Fault prediction model training method and photovoltaic equipment inspection method

Abstract

The invention discloses a fault prediction model training method and a photovoltaic equipment inspection method, which are applied to the field of artificial intelligence, and comprise the following steps: obtaining historical data and an operation state of photovoltaic equipment in a distributed photovoltaic field station; training the self-organizing mapping model by using historical data to obtain a trained self-organizing mapping model and output data; the output data is a feature identification vector corresponding to each piece of historical data; training a radial basis function neural network model by using the output data, the historical data and the operation state to obtain a trained radial basis function neural network model; the trained self-organizing mapping model and the trained radial basis function neural network model jointly construct a fault prediction model, and the output of the trained self-organizing mapping model serves as the input of the trained radial basis function neural network. According to the method, the complexity and the calculation amount are reduced, the accuracy of fault identification can be improved, the inspection efficiency and accuracy are further improved, the operation and maintenance cost is reduced, and the safety is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a fault prediction model training method and a photovoltaic equipment inspection method. Background Art

[0002] Since distributed photovoltaic stations are widely distributed and require a lot of inspection and maintenance work, traditional inspection methods require a lot of manpower and time, are accompanied by great duplication of work, and have certain safety risks.

[0003] Therefore, how to efficiently and accurately predict the faults of photovoltaic equipment in distributed photovoltaic stations, and then only need to inspect the photovoltaic equipment with failure risks, which can reduce a certain amount of manpower and material resources, save time and economic costs, and improve inspection efficiency. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a fault prediction model training method and a photovoltaic equipment inspection method, which solves the problems of poor fault identification accuracy and low inspection efficiency in the prior art.

[0005] In order to solve the above technical problems, the present invention provides a fault prediction model training method, comprising:

[0006] Obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations;

[0007] Using the historical data to train the self-organizing map model, to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data;

[0008] Using the output data, the historical data and the operating status to train a radial basis neural network model to obtain a trained radial basis neural network model;

[0009] The trained self-organizing map model and the trained radial basis function neural network model are jointly constructed to obtain a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

[0010] Optionally, after obtaining the historical data and operating status of photovoltaic equipment in the distributed photovoltaic station, the method further includes:

[0011] Analyze and make decisions on each photovoltaic device according to the historical data and the preset analysis strategy to determine the inefficient photovoltaic devices; the preset analysis strategy is any one of the power generation plan analysis strategy, the equivalent hours analysis strategy and the equipment inefficiency ranking strategy;

[0012] Correspondingly, the historical data and operating status are the historical data and operating status of the inefficient photovoltaic device.

[0013] Optionally, the self-organizing map model is trained using the historical data to obtain a trained self-organizing map model and output data, including:

[0014] Step 1: randomly assigning an initial weight vector to each neuron in the self-organizing map model to obtain a neuron weight vector corresponding to each neuron;

[0015] Step 2: Input the historical data into the self-organizing map model, calculate the distance between each historical data and each neuron weight vector, and take the neuron corresponding to the minimum distance as the target neuron;

[0016] Step 3: Determine the neighborhood range of the target neuron, and update the neuron weight vector within the neighborhood range according to the neighborhood function and the learning rate;

[0017] Step 4: Repeat steps 2 to 3 until a preset number of training times or convergence conditions are reached, and the training is completed to obtain the trained self-organizing map model and the feature identification vectors corresponding to each of the historical data.

[0018] Optionally, the output data, the historical data and the operating status are used to train a radial basis function neural network model to obtain a trained radial basis function neural network model, including:

[0019] Training the radial basis function neural network model based on the historical data and the operating status to obtain an initial radial basis function neural network model;

[0020] Training the initial radial basis function neural network model based on the output data and the operating status data to obtain the trained radial basis function neural network model;

[0021] The radial basis function of the radial basis neural network model is a Gaussian function.

[0022] Optionally, historical data of photovoltaic equipment in distributed photovoltaic stations is obtained, including:

[0023] Acquire historical operation data through edge devices; the historical operation data includes at least voltage, current, and power generation;

[0024] Acquire historical environmental data through on-site sensor equipment; the historical environmental data at least includes irradiance, temperature, wind force, and wind direction;

[0025] The historical occlusion data is acquired through video monitoring; the historical occlusion data at least includes shadow occlusion and dust occlusion data.

[0026] The present invention also provides a photovoltaic equipment inspection method, comprising:

[0027] Obtain real-time data of photovoltaic equipment in distributed photovoltaic stations;

[0028] The real-time data is input into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained using the above-mentioned fault prediction model training method;

[0029] The photovoltaic equipment corresponding to the fault prediction result greater than the preset threshold is used as the inspection equipment;

[0030] An inspection route is formulated according to the position of the inspection equipment, and each of the inspection equipment is inspected in turn according to the inspection route.

[0031] The present invention also provides a fault prediction model training device, comprising:

[0032] An acquisition module is used to obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations;

[0033] A first training module is used to train the self-organizing map model using the historical data to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data;

[0034] A second training module is used to train the radial basis neural network model using the output data, the historical data and the operating status to obtain a trained radial basis neural network model;

[0035] The model construction module is used to construct a fault prediction model by jointly using the trained self-organizing map model and the trained radial basis function neural network model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

[0036] The present invention also provides a photovoltaic equipment inspection device, comprising:

[0037] Real-time data acquisition module, used to obtain real-time data of photovoltaic equipment in distributed photovoltaic stations;

[0038] A prediction module, used to input the real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method;

[0039] An inspection equipment determination module is used to select photovoltaic equipment corresponding to a fault prediction result greater than a preset threshold as an inspection equipment;

[0040] The inspection path determination module is used to formulate an inspection path according to the position of the inspection device, and inspect each of the inspection devices in turn according to the inspection path.

[0041] The present invention also provides an electronic device, comprising:

[0042] Memory for storing computer programs;

[0043] A processor is used to implement the above method when executing the computer program.

[0044] The present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, the above method is implemented.

[0045] It can be seen that the present invention obtains the historical data and operating status of the photovoltaic equipment in the distributed photovoltaic field station; uses the historical data to train the self-organizing map model to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data; uses the output data, the historical data and the operating status to train the radial basis neural network model to obtain a trained radial basis neural network model; the trained self-organizing map model and the trained radial basis neural network model jointly construct a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis neural network. The present invention uses the self-organizing map model to map high-dimensional data to low-dimensional space, while retaining the topological structure of the data, which can reduce the complexity and amount of calculation of subsequent processing; the data after the self-organizing map model is reduced in dimension is used as the input of the radial basis neural network model. Since the dimension of the input data is reduced, the number of hidden layer nodes of the radial basis neural network model can also be reduced accordingly, making it easier to learn and identify the characteristics of the data, thereby simplifying the network structure and improving the training efficiency. In this way, fault prediction is carried out on photovoltaic equipment in distributed photovoltaic stations based on the fault prediction model, and inspections are carried out based on the prediction results, which improves inspection efficiency and accuracy, reduces operation and maintenance costs, and enhances safety.

[0046] In addition, the present invention also provides a photovoltaic equipment inspection method, a fault prediction model training device, a photovoltaic equipment inspection device, an electronic device and a computer-readable storage medium, which also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0048] Figure 1 A flowchart of a fault prediction model training method provided by an embodiment of the present invention;

[0049] Figure 2 A flow chart of a photovoltaic equipment inspection method provided by an embodiment of the present invention;

[0050] Figure 3 An example diagram of a photovoltaic equipment inspection system based on a fault prediction model provided by an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of the structure of a fault prediction model training device provided by an embodiment of the present invention;

[0052] Figure 5 A schematic diagram of the structure of a photovoltaic equipment inspection device provided by an embodiment of the present invention;

[0053] Figure 6 A schematic diagram of the structure of a fault prediction model training device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Please refer to Figure 1 , Figure 1 A flowchart of a fault prediction model training method provided by an embodiment of the present invention. The method may include:

[0056] S101: Obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations.

[0057] The photovoltaic equipment in this embodiment is the photovoltaic equipment in each distributed photovoltaic field station. This embodiment does not limit the historical data and operating status. It can be understood that the operating status is a fault state and a normal state. The historical data in this embodiment may include historical equipment data, weather data, etc. Among them, the equipment data may include inverter three-phase voltage, current, frequency, active power, reactive power, power factor, internal temperature, string voltage, current, transformer voltage, current, power generation data, etc.; weather data may include irradiance, wind force, wind speed and other environmental index data.

[0058] Furthermore, since there is non-standardized collected data and the units and dimensions of the data are different, direct use will not only affect the prediction accuracy, but also cause the training time to increase geometrically, so the data needs to be normalized so that its values ​​are between [0,1]. This embodiment can use Z-score normalization:

[0059] ;

[0060] in, is the normalized sample data, , are the mean and standard deviation respectively.

[0061] Furthermore, based on the standardized and normalized data, the eigenvalues ​​can be selected and extracted through correlation analysis and principal component analysis (PCA), and the selected and extracted data can be used to train the model.

[0062] Furthermore, obtaining historical data of photovoltaic equipment in a distributed photovoltaic station may specifically include the following steps:

[0063] Obtain historical operation data through edge devices; the historical operation data includes at least voltage, current, and power generation;

[0064] Acquire historical environmental data through on-site sensor equipment; the historical environmental data at least includes irradiance, temperature, wind force, and wind direction;

[0065] The historical occlusion data is obtained through video monitoring; the historical occlusion data at least includes shadow occlusion and dust occlusion data.

[0066] This embodiment obtains various data from multiple angles, such as operating data, environmental data, and shading data, and analyzes the operating status of each photovoltaic device in the distributed photovoltaic field station and the influencing factors of various operating statuses from multiple aspects. Specifically, historical operating data such as voltage, current, power generation, temperature, etc. can be obtained from edge devices through the Internet of Things protocol, and irradiance, temperature, wind force, wind direction and other data can be obtained through sensor devices installed in distributed photovoltaic fields. Data such as shadow shading and dust shading can be obtained through monitoring installed in distributed photovoltaic fields. Among them, edge devices usually refer to those devices located at the edge of the system, responsible for collecting and preliminarily processing data, and then transmitting the data to the central processing system.

[0067] Furthermore, after obtaining the historical data and operating status of the photovoltaic equipment in the distributed photovoltaic station, the following steps may also be included:

[0068] An analysis decision is made on each photovoltaic device based on historical data and a preset analysis strategy to determine inefficient photovoltaic devices; the preset analysis strategy is any one of a power generation plan analysis strategy, an equivalent hours analysis strategy, and an equipment inefficiency ranking strategy; correspondingly, the historical data and operating status are the historical data and operating status of the inefficient photovoltaic devices.

[0069] It should be noted that this embodiment first screens each photovoltaic device in the distributed photovoltaic station to screen out inefficient devices, so that targeted analysis can be performed on inefficient devices. For the criteria for distinguishing inefficient devices, that is, the specific analysis decision, it can be a power generation plan analysis strategy; it can also be an equivalent hours analysis strategy, or it can also be an equipment equivalent ranking strategy. Specifically, the power generation plan analysis strategy refers to obtaining a power generation forecast value through a historical theoretical power generation algorithm to determine whether the current power generation completion rate meets the standard. The equivalent hours analysis strategy refers to sorting the daily equivalent hours of all stations in descending order, taking the average value of the daily equivalent hours of 45%-50% in order * coefficient (configurable), and obtaining a judgment value, and defining the power station that is less than the judgment value as an inefficient device. The equipment inefficiency ranking strategy refers to sorting the power generation efficiency of the equipment from high to low, and treating the following 20%-30% of the power stations as inefficient equipment.

[0070] S102: Use historical data to train the self-organizing map model to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each historical data.

[0071] The reasons for the inefficiency of distributed photovoltaic stations are complex and diverse, such as component damage, line failure, shadow occlusion, cloudy, rainy weather and other nonlinear relationships. This embodiment uses SOM (Self-Organizing Map) to pre-process historical data, map high-dimensional data to low-dimensional space, and retain the topological structure of the data, which can reduce the complexity and amount of calculation of subsequent processing. The data after dimensionality reduction of the self-organizing map model is used as the input of the radial basis neural network model. Since the dimension of the input data is reduced, the number of hidden layer nodes of the radial basis neural network model can also be reduced accordingly, making it easier to learn and identify the characteristics of the data, thereby simplifying the network structure and improving the training efficiency. The self-organizing map model is trained using historical data. The historical data in this embodiment can be the historical data of photovoltaic equipment in all distributed photovoltaic stations; of course, it can further be the historical data of inefficient equipment in distributed photovoltaic stations. The self-organizing map model is a relatively special competitive learning type unsupervised neural network. During the training process, it can automatically learn the characteristics of the input data and map similar features to adjacent neurons. This self-organizing characteristic makes SOM an effective feature extraction tool; the output of the model is the feature identification vector corresponding to each input data.

[0072] Furthermore, the above-mentioned use of historical data to train the self-organizing map model to obtain a trained self-organizing map model and output data includes:

[0073] Step 1: Randomly assign an initial weight vector to each neuron in the self-organizing map model to obtain the neuron weight vector corresponding to each neuron;

[0074] Step 2: Input the historical data into the self-organizing map model, calculate the distance between each historical data and each neuron weight vector, and take the neuron corresponding to the minimum distance as the target neuron;

[0075] Step 3: Determine the neighborhood range of the target neuron, and update the neuron weight vector within the neighborhood range according to the neighborhood function and learning rate;

[0076] Step 4: Repeat steps 2 to 3 until the preset number of training times or convergence conditions are reached, the training is completed, and a trained self-organizing map model and a feature identification vector corresponding to each historical data are obtained.

[0077] Specifically, the two-dimensional grid topology of the self-organizing map model can be determined according to the size and complexity of the data set, and the weight vector of each neuron can be randomly initialized; wherein, each neuron represents a data feature. The competitive learning process in the training process of the model is mainly divided into three steps. First, an input vector (i.e., input historical data) is randomly selected from the data set as the selected input sample, and then the distance between the input vector and all neuron weight vectors (referring to the Euclidean distance) is calculated, and the neuron with the smallest distance is selected as the target neuron (Best Matching Unit, BMU). Subsequently, a neighborhood function is defined with BMU as the center (the neighborhood function also represents the neighborhood range). As the distance increases, the weight gradually decreases, and the weights of the neurons in the BMU and its neighborhood are updated to make them closer to the input vector. The above steps are repeated until the predetermined number of iterations is reached or the weight change is less than a certain threshold, then the training ends and the trained self-organizing map model is obtained. Correspondingly, the output data of the self-organizing map model, i.e., the feature identification vector corresponding to each historical data, can be obtained according to the trained self-organizing map model and historical data.

[0078] Among them, the neuron weight vector formula is as follows:

[0079] ;

[0080] in, is the learning rate, which decreases gradually over time; is the neighborhood function; is the input vector; is the weight vector of the previous neuron, is the updated weight vector of the next neuron.

[0081] According to the feature identification vector output by the self-organizing map model, the clustering results of each input sample can be reflected, and a U-Matrix diagram can be drawn to show the distance between neurons. The data point features in each cluster are analyzed, and each feature identification vector is linked to the operating status and failure mode of the photovoltaic equipment to determine which clusters represent normal conditions and which clusters may represent potential failures or abnormal conditions, such as component attenuation, dust, rain shielding, temperature reduction, transformer power loss, anti-reverse flow protection control, etc. Subsequently, the feature identification vector corresponding to each historical data is used as the input of the radial basis neural network model to help the radial basis neural network model focus more on the classification task itself without spending too much energy on feature extraction.

[0082] S103: Train the radial basis function neural network model using the output data, historical data and operating status to obtain a trained radial basis function neural network model.

[0083] The nonlinear relationship of the sample data in this embodiment is complex, and the radial basis neural network model has a strong nonlinear fitting ability, can map any complex nonlinear relationship, has a simple learning rule, is easy to implement by computer, and is suitable for building a prediction model. Therefore, this embodiment adopts a radial basis neural network model. Specifically, RBFNN (Radial Basis Function Neural Network) is a feedforward neural network with excellent performance, which can approximate any nonlinear function with arbitrary accuracy, has global approximation ability, compact topological structure, and structural parameters can be separated for learning, with fast convergence speed. In addition, compared with other neural networks, the radial basis neural network model has the following advantages: (1) It has relatively good approximation characteristics and no local minimum problem exists. (2) It has a strong input and output mapping function. (3) The network connection weight is linearly related to the output. (4) The classification ability is good. (5) The learning process converges quickly and is easy to quickly train and put into the environment.

[0084] Furthermore, the above-mentioned use of the output data, historical data and operating status to train the radial basis function neural network model to obtain a trained radial basis function neural network model may specifically include the following steps:

[0085] The radial basis function neural network model is trained based on historical data and operating status to obtain an initial radial basis function neural network model;

[0086] Training the initial radial basis function neural network model based on the output data and the operation status data to obtain the trained radial basis function neural network model;

[0087] The radial basis function of the radial basis neural network model is a Gaussian function.

[0088] This embodiment first uses historical data and operating status data to train the radial basis neural network model to obtain an initial radial basis neural network model, and then further uses the output data and operating status data output by the self-organizing map model for secondary training to further improve the accuracy of the radial basis neural network model and optimize its network structure.

[0089] The radial basis neural network is a three-layer feedforward neural network, including an input layer, a hidden layer and an output layer. The input layer is mainly used to receive data from the outside and pass it to the hidden layer. The data input in this embodiment comes from the output of the self-organizing map model and historical data. The hidden layer is the core part of the radial basis neural network model, which is composed of multiple radial basis neurons, each of which is associated with a radial basis function (nonlinear function). The input data is transformed nonlinearly by the radial basis function, and the linearly inseparable data in the low-dimensional space is mapped to the high-dimensional space, so that it becomes linearly separable. Each radial basis function has a center vector and a standard deviation to determine the shape and size of the function. When the input vector enters the hidden layer, each neuron calculates the distance between the input vector and its own center vector, and obtains the output through the radial basis function conversion. The output layer maps the high-dimensional features of the hidden layer to the output space through linear transformation, and adjusts the weight coefficient through continuous training to minimize the error between the actual output and the expected output. The output of the hidden layer is weighted and summed to output the linear fault prediction result.

[0090] Among them, the radial basis function is a real-valued function whose value depends only on the distance from the origin, that is, Φ(x)=Φ(‖x‖), or it can also be the distance to any point c, point c is called the center point, that is, Φ(x,c)=Φ(‖xc‖). Any function Φ that satisfies the characteristic of Φ(x)=Φ(‖x‖) is called a radial basis function. This embodiment uses a Gaussian function with better performance.

[0091] S104: The trained self-organizing map model and the trained radial basis function neural network model are jointly constructed to obtain a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

[0092] In this embodiment, the fault prediction model is composed of a self-organizing map model and a radial basis function neural network model, and the output of the self-organizing map model serves as the input of the radial basis function neural network model.

[0093] The fault prediction model training method provided by the embodiment of the present invention is applied, by obtaining the historical data and operating status of the photovoltaic equipment in the distributed photovoltaic field station; using the historical data to train the self-organizing map model to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data; using the output data, the historical data and the operating status to train the radial basis neural network model to obtain a trained radial basis neural network model; the trained self-organizing map model and the trained radial basis neural network model are jointly constructed to obtain a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis neural network. The present invention uses the self-organizing map model to map high-dimensional data to a low-dimensional space while retaining the topological structure of the data, which can reduce the complexity and amount of calculation of subsequent processing; the data after the dimensionality reduction of the self-organizing map model is used as the input of the radial basis neural network model. Since the dimension of the input data is reduced, the number of hidden layer nodes of the radial basis neural network model can also be reduced accordingly, making it easier to learn and identify the characteristics of the data, thereby simplifying the network structure and improving the training efficiency. In this way, fault prediction is carried out on photovoltaic equipment in distributed photovoltaic stations based on the fault prediction model, and inspections are carried out based on the prediction results, which improves inspection efficiency and accuracy, reduces operation and maintenance costs, and enhances safety.

[0094] Please refer to Figure 2 , Figure 2 A flowchart of a photovoltaic device inspection method provided by an embodiment of the present invention. The method may include:

[0095] S201: Acquire real-time data of photovoltaic equipment in distributed photovoltaic stations.

[0096] The executor of this embodiment is a terminal. The terminal in this embodiment can be an operation and maintenance platform. The operation and maintenance platform collects real-time operation data of photovoltaic equipment in distributed photovoltaic stations from microgrid controllers or inverters through the Internet of Things protocol. Common protocols include MQTT, Modbus, IEC104, etc. Specifically, real-time data such as voltage, current, power generation, temperature, etc. are obtained from edge devices through the Internet of Things protocol, and real-time data such as irradiance, temperature, wind force, wind direction, etc. are obtained through sensor equipment installed in distributed photovoltaic stations. Real-time data such as shadow occlusion and dust occlusion can be obtained through monitoring installed in distributed photovoltaic stations.

[0097] S202: inputting real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method.

[0098] The data acquired in the above step S101 is input into the fault prediction model obtained by the above fault prediction model training method, and a fault prediction result is output. It should be noted that the fault prediction result in this embodiment is a fault prediction probability value.

[0099] S203: The photovoltaic devices corresponding to the fault prediction results greater than the preset threshold are used as inspection devices.

[0100] This embodiment does not limit the preset threshold value, and it can be set according to actual conditions. The preset threshold value is a fault alarm threshold value. If it exceeds the threshold value, it means that the fault risk is too high, and an alarm prompt needs to be issued and certain measures need to be taken. Specifically, it is used as a patrol inspection device.

[0101] S204: Formulate an inspection route according to the location of the inspection device, and inspect each inspection device in turn according to the inspection route.

[0102] Specifically, the operation and maintenance platform can automatically push alarm information that exceeds the preset threshold to the operation and maintenance administrator. The operation and maintenance administrator can log in to the cloud platform to realize the linkage between alarms and inspection work orders, and set inspection routes according to the equipment failure rate. The operation and maintenance personnel complete the inspection tasks according to the inspection work order tasks issued, and submit the inspection records. The operation and maintenance administrator can track the inspection progress in real time.

[0103] The inspection method for photovoltaic equipment provided by the embodiment of the present invention is applied, by obtaining real-time data of photovoltaic equipment in distributed photovoltaic stations; inputting real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method; the photovoltaic equipment corresponding to the fault prediction result greater than the preset threshold is used as the inspection equipment; the inspection path is formulated according to the location of the inspection equipment, and each inspection equipment is inspected in turn according to the inspection path. The present invention can make fault warnings in advance according to real-time data through the fault prediction model, and notify the user to solve it in time, effectively reducing the inefficiency of equipment caused by faults and improving the power generation efficiency of the station. At the same time, the alarm push can help the operation and maintenance personnel accurately locate the fault location, reduce the frequency and intensity of manual inspections, and reduce the operation and maintenance costs. For some remote areas, when the fault prediction model predicts that there is no risk of failure, there is no need to conduct manual inspections on the station, avoiding the risk of personnel working at high altitudes.

[0104] Furthermore, this embodiment can automatically record the operation data and maintenance records of the station, and generate detailed inspection reports to provide data support for station managers. Through in-depth analysis of the data, the system can provide a scientific basis for the operation and maintenance decisions of the station, optimize the operation and maintenance strategy, and improve the overall operational efficiency of the station. The successful application of this system will provide other photovoltaic stations with experience and demonstration effects that can be used as reference, and promote the entire photovoltaic industry to develop in a more intelligent and automated direction.

[0105] In order to make the present invention easier to understand, please refer to Figure 3 , Figure 3 An example diagram of a photovoltaic equipment inspection system based on a fault prediction model provided in an embodiment of the present invention may specifically include: a data acquisition module, a decision analysis module, an intelligent prediction module and an operation and maintenance management module.

[0106] (1) The data acquisition module includes site management, equipment management, point management, communication components, space model and gateway management. It supports common IoT protocols and edge devices for data delivery, including MQTT, Modbus, IEC104, etc. Specifically, the IoT protocol is used to collect real-time operation data of photovoltaic equipment from the microgrid controller or inverter, including inverter three-phase voltage, current, frequency, active power, reactive power, power factor, internal temperature, string voltage, current, transformer voltage, current, power generation, irradiance, wind force, wind speed and other environmental indicators.

[0107] (2) The decision analysis module includes power station monitoring (photovoltaic power station monitoring), equipment monitoring, power station analysis report, power generation plan analysis, equivalent hours analysis, hourly power analysis, and data curves. Specifically, the decision analysis module selects inefficient equipment as the equipment input of the intelligent prediction module through power generation plan analysis, equivalent hours analysis, and equipment inefficiency ranking.

[0108] (3) The intelligent prediction module includes a fault prediction model, which includes a self-organizing map model and a radial basis neural network model. The intelligent prediction module first standardizes and normalizes the inefficient equipment data, and then selects and extracts the feature values ​​of the data through the self-organizing map model to form cluster analysis data. The feature data output of the self-organizing map model is used as the input of the radial basis neural network model to help the radial basis neural network model learn and identify the characteristics of the data more easily. The radial basis neural network model uses the radial basis function to perform nonlinear transformation on the input data, mapping the linearly inseparable data in the low-dimensional space to the high-dimensional space, thereby making it linearly separable. Finally, the output layer weights the output of the hidden layer and outputs the linear fault prediction result.

[0109] (4) The operation and maintenance management module includes functions such as personnel certificate management, alarm management, inspection tasks, inspection routes, and work order management. Thresholds are set for predicted fault results. If the threshold is exceeded, an alarm will be generated and pushed to the operation and maintenance administrator; the operation and maintenance administrator assigns the inspection work order to the operation and maintenance personnel by converting the alarm information into a work order, and generates an inspection route based on the location of the fault site; the operation and maintenance personnel perform inspection tasks according to the inspection route and submit inspection records. If the equipment is damaged and needs to be replaced, the spare parts application must be submitted for entry and exit. The operation and maintenance administrator can track the inspection status online in real time and approve the spare parts plan.

[0110] The system supports artificial intelligence prediction as the main means, supplemented by manual configuration and adding modules. It can quickly locate photovoltaic equipment that needs to be inspected for enterprises, so as to achieve the purpose of selective inspection, thereby reducing time and space waste, improving efficiency and saving labor costs; helping users improve inspection efficiency and accuracy, reduce operation and maintenance costs, and enhance operation and maintenance safety. Among them, normalization processing can eliminate the dimension and unit differences in the data; the self-organizing mapping model method is used for competitive clustering to obtain the optimal data characteristics. The radial basis neural network has the advantages of better approximation characteristics and faster training, so it is more suitable for non-long-term prediction applications. The photovoltaic equipment inspection system and method based on the fault prediction model have brought significant beneficial effects in improving inspection efficiency and accuracy, reducing operation and maintenance costs, enhancing safety, data management and decision support, and promoting the intelligent development of the photovoltaic industry.

[0111] The following is an introduction to a fault prediction model training device provided in an embodiment of the present invention. The fault prediction model training device described below and the fault prediction model training method described above can be referenced to each other.

[0112] Please refer to Figure 4 , Figure 4 A structural diagram of a fault prediction model training device provided by an embodiment of the present invention may include:

[0113] The acquisition module 100 is used to acquire the historical data and operating status of photovoltaic equipment in the distributed photovoltaic station;

[0114] A first training module 200 is used to train the self-organizing map model using the historical data to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data;

[0115] A second training module 300 is used to train the radial basis neural network model using the output data, the historical data and the operating status to obtain a trained radial basis neural network model;

[0116] The model construction module 400 is used to construct a fault prediction model by jointly constructing the trained self-organizing map model and the trained radial basis function neural network model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

[0117] Based on the above embodiment, the fault prediction model training device may further include:

[0118] The analysis and decision-making module is used to analyze and make decisions on each photovoltaic device according to the historical data and the preset analysis strategy to determine the inefficient photovoltaic devices; the preset analysis strategy is any one of the power generation plan analysis strategy, the equivalent hours analysis strategy and the equipment inefficiency ranking strategy; correspondingly, the historical data and operating status are the historical data and operating status of the inefficient photovoltaic devices.

[0119] Based on the above embodiment, the first training module 200 may include:

[0120] An initialization unit is used to perform step 1: randomly assign an initial weight vector to each neuron in the self-organizing map model to obtain a neuron weight vector corresponding to each neuron;

[0121] A distance calculation unit, used to perform step 2: input the historical data into the self-organizing map model, calculate the distance between each historical data and each neuron weight vector, and take the neuron corresponding to the minimum distance as the target neuron;

[0122] An updating unit, configured to execute step 3: determining a neighborhood range of the target neuron, and updating a neuron weight vector within the neighborhood range according to a neighborhood function and a learning rate;

[0123] The training end unit is used to execute step 4: repeatedly execute steps 2 to 3 until a preset number of training times or convergence conditions are reached, and the training ends to obtain the trained self-organizing map model and the feature identification vectors corresponding to each of the historical data.

[0124] Based on the above embodiment, the second training module 300 may include:

[0125] A first training unit is used to train the radial basis function neural network model based on the historical data and the operating state to obtain an initial radial basis function neural network model;

[0126] The second training unit is used to train the initial radial basis neural network model based on the output data and the operating status data to obtain the trained radial basis neural network model; wherein the radial basis function of the radial basis neural network model is a Gaussian function.

[0127] Based on the above embodiment, the acquisition module 100 may include:

[0128] A first acquisition unit is used to acquire historical operation data through an edge device; the historical operation data at least includes voltage, current, and power generation;

[0129] A second acquisition unit is used to acquire historical environmental data through on-site sensor equipment; the historical environmental data at least includes irradiance, temperature, wind force, and wind direction;

[0130] The third acquisition unit is used to acquire historical occlusion data through video monitoring; the historical occlusion data at least includes shadow occlusion and dust occlusion data.

[0131] It should be noted that the order of the modules and units in the above-mentioned fault prediction model training device can be changed without affecting the logic.

[0132] The fault prediction model training device provided by the embodiment of the present invention is used to obtain the historical data and operating status of the photovoltaic equipment in the distributed photovoltaic field station through the acquisition module 100; the first training module 200 is used to train the self-organizing map model using the historical data to obtain the trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data; the second training module 300 is used to train the radial basis neural network model using the output data, the historical data and the operating status to obtain the trained radial basis neural network model; the model construction module 400 is used to jointly construct the trained self-organizing map model and the trained radial basis neural network model to obtain a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis neural network. This device uses a self-organizing map model to map high-dimensional data to a low-dimensional space while retaining the topological structure of the data, which can reduce the complexity and amount of calculation of subsequent processing; the data after the self-organizing map model is reduced in dimension is used as the input of the radial basis neural network model. Since the dimension of the input data is reduced, the number of hidden layer nodes of the radial basis neural network model can also be reduced accordingly, making it easier to learn and identify the characteristics of the data, thereby simplifying the network structure and improving the training efficiency. In this way, the fault prediction model is used to predict the faults of photovoltaic equipment in distributed photovoltaic stations, and inspections are carried out based on the prediction results, which improves the efficiency and accuracy of inspections, reduces operation and maintenance costs, and enhances safety.

[0133] The following is an introduction to the inspection device of the photovoltaic equipment provided by the embodiment of the present invention. The inspection device of the photovoltaic equipment described below and the inspection method of the photovoltaic equipment described above can be referred to each other.

[0134] Please refer to Figure 5 , Figure 5 A schematic structural diagram of a photovoltaic equipment inspection device provided by an embodiment of the present invention may include:

[0135] A real-time data acquisition module 500 is used to acquire real-time data of photovoltaic equipment in a distributed photovoltaic station;

[0136] The prediction module 600 is used to input the real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method;

[0137] Inspection equipment determination module 700, used to select photovoltaic equipment corresponding to a fault prediction result greater than a preset threshold as an inspection equipment;

[0138] The inspection path determination module 800 is used to formulate an inspection path according to the position of the inspection device, and inspect each of the inspection devices in turn according to the inspection path.

[0139] It should be noted that the order of the modules and units in the inspection device for the photovoltaic equipment described above can be changed without affecting the logic.

[0140] The inspection device for photovoltaic equipment provided by the embodiment of the present invention comprises a real-time data acquisition module 500 for acquiring real-time data of photovoltaic equipment in a distributed photovoltaic field station; a prediction module 600 for inputting the real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method; an inspection device determination module 700 for taking the photovoltaic equipment corresponding to the fault prediction result greater than a preset threshold as the inspection device; and an inspection path determination module 800 for formulating an inspection path according to the position of the inspection device, and inspecting each of the inspection devices in turn according to the inspection path. The device uses a self-organizing map model to map high-dimensional data to a low-dimensional space while retaining the topological structure of the data, which can reduce the complexity and amount of calculation of subsequent processing; the data after the dimensionality reduction of the self-organizing map model is used as the input of the radial basis neural network model. Since the dimension of the input data is reduced, the number of hidden layer nodes of the radial basis neural network model can also be reduced accordingly, making it easier to learn and identify the characteristics of the data, thereby simplifying the network structure and improving the training efficiency. In this way, the fault prediction model is used to predict the faults of photovoltaic equipment in distributed photovoltaic stations, and inspections are carried out based on the prediction results, which improves the efficiency and accuracy of inspections, reduces operation and maintenance costs, and enhances safety. In addition, based on the fault prediction model, fault warnings are made in advance according to the collected data, and users are notified to solve them in time, effectively reducing the inefficiency of equipment caused by faults and improving the power generation efficiency of power stations. At the same time, alarm push can help operation and maintenance personnel accurately locate the fault location, reduce the frequency and intensity of manual inspections, and reduce operation and maintenance costs. For some areas that are difficult to reach manually, fault prediction replaces manual inspections to avoid the risk of personnel working at heights.

[0141] The electronic device provided by the embodiment of the present invention is introduced below. The electronic device described below and the fault prediction model training method and / or the inspection method of photovoltaic equipment described above can be referenced to each other.

[0142] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention may include:

[0143] A memory 10, used for storing computer programs;

[0144] The processor 20 is used to execute a computer program to implement the above-mentioned fault prediction model training method and / or photovoltaic equipment inspection method.

[0145] The memory 10 , the processor 20 , and the communication interface 31 all communicate with each other via the communication bus 32 .

[0146] In the embodiment of the present invention, the memory 10 is used to store one or more programs, and the program may include program code, and the program code includes computer operation instructions. In the embodiment of the present invention, the memory 10 may store programs for implementing the following functions:

[0147] Obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations;

[0148] The self-organizing map model is trained using historical data to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each historical data;

[0149] The radial basis function neural network model is trained by using the output data, historical data and operating status to obtain a trained radial basis function neural network model;

[0150] The trained self-organizing map model and the trained radial basis function neural network model are used together to construct a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network;

[0151] and / or;

[0152] Obtain real-time data of photovoltaic equipment in distributed photovoltaic stations;

[0153] Inputting real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained according to the above-mentioned fault prediction model training method;

[0154] The photovoltaic equipment corresponding to the fault prediction result greater than the preset threshold is used as the inspection equipment;

[0155] An inspection route is formulated according to the location of the inspection equipment, and each inspection equipment is inspected in turn according to the inspection route.

[0156] In a possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function, etc.; the data storage area may store data created during use.

[0157] In addition, the memory 10 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include an NVRAM. The memory stores an operating system and operating instructions, executable modules or data structures, or a subset thereof, or an extended set thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0158] The processor 20 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic device, a microprocessor or any conventional processor, etc. The processor 20 may call a program stored in the memory 10 .

[0159] The communication interface 31 may be an interface of a communication module, and is used to connect to other devices or systems.

[0160] Of course, it should be noted that Figure 6 The structure shown does not constitute a limitation on the electronic device in the embodiment of the present invention. In actual applications, the electronic device may include Figure 6 More or fewer components than shown, or combinations of certain components.

[0161] The computer-readable storage medium provided in the embodiments of the present invention is introduced below. The computer-readable storage medium described below and the method described above can be referenced to each other.

[0162] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0163] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0164] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0165] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0166] Finally, it should be noted that, in this article, relationships such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0167] The above is a detailed introduction to a fault prediction model training method, a photovoltaic equipment inspection method, a device, a device and a computer-readable storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A fault prediction model training method, characterized in that: include: Obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations; Using the historical data to train the self-organizing map model, to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data; Using the output data, the historical data and the operating status to train a radial basis neural network model to obtain a trained radial basis neural network model; The trained self-organizing map model and the trained radial basis function neural network model are jointly constructed to obtain a fault prediction model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

2. The fault prediction model training method according to claim 1, characterized in that: After obtaining the historical data and operating status of photovoltaic equipment in distributed photovoltaic stations, it also includes: Analyze and make decisions on each photovoltaic device according to the historical data and the preset analysis strategy to determine the inefficient photovoltaic devices; the preset analysis strategy is any one of the power generation plan analysis strategy, the equivalent hours analysis strategy and the equipment inefficiency ranking strategy; Correspondingly, the historical data and operating status are the historical data and operating status of the inefficient photovoltaic device.

3. The fault prediction model training method according to claim 1, characterized in that: The self-organizing map model is trained using the historical data to obtain a trained self-organizing map model and output data, including: Step 1: randomly assigning an initial weight vector to each neuron in the self-organizing map model to obtain a neuron weight vector corresponding to each neuron; Step 2: Input the historical data into the self-organizing map model, calculate the distance between each historical data and each neuron weight vector, and take the neuron corresponding to the minimum distance as the target neuron; Step 3: Determine the neighborhood range of the target neuron, and update the neuron weight vector within the neighborhood range according to the neighborhood function and the learning rate; Step 4: Repeat steps 2 to 3 until a preset number of training times or convergence conditions are reached, and the training is completed to obtain the trained self-organizing map model and the feature identification vectors corresponding to each of the historical data.

4. The fault prediction model training method according to claim 1, characterized in that: The radial basis function neural network model is trained using the output data, the historical data and the operating status to obtain a trained radial basis function neural network model, including: Training the radial basis function neural network model based on the historical data and the operating status to obtain an initial radial basis function neural network model; Training the initial radial basis function neural network model based on the output data and the operating status data to obtain the trained radial basis function neural network model; The radial basis function of the radial basis neural network model is a Gaussian function.

5. The fault prediction model training method according to claim 1, characterized in that: Obtain historical data of photovoltaic equipment in distributed photovoltaic sites, including: Acquire historical operation data through edge devices; the historical operation data includes at least voltage, current, and power generation; Acquire historical environmental data through on-site sensor equipment; the historical environmental data at least includes irradiance, temperature, wind force, and wind direction; The historical occlusion data is acquired through video monitoring; the historical occlusion data at least includes shadow occlusion and dust occlusion data.

6. A method for inspecting photovoltaic equipment, characterized in that: include: Obtain real-time data of photovoltaic equipment in distributed photovoltaic stations; Inputting the real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained by the fault prediction model training method according to any one of claims 1 to 5; The photovoltaic equipment corresponding to the fault prediction result greater than the preset threshold is used as the inspection equipment; An inspection route is formulated according to the position of the inspection equipment, and each of the inspection equipment is inspected in turn according to the inspection route.

7. A fault prediction model training device, characterized in that: include: An acquisition module is used to obtain historical data and operating status of photovoltaic equipment in distributed photovoltaic stations; A first training module is used to train the self-organizing map model using the historical data to obtain a trained self-organizing map model and output data; the output data is a feature identification vector corresponding to each of the historical data; A second training module is used to train the radial basis neural network model using the output data, the historical data and the operating status to obtain a trained radial basis neural network model; The model construction module is used to construct a fault prediction model by jointly using the trained self-organizing map model and the trained radial basis function neural network model, and the output of the trained self-organizing map model is used as the input of the trained radial basis function neural network.

8. A photovoltaic equipment inspection device, characterized in that: include: Real-time data acquisition module, used to obtain real-time data of photovoltaic equipment in distributed photovoltaic stations; A prediction module, used for inputting the real-time data into a fault prediction model to obtain a fault prediction result; the fault prediction model is a model trained by the fault prediction model training method according to any one of claims 1 to 5; An inspection equipment determination module is used to select photovoltaic equipment corresponding to a fault prediction result greater than a preset threshold as an inspection equipment; The inspection path determination module is used to formulate an inspection path according to the position of the inspection device, and inspect each of the inspection devices in turn according to the inspection path.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the method according to any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images