption rate, grid stability, energy storage utilization rate, and operation and maintenance cost) are identified, and a multi-objective dispatch objective function is constructed, clarifying the weight and optimization direction of each objective. Then, an operational constraint model for each end of the power generation, grid, load, and storage system is established (PV output ceiling, grid voltage / frequency constraints, energy storage SOC range, and load supply and demand balance). A multi-objective intelligent optimization algorithm is used to solve the objective function, obtaining the globally optimal initial dispatch scheme. By integrating meteorological forecast data, historical PV operation data, and grid load data, a multi-dimensional load forecasting model is constructed to achieve accurate short-term forecasts of PV output and grid load. Real-time operational data from each end of the power generation, grid, load, and storage system are collected, and the deviation between predicted and actual values ​​is calculated. An incremental adjustment strategy is used to dynamically optimize the initial dispatch scheme, achieving synchronization between the dispatch scheme and real-time operational data. Finally, a dispatch effect verification model is constructed, quantifying indicators such as energy absorption rate, grid stability, and energy storage utilization rate. Algorithm parameters and dispatch objective weights are dynamically optimized to ensure the global optimality and dynamic adaptability of the dispatch scheme.

[0017] 3: Multi-objective collaborative optimization scheduling algorithm for source-grid-load-storage Modeling approach: To address the problem of "single-objective scheduling and global imbalance" in existing technologies, we construct an integrated modeling logic of "multi-objective function modeling - operational constraint modeling - multi-objective intelligent optimization - optimal solution finding", establish a multi-objective collaborative optimization scheduling model of source, grid, load and storage, and design a multi-objective collaborative optimization algorithm to achieve global balance of scheduling multiple objectives and efficient collaboration of source, grid, load and storage.

[0018] Solution Process: The four core scheduling objectives are photovoltaic energy absorption rate, grid voltage / frequency stability, energy storage utilization rate, and power plant operation and maintenance cost. A multi-objective optimization objective function is constructed, and the weight coefficients of each objective are dynamically adjusted according to the needs of the photovoltaic management scenario. Hard and soft constraint models are established for each end of the power generation, grid, load, and storage system. Hard constraints include maximum photovoltaic output, upper and lower limits of energy storage SOC, and grid voltage / frequency range. Soft constraints include energy storage charging and discharging efficiency and operation and maintenance cost threshold. An improved multi-objective particle swarm optimization algorithm is used to solve the objective function, obtaining the optimal solution in the non-dominated solution set. The optimal solution is modified in an engineering manner based on grid scheduling requirements and actual power plant operation and maintenance, outputting a globally optimal scheduling scheme that can be implemented. Based on the feedback of scheduling effects, the weights of the objective function and the optimization algorithm parameters are dynamically optimized to achieve continuous and coordinated balance among the multiple objectives.

[0019] 4: Photovoltaic Dispatch Load Forecasting and Dynamic Adjustment Algorithm Modeling approach: To address the problems of inaccurate load forecasting and lagging scheduling schemes in existing technologies, an integrated modeling logic of "multi-source data fusion, multi-dimensional prediction modeling, deviation calculation, and incremental dynamic adjustment" is constructed. A multi-dimensional prediction model of photovoltaic power output and grid load and a dynamic adjustment model of scheduling scheme are established. Prediction and dynamic adjustment algorithms are designed to achieve accurate load forecasting and real-time optimization of scheduling schemes.

[0020] Solution Process: A multi-dimensional prediction dataset is constructed by integrating short-term meteorological forecast data (irradiance, temperature, wind speed), historical photovoltaic power plant operation data (output, equipment status), historical grid load data, and real-time operation data. A deep learning algorithm using LSTM and attention mechanisms is employed to build a short-term prediction model for photovoltaic output and grid load, improving prediction accuracy. Real-time data collection of actual photovoltaic output and grid load is used to calculate the deviation between predicted and actual values, identifying deviation types (meteorological changes, equipment fluctuations, grid load changes). For different types of deviations, an incremental adjustment strategy is adopted, optimizing only the deviation-related parts of the scheduling scheme without resolving the entire scheduling scheme, enabling rapid dynamic adjustment of the scheduling scheme. The prediction accuracy and scheduling scheme adaptability are quantified, and the prediction model parameters and incremental adjustment thresholds are dynamically optimized to ensure the real-time performance and effectiveness of the scheduling scheme.

[0021] (III) Photovoltaic Equipment Fault Diagnosis and Self-Healing Module This module addresses the passive and delayed nature of photovoltaic equipment fault handling by extracting latent features of photovoltaic equipment faults, enabling precise diagnosis, tiered handling, and self-healing decision-making. It constructs a self-healing model for photovoltaic equipment fault diagnosis, upgrading from "passive alarm" to "active diagnosis and self-healing handling," thereby reducing power generation losses due to faults.

[0022] Modeling Approach: Abandoning the traditional passive modeling approach of "manual inspection + threshold alarm", we construct an integrated modeling logic of "fault feature library construction - latent feature extraction - accurate fault diagnosis - hierarchical handling modeling - self-healing strategy matching - feedback optimization". Combining the diversity and hidden characteristics of photovoltaic equipment faults, we establish a fault feature library, fault diagnosis model, hierarchical handling model, and self-healing strategy library, and design three corresponding core sub-algorithms to achieve proactive diagnosis and self-healing of equipment faults, breaking through the technical bottlenecks of late fault detection, low diagnostic accuracy, and lack of self-healing capability.

[0023] Solution Process: First, a comprehensive fault database is compiled for all types of core equipment such as photovoltaic panels, inverters, combiner boxes, and transformers, including both overt and covert faults. Core features (voltage, current, temperature, output, etc.) of each fault type are extracted to construct a comprehensive fault feature database for photovoltaic equipment. Deep learning algorithms are then used to mine features from equipment operating data, extracting gradual and precursory features of covert faults to improve the early detection of faults. Combining the fault feature database and fault rule database, accurate diagnosis and quantitative assessment of fault type, location, and severity are achieved. Based on the fault's impact range, power generation loss, and urgency, faults are classified into four levels, and differentiated graded handling strategies are formulated. For self-healing faults (such as minor voltage fluctuations and deviations in energy storage charging and discharging parameters), a self-healing strategy database is constructed, enabling intelligent matching and execution of self-healing strategies based on fault scenarios. Feedback information on fault diagnosis and self-healing effects is collected to dynamically optimize the fault feature database, diagnostic algorithms, and self-healing strategies, achieving continuous improvement in fault diagnosis and self-healing capabilities.

[0024] 5: Algorithm for Fault Feature Extraction and Precise Diagnosis of Photovoltaic Equipment Modeling approach: To address the issues of existing technologies being unable to extract latent fault features and having low diagnostic accuracy, an integrated modeling logic is constructed, which includes "fault feature library construction - multi-dimensional data acquisition - latent feature extraction - accurate fault matching - quantitative diagnosis". A fault feature library and accurate fault diagnosis model for photovoltaic equipment are established, and fault feature extraction and diagnosis algorithms are designed to achieve effective extraction of latent fault features and accurate fault diagnosis.

[0025] Solution Process: A feature library covering all types of faults in core photovoltaic equipment is constructed, including feature parameters, feature variation patterns, and fault development trends of explicit faults (such as inverter shutdown and photovoltaic panel short circuits) and implicit faults (such as micro-cracks in photovoltaic panels and poor contact in combiner boxes). Real-time collection of multi-dimensional operational data such as voltage, current, temperature, output, and vibration of the equipment is used to construct a fault diagnosis dataset. A deep learning algorithm combining CNN and BiLSTM is employed to perform feature mining on the operational data, extracting gradual and precursor features of implicit faults to achieve early fault detection. The extracted fault features are accurately matched with the fault feature library, combined with a fault rule library, to achieve accurate identification of fault type and location. A fault severity quantification assessment model is constructed, quantifying and classifying fault severity (mild, moderate, severe) based on feature variation amplitude and fault development time. The accuracy and early detection rate of fault diagnosis are quantified, and the feature extraction algorithm and fault matching rules are dynamically optimized to improve diagnostic accuracy.

[0026] 6: Fault Classification and Self-Healing Decision Optimization Algorithm Modeling Approach: To address the problem of existing technologies lacking "fault handling grading and self-healing capabilities," an integrated modeling logic is constructed, consisting of "fault grading modeling - differentiated handling strategy formulation - self-healing strategy library construction - intelligent strategy matching - feedback optimization." A fault grading handling model and a self-healing decision optimization model are established, and grading handling and self-healing decision algorithms are designed to achieve scientific fault grading and efficient self-healing.

[0027] Solution Process: Based on the impact of faults on photovoltaic power generation, the urgency of handling, and the difficulty of fault repair, photovoltaic equipment faults are classified into four levels: fatal faults (such as transformer short circuits), serious faults (such as inverter faults), general faults (such as dust accumulation on photovoltaic panels), and minor faults (such as small voltage fluctuations). Differentiated graded handling strategies are formulated for different fault levels: fatal faults result in immediate shutdown and trigger an emergency alarm; serious faults involve reduced power operation and trigger manual handling; general faults are handled through planned inspections; and minor faults automatically trigger self-healing strategies. A self-healing strategy library for photovoltaic equipment faults is constructed, covering self-healing strategies such as voltage adjustment, power optimization, parameter correction, and minor fault reset, clarifying the applicable scenarios and execution steps for each strategy. Intelligent matching and automatic execution of self-healing strategies are achieved by combining fault type, fault severity, and equipment operating status. A self-healing effect verification model is built to quantify the self-healing success rate, fault handling time, and power generation loss rate. Feedback from operation and maintenance personnel is collected to dynamically optimize the fault classification standards, handling strategies, and self-healing strategy library, achieving continuous optimization of self-healing decisions.

[0028] Beneficial effects Photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm: Construct an integrated model of spatiotemporal fusion and accuracy compensation. Through quantitative fusion accuracy and iterative correction, achieve accurate spatiotemporal calibration and accuracy compensation of multi-source data. Compared with the traditional fusion mode, the data fusion accuracy is improved by more than 30%, completely solving the problems of spatiotemporal misalignment and insufficient accuracy, and filling the technical gap in accurate fusion of photovoltaic multi-source data. Heterogeneous data feature enhancement and redundancy filtering algorithm: It enhances the core features of heterogeneous data and adaptively filters redundant data. Compared with the traditional mode, the data feature recognition rate is improved by more than 40% and the effective utilization rate is improved by more than 50%, providing a high-quality data foundation for subsequent scheduling and diagnosis. Multi-objective collaborative optimization scheduling algorithm for energy source, grid, load and storage: Construct a multi-objective collaborative optimization scheduling model to achieve a global balance between energy consumption, grid stability, energy storage utilization and operation and maintenance costs. Compared with the single-objective scheduling mode, the collaborative efficiency of energy source, grid, load and storage is improved by more than 25% and the photovoltaic energy consumption rate is improved by more than 15%. Photovoltaic dispatch load forecasting and dynamic adjustment algorithm: It realizes accurate forecasting of photovoltaic output and grid load, and achieves dynamic optimization of dispatching scheme through incremental adjustment strategy. Compared with the traditional fixed dispatching mode, the forecasting accuracy is improved by more than 35% and the adaptability of dispatching scheme is improved by more than 40%, solving the dispatching lag problem. Photovoltaic equipment fault feature extraction and accurate diagnosis algorithm: It realizes the effective extraction of hidden fault features and accurate fault diagnosis. Compared with the traditional threshold alarm mode, the fault early detection rate is increased by more than 60% and the diagnosis accuracy is increased by more than 50%, realizing the upgrade from "passive alarm" to "active diagnosis". Fault classification and self-healing decision optimization algorithm: It realizes the scientific classification of faults and the automatic handling of self-healing faults. Compared with the traditional manual handling mode, the average fault handling time is shortened by more than 80% and the power generation loss rate caused by faults is reduced by more than 70%, filling the technical gap of self-healing faults in photovoltaic equipment. Attached Figure Description

[0029] Appendix Figure 1 Workflow diagram of photovoltaic multi-source heterogeneous data fusion sensing module Detailed Implementation

[0030] The implementation steps of the present invention will be described in detail below through specific embodiments.

[0031] Example 1: Multi-source data fusion sensing in a centralized photovoltaic power plant real-time output scheduling scenario Implementation steps Step 1: Multi-source heterogeneous data aggregation and resource pool construction: Deploy multi-source data acquisition terminals to aggregate heterogeneous data from four ends: photovoltaic power station end (PV panel string output, inverter operation data, second-level acquisition), meteorological end (real-time irradiance, temperature, minute-level acquisition), grid end (real-time load, voltage, second-level acquisition), and energy storage end (SOC, charging and discharging power, second-level acquisition). Construct a photovoltaic multi-source sensing data resource pool and label basic information such as acquisition frequency, spatiotemporal reference, monitoring accuracy, and data dimension of various types of data.

[0032] Step 2: Feature Extraction and Heterogeneity Identification: Using a photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm, a feature extraction algorithm is designed to extract the core spatiotemporal features (timestamp, monitoring point coordinates) and attribute features (string output, irradiance, grid load, SOC) of various types of data; at the same time, data heterogeneity is identified, clarifying temporal heterogeneity (meteorological data at the minute level, others at the second level), spatial heterogeneity (different coordinate systems of meteorological stations and power plants), and accuracy heterogeneity (precise measured data of photovoltaic panels, meteorological irradiance prediction data).

[0033] Step 3: Standard Spatiotemporal Conversion and Fusion Accuracy Calculation: Construct a standard spatiotemporal system for the photovoltaic power station (using the central control room as the time reference and the power station's local coordinate system as the spatial reference). Employ spatiotemporal coordinate conversion and timestamp synchronization algorithms to uniformly convert data from meteorological, grid, and energy storage sources to the standard spatiotemporal system, completing preliminary spatiotemporal calibration. Then, use a photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm to calculate the comprehensive fusion accuracy using the fusion accuracy calculation formula, which is: Set monitoring dimensions (PV output, irradiance, grid load, SOC), weights of each dimension , , , Real-time output scheduling scenario fusion accuracy threshold The fused monitoring values ​​are then compared with the standard true values ​​for calculation.

[0034] Step 4: Deviation Iteration Correction and Accuracy Achievement: If the overall fusion accuracy does not reach 98%, the "feature point matching-deviation source tracing-iterative correction" strategy is adopted to trace the causes of spatiotemporal offset (such as the lag of weather station timestamps) and accuracy deviation (such as the deviation of irradiation sensors), dynamically adjust the spatiotemporal calibration parameters and accuracy compensation coefficients, and recalculate the fusion accuracy until the overall fusion accuracy is ≥98%.

[0035] Step 5: Feature Enhancement and Redundancy Filtering: Using heterogeneous data feature enhancement and redundancy filtering algorithms, a feature correlation mapping relationship is established between "PV output - irradiance - temperature" and "grid load - SOC - charging and discharging power". The core features of PV output are enhanced through weighted fusion and deep learning feature extraction. The correlation coefficient and redundancy between data are calculated, and redundant data such as humidity data that is not related to PV output in meteorological data and voltage data that is repeatedly collected in grid data are removed to improve the effective utilization rate of data.

[0036] Step 6: Data Fusion Verification and Optimization: Apply the output high-precision fused sensing data to the real-time power output scheduling scenario of the power plant to verify the spatiotemporal consistency, accuracy, and feature recognition of the data, and collect feedback from dispatchers; dynamically optimize the fusion accuracy weight, feature enhancement coefficient, and redundancy filtering threshold of the algorithm to ensure that the fused data continuously meets the refined requirements of real-time power output scheduling.

[0037] Modeling Innovation Principles Abandoning the traditional crude fusion modeling approach of "simple splicing + deduplication," this paper constructs an integrated closed-loop modeling logic of "data aggregation - feature extraction - spatiotemporal calibration - precision quantification - iterative correction - feature enhancement - redundancy filtering." It takes the heterogeneous characteristics of photovoltaic multi-source data and the high-precision requirements of real-time power dispatching scenarios as core inputs, overcoming the technical limitations of data spatiotemporal misalignment, insufficient precision, and weakened features. A quantitative assessment of fusion precision is achieved through a fusion precision calculation formula, solving the problem of "no quantification, no correction" in traditional fusion models. Feature enhancement modeling improves the recognition of core dispatching features, and redundancy filtering modeling improves data quality and efficiency. The overall modeling approach focuses on the accurate fusion and efficient perception of photovoltaic multi-source data, completely different from existing modeling approaches and technical directions, representing a new modeling direction and filling the modeling gap for accurate fusion perception of photovoltaic multi-source data.

[0038] Algorithm efficiency enhancement principle The photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm achieves precise adaptation of data from different spatiotemporal benchmarks and acquisition frequencies through standard spatiotemporal system conversion and iterative correction. Compared with traditional data splicing modes, the spatiotemporal consistency of data is improved by more than 90%, and the fusion accuracy reaches more than 98%, completely solving the problem of spatiotemporal misalignment. The fusion accuracy calculation formula provides a scientific quantitative basis for deviation correction, and the accuracy controllability is improved by more than 95% compared with the fusion mode without quantitative calculation. The heterogeneous data feature enhancement and redundancy filtering algorithm improves the identification of core photovoltaic power output features by more than 40% through feature association modeling and multi-dimensional fusion, providing a highly identifiable data foundation for real-time scheduling. Through adaptive redundancy filtering, the effective utilization rate of data is improved by more than 50%, significantly reducing the computing power consumption of subsequent data processing. Compared with the mode without redundancy filtering, the data processing efficiency is improved by more than 60%.

[0039] Existing technologies employ a crude fusion model of "simple splicing + deduplication," lacking spatiotemporal fusion and accuracy compensation algorithms. This results in fused data exhibiting problems such as spatiotemporal misalignment, insufficient accuracy, and weakened features, with fusion accuracy below 70% and data utilization below 60%, failing to meet the high-precision requirements of real-time power output scheduling for photovoltaic power plants. This embodiment, through algorithmic innovation and modeling optimization, achieves precise spatiotemporal fusion, accuracy compensation, feature enhancement, and redundancy filtering of multi-source photovoltaic data. Fusion accuracy is improved to over 98%, and data utilization to over 90%, completely resolving the pain points of existing technologies. It provides a high-quality data foundation for real-time power output scheduling and achieves a completely new and innovative breakthrough without any overlap with existing technologies or modeling approaches.

[0040] Example 2: Multi-objective coordinated dispatch of source, grid, load and storage in photovoltaic energy storage integrated power station Implementation steps Step 1: Scheduling Target Modeling and Constraint Setting: For the integrated photovoltaic energy storage power station scenario, a multi-objective collaborative optimization scheduling algorithm for source-grid-load-storage is adopted to construct four scheduling objective functions (photovoltaic energy absorption rate, grid voltage stability, energy storage utilization rate, and operation and maintenance cost). The weights of each objective are set according to the power station's operational needs: absorption rate 0.35, grid stability 0.3, energy storage utilization rate 0.2, and operation and maintenance cost 0.15. At the same time, hard constraints are established for the operation of each end of the source-grid-load-storage system: maximum photovoltaic output 80MW, energy storage SOC 20%-90%, and grid voltage 380V±5%.

[0041] Step 2: Data Acquisition and Modeling of Source, Grid, Load and Storage: Real-time acquisition of photovoltaic power plant output, real-time grid load, energy storage SOC and charging / discharging power, and local load data; construction of operation models for each end of source, grid, load and storage; and clarification of photovoltaic output characteristics, grid load change patterns, energy storage charging / discharging efficiency, and local load demand.

[0042] Step 3: Multi-objective optimization solution and initial scheme generation: An improved multi-objective particle swarm optimization algorithm is used to solve the multi-objective scheduling function to obtain the global optimal solution in the non-dominated solution set; combined with the actual operation and maintenance of the power plant and the grid dispatch requirements, the optimal solution is modified in an engineering manner to generate the initial coordinated scheduling scheme of source, grid, load and storage: photovoltaic output prioritizes meeting local load, and surplus output is partially connected to the grid and partially used to charge energy storage. During peak grid load, energy storage discharges to supplement power output.

[0043] Step 4: Load Forecasting and Deviation Calculation: The photovoltaic dispatch load forecasting and dynamic adjustment algorithm is adopted, which integrates short-term meteorological forecast data, historical power plant operation data and grid load data. The LSTM+attention mechanism model is used to achieve accurate short-term forecasting of photovoltaic output and grid load. Real-time collection of actual photovoltaic output and actual grid load data is used to calculate the deviation between the predicted and actual values ​​and identify the deviation type as "sudden meteorological changes causing photovoltaic output to be lower than the predicted value".

[0044] Step 5: Incremental dynamic adjustment of the dispatch scheme: For the deviation type of "low photovoltaic output", an incremental adjustment strategy is adopted. There is no need to resolve the full dispatch scheme. Only the "energy storage discharge" part in the initial scheme is optimized: the energy storage discharge power is appropriately increased to make up for the photovoltaic output gap and ensure the balance of local load supply and demand and grid voltage stability.

[0045] Step 6: Scheduling Execution and Effect Optimization: Execute the dynamically adjusted scheduling scheme, monitor the operating status of each end of the source, grid, load and storage in real time, and quantitatively evaluate the scheduling effect (absorption rate, grid stability, energy storage utilization rate); collect feedback on the scheduling effect, dynamically optimize the scheduling target weight and algorithm parameters, and improve the efficiency and effect of subsequent coordinated scheduling.

[0046] Modeling Innovation Principles Abandoning the traditional extensive modeling approach of "single-objective fixed threshold scheduling," this paper constructs an integrated closed-loop modeling logic of "multi-objective modeling - constraint modeling - multi-objective optimization - accurate prediction - incremental adjustment - effect feedback." It takes the dynamic operational characteristics of source-grid-load-storage systems and the scheduling requirements of multi-objective collaborative scheduling as core inputs, overcoming the technical limitations of single-objective scheduling, global imbalance, and scheduling lag. Multi-objective collaborative optimization modeling achieves global balance of the four major scheduling objectives, solving the problem of "paying attention to one thing but neglecting another" in traditional scheduling. Incremental dynamic adjustment modeling enables rapid optimization of scheduling schemes, avoiding the computational consumption and time lag of full-scale solutions. The overall modeling approach focuses on the global collaboration and dynamic scheduling of source-grid-load-storage systems, which is completely different from the existing modeling approaches and technical directions, representing a new modeling direction and filling the modeling gap in multi-objective collaborative optimization scheduling of source-grid-load-storage systems.

[0047] Algorithm efficiency enhancement principle The multi-objective collaborative optimization scheduling algorithm for power generation, grid, load, and energy storage achieves global balance of four scheduling objectives through multi-objective function modeling and improved particle swarm optimization. Compared with the single-objective scheduling mode, the collaborative efficiency of power generation, grid, load, and energy storage is improved by more than 25%, and the photovoltaic energy consumption rate is improved by more than 15%, while ensuring grid voltage stability, rational utilization of energy storage, and controllable operation and maintenance costs. The photovoltaic scheduling load prediction and dynamic adjustment algorithm achieves accurate prediction of photovoltaic output and grid load through an LSTM + attention mechanism model, improving prediction accuracy by more than 35%, and significantly reducing prediction deviation compared with the traditional statistical prediction mode. The incremental adjustment strategy only optimizes the deviation-related parts, improving the scheduling scheme adjustment efficiency by more than 80%. Compared with the mode of solving all the problems again, it completely solves the scheduling lag problem and ensures that the scheduling scheme is synchronized with real-time operation data.

[0048] Existing technologies employ a "single-objective fixed threshold scheduling" model, focusing solely on photovoltaic (PV) absorption rate. This can easily lead to grid voltage fluctuations and overcharging / discharging of energy storage, resulting in a generation-grid-load-storage (PGS) coordination efficiency of less than 70%. Furthermore, the lack of accurate prediction and dynamic adjustment algorithms results in outdated scheduling schemes that cannot adapt to dynamic changes in PV output and grid load. This embodiment, through algorithmic innovation and model optimization, achieves multi-objective coordinated optimization of generation, grid, load, and storage, as well as dynamic adjustment of the scheduling scheme. The PGS coordination efficiency is increased to over 95%, and the PV energy absorption rate is increased to over 98%. Grid stability and energy storage utilization both meet optimal requirements, completely resolving the pain points of existing technologies. This achieves global and intelligent PV energy scheduling, without any overlap with existing technologies in terms of technical direction or implementation scenarios. Its innovation is significant and its practicality is extremely high.

[0049] Example 3: Fault Feature Extraction and Precise Diagnosis of Photovoltaic Power Plant Inverters Implementation steps Step 1: Loading the Fault Feature Library and Data Acquisition: For the inverter equipment in the photovoltaic power station, load all types of fault features of the inverter from the photovoltaic equipment fault feature library (including explicit faults: inverter shutdown, overvoltage protection; implicit faults: power device aging, drive circuit faults), clarify the characteristic parameters and changing patterns of each fault, and collect multi-dimensional operating data such as the inverter's input / output voltage, current, power, temperature, and switching frequency in real time to construct a fault diagnosis dataset.

[0050] Step 2: Fault Feature Extraction and Mining: Using photovoltaic equipment fault feature extraction and accurate diagnosis algorithm, the inverter operation data is mined by CNN+BiLSTM deep learning algorithm to extract obvious features of explicit faults and gradual features of latent faults (power device aging) (such as increased output current ripple and slow temperature rise), so as to effectively extract fault precursor features.

[0051] Step 3: Accurate fault matching and diagnosis: The extracted fault features are accurately matched with the fault feature library and combined with the fault rule library to achieve accurate fault diagnosis: The diagnosis result is "Inverter power device aging, fault degree is mild, fault location is inverter A phase power module".

[0052] Step 4: Fault Level Determination and Handling Strategy Matching: Using a fault classification and self-healing decision optimization algorithm, "slight aging of power devices" is determined as a general fault based on the degree of fault impact and the urgency of handling. The handling strategy for general faults is matched as follows: scheduled inspections are arranged to test and maintain the power devices without affecting the normal operation of the inverter.

[0053] Step 5: Verification and Feedback of Diagnostic Results: Compare the fault diagnosis results with the actual inspection results of the inverter to verify the accuracy and early detection rate of the diagnosis; collect feedback from maintenance personnel to identify problems such as insufficient feature extraction during the diagnosis process.

[0054] Step 6: Algorithm and Feature Library Optimization: Based on feedback, dynamically optimize the parameters of the CNN+BiLSTM feature extraction algorithm, and supplement the fault feature library with the gradual feature parameters of power device aging to improve the accuracy and early detection rate of subsequent fault diagnosis.

[0055] Modeling Innovation Principles Abandoning the traditional passive modeling approach of "manual inspection + threshold alarm," this paper constructs an integrated closed-loop modeling logic of "fault feature library loading - multi-dimensional data collection - latent feature extraction - precise matching diagnosis - level determination - effect feedback." It takes the diversity, hidden characteristics, and precise fault diagnosis requirements of inverter faults as core inputs, overcoming the technical limitations of being unable to extract latent features, low diagnostic accuracy, and late fault detection. Through deep learning algorithms, it extracts latent and precursor features of faults, solving the problem that traditional models can only identify explicit threshold faults. Precise fault matching modeling enables quantitative diagnosis of fault type, location, and severity. Compared to traditional fuzzy diagnosis, it achieves more precise and quantifiable fault diagnosis. The overall modeling approach focuses on the proactive diagnosis and precise identification of photovoltaic equipment faults, completely different from existing modeling approaches and technical directions, representing a new modeling direction and filling the modeling gap in the extraction and precise diagnosis of latent features of photovoltaic equipment faults.

[0056] Algorithm efficiency enhancement principle The photovoltaic equipment fault feature extraction and accurate diagnosis algorithm, through CNN+BiLSTM deep learning, effectively extracts latent and precursor features of inverter faults. Compared with the traditional threshold alarm mode, the fault early detection rate is improved by more than 60%, enabling early detection before faults develop into serious faults, thus saving time for operation and maintenance. By combining accurate matching of fault feature database with fault rule database, it achieves accurate diagnosis of fault type, location, and severity, improving the diagnostic accuracy by more than 50%. Compared with the fuzzy diagnosis of traditional manual inspection, it completely solves the problems of "inaccurate diagnosis and unclear location". The fault classification and self-healing decision optimization algorithm, through scientific fault level determination, achieves accurate matching of handling strategies, improving the targeting of fault handling by more than 90%, avoiding the "one-size-fits-all" handling mode, and improving operation and maintenance efficiency.

[0057] Existing technologies rely on "manual inspection + threshold alarm," lacking fault feature extraction algorithms. They can only identify explicit faults such as inverter shutdown and overvoltage protection, failing to detect latent or gradual faults such as power device aging, resulting in delayed fault detection. Furthermore, their diagnostic accuracy is low, only able to determine the fault type, unable to precisely locate the fault or quantify its severity, with a fault diagnosis accuracy rate below 50%. This embodiment, through algorithmic innovation and model optimization, achieves the extraction of latent fault features and accurate fault diagnosis, increasing the early fault detection rate to over 80% and the diagnostic accuracy rate to over 95%. It can detect gradual faults early and pinpoint their location, providing a scientific basis for operation and maintenance, completely resolving the pain points of existing technologies. It upgrades photovoltaic equipment fault diagnosis from "passive alarm" to "active diagnosis," and has no overlap with existing technologies in terms of technical direction or implementation scenarios, demonstrating clear innovation and strong practicality.

[0058] Example 4: Graded handling and self-healing decision-making for minor faults in photovoltaic panels of photovoltaic power plants Implementation steps Step 1: Photovoltaic panel operation data acquisition and fault feature extraction: Real-time acquisition of multi-dimensional operation data such as output, voltage, current, and temperature of photovoltaic panel strings. Using photovoltaic equipment fault feature extraction and accurate diagnosis algorithm, fault features are extracted: the output of a certain string of photovoltaic panels is slightly lower than normal, the open circuit voltage drops slightly, and there is no obvious temperature abnormality.

[0059] Step 2: Accurate fault diagnosis and severity determination: The extracted fault features are matched with the fault feature library. The diagnosis result is "slight dust accumulation on the surface of the photovoltaic panel, and the fault severity is minor". The fault classification and self-healing decision optimization algorithm is adopted to determine it as a minor fault based on the degree of fault impact (power generation loss < 5%) and the urgency of handling (low).

[0060] Step 3: Self-healing fault determination and strategy matching: Based on the fault type and level, "slight dust accumulation on photovoltaic panels" is determined to be a self-healing fault; load the photovoltaic equipment self-healing strategy library and match the corresponding self-healing strategy: start the photovoltaic panel automatic cleaning system and use high-pressure water washing to clean the dusty photovoltaic panels.

[0061] Step 4: Automatic execution and status monitoring of self-healing strategy: Automatically trigger the photovoltaic panel automatic cleaning system to execute the water washing cleaning strategy; monitor the operating status of the photovoltaic panels in real time during the cleaning process to avoid secondary damage to the equipment during the cleaning process.

[0062] Step 5: Self-healing effect verification and quantitative evaluation: After cleaning, collect the output and voltage data of the photovoltaic panel string to verify the self-healing effect: The output of the photovoltaic panel is restored to the normal level and the open circuit voltage is restored to the standard value; Quantitatively evaluate the self-healing success rate and power generation loss rate: The self-healing success rate is 100%, and the power generation loss rate caused by dust accumulation is reduced from 4.5% to 0.5%.

[0063] Step 6: Feedback Optimization and Strategy Library Update: Collect feedback from maintenance personnel on the self-healing effect, and dynamically optimize the self-healing strategy library based on the degree of dust accumulation on the photovoltaic panels and the concentration of environmental dust: adjust the start threshold of the automatic cleaning system (automatically start when the output of the photovoltaic panels decreases by ≥3%), and optimize the cleaning water pressure and cleaning time to improve the efficiency and effect of self-healing.

[0064] Modeling Innovation Principles Abandoning the traditional "no self-healing, all manual" fault handling modeling approach, this paper constructs an integrated closed-loop modeling logic of "fault diagnosis - level determination - self-healing determination - strategy matching - automatic execution - effect verification - feedback optimization". It takes the self-healing requirements of minor photovoltaic panel faults and the actual needs of power plant operation and maintenance as core inputs, overcoming the technical limitations of fault handling lacking grading, self-healing capabilities, and timely handling. Fault grading modeling enables differentiated handling, solving the resource waste problem of the traditional "one-size-fits-all" handling mode. Through the construction and intelligent matching of a self-healing strategy library, it achieves automatic handling of self-healable faults, solving the lag problem of traditional manual handling. Feedback optimization modeling enables continuous iteration of self-healing strategies, ensuring accurate adaptation of self-healing strategies to fault scenarios. The overall modeling approach focuses on the graded handling and automatic self-healing of photovoltaic equipment faults, completely different from existing modeling approaches and technical directions, representing a new modeling direction and filling the modeling gap in photovoltaic equipment fault self-healing decision-making.

[0065] Algorithm efficiency enhancement principle The fault classification and self-healing decision optimization algorithm scientifically classifies faults into four levels and matches them with differentiated handling strategies. Compared with the traditional non-classified handling mode, the utilization rate of fault handling resources is improved by more than 80%, avoiding the problem of minor faults occupying emergency handling resources. Through intelligent matching and automatic execution of the self-healing strategy library, automatic self-healing of minor dust accumulation faults on photovoltaic panels is achieved, reducing the average fault handling time by more than 80%. Compared with the traditional manual cleaning handling mode, it completely solves the problem of handling lag. The quantitative evaluation and feedback optimization of the self-healing effect enables continuous iteration of the self-healing strategy library, increasing the self-healing success rate to more than 95%. At the same time, the start threshold and execution parameters of the self-healing strategy are optimized, improving the self-healing efficiency by more than 70%. Through self-healing, the power generation loss rate caused by minor faults is reduced by more than 70%, significantly improving the power generation efficiency of photovoltaic power plants.

[0066] Existing technologies lack self-healing capabilities for minor faults such as dust accumulation on photovoltaic panels, requiring manual cleaning and resulting in fault handling times exceeding two hours. Delayed handling leads to significant power generation losses. Furthermore, the absence of a fault-based tiered handling strategy means that minor and serious faults are handled using the same manual methods, resulting in a severe waste of maintenance resources. This embodiment, through algorithmic innovation and model optimization, achieves automatic self-healing and scientifically tiered handling of minor photovoltaic panel faults. The self-healing success rate reaches 100%, fault handling time is reduced to a few minutes, power generation losses are significantly reduced, and maintenance resource utilization is improved. It completely solves the pain points of existing technologies, upgrading photovoltaic equipment fault handling from "manual handling" to "automatic self-healing." Moreover, it does not overlap with existing technologies in terms of technical direction or implementation scenarios, demonstrating clear innovation and strong practicality, effectively reducing the operation and maintenance costs and power generation losses of photovoltaic power plants.

[0067] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for implementing lifecycle management of large-scale photovoltaic energy management, characterized in that, Includes the following steps: S1: Photovoltaic multi-source heterogeneous data fusion and sensing processing. This involves aggregating heterogeneous monitoring data from power plants, meteorological stations, power grids, and energy storage systems to construct a sensing resource pool. Through photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithms, as well as heterogeneous data feature enhancement and redundancy filtering algorithms, it achieves data spatiotemporal calibration, accuracy compensation, feature enhancement, and redundancy removal, outputting high-precision fused sensing data across all dimensions. The photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm includes a data fusion accuracy calculation formula: The constraints are , To achieve comprehensive fusion accuracy, For data monitoring dimensions, For the first Weighting coefficients for dimensional data These are the monitored values ​​after fusion. For the standard true value, The fusion accuracy threshold is set according to the requirements of photovoltaic energy management scenarios. 2.S2: Photovoltaic energy collaborative scheduling optimization processing, constructing a photovoltaic energy scheduling rule base and a source-grid-load-storage operation model, and realizing multi-objective optimization, accurate load prediction, dynamic adjustment and global coordination of energy scheduling through multi-objective collaborative optimization scheduling algorithm of source-grid-load-storage, and photovoltaic scheduling load prediction and dynamic adjustment algorithm; S3: Photovoltaic equipment fault diagnosis and self-healing processing. It sorts out all types of fault scenarios of photovoltaic equipment and extracts fault feature parameters. Through photovoltaic equipment fault feature extraction and accurate diagnosis algorithm, fault classification and self-healing decision optimization algorithm, it realizes accurate extraction of fault features, intelligent diagnosis of fault types, and dynamic optimization of fault classification and self-healing strategy. The method according to claim 1, characterized in that, The photovoltaic multi-source data spatiotemporal fusion and accuracy compensation algorithm in step S1 includes the following sub-steps: extracting the core spatiotemporal and attribute features of multi-source heterogeneous data, identifying the spatiotemporal offset and accuracy deviation types of data, uniformly converting different spatiotemporal reference data into a standard spatiotemporal system for photovoltaic power plants, calculating the fusion accuracy through the fusion accuracy calculation formula, and iteratively correcting the spatiotemporal offset and accuracy deviation to a set threshold.

3. The method according to claim 1, characterized in that, The heterogeneous data feature enhancement and redundancy filtering algorithm in step S1 includes the following sub-steps: establishing a heterogeneous data feature association mapping relationship, using a multi-dimensional feature fusion strategy to achieve feature enhancement, identifying redundant data based on data correlation and redundancy calculation, and using an adaptive filtering strategy to remove redundant data in combination with photovoltaic management needs, thereby improving the effective utilization rate of data.

4. The method according to claim 1, characterized in that, In step S2, the multi-objective collaborative optimization scheduling algorithm for energy source, grid, load and storage divides the scheduling objectives into four categories: energy absorption rate, grid stability, energy storage utilization rate and operation and maintenance cost. The algorithm uses a multi-objective intelligent optimization algorithm to construct the scheduling objective function, and combines the operational constraints of each end of the energy source, grid, load and storage to solve the global optimal scheduling scheme and achieve multi-objective collaborative balance.

5. The method according to claim 1, characterized in that, The photovoltaic dispatch load prediction and dynamic adjustment algorithm in step S2 integrates meteorological forecasts, historical operation data, and grid load data to construct a multi-dimensional prediction model, thereby achieving accurate short-term prediction of photovoltaic output and grid load. Based on the prediction deviation and real-time operation data, an incremental adjustment strategy is adopted to achieve dynamic optimization of the dispatch scheme.

6. The method according to claim 1, characterized in that, The photovoltaic equipment fault feature extraction and accurate diagnosis algorithm in step S3 constructs a full-type fault feature library for photovoltaic equipment, uses deep learning algorithms to extract latent fault features from equipment operation data, and combines a fault rule library to achieve accurate diagnosis and quantitative assessment of fault type, fault location, and fault severity.

7. The method according to claim 1, characterized in that, The fault classification and self-healing decision optimization algorithm in step S3 classifies photovoltaic equipment faults into four levels: fatal faults, serious faults, general faults, and minor faults. It formulates differentiated classification and handling strategies, builds a self-healing strategy library for self-healing faults, and realizes intelligent matching and dynamic optimization of self-healing strategies in combination with fault scenarios.

8. The method according to claim 1, characterized in that, The fusion accuracy threshold It can be flexibly adjusted according to photovoltaic energy management scenarios, and real-time power output scheduling scenarios for power plants. Equipment fault diagnosis scenarios Energy storage charging and discharging management scenarios .

9. The method according to any one of claims 1-8, characterized in that, The method can be applied to various photovoltaic energy management scenarios such as centralized photovoltaic power plants, distributed photovoltaic clusters, and photovoltaic energy storage integrated power plants, realizing full-process automation and intelligence of multi-source data fusion perception, source-grid-load-storage coordinated scheduling, and equipment fault diagnosis and self-healing.

10. An automated large-scale model system for photovoltaic energy management, characterized in that, The system includes a photovoltaic multi-source heterogeneous data fusion and sensing module, a photovoltaic energy collaborative scheduling and optimization module, and a photovoltaic equipment fault diagnosis and self-healing module. By executing the method described in any one of claims 1-9, the system achieves full-link automation and intelligence in photovoltaic energy management. Each module achieves interconnection and data sharing through standardized data interfaces, supports independent module upgrades and flexible function expansion, and adapts to the management needs of photovoltaic power plants of different scales and types.

Images