Photovoltaic power station intelligent analysis and fault intelligent diagnosis method and system

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a photovoltaic power station intelligent analysis and fault intelligent diagnosis method and system. The method comprises the steps of collecting electrical and environmental data of a photovoltaic power station through a sensor network, transmitting the electrical and environmental data to a data processing unit for preprocessing and standardization, and extracting multi-dimensional features to construct a feature matrix; and inputting the feature matrix into a multi-scale pulse neural network for fault diagnosis to obtain fault type and position information. And evaluating the severity of the fault according to the fault result, generating a maintenance scheme, and finally sending the maintenance scheme to the terminal equipment through the management system. The intelligent fault diagnosis system based on the pulse neural network is introduced, and the fault propagation analysis, the maintenance priority distribution, the path planning and the resource matching optimization are combined, so that the problems of low fault diagnosis precision and low maintenance resource scheduling efficiency in the prior art are solved, and the operation and maintenance efficiency and the equipment reliability of the photovoltaic power station are improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a method and system for intelligent analysis and intelligent fault diagnosis of photovoltaic power stations. Background Art

[0002] With the continuous advancement of photovoltaic power generation technology, PV power plants have become a vital component of renewable energy. To ensure the efficient operation of PV power plants, monitoring and maintenance systems are crucial. Traditional PV power plant management relies on manual inspections, regular maintenance, and simple monitoring systems. These traditional methods typically include monitoring basic parameters such as voltage, current, and temperature, often relying on manual inspections and fault diagnosis. While this approach can ensure equipment operation to a certain extent, it is limited in its ability to promptly detect and accurately locate faults. Furthermore, it is difficult to respond quickly to large-scale faults, resulting in prolonged downtime and losses. Furthermore, existing PV power plant maintenance methods lack intelligence and systematization, making it difficult to optimize the allocation of maintenance resources in real time, reducing maintenance efficiency and effectiveness.

[0003] However, with the continuous expansion and increasing complexity of photovoltaic power plants, traditional maintenance methods have exposed significant shortcomings. On the one hand, fault diagnosis and repair rely on manual experience, resulting in inaccurate fault identification and long response times. On the other hand, traditional maintenance lacks an efficient resource scheduling mechanism, resulting in the inability to rationally allocate maintenance task priorities based on the actual severity of the fault, often leading to wasted maintenance resources or delays. Existing monitoring systems are often unable to process large amounts of data in real time, unable to effectively integrate equipment failure modes with maintenance plans, and rely on manual intervention to schedule maintenance tasks, resulting in low maintenance efficiency. To improve the operational efficiency of photovoltaic power plants, it is necessary to overcome these problems, enhance the intelligence level of fault diagnosis, and optimize maintenance resource scheduling and maintenance efficiency. Summary of the Invention

[0004] The present application provides a method and system for intelligent analysis and intelligent fault diagnosis of photovoltaic power stations, which solves the problems of low fault diagnosis accuracy and low maintenance resource scheduling efficiency in the existing technology by introducing an intelligent fault diagnosis system based on a pulse neural network and combining it with fault propagation analysis, maintenance priority allocation, path planning and resource matching optimization, thereby improving the operation and maintenance efficiency and equipment reliability of photovoltaic power stations.

[0005] In the first aspect, the present application provides a method for intelligent analysis and intelligent fault diagnosis of a photovoltaic power station, which includes: collecting photovoltaic power station operation data through a sensor network to obtain an electrical parameter data set and an environmental parameter data set; transmitting the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set; extracting multidimensional features from the standardized data set, constructing correlation features, and obtaining a feature matrix; inputting the feature matrix into a multi-scale pulse neural network fault diagnosis model, wherein the multi-scale pulse neural network fault diagnosis model includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer, and dynamically adjusts the network connection weights through a biological synaptic plasticity mechanism to obtain a fault type result and fault location information; performing a fault assessment based on the fault type result and the fault location information to obtain a fault severity level; generating a maintenance plan based on the fault severity level and the fault type result, and sending it to the terminal device through a management system.

[0006] In a second aspect, the present application provides a photovoltaic power station intelligent analysis and fault intelligent diagnosis system, the photovoltaic power station intelligent analysis and fault intelligent diagnosis system comprising:

[0007] The acquisition module is used to collect the operation data of the photovoltaic power station through the sensor network to obtain the electrical parameter data set and the environmental parameter data set;

[0008] A transmission module, configured to transmit the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set;

[0009] A construction module is used to extract multidimensional features from the standardized data set, construct correlation features, and obtain a feature matrix;

[0010] An input module is used to input the feature matrix into a multi-scale spiking neural network fault diagnosis model. The multi-scale spiking neural network fault diagnosis model includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The network connection weights are dynamically adjusted through a biological synaptic plasticity mechanism to obtain fault type results and fault location information.

[0011] An evaluation module, configured to perform a fault evaluation based on the fault type result and the fault location information to obtain a fault severity level;

[0012] A generation module is used to generate a maintenance plan based on the fault severity level and the fault type result, and send the plan to the terminal device through the management system.

[0013] In a third aspect, a photovoltaic power station intelligent analysis and fault intelligent diagnosis device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the photovoltaic power station intelligent analysis and fault intelligent diagnosis device executes the above-mentioned photovoltaic power station intelligent analysis and fault intelligent diagnosis method.

[0014] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned photovoltaic power station intelligent analysis and fault intelligent diagnosis method.

[0015] The technical solution provided in this application significantly improves the fault detection accuracy of photovoltaic power plants by combining sensor network data collection, intelligent analysis, precise fault diagnosis, and optimized maintenance decisions. It also significantly enhances the operational efficiency and stability of photovoltaic power plants through efficient maintenance resource scheduling and optimized path planning. First, a distributed sensor network deployed at key locations in the photovoltaic power plant enables real-time collection of photovoltaic string, inverter, and environmental monitoring data. Using an adaptive sensor sampling adjustment strategy, the sampling frequency can be flexibly increased under different environmental conditions, ensuring comprehensive and high-frequency data collection. During data transmission, differential transmission technology is applied to reduce the transmission of redundant data, lowering the network burden and ensuring real-time and accurate data transmission. Furthermore, a triple filtering algorithm is used to denoise the collected data, and an outlier detection mechanism is used to correct abnormal data. Ultimately, a standardized data set is formed, providing an accurate and stable foundation for subsequent intelligent analysis. Based on this standardized, high-quality data, multidimensional feature extraction and feature engineering techniques are used to extract key features of photovoltaic strings, inverters, and environmental factors. By constructing time series features and inter-device correlation features, the accuracy of fault diagnosis is effectively enhanced. By inputting into a multi-scale spiking neural network (SNN) model, this method can convert continuous-valued features into time-coded pulse sequences, accurately identifying different fault modes. Through a hierarchical decision fusion mechanism, it integrates feature information at different scales to output the final fault type and location information. Simultaneously, the pulse time-dependent plasticity mechanism dynamically adjusts the synaptic weights between neurons, effectively capturing the fault propagation path between devices, accurately locating the fault source, and assessing the risk of fault spread based on the fault characteristics, thereby achieving global fault propagation analysis. During the maintenance decision-making phase, the present invention uses semantic matching retrieval from a fault-repair knowledge base to quickly match corresponding standardized maintenance plans based on different fault types. Maintenance priorities are then assigned based on fault severity, ensuring that the most urgent faults are addressed promptly. Through resource matching and path planning optimization, the system can rationally schedule maintenance personnel's work and ensure the efficient execution of maintenance tasks. The optimized maintenance plan not only takes into account factors such as maintenance personnel's skill level, spare parts inventory, and weather conditions, but also avoids unnecessary movement of maintenance personnel by calculating the shortest maintenance path, further improving maintenance efficiency. Finally, a standardized work order is generated based on the optimized maintenance plan and sent to on-site staff in a timely manner through multiple communication channels. It is also equipped with augmented reality assistance functions to help maintenance personnel accurately identify the fault location and improve the accuracy and efficiency of maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a schematic diagram of an embodiment of a method for intelligent analysis and fault diagnosis of a photovoltaic power station in an embodiment of the present application;

[0018] Figure 2 This is a schematic diagram of an embodiment of the photovoltaic power station intelligent analysis and fault intelligent diagnosis system in the embodiment of the present application;

[0019] Figure 3 It is a schematic block diagram of the structure of the intelligent analysis and fault intelligent diagnosis equipment for photovoltaic power stations in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present application provide a method and system for intelligent analysis and fault intelligent diagnosis of photovoltaic power plants. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the photovoltaic power station intelligent analysis and fault intelligent diagnosis method of the present application, the method includes:

[0022] Step S101: Collecting photovoltaic power station operation data through a sensor network to obtain an electrical parameter data set and an environmental parameter data set;

[0023] Step S102: transmitting the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set;

[0024] Step S103: extract multidimensional features from the standardized data set, construct correlation features, and obtain a feature matrix;

[0025] Step S104: Input the feature matrix into a multi-scale spiking neural network fault diagnosis model. The multi-scale spiking neural network fault diagnosis model includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The network connection weights are dynamically adjusted through the biological synaptic plasticity mechanism to obtain fault type results and fault location information.

[0026] Step S105: perform fault assessment based on the fault type result and the fault location information to obtain the fault severity level;

[0027] Step S106: Generate a maintenance plan based on the fault severity level and fault type results, and send it to the terminal device through the management system.

[0028] It is understandable that the execution subject of this application can be a photovoltaic power station intelligent analysis and fault intelligent diagnosis system, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0029] Specifically, a distributed sensor network collects operational data from the PV power plant, generating electrical parameter and environmental parameter datasets. Specifically, the sensor network includes three types of sensors: string-level sensors, inverter-embedded sensors, and environmental monitoring sensors. String-level sensors collect real-time voltage, current, and power parameters for each PV string. Inverter-embedded sensors collect inverter input voltage, output current, temperature, and conversion efficiency. Environmental monitoring sensors measure ambient temperature, humidity, light intensity, and dust accumulation. Data from each sensor type is timestamped (i.e., the specific time of data collection) and a unique device identifier, forming a structured dataset with temporal associations. This structured processing facilitates timeline arrangement and analysis in subsequent processing, and allows for rapid identification of data sources based on device numbers, ensuring data traceability. For example, if a string sensor records 60 sets of voltage data within a one-minute sampling period, each set is associated with a timestamp and string number, forming a string electrical characteristic dataset with {string number, time, voltage value} as the basic unit.

[0030] The collected electrical parameter and environmental parameter datasets are transmitted to the data processing unit via a hierarchical data transmission system. This hierarchical data transmission system includes a fieldbus network (such as RS-485 or CAN bus) that first aggregates sensor data to an on-site data concentrator. The data is then transmitted to the station control server via a wired or wireless local area network (LAN) and finally to the cloud processing center via an encrypted tunnel mechanism. During transmission, differential transmission technology is employed to reduce bandwidth load. This means that only fields that have changed compared to the previous data are transmitted each time. For example, if the string current value remains unchanged, it is not transmitted repeatedly; only string data with voltage changes is uploaded. Furthermore, after the data is transmitted to the cloud, it is first denoised using a triple filtering algorithm. This triple filtering algorithm typically includes a median filter (to remove impulse noise), a mean filter (to smooth small noise), and a Kalman filter (to dynamically estimate the true signal). When used in combination, this significantly improves the stability and reliability of the data. Afterwards, outlier detection methods based on statistical distributions are applied. For example, by calculating the mean and standard deviation, data points that deviate beyond ±3σ are marked as outliers. Corrections are then made based on interpolation of historical time series data to ensure consistency in data trends before and after correction. All denoised and corrected data are then normalized. Data of different dimensions (such as current in A, voltage in V, and light intensity in lx) are uniformly scaled to the interval [-1, 1]. This is accomplished using a linear scaling formula, and the normalized data constitutes a standardized dataset. Standardization ensures that all features are numerically on the same scale when subsequently fed into the neural network, avoiding gradient imbalances caused by dimensional differences during model training.

[0031] The standardized dataset was used for multidimensional feature extraction and feature engineering. First, based on the different data sources, key data points of the PV string's current-voltage (IV) characteristics (such as open-circuit voltage (Voc), short-circuit current (Isc), maximum power point (Vmp), and Imp) were extracted, along with inverter operating status characteristics (such as conversion efficiency curves and internal temperature fluctuations), and environmental characteristics (such as the daily rate of change of light intensity and temperature gradients). Next, time series features were constructed by expanding the historical values ​​of each parameter using a sliding window method to extract features such as trend changes, periodic fluctuations, and the number of mutation points, thereby forming temporal pattern features. For example, by processing inverter temperature data using a 10-minute sliding window, the temperature rise slope can be extracted, reflecting the operating status of the inverter's cooling system. Inter-device correlation features were also constructed, such as by calculating the Pearson correlation coefficient between the power output of different strings to identify the presence of local shadows or local faults. Furthermore, feature selection was performed using Lasso regression, using sparse regularization to remove redundant features and retain only a subset of features with discriminative power. Ultimately, these features were integrated into a high-dimensional feature matrix.

[0032] The above feature matrix is ​​input into a designed multi-scale spiking neural network fault diagnosis model. This model consists of a time-domain perception layer, a spatial-domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The time-domain perception layer encodes continuous feature values ​​into a pulse sequence using a biomimetic Hodgkin-Huxley neuron model, where the interval between each pulse represents the magnitude of the input feature. The spatial-domain mapping layer associates data from different sensor nodes onto a spatial mapping plane using a topology-preserving mapping (such as a SOM network), preserving the physical connections between devices. The fault mode encoding layer encodes a specific fault mode into a unique set of pulse patterns using a synaptic integration-trigger mechanism (i.e., triggering discharge based on the sum of the input pulses exceeding the neuron threshold). The decision fusion layer uses a hierarchical pulse voting mechanism to integrate the classification judgments of multiple scale features and ultimately outputs the fault type and fault location results. During the inference process, the strength of the neural connections is dynamically adjusted through a biological synaptic plasticity mechanism (i.e., spike time-dependent plasticity (STDP)), resulting in a stronger response to frequently occurring fault modes and significantly improving diagnostic accuracy.

[0033] After the diagnosis is complete, a fault assessment is performed based on the fault type and fault location information obtained. First, through electrical topology analysis methods, the fault impact range is further subdivided into specific components or connection points. For example, the possible fault transmission path is derived based on the location of the abnormal string voltage point and the electrical string parallel relationship. Subsequently, the impact of power generation loss is calculated. The percentage reduction is obtained by comparing the difference in string power generation per unit time before and after the fault occurs. At the same time, the fault propagation risk is assessed based on the interconnection relationship between components. For example, it is analyzed whether a failed string may cause the voltage drop at the inverter input to exceed the limit, triggering the inverter protection to shut down. Finally, the maintenance complexity is evaluated based on historical maintenance data statistics. For example, replacing the IGBT module inside a certain inverter model requires two people for four hours, while repairing the junction box only takes one person 30 minutes. This results in the difference in maintenance time cost. The above three dimensions are scored and the final fault severity level is calculated through weighted average calculation.

[0034] A customized repair plan is generated based on the fault type and severity level. First, the fault-repair knowledge base is searched for the most relevant standard repair process for the current fault, using semantic matching principles. For example, for a string wiring fault, the repair steps of "string terminal tightening + insulation testing + resealing" are selected. All pending faults are then prioritized based on fault severity, with repairs for those with large impacts or safety hazards prioritized. Furthermore, considering the availability of repair resources (such as the number of available engineers and spare parts inventory), the optimal repair route is replanned within resource constraints. For example, multiple minor repairs in adjacent areas can be completed consecutively within a single day to reduce repeated travel and time waste. Finally, a structured repair work order is generated, detailing the fault description, fault location, required repair tools, operating procedures, and safety precautions. The work order is then pushed to the on-site maintenance personnel's mobile device, along with a display of equipment structure and wiring diagrams, enabling them to quickly understand the fault context and efficiently execute repair operations.

[0035] In the embodiments of the present application, by combining sensor network data collection, intelligent analysis, precise fault diagnosis, and optimized maintenance decisions, not only is the fault detection accuracy of photovoltaic power stations significantly improved, but also the operation and maintenance efficiency of photovoltaic power stations and the operational stability of equipment are greatly improved through efficient maintenance resource scheduling and path planning optimization. First, by deploying a distributed sensor network at key locations in the photovoltaic power station, it is possible to collect photovoltaic strings, inverters, and environmental monitoring data in real time. Using the sensor's adaptive sampling adjustment strategy, the sampling frequency can be flexibly increased under different environmental conditions, thereby ensuring the comprehensiveness and high frequency of data collection. During the data transmission process, the application of differential transmission technology reduces the transmission of redundant data, reduces the network burden, and ensures the real-time and accuracy of data transmission. On this basis, the collected data is denoised using a triple filtering algorithm, and abnormal data is corrected by combining an outlier detection mechanism, ultimately forming a standardized data set, providing an accurate and stable foundation for subsequent intelligent analysis. Based on this standardized, high-quality data, multidimensional feature extraction and feature engineering techniques are used to extract key features of photovoltaic strings, inverters, and environmental factors. By constructing time series features and inter-device correlation features, the accuracy of fault diagnosis is effectively enhanced. By inputting into a multi-scale spiking neural network (SNN) model, this method can convert continuous-valued features into time-coded pulse sequences, accurately identifying different fault modes. Through a hierarchical decision fusion mechanism, it integrates feature information at different scales to output the final fault type and location information. Simultaneously, the pulse time-dependent plasticity mechanism dynamically adjusts the synaptic weights between neurons, effectively capturing the fault propagation path between devices, accurately locating the fault source, and assessing the risk of fault spread based on the fault characteristics, thereby achieving global fault propagation analysis. During the maintenance decision-making phase, the present invention uses semantic matching retrieval from a fault-repair knowledge base to quickly match corresponding standardized maintenance plans based on different fault types. Maintenance priorities are then assigned based on fault severity, ensuring that the most urgent faults are addressed promptly. Through resource matching and path planning optimization, the system can rationally schedule maintenance personnel's work and ensure the efficient execution of maintenance tasks. The optimized maintenance plan not only takes into account factors such as maintenance personnel's skill level, spare parts inventory, and weather conditions, but also avoids unnecessary movement of maintenance personnel by calculating the shortest maintenance path, further improving maintenance efficiency. Finally, a standardized work order is generated based on the optimized maintenance plan and sent to on-site staff in a timely manner through multiple communication channels. It is also equipped with augmented reality assistance functions to help maintenance personnel accurately identify the fault location and improve the accuracy and efficiency of maintenance.

[0036] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0037] The voltage, current and power parameters of the photovoltaic strings are collected in real time through string-level sensors to obtain string electrical characteristic data;

[0038] The built-in sensors of the inverter monitor the input voltage, output current, conversion efficiency and internal temperature of the inverter to obtain the inverter operation data;

[0039] Environmental monitoring sensors are used to collect data on ambient temperature, humidity, light intensity, and dust deposition to obtain data on environmental influencing factors;

[0040] Time-stamp the string electrical characteristic data, add the acquisition timestamp and device identification information, and obtain a time-series-correlated string data set;

[0041] Time-stamp the inverter operation data, add the acquisition timestamp and device identification information, and obtain a time-series-correlated inverter data set;

[0042] The time-series-associated string data set, time-series-associated inverter data set, and environmental influencing factor data are integrated to form structured electrical parameter data sets and environmental parameter data sets.

[0043] Specifically, in the intelligent analysis and fault diagnosis method for photovoltaic power plants, string-level sensors first collect key electrical parameters of each photovoltaic string in real time, including voltage (V), current (I), and power (P). String-level sensors are typically installed at the output end of each photovoltaic string circuit to monitor the operating status of each string in real time. Voltage data refers to the DC voltage value at the string output end at a specific moment, current data refers to the current intensity flowing through the string circuit, and power is calculated by multiplying voltage and current. This real-time data collection not only reflects the immediate power generation performance of a single string but also allows for timely detection of fault signs such as power drop and output anomalies by comparing with historical data.

[0044] Built-in sensors in the inverter continuously monitor the inverter's input voltage, output current, conversion efficiency, and internal temperature. The input voltage is typically the total DC voltage provided by multiple strings connected in parallel, and the output current is the AC output current after the inverter converts DC to AC. Conversion efficiency is defined as the ratio of output AC power to input DC power and is commonly used to evaluate the inverter's energy conversion performance. Internal temperature monitoring provides early warning of inverter overheating, as excessive temperatures can directly impact the inverter's lifespan and performance stability. For example, if the internal temperature of an inverter continues to rise and exceeds the rated safety threshold during the midday heat period, this may indicate cooling fan failure or cooling channel blockage, requiring immediate intervention. Furthermore, environmental monitoring sensors are installed in key areas of the power plant to collect real-time information on ambient temperature, air humidity, light intensity, and dust accumulation. Ambient temperature directly affects module efficiency, while humidity fluctuations can degrade electrical insulation performance. Light intensity is the most direct environmental factor affecting power generation, and dust accumulation can obscure the surface of photovoltaic panels, reducing light utilization. Light intensity is usually measured in lx (lux), and the degree of dust deposition can be measured by special sensors to measure the obstruction rate or regularly evaluate the surface contamination of components.

[0045] After data collection is complete, the string electrical characteristic data needs to be time-stamped. This means adding a data collection timestamp to each data record to identify the specific collection moment, ensuring accurate sorting and traceability during subsequent processing. Furthermore, each data entry must include device identification information, such as the string number, inverter number, or sensor number. This allows for quick identification of the specific data source within a large dataset. For example, for a string sample, the record structure should include fields such as the string number (e.g., String_001), timestamp (e.g., 2025-04-30 12:00:00), voltage value (e.g., 600V), current value (e.g., 9A), and power value (e.g., 5400W). This data with time and device identification is called time-series-associated data, which provides fundamental support for subsequent time series analysis and device traceability.

[0046] Similarly, inverter operating data also needs to undergo the same time stamping and device identification operations after collection. The recording format includes key fields such as the inverter number (such as Inv_01), collection time (such as 2025-04-30 12:00:00), input voltage (such as 600V), output current (such as 50A), conversion efficiency (such as 96%), and internal temperature (such as 45°C). Through such a unified and standardized data structure, consistency can be ensured between data collected by different data sources and different sensors, avoiding data splicing errors caused by field inconsistencies during subsequent processing.

[0047] Finally, the time-series-correlated string data sets, inverter data sets, and environmental influencing factor data are integrated. The integration process aligns data from all different sources using a unified time base, typically the shortest sampling period. For example, if string electrical data is sampled every minute and environmental data is sampled every five minutes, a time interpolation algorithm is used to interpolate the environmental data to every minute to ensure that the string, inverter, and environmental data are fully aligned at each point in time. This interpolation can be performed using linear interpolation or nearest neighbor interpolation. After data alignment, a structured data set is formed, where each time segment contains data on three types of information: string electrical characteristics, inverter operating status, and environmental parameters. This creates a unified electrical parameter data set and environmental parameter data set, which serves as the basis for subsequent feature extraction, modeling, and analysis.

[0048] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0049] The electrical parameter data set and the environmental parameter data set are transmitted from the fieldbus network to the data concentrator in the station through a hierarchical transmission architecture to obtain aggregated data;

[0050] Perform differential transmission processing on the aggregated data, only transmit the changed data, and obtain a compressed data stream;

[0051] The compressed data stream is transmitted to the cloud server through an encrypted tunnel mechanism to obtain the original cloud data;

[0052] Apply triple filtering algorithm to the original cloud data for denoising to obtain denoised data;

[0053] Perform outlier detection on the denoised data, using a dual detection mechanism of statistical methods and machine learning to identify and correct data points that deviate from the normal range to obtain cleaned data;

[0054] A standardized transformation is performed on the cleaned data, and the monitoring data of different dimensions are uniformly converted to dimensionless values ​​in the interval [-1,1] to obtain a standardized data set.

[0055] Specifically, after completing data collection from string-level sensors, inverter sensors, and environmental sensors, the electrical and environmental parameter datasets are sequentially transmitted from the fieldbus network to the in-station data concentrator via a hierarchical transmission architecture. The fieldbus network utilizes industry-standard protocols such as RS-485 or CAN to aggregate data from distributed PV strings, inverters, and environmental monitoring points to local collection nodes via wired connections. These collection nodes are responsible for initial data aggregation and caching before transmitting it to the in-station data concentrator. The data concentrator is capable of processing large volumes of real-time data, receiving multiple sensor data streams and organizing and storing them in chronological order. This hierarchical structure effectively reduces latency and packet loss associated with direct, long-distance communication along the data transmission path. It also enables local initial caching to mitigate network failures or delays.

[0056] After the in-station data concentrator receives aggregated data, it undergoes differential transmission. Differential transmission means that only the fields that have changed between sampling cycles are transmitted, rather than transmitting all fields in full each time. In this implementation, each sensor-collected data field is compared with the field value at the previous point in time. If the change is less than a preset threshold, it is recorded as "no change" and not transmitted. If the change exceeds the threshold, only the changed field and its new value are recorded and transmitted. For example, if the voltage of a string was 600 volts sampled one minute ago and 602 volts this time, if the threshold is set to 1 volt, this field will be selected for differential transmission because the change of 2 volts exceeds the threshold. Fields with minor changes, such as current fluctuations of less than 0.01 amperes, are ignored and not transmitted. This approach significantly reduces the amount of data transmitted, improving overall transmission efficiency and conserving bandwidth resources.

[0057] After differential processing, the resulting compressed data stream is transmitted to the cloud server via an encrypted tunnel. This tunnel utilizes standard encryption protocols such as IPSec or TLS to ensure confidentiality and integrity during data transmission over public networks. Each data packet undergoes end-to-end encryption before transmission, and the cloud server performs the corresponding decryption and recovery at the receiving end. If packet loss or verification errors are detected during transmission, the system automatically requests retransmission based on a built-in checksum mechanism to ensure data accuracy. All successfully received data is aggregated by the cloud server, forming the raw, unprocessed cloud data.

[0058] To improve the quality of raw cloud data, a triple filtering algorithm must first be applied to the data for denoising. Triple filtering consists of three stages: median filtering, mean filtering, and Kalman filtering. Median filtering primarily addresses sudden noise spikes. For example, if a voltage reading is abnormally high or low due to a transient disturbance, median filtering eliminates these spikes by selecting the middle value within the sampling window. Mean filtering uses the average value of a sliding window to smooth data trends and reduce high-frequency, small jitter for continuous data with small fluctuations during normal operation. Kalman filtering utilizes the estimated state at the previous moment and the current observed state to perform optimal estimation through weighted fusion, effectively eliminating slow drift noise. This method is particularly suitable for processing slowly changing data such as light intensity and ambient temperature. Triple filtering significantly reduces occasional interference, stability drift, and measurement random noise, making the data closer to the actual physical state.

[0059] After denoising, the data needs to be detected and processed for outliers. This outlier detection utilizes both statistical and machine learning methods. Statistical methods establish a parameter model based on historical normal data. For example, by calculating the mean and standard deviation of each field, the current sampled data is determined to be outside a reasonable range. Machine learning methods employ the isolation forest algorithm to train a model for detecting rare patterns, identifying data points that fall within sparsely distributed areas. Detected outliers are corrected based on the location and temporal context of the outlier. For example, if the inverter output current at a certain point in time drops below half of the normal range, while the surrounding data is stable, correction can be performed by linearly interpolating the data from the preceding and following moments to restore continuity and plausibility to the data series. Interpolation methods linearly estimate the reasonable value at the time of the anomaly based on the time interval and amplitude change between the preceding and following normal values, ensuring that the corrected data does not introduce new discontinuities or abnormal jumps.

[0060] After cleaning, the data still suffers from inconsistent dimensions, making subsequent modeling and analysis inconvenient. Therefore, it's necessary to perform a normalization transformation on the cleaned data, converting all data to a unified dimensionless standard interval, i.e., [-1, 1]. During the normalization process, the historical minimum and maximum values ​​of each feature are first calculated. Then, based on the position of each sampling point within the feature's historical range, the data is scaled to between -1 and 1. For example, if the historical minimum value of the string voltage is 580 volts and the maximum value is 620 volts, and the voltage of a single sampling point is 600 volts, then this value, after scaling, corresponds to the middle of the interval. Through this standardization, various physical quantities in PV power plants (such as voltage, light, and temperature) no longer affect subsequent feature extraction and model training due to differences in units and numerical levels, ensuring the comparability and uniformity of the numerical processing of each feature.

[0061] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0062] Extract key points of the current-voltage characteristic curve of the photovoltaic string from the standardized data set, calculate the open circuit voltage, short circuit current and maximum power point, and obtain the basic feature set of the string;

[0063] Extracting the efficiency, temperature, spectrum and stability features of the inverter from the standardized data set to obtain the inverter feature set;

[0064] Extract the ambient light intensity change rate, temperature gradient and temperature-light coupling features from the standardized data set to obtain the environmental feature set;

[0065] Based on the basic string feature set, inverter feature set and environmental feature set, time series features are constructed. The time trend, periodicity and mutation of each parameter are calculated through a sliding window to obtain the time series pattern features.

[0066] Based on the string basic feature set, inverter feature set, and environmental feature set, the device correlation features are constructed. The Pearson correlation coefficient and mutual information index between different device parameters are calculated to obtain the device correlation features.

[0067] Principal component analysis and Lasso regression are used to evaluate and screen the importance of time series pattern features and device association features, extract the most discriminative feature subset, and form a high-dimensional feature matrix.

[0068] Specifically, key points in the current-voltage characteristic curve for photovoltaic strings are extracted from a standardized dataset, including open-circuit voltage, short-circuit current, and maximum power point. Open-circuit voltage refers to the maximum voltage value measured across the string when the load is disconnected under sunlight conditions. Short-circuit current refers to the maximum current value when the string is short-circuited. The maximum power point corresponds to the voltage-current combination point with the maximum output power. To extract these features, a voltage-current relationship curve is constructed based on the actual sampling points. The points with maximum voltage, maximum current, and maximum power are located, and their corresponding values ​​are recorded, thereby constructing a basic feature set for the string.

[0069] Four types of features are extracted from the standardized dataset for inverter data. The efficiency feature describes the changing trend of the ratio of the inverter's input DC power to its output AC power. By statistically analyzing the efficiency values ​​during the sampling period, metrics such as the mean and rate of change are extracted. The temperature feature refers to the measurement results of temperature sensors at various locations within the inverter, extracting the maximum, minimum, and fluctuation amplitude. The spectrum feature is based on frequency analysis of the inverter's output current. Fourier transforms are applied to obtain the amplitudes of each harmonic order and calculate the total harmonic distortion. The stability feature is reflected by calculating the standard deviation and fluctuation amplitude of parameters such as the inverter's output current and output power over a period of time. These features are combined to form the inverter feature set. For environmental data, the light intensity change rate, temperature gradient, and temperature-light coupling features are extracted from the standardized dataset. The light intensity change rate refers to the rate of change of light intensity per unit time, the temperature gradient refers to the rate of change of ambient temperature across different regions in space, and the temperature-light coupling feature reflects the dependency between temperature and light data by calculating the correlation coefficient between them. These features are grouped together to form the environmental feature set.

[0070] When constructing time series features, a sliding window approach is used to process data based on the aforementioned string basic feature set, inverter feature set, and environmental feature set. A sliding window approach extracts statistical characteristics of data within a fixed time period, such as mean, maximum, minimum, and rate of change, while also identifying periodic patterns and mutation points. For example, using a 10-minute window length and sliding the window for 1 minute, trend analysis is performed on the feature sequence within each time period, extracting the linear growth rate, periodic frequency, and mutation frequency. These calculation results form time series pattern features, describing the dynamic characteristics of parameter changes over time.

[0071] When constructing inter-device correlation features, the Pearson correlation coefficient and mutual information index are calculated between different device parameters based on the basic string feature set, inverter feature set, and environmental feature set. The Pearson correlation coefficient measures the linear correlation between two variables, while the mutual information index measures the nonlinear correlation and information sharing between two variables. By calculating the correlation coefficients and mutual information values ​​for multiple combinations, such as between string power and inverter output power, and between string current and ambient light intensity, feature pairs with potential strong correlations between devices are screened, further enriching the feature representation capabilities.

[0072] Principal component analysis and Lasso regression are used to evaluate and screen the importance of extracted time series pattern features and device-related features. Principal component analysis reduces redundant information and improves the efficiency of feature set representation by extracting the directions with the highest variance in the feature space. Lasso regression introduces a sparse regularization term into the feature selection process, automatically zeroing out the coefficients of unimportant features to select the features with the greatest impact on the output. Combining these two methods, they select the most discriminative feature subset from the large initial feature set, ultimately forming a well-structured, information-rich, high-dimensional feature matrix.

[0073] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0074] The feature matrix is ​​input into the time domain perception layer, and the continuous value features are converted into time-coded pulse sequences through the Hodgkin-Huxley neuron model to obtain a time series pulse representation;

[0075] The temporal pulse representation is input into the spatial domain mapping layer, and the spatial relationship between devices is captured through the topology-preserving mapping network to obtain the spatially correlated pulse representation;

[0076] The spatial correlation pulse representation is input into the fault pattern encoding layer, and the specific fault mode is encoded into a unique pulse firing pattern through the synaptic integration-trigger mechanism to obtain the fault pattern pulse code;

[0077] The fault mode pulse code is input into the decision fusion layer, and the multi-scale features are integrated through the hierarchical voting mechanism to obtain the fault type judgment result;

[0078] Based on the fault type judgment result, the synaptic weights between neurons are dynamically adjusted through the pulse time-dependent plasticity rule to perform fault propagation analysis and obtain the fault location information;

[0079] The fault type judgment result and fault location information are assigned confidence scores, and the optimal diagnosis result is selected through the Winner-Take-All competition mechanism to obtain the final fault type result and fault location information.

[0080] Specifically, a feature matrix consisting of a selected subset of important features is first passed as input to the time-domain perception layer, which performs preliminary pulse encoding processing. The time-domain perception layer uses the Hodgkin-Huxley neuron model to convert the continuous-valued features in the feature matrix into a time-encoded pulse train. The Hodgkin-Huxley neuron model is a dynamic system model based on the membrane potential variation of real biological neurons, capable of dynamically adjusting the timing and frequency of pulse firing based on the input current. During this process, each feature dimension is mapped to a corresponding neuron node. The continuous value influences the neuron's firing time by regulating the charging rate of the membrane potential: larger values ​​result in faster charging and earlier pulse firing. Consequently, the originally static feature matrix is ​​converted into a sparsely distributed stream of pulse events in the time domain, forming a time-series pulse representation. This time-series pulse representation is then input to the spatial mapping layer for processing using a topology-preserving mapping network. This topology-preserving mapping network self-organizes to arrange the pulse events in a low-dimensional space, preserving the spatial relationships and relative distances between input features, allowing correlated device feature pulses to be mapped to adjacent neuron nodes. For example, pulses from different strings of the same inverter are mapped to adjacent locations in the spatial domain, reflecting their physical proximity. This step captures the structural spatial connections between devices, laying the foundation for subsequent analysis of device fault propagation paths.

[0081] The spatially correlated pulse representation is then passed to the fault mode encoding layer. This layer applies a synaptic integration-trigger mechanism to accumulate synaptic charge within neuron nodes and trigger pulse emission when the accumulated charge exceeds a set threshold. Different fault modes manifest as distinct pulse emission patterns in feature space. For example, a characteristic anomaly caused by an open-circuit component failure will cause a group of neurons to emit pulses in a specific time sequence and frequency, while a hot spot effect failure exhibits a distinct spatiotemporal pulse pattern. This mechanism dynamically encodes complex characteristic patterns into easily distinguishable pulse stream sequences, resulting in a fault mode pulse encoding.

[0082] The fault mode pulse code is then fed into the decision fusion layer. This layer employs pulse voting mechanisms at different levels based on the granularity of multi-scale features, integrating the judgment results from pulse patterns at different time and spatial scales. Specifically, the lower layers make preliminary judgments based on small-grained features within a short time window, while the higher layers make comprehensive judgments based on longer time windows and multi-device correlation features, ultimately forming a globally consistent judgment of the fault type. The output pulses from each layer are fused through a weighted voting mechanism to improve the robustness and accuracy of the diagnosis.

[0083] After the initial fault type determination is completed, pulse time-dependent plasticity adjustments are further performed based on the results to dynamically correct the synaptic weights between neurons. The pulse time-dependent plasticity rule states that if the pulse emission of one neuron occurs closely before the pulse emission of another neuron, the connection weight between the two is strengthened; if the emission order is reversed, the connection is weakened. This mechanism allows the model to adaptively adjust the network structure based on the temporal sequence of fault characteristics, thereby further inferring the fault propagation path and calculating the most likely fault starting point. By comprehensively analyzing the temporal sequence of pulse emission at each node and the trend of synaptic weight changes, the specific fault source device or string unit can be located to form fault location information.

[0084] Finally, a confidence score is assigned to each diagnostic result, combining the fault type judgment result with the fault location information. The confidence score is a comprehensive assessment based on multiple indicators such as pulse consistency, synaptic weight change amplitude, and local pulse density, reflecting the credibility of the result. Among multiple candidate diagnostic results, a Winner-Take-All competition mechanism is used to select the result with the highest confidence as the final output. The Winner-Take-All mechanism compares the confidence scores of all candidate neuron groups, retaining only the ones with the highest scores and suppressing the rest, ensuring that the final diagnostic result is unique and optimal. Through this coherent data processing process, fault type identification and fault location positioning can be completed efficiently and accurately, providing reliable support for the intelligent operation and maintenance of photovoltaic power stations.

[0085] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0086] Determine the fault scope based on the fault type and location information, and locate the fault to a specific device unit through electrical topology analysis to obtain accurate fault location results.

[0087] Establish a fault propagation diagram based on the precise fault location results, analyze the fault signal propagation path, and obtain the fault impact area;

[0088] The power generation loss impact is calculated based on the fault-affected area, and the power generation loss index is obtained by comparing the percentage difference in power generation before and after the fault.

[0089] Perform a diffusion risk assessment on the fault type results, calculate the fault diffusion probability based on the fault characteristics and device connection relationship, and obtain a diffusion risk score;

[0090] Calculate the maintenance complexity based on accurate fault location results and historical maintenance data, and obtain a maintenance complexity score by evaluating the technical difficulty, required manpower, material resources, and time cost;

[0091] By weighted fusion of power generation loss indicators, diffusion risk scores, and maintenance complexity scores, a multi-dimensional comprehensive score calculation is performed to obtain the fault severity level.

[0092] Specifically, by establishing a complete topological diagram of the power station's electrical system, the connections between the strings, inverters, combiner boxes, and step-up transformers are clearly defined. Combined with the fault type, the initially located faulty string or inverter node is traced back within the topological diagram, and the specific faulty device unit is determined based on the electrical connections. For example, if an inverter's output is abnormal, all corresponding strings and connecting cables are checked step by step along the input side. The specific fault point is further confirmed by the consistency or discontinuity of the voltage and current data. This process allows the fault location to be precisely pinpointed at the topological level, resulting in accurate fault location results.

[0093] After obtaining a precise location, a fault propagation graph is constructed based on the localization results. This graph uses device nodes as vertices and electrical connections as edges, identifying the paths that may spread from the fault source to other devices. For example, if a string short circuits, the graph will start at that string and expand along its connected combiner boxes and inverters, identifying potentially affected downstream devices. By analyzing the signal propagation path and electrical protection settings, it is possible to infer which devices will experience a chain reaction due to the fault, forming the fault-affected area. Based on the fault-affected area, the power generation loss impact is calculated. Specifically, the power generation per unit time before and after the fault is compared and the difference is calculated as a percentage of the normal power generation before the fault, thereby quantifying the degree of power generation capacity reduction. For example, if the power generation of a power plant is 500 kWh in the hour before the fault and drops to 450 kWh in the same period after the fault, the power generation loss impact is (500 - 450) / 500, or 10%. This metric directly reflects the extent of the fault's negative impact on the power plant's overall power generation performance.

[0094] Based on the fault type results, a diffusion risk assessment is conducted. This assessment considers the characteristics of the fault itself and the device connection relationships. If the fault is a string-level short circuit, and the connection structure is multiple strings connected in parallel to the same inverter, the short-circuit current may trigger inverter protection or even damage it, thus posing a higher diffusion risk. If the fault only causes power attenuation of a single component and has no parallel impact, the diffusion risk is lower. The assessment comprehensively scores the fault propagation path length, the fault electrical impact radius, and the response capability of the protection device to determine the fault diffusion probability, which in turn forms a diffusion risk score.

[0095] A repair complexity score is calculated based on accurate fault location results and historical repair data. Repair complexity primarily assesses the technical difficulty, manpower, material consumption, and time required to repair the fault. For example, a faulty IGBT module within the inverter requires specialized engineers to disassemble and repair it, along with expensive component replacement, resulting in a high repair complexity. On the other hand, a loose component connector only requires on-site inspection and reinforcement, resulting in a low repair complexity. The repair complexity of this fault is comprehensively assessed by referring to metrics such as labor hours, failure rates, and return rates from similar repairs in the past.

[0096] After obtaining the power loss index, diffusion risk score, and maintenance complexity score, a weighted fusion is performed to calculate the final fault severity level. This weighted fusion process uses the pre-set weighting ratios of the power plant operation and maintenance management system (e.g., 40% for the power loss index, 30% for the diffusion risk score, and 30% for the maintenance complexity score). Each score is multiplied by its corresponding weight and summed to produce a multi-dimensional comprehensive score. This comprehensive score range is categorized as mild, moderate, and severe, which is used to determine maintenance priorities and resource scheduling decisions.

[0097] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0098] Based on the fault type results, semantic matching retrieval is performed in the fault-repair knowledge base to extract the basic repair plan with the highest matching degree and obtain the initial repair plan;

[0099] Prioritize the initial maintenance plans according to the fault severity level, perform maintenance work queue optimization sorting, and obtain priority-marked maintenance plans;

[0100] Resource matching is performed based on the maintenance plan with priority marking, and maintenance plans under resource constraints are obtained by combining the maintenance personnel's skill level, spare parts inventory status and weather forecast information.

[0101] Perform path planning optimization on the maintenance plan under resource constraints, calculate the shortest maintenance path, and obtain the optimized maintenance plan;

[0102] Generate a standardized work order based on the optimized maintenance plan, including fault description, location information, treatment method, required tools, estimated working hours and safety precautions, to obtain a structured maintenance work order;

[0103] The structured maintenance work order is sent to the terminal devices of relevant staff through multiple communication channels of the operation and maintenance management system, and augmented reality auxiliary display data of equipment structure diagrams, wiring diagrams and maintenance guidance information is generated at the same time.

[0104] Specifically, a semantic matching search is performed within the fault-repair knowledge base based on the fault type, extracting the most suitable basic repair plan and generating a preliminary repair plan. The fault-repair knowledge base is a database containing historical fault cases, repair methods, required tools, equipment information, and repair steps. Using a semantic matching algorithm, the system compares the fault type description with the faults and repair plans in the knowledge base to identify the repair plan with the highest similarity. For example, if the fault type is "inverter overtemperature protection," the system retrieves the repair steps related to "inverter overtemperature" from the knowledge base and extracts the corresponding basic repair plan. This method automatically and quickly matches a preliminary repair plan for each fault type, providing a foundation for subsequent handling. Based on the fault severity level, the system prioritizes the preliminary repair plans. The repair priority is determined based on a comprehensive assessment of factors such as the fault's impact range, the impact of power generation losses, and the risk of escalation. For example, a large-scale fault affecting the entire power plant output will clearly have a higher priority than a minor fault affecting a single device. For faults with higher severity levels, repair work is marked as "urgent" or "priority" to ensure that the most urgent repair tasks are completed first. At the same time, the repair work queue is sorted according to the priority of each task. The system uses an optimization algorithm to assign maintenance personnel a reasonable work sequence to avoid handling multiple high-priority tasks simultaneously and ensure the rational use of resources.

[0105] After completing priority assignment, the system matches resources based on current resource availability. Resource matching for maintenance plans involves multiple factors, including the skill level of maintenance personnel, spare parts inventory levels, and environmental factors (such as weather forecasts). For example, if a fault requires highly skilled inverter repair personnel, but only low-skilled personnel are currently available, the system will match them based on their skill requirements, potentially mobilizing highly skilled personnel from other power plants. If the required spare parts are insufficient, the system will automatically connect with the inventory management system to trigger an automated procurement process to ensure timely delivery of the required parts. Weather forecasts also influence resource matching. If the forecast indicates heavy rain in the coming days, the system may prioritize equipment repairs during good weather windows to avoid delays caused by inclement weather. Once resource matching is complete, the system performs route planning optimization to determine the shortest repair route. Route planning optimization aims to reduce travel time and distance for maintenance personnel, thereby improving efficiency. This process calculates the shortest path between the repair task location and the maintenance personnel's current location, ensuring that the maintenance personnel arrive at the fault site with the least possible time and cost. For example, if multiple fault points in a power plant are relatively scattered, the system will arrange maintenance personnel to optimize their work paths based on the location of the fault points and task priorities to avoid unnecessary repeated movements.

[0106] After path planning optimization, the system will generate a standardized work order. This work order includes a detailed description of the fault, location information, handling method, required tools, estimated working time, and safety precautions. Standardized work orders ensure that each maintenance task has clear instructions and records, ensuring that maintenance personnel can operate according to established procedures and safety regulations. For example, a work order may include "Inverter fault: overtemperature protection triggered; Location: Inverter location is area B2; Handling method: Check the fan for damage and clean dust; Tools: Screwdriver, thermometer, hair dryer; Estimated working time: 1 hour; Safety precautions: Wear insulating gloves to avoid electric shock." Through the standardized format, all maintenance personnel can quickly understand the work content and follow the standardized process.

[0107] The system sends structured maintenance work orders to the terminal devices of relevant staff through various communication channels of the operation and maintenance management system. These channels include mobile application push, text messages, emails, etc., to ensure that staff can obtain the latest maintenance tasks at any time. At the same time, the system will automatically generate augmented reality (AR) auxiliary display data of equipment structure diagrams, wiring diagrams and maintenance guidance information to help on-site maintenance personnel understand the equipment structure and fault location more intuitively. Through AR technology, maintenance personnel can use mobile devices to overlay the display of equipment structure diagrams, wiring diagrams and specific maintenance steps in real time, making maintenance work more accurate and efficient. For example, when handling inverter maintenance, AR equipment can display the internal structure diagram of the inverter on the screen in real time, and highlight the parts that need to be inspected or replaced, providing real-time operation guidance, reducing human errors and improving maintenance efficiency.

[0108] The above describes the photovoltaic power station intelligent analysis and fault intelligent diagnosis method in the embodiment of the present application. The following describes the photovoltaic power station intelligent analysis and fault intelligent diagnosis system in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the photovoltaic power station intelligent analysis and fault intelligent diagnosis system includes:

[0109] The acquisition module 201 is used to collect the photovoltaic power station operation data through the sensor network to obtain the electrical parameter data set and the environmental parameter data set;

[0110] A transmission module 202 is used to transmit the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set;

[0111] A construction module 203 is used to extract multidimensional features from the standardized data set, construct correlation features, and obtain a feature matrix;

[0112] Input module 204 is used to input the feature matrix into a multi-scale spiking neural network fault diagnosis model. The multi-scale spiking neural network fault diagnosis model includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The network connection weights are dynamically adjusted through the biological synaptic plasticity mechanism to obtain fault type results and fault location information.

[0113] An evaluation module 205 is configured to perform a fault evaluation based on the fault type result and the fault location information to obtain a fault severity level;

[0114] The generating module 206 is configured to generate a maintenance plan based on the fault severity level and the fault type result, and send the plan to the terminal device via the management system.

[0115] Through the collaborative efforts of these components, combined with sensor network data collection, intelligent analysis, precise fault diagnosis, and optimized maintenance decisions, this system not only significantly improves the fault detection accuracy of photovoltaic power plants, but also significantly enhances the operational efficiency and stability of these plants through efficient maintenance resource scheduling and optimized path planning. First, a distributed sensor network deployed at key locations in the PV power plant enables real-time collection of PV strings, inverters, and environmental monitoring data. Using an adaptive sensor sampling adjustment strategy, the sampling frequency can be flexibly increased under varying environmental conditions, ensuring comprehensive and high-frequency data collection. During data transmission, differential transmission technology is applied to reduce redundant data transmission, lowering network burden and ensuring real-time and accurate data transmission. Furthermore, a triple filtering algorithm is used to denoise the collected data, and an outlier detection mechanism is used to correct abnormal data. Ultimately, a standardized dataset is generated, providing an accurate and stable foundation for subsequent intelligent analysis. Based on this standardized, high-quality data, multidimensional feature extraction and feature engineering techniques are employed to extract key features of the PV strings, inverters, and environmental factors. By constructing time series features and inter-device correlation features, the accuracy of fault diagnosis is effectively enhanced. By inputting into a multi-scale spiking neural network (SNN) model, this method can convert continuous-valued features into time-coded pulse sequences, accurately identifying different fault modes. Through a hierarchical decision fusion mechanism, it integrates feature information at different scales to output the final fault type and location information. Simultaneously, the pulse time-dependent plasticity mechanism dynamically adjusts the synaptic weights between neurons, effectively capturing the fault propagation path between devices, accurately locating the fault source, and assessing the risk of fault spread based on the fault characteristics, thereby achieving global fault propagation analysis. During the maintenance decision-making phase, the present invention uses semantic matching retrieval from a fault-repair knowledge base to quickly match corresponding standardized maintenance plans based on different fault types. Maintenance priorities are then assigned based on fault severity, ensuring that the most urgent faults are addressed promptly. Through resource matching and path planning optimization, the system can rationally schedule maintenance personnel's work and ensure the efficient execution of maintenance tasks. The optimized maintenance plan not only takes into account factors such as maintenance personnel's skill level, spare parts inventory, and weather conditions, but also avoids unnecessary movement of maintenance personnel by calculating the shortest maintenance path, further improving maintenance efficiency. Finally, a standardized work order is generated based on the optimized maintenance plan and sent to on-site staff in a timely manner through multiple communication channels. It is also equipped with augmented reality assistance functions to help maintenance personnel accurately identify the fault location and improve the accuracy and efficiency of maintenance.

[0116] above Figure 2The photovoltaic power station intelligent analysis and fault intelligent diagnosis system in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The photovoltaic power station intelligent analysis and fault intelligent diagnosis device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0117] Figure 3 FIG3 is a schematic diagram of the structure of a photovoltaic power station intelligent analysis and fault diagnosis device provided by an embodiment of the present invention. The photovoltaic power station intelligent analysis and fault diagnosis device 300 may vary significantly depending on configuration or performance. The device may include one or more central processing units (CPUs) 310 (e.g., one or more processors), memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instruction operations within the photovoltaic power station intelligent analysis and fault diagnosis device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, executing the series of instruction operations stored in the storage medium 330 on the photovoltaic power station intelligent analysis and fault diagnosis device 300 to implement the steps of the photovoltaic power station intelligent analysis and fault diagnosis method described above.

[0118] The photovoltaic power station intelligent analysis and fault intelligent diagnosis device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the photovoltaic power station intelligent analysis and fault intelligent diagnosis equipment shown does not constitute a limitation on the photovoltaic power station intelligent analysis and fault intelligent diagnosis equipment provided by the present invention, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0119] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the steps of the method for intelligent analysis and intelligent fault diagnosis of photovoltaic power stations.

[0120] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a photovoltaic power station intelligent analysis and fault intelligent diagnosis device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc., various media that can store program code.

[0122] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for intelligent analysis and fault diagnosis of photovoltaic power stations, characterized in that: The method comprises: Collect photovoltaic power station operation data through sensor networks to obtain electrical parameter data sets and environmental parameter data sets; Transmitting the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set; Extracting multidimensional features from the standardized data set, constructing correlation features, and obtaining a feature matrix; Inputting the feature matrix into a multi-scale spiking neural network fault diagnosis model, the multi-scale spiking neural network fault diagnosis model comprises a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer, dynamically adjusting network connection weights through a biological synaptic plasticity mechanism to obtain fault type results and fault location information; Performing a fault assessment based on the fault type result and the fault location information to obtain a fault severity level; A maintenance plan is generated based on the fault severity level and the fault type, and is sent to the terminal device through the management system.

2. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1 is characterized in that: The photovoltaic power station operation data is collected through the sensor network to obtain the electrical parameter data set and the environmental parameter data set, including: The voltage, current and power parameters of the photovoltaic strings are collected in real time through string-level sensors to obtain string electrical characteristic data; The built-in sensors of the inverter monitor the input voltage, output current, conversion efficiency and internal temperature of the inverter to obtain the inverter operation data; Environmental monitoring sensors are used to collect data on ambient temperature, humidity, light intensity, and dust deposition to obtain data on environmental influencing factors; Time-stamping the string electrical characteristic data, adding a collection timestamp and device identification information, and obtaining a time-series-associated string data set; Time-stamping the inverter operation data, adding a collection timestamp and device identification information, and obtaining a time-series-associated inverter data set; The time-series-associated string data set, the time-series-associated inverter data set, and the environmental influencing factor data are integrated to form a structured electrical parameter data set and an environmental parameter data set.

3. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1, characterized in that: The step of transmitting the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set includes: Transmitting the electrical parameter data set and the environmental parameter data set from the fieldbus network to an in-station data concentrator through a hierarchical transmission architecture to obtain aggregated data; Performing differential transmission processing on the collected data, transmitting only the changed data, and obtaining a compressed data stream; Transmitting the compressed data stream to the cloud server through an encrypted tunnel mechanism to obtain the original cloud data; Applying a triple filtering algorithm to the original cloud data to perform denoising processing to obtain denoised data; Performing outlier detection on the denoised data, identifying and correcting data points that deviate from the normal range based on a dual detection mechanism of statistical methods and machine learning, to obtain cleaned data; A standardized transformation is performed on the cleaned data to uniformly convert the monitoring data of different dimensions into dimensionless values ​​within the interval [-1, 1] to obtain a standardized data set.

4. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1, characterized in that: The step of extracting multidimensional features from the standardized data set, constructing associated features, and obtaining a feature matrix includes: Extracting key points of the current-voltage characteristic curve of the photovoltaic string from the standardized data set, calculating the open circuit voltage, short circuit current and maximum power point, and obtaining a basic feature set of the string; Extracting efficiency features, temperature features, spectrum features, and stability features of the inverter from the standardized data set to obtain an inverter feature set; Extracting the ambient light intensity change rate, temperature gradient, and temperature-light coupling features from the standardized data set to obtain an environmental feature set; Constructing a time series feature based on the string basic feature set, the inverter feature set, and the environmental feature set, calculating the time trend, periodicity, and mutation of each parameter through a sliding window to obtain a time series pattern feature; Building inter-device correlation features based on the string basic feature set, the inverter feature set, and the environmental feature set, calculating the Pearson correlation coefficient and mutual information index between different device parameters, and obtaining device correlation features; The importance of the time series pattern features and the device association features are evaluated and screened through principal component analysis and Lasso regression, and the most discriminative feature subset is extracted to form a high-dimensional feature matrix.

5. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1, characterized in that: The feature matrix is ​​input into a multi-scale spiking neural network fault diagnosis model, which includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The network connection weights are dynamically adjusted through the biological synaptic plasticity mechanism to obtain fault type results and fault location information, including: Inputting the feature matrix into the time domain perception layer, converting the continuous value features into a time-coded pulse sequence through the Hodgkin-Huxley neuron model to obtain a time series pulse representation; Inputting the time-series pulse representation into the spatial domain mapping layer, capturing the spatial relationship between devices through a topology-preserving mapping network, and obtaining a spatially correlated pulse representation; Inputting the spatially correlated pulse representation into the fault mode encoding layer, encoding the specific fault mode into a unique pulse emission pattern through a synaptic integration-triggering mechanism to obtain a fault mode pulse code; The fault mode pulse code is input into the decision fusion layer, and multi-scale features are integrated through a hierarchical voting mechanism to obtain a fault type judgment result; Based on the fault type judgment result, dynamically adjusting the synaptic weights between neurons through the pulse time dependent plasticity rule, performing fault propagation analysis, and obtaining fault location information; A confidence score is assigned to the fault type judgment result and the fault location information, and the optimal diagnosis result is selected through a Winner-Take-All competition mechanism to obtain the final fault type result and fault location information.

6. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1, characterized in that: The performing a fault assessment based on the fault type result and the fault location information to obtain a fault severity level includes: Determine the fault scope based on the fault type result and the fault location information, locate the fault to a specific device unit through electrical topology analysis, and obtain an accurate fault location result; Establishing a fault propagation diagram based on the precise fault location result, analyzing the fault signal propagation path, and obtaining the fault impact area; Calculating the power generation loss impact based on the fault affected area, and obtaining a power generation loss index by comparing the power generation difference percentage before and after the fault; Performing a diffusion risk assessment on the fault type results, calculating the fault diffusion probability based on the fault characteristics and device connection relationships, and obtaining a diffusion risk score; Calculating the maintenance complexity based on the precise fault location results and historical maintenance data, and obtaining a maintenance complexity score by evaluating the technical difficulty, required manpower, material resources, and time cost; By weighted fusion of the power generation loss index, the diffusion risk score and the maintenance complexity score, a multi-dimensional comprehensive score calculation is performed to obtain a fault severity level.

7. The photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to claim 1, characterized in that: Generating a maintenance plan based on the fault severity level and the fault type result and sending the plan to the terminal device through the management system includes: Perform semantic matching retrieval in the fault-repair knowledge base based on the fault type result, extract the basic repair plan with the highest matching degree, and obtain an initial repair plan; Assigning priorities to the initial maintenance plans according to the fault severity levels, performing maintenance work queue optimization sorting, and obtaining priority-marked maintenance plans; Perform resource matching based on the maintenance plan marked with the priority, and combine the maintenance personnel's skill level, spare parts inventory status and weather forecast information to obtain a maintenance plan under resource constraints; Optimizing the path planning for the maintenance plan under the resource constraints, calculating the shortest maintenance path, and obtaining an optimized maintenance plan; Generate a standardized work order based on the optimized maintenance plan, which includes fault description, location information, treatment method, required tools, estimated working hours and safety precautions, to obtain a structured maintenance work order; The structured maintenance work order is sent to the terminal devices of relevant staff through multiple communication channels of the operation and maintenance management system, and augmented reality auxiliary display data of equipment structure diagram, wiring diagram and maintenance guidance information is generated at the same time.

8. A photovoltaic power station intelligent analysis and fault intelligent diagnosis system, characterized in that: For implementing the photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to any one of claims 1 to 7, the photovoltaic power station intelligent analysis and fault intelligent diagnosis system comprises: The acquisition module is used to collect the operation data of the photovoltaic power station through the sensor network to obtain the electrical parameter data set and the environmental parameter data set; A transmission module, configured to transmit the electrical parameter data set and the environmental parameter data set to a data processing unit for data preprocessing to obtain a standardized data set; A construction module is used to extract multidimensional features from the standardized data set, construct correlation features, and obtain a feature matrix; An input module is used to input the feature matrix into a multi-scale spiking neural network fault diagnosis model. The multi-scale spiking neural network fault diagnosis model includes a time domain perception layer, a spatial domain mapping layer, a fault mode encoding layer, and a decision fusion layer. The network connection weights are dynamically adjusted through a biological synaptic plasticity mechanism to obtain fault type results and fault location information. An evaluation module, configured to perform a fault evaluation based on the fault type result and the fault location information to obtain a fault severity level; A generation module is used to generate a maintenance plan based on the fault severity level and the fault type result, and send the plan to the terminal device through the management system.

9. A photovoltaic power station intelligent analysis and fault intelligent diagnosis device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for intelligent analysis and fault intelligent diagnosis of a photovoltaic power station according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the photovoltaic power station intelligent analysis and fault intelligent diagnosis method according to any one of claims 1 to 7.

Images