Photovoltaic module power generation system data acquisition control method and system

Abstract

The present application provides a data acquisition control method and system for a photovoltaic module power generation system. According to the fault types corresponding to Y fault protection trigger events and the order in which the Y fault protection trigger events are triggered, a fault protection trigger graph structure including X graph vertices is constructed with the fault types as graph vertices. For a fault protection trigger event tuple consisting of the mth fault protection trigger event and the nth fault protection trigger event, when the mth fault protection trigger event is the first fault type, and the nth fault protection trigger event is the second fault type, there is a target connection line from the first graph vertex to the second graph vertex in the fault protection trigger graph structure, and the first graph vertex is used to indicate the first fault type, and the second graph vertex is used to indicate the second fault type. The present application can improve the accuracy of identifying false triggering of protection actions.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a data acquisition control method and system for a photovoltaic module power generation system. Background Art

[0002] As the global demand for clean energy continues to grow, photovoltaic power generation systems have been widely used as a sustainable energy solution. Photovoltaic power generation systems provide a renewable energy source by converting solar energy into electrical energy. However, photovoltaic power generation systems face various challenges in actual operation, among which fault protection is particularly critical. Since photovoltaic power generation systems are composed of many photovoltaic modules, inverters, combiner boxes, connecting lines and other components, their structure is complex and the operating environment is changeable, which makes the system prone to various faults. These faults not only affect the power generation efficiency, but may also damage the equipment and even cause the entire power generation system to stop operating.

[0003] In order to ensure the stable operation of photovoltaic module power generation systems, various fault protection mechanisms are usually set up. These fault protection mechanisms are designed to trigger corresponding protection actions in time when a fault occurs, such as cutting off the circuit, adjusting the working state of the module, etc., to avoid further expansion of the fault. However, in actual applications, the fault protection mechanism does not always work accurately.

[0004] At present, many existing fault protection false triggering identification methods adopt relatively independent evaluation strategies. These strategies often view each fault protection triggering event in isolation, and determine whether a protection action is falsely triggered based only on the characteristics of a single event or a simple threshold judgment. For example, whether a protection action is falsely triggered is determined only based on whether a certain parameter of a certain device exceeds a preset fixed threshold, without considering the propagation relationship of the fault in the entire system and the correlation between different faults.

[0005] This independent evaluation strategy has great limitations. Since the photovoltaic module power generation system is a complex whole, the various components are interrelated and mutually influential. The failure of one component may cause abnormalities in other components, resulting in a series of fault protection triggering events. Analyzing each event in isolation cannot accurately reflect the true propagation path of the fault in the system and the inherent connection between different faults, so it is easy to lead to misjudgment.

[0006] Some existing methods fail to fully consider the correlation between faults when analyzing fault protection triggering events. For example, after a fault occurs, according to the system's operating logic, there is no in-depth study of which faults may be triggered next, as well as the triggering sequence and correlation between these faults. This lack of comprehensive fault correlation analysis makes it impossible to accurately determine whether a certain protection action is triggered under reasonable fault correlation conditions when identifying false triggering of protection actions. Summary of the invention

[0007] The purpose of the present invention is to provide a data acquisition control method and system for a photovoltaic module power generation system. The present application is implemented as follows:

[0008] In a first aspect, the present application provides a data acquisition control method for a photovoltaic module power generation system, the method comprising: obtaining Y fault protection trigger events corresponding to a target photovoltaic module power generation system in a data acquisition cycle; constructing a fault protection trigger graph structure including X graph vertices according to the fault types corresponding to the Y fault protection trigger events and the order in which the Y fault protection trigger events are triggered in the data acquisition cycle, wherein different graph vertices are used to indicate different fault types, and according to the order in which they are triggered, for a fault protection trigger event binary consisting of an mth fault protection trigger event and an nth fault protection trigger event, wherein n=m+1; if the mth fault protection trigger event is a first fault type class, and the nth fault protection trigger event is a second fault type, then there is a target connection line from the first graph vertex to the second graph vertex in the fault protection trigger graph structure, the first graph vertex is used to indicate the first fault type, the second graph vertex is used to indicate the second fault type, and the influence coefficient of the target connection line is used to indicate the trigger parameter of the fault protection trigger event binary group; according to the fault protection trigger graph structure and the fault type vector representations corresponding to the X graph vertices respectively, the target photovoltaic component power generation system is identified for protection false triggering based on the target false triggering identification model, and the identification information indicating whether the target photovoltaic component power generation system has a protection action false triggering in the data collection cycle is obtained.

[0009] In a second aspect, the present application provides a computer system comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method as described above is implemented.

[0010] The technical effects of the present application include at least: in the embodiment of the present application, for the Y fault protection trigger events corresponding to the target photovoltaic module power generation system in the data collection cycle, based on the fault types corresponding to the Y fault protection trigger events and the order in which the Y fault protection trigger events are triggered, a fault protection trigger graph structure including X graph vertices is constructed with the fault types as graph vertices. Under the order in which they are triggered, for the fault protection trigger event binary consisting of the mth fault protection trigger event and the nth fault protection trigger event, when the mth fault protection trigger event is the first fault type and the nth fault protection trigger event is the second fault type, there is a target connection line from the first graph vertex to the second graph vertex in the fault protection trigger graph structure, the first graph vertex is used to indicate the first fault type, the second graph vertex is used to indicate the second fault type, and the influence coefficient of the target connection line is used to indicate the trigger parameter of the fault protection trigger event binary. Therefore, according to the super direction of the connecting line, it can be indicated that after the fault protection trigger event of one fault type is triggered and responded, which fault protection trigger events of which fault types will be triggered and responded, and which fault protection trigger events of which fault types will not be triggered and responded. At the same time, according to the influence coefficient of the connecting line, it can be indicated that after the fault protection trigger event of one fault type is triggered and responded, the trigger response related to the fault protection trigger event of another fault type is triggered and responded. It reflects the trigger response of Y fault protection trigger events in the overall situation. Compared with the independent protection action false trigger evaluation strategy, this application identifies the protection false trigger based on the trigger response of Y fault protection trigger events in the overall situation, and the identification accuracy is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0012] Figure 1 It is a flow chart of a data acquisition and control method for a photovoltaic module power generation system provided in an embodiment of the present application.

[0013] Figure 2 It is a schematic diagram of the composition of a computer system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] The executor of the photovoltaic module power generation system data acquisition control method in the embodiment of the present application is a computer system, including but not limited to a server, a personal computer, a laptop computer, a tablet computer, a smart phone, etc. The embodiment of the present application provides a photovoltaic module power generation system data acquisition control method, such as Figure 1 As shown, the method includes:

[0015] Step S100: obtaining Y fault protection triggering events corresponding to the target photovoltaic module power generation system in a data collection cycle.

[0016] The target photovoltaic module power generation system is a power generation unit composed of multiple photovoltaic modules, related electrical equipment (such as inverters, combiner boxes, etc.) and connecting lines, and its function is to convert solar energy into electrical energy. In the actual operation process, due to the influence of various internal and external factors, faults may occur, and fault protection mechanisms are pre-set for these faults. When certain conditions are met, the fault protection mechanism will be triggered, thereby generating a fault protection trigger event.

[0017] The data collection cycle is a pre-set time interval used to collect and analyze the operating data of the target photovoltaic module power generation system. The setting of this cycle can be determined according to specific needs and system characteristics. For example, the data collection cycle can be set to one day, one week, or one month based on factors such as system stability, frequency of failures, and maintenance plans.

[0018] A fault protection trigger event refers to an event in which a pre-set fault protection mechanism is triggered due to some abnormal situation during the operation of the target photovoltaic module power generation system. For example, when the temperature of the photovoltaic module is too high, the over-temperature protection mechanism may be triggered, which is a fault protection trigger event; or when the output voltage of the inverter exceeds the normal range, the corresponding voltage protection mechanism will be triggered, which is also a fault protection trigger event.

[0019] When obtaining these fault protection trigger events, multiple implementation methods can be used. One feasible implementation method is to implement it through a sensor network. In the target photovoltaic module power generation system, sensors are installed at various key parts, such as temperature sensors installed on photovoltaic modules, voltage sensors and current sensors installed on inverters. These sensors can monitor relevant parameters in real time and transmit the monitored data to the computer system. The computer system analyzes and processes these data. When the data meets the pre-set fault trigger conditions, it is determined that a fault protection trigger event has occurred and recorded. For example, for the data transmitted by the temperature sensor, if the normal operating temperature range of the photovoltaic module is set to [20℃, 60℃], when the data transmitted by the sensor indicates that the temperature reaches 70℃, the computer system will determine that an over-temperature fault protection trigger event has occurred.

[0020] Another way to achieve this is through a monitoring system. The monitoring system can monitor the operating status of the target photovoltaic module power generation system in real time, such as monitoring the communication status between the devices, the operating status indicator lights of the devices, etc. When a communication interruption is found or the device status indicator light shows an abnormality, the computer system can also determine that a fault protection trigger event has occurred. For example, if the communication between a photovoltaic module and the combiner box is interrupted, the monitoring system detects this situation and notifies the computer system, and the computer system will record it as a fault protection trigger event.

[0021] During the entire data collection cycle, the computer system continuously monitors and analyzes data, continuously discovers and records fault protection trigger events, until the end of the data collection cycle, and finally obtains Y fault protection trigger events. Here, Y is a variable representing the number of fault protection trigger events, and its specific value depends on the number of fault protection trigger events that actually occurred during the data collection cycle. Different data collection cycles may have different Y values.

[0022] The Y fault protection trigger events obtained through step S100 will serve as the basic data for subsequent steps such as fault analysis, construction of fault protection trigger graph structure, and identification of false protection triggering. These data can comprehensively reflect the fault protection triggering status of the target photovoltaic module power generation system during the data collection cycle, which is of great significance for ensuring the stable operation of the power generation system and improving the accuracy and efficiency of fault diagnosis.

[0023] Step S200: According to the fault types corresponding to the Y fault protection trigger events and the order in which the Y fault protection trigger events are triggered in the data collection cycle, a fault protection trigger graph structure including X graph vertices is constructed, where different graph vertices are used to indicate different fault types. According to the order in which the fault protection trigger events are triggered, for a fault protection trigger event tuple consisting of an m-th fault protection trigger event and an n-th fault protection trigger event, where n=m+1; if the m-th fault protection trigger event is the first fault type and the n-th fault protection trigger event is the second fault type, then there is a target connection line from the first graph vertex to the second graph vertex in the fault protection trigger graph structure, the first graph vertex is used to indicate the first fault type, the second graph vertex is used to indicate the second fault type, and the influence coefficient of the target connection line is used to indicate the trigger parameter of the fault protection trigger event tuple.

[0024] Fault protection trigger events correspond to different fault types. Fault types are the classification definitions of different fault conditions in the target photovoltaic module power generation system. For example, in a photovoltaic module power generation system, possible fault types include photovoltaic module short circuit fault, open circuit fault, inverter efficiency reduction fault, combiner box overcurrent fault, etc. These fault types have different manifestations and impact mechanisms on the operation of the power generation system.

[0025] When the computer system constructs the fault protection trigger graph structure, it must first analyze the order in which the Y fault protection trigger events are triggered in the data collection cycle. Assume that in a data collection cycle, the computer system obtains 5 fault protection trigger events (Y=5), which are the short circuit fault of the photovoltaic module, the efficiency reduction fault of the inverter, the overcurrent fault of the junction box, the open circuit fault of the photovoltaic module, and the overtemperature fault of the inverter, and they are triggered in the above order.

[0026] The fault protection trigger graph structure contains X graph vertices, where X refers to the number of different fault types. Each graph vertex is used to indicate a different fault type. Continuing with the above example, if the five fault protection trigger events cover four different fault types (assuming that short circuit fault and open circuit fault are two different fault types, inverter efficiency reduction fault and overtemperature fault are two different fault types, and combiner box overcurrent fault is one fault type, that is, X=4), then there will be four graph vertices corresponding to these four fault types.

[0027] For a fault protection trigger event tuple consisting of the mth fault protection trigger event and the nth fault protection trigger event (n=m+1), the computer system will perform the following operations. For example, m=1 corresponds to a short circuit fault of a photovoltaic module (the first fault type), and n=2 corresponds to a reduced efficiency fault of an inverter (the second fault type). If this is true, then in the fault protection trigger graph structure, there will be a target connection line from the first graph vertex (corresponding to the short circuit fault of the photovoltaic module) to the second graph vertex (corresponding to the reduced efficiency fault of the inverter).

[0028] The target link has an important attribute, the influence coefficient, which is a weight used to indicate the trigger parameter of the fault protection trigger event binary. The trigger parameter can reflect some quantitative characteristics of the association between the two fault protection trigger events.

[0029] For the determination of trigger parameters, the computer system can adopt different implementation methods. One possible implementation method is event-based statistical analysis. Assume that the triggering number of the fault protection trigger event binary group is used as the trigger parameter. The computer system can count each fault protection trigger event binary group during the data collection cycle. For example, during the entire data collection cycle, the short circuit fault of the photovoltaic module and the inverter efficiency reduction fault occurred three times, then these three times can be used as the trigger parameter of the fault protection trigger event binary group, and then the influence coefficient of the target connection line is determined. Although no specific formula is given here, a simple way to set the influence coefficient is to directly use the triggering number as the influence coefficient. Of course, it can also be normalized or other mathematical transformations can be performed according to actual needs, such as influence coefficient = triggering number / total possible triggering number (assuming that the total possible triggering number is a fixed value obtained based on the system operation history data or theoretical analysis).

[0030] Another implementation method is based on a physical model or empirical model of fault propagation. For example, in a photovoltaic module power generation system, there is a certain causal relationship or degree of correlation between certain faults. If there is a correlation based on circuit principles and device characteristics from the short-circuit fault of the photovoltaic module to the efficiency reduction fault of the inverter, the computer system can determine the trigger parameter based on the strength of this correlation. This correlation strength may be based on the analysis of changes in device parameters, such as the relationship between the rate of change of the input power of the inverter after the photovoltaic module short-circuit fault occurs and the inverter efficiency reduction. Assume that through experiments or theoretical analysis, when the photovoltaic module short-circuit fault occurs, the inverter input power drops by more than a certain proportion (such as 20%), there will be a high probability that the inverter efficiency will be reduced. According to historical data statistics, the credibility of this situation is 0.6 (this 0.6 can be used as a trigger parameter to determine the influence coefficient, for example, the influence coefficient can be directly set to this credibility value).

[0031] After the fault protection trigger graph structure is constructed in this way, it can intuitively reflect the triggering order relationship between different fault types and the quantitative degree of association between them. This graph structure provides an important basis for the subsequent protection false triggering identification of the target photovoltaic module power generation system based on the target false triggering identification model. It is different from viewing each fault protection triggering event in isolation, but all fault protection triggering events are considered in an overall structure. Through the connecting lines and influence coefficients in the graph structure, the propagation path and mutual influence relationship of the fault in the system can be deeply analyzed, so as to more accurately identify whether there is a false triggering of the protection action. For example, if it is found in the graph structure that a certain fault type rarely triggers the next fault type in the order represented by the connecting lines in the graph (that is, the influence coefficient is very low), but it actually triggers the corresponding protection action frequently, this may indicate that there is a false triggering of the protection action.

[0032] The process of constructing a fault protection trigger graph structure by a computer system is a complex process that comprehensively considers the type of fault, the sequence of fault protection trigger events, and the degree of correlation between them. It requires the use of multiple implementation methods to analyze and process a large amount of operating data to ensure that the constructed graph structure can accurately reflect the fault protection trigger characteristics of the target photovoltaic module power generation system, thereby providing effective support for the stable operation and fault management of the entire power generation system.

[0033] Step S300: According to the fault protection trigger graph structure and the fault type vector representations corresponding to the X graph vertices, protection false triggering identification is performed on the target photovoltaic module power generation system based on the target false triggering identification model to obtain identification information indicating whether the target photovoltaic module power generation system has a protection action false triggering during the data collection cycle.

[0034] In the fault analysis process of the target photovoltaic module power generation system, step S300 is executed by the computer system. This step is intended to use the target false triggering identification model to identify the false triggering of protection of the target photovoltaic module power generation system based on the previously constructed fault protection trigger graph structure and the fault type vector representation corresponding to the graph vertex, so as to obtain identification information indicating whether the target photovoltaic module power generation system has a false triggering of protection action during the data collection cycle.

[0035] The target false trigger identification model is a model specially designed to analyze the fault protection triggering situation of the photovoltaic module power generation system to determine whether the protection action is falsely triggered. It can be a variety of types of models, such as a graph neural network model. The graph neural network model can process graph structure data well. In this case, the fault protection trigger graph structure can be effectively analyzed as input data. Taking the graph neural network model as an example, it can learn the relationship between the nodes in the graph structure (i.e., graph vertices, corresponding to different fault types) and the characteristics of the nodes themselves (fault type vector representation), so as to mine the rules hidden behind the fault protection triggering event.

[0036] The fault protection trigger graph structure plays a key role in this step. It contains multiple graph vertices, each of which indicates a type of fault, and the connecting lines between the graph vertices and their influence coefficients can reflect the triggering order relationship between different fault types and the quantitative degree of association. For example, in a simple fault protection trigger graph structure, there are three graph vertices representing PV module short circuit fault, inverter efficiency reduction fault and combiner box overcurrent fault. If there is a connecting line from the PV module short circuit fault graph vertex to the inverter efficiency reduction fault graph vertex, its influence coefficient is 0.5, which means that in actual operation, there is a certain probability (quantized as 0.5 here) of triggering the inverter efficiency reduction fault after the PV module short circuit fault occurs. This structural information provides the computer system with an intuitive representation of the fault propagation relationship in the system.

[0037] The fault type vector representation corresponding to the graph vertex is a way to quantitatively describe the fault type. For example, for a photovoltaic module short-circuit fault, its fault type vector representation may contain information in multiple dimensions, such as the location of the fault (expressed by coordinates or component numbers), the system current value when the fault occurs, and the degree of impact of the fault on the power generation. Each element in these vector representations has its specific meaning and value range. In this way, the various characteristics of the fault type are integrated together so that the computer system can perform unified analysis and processing.

[0038] When the computer system performs protection false triggering identification, multiple implementation methods can be used. One possible implementation method is a method based on probability statistics. Assume that in the historical operation data of the target photovoltaic module power generation system, a large number of records about fault protection triggering events and whether the protection action is correctly triggered have been accumulated. The computer system can build a probability model based on these historical data. For example, for a certain fault protection triggering graph structure pattern (i.e., a specific graph vertex connection relationship and influence coefficient combination), if the probability of false triggering of the protection action in this pattern is found to be 0.1 in the historical data, then when a similar fault protection triggering graph structure appears again in the current data collection cycle, the computer system can preliminarily judge that the possibility of false triggering of the protection action is 0.1. Although this method based on probability statistics is simple and direct, it may have certain limitations because it relies on a large amount of historical data and assumes that the situation in the historical data is similar to the current situation.

[0039] Another implementation method is feature analysis based on machine learning algorithms. The computer system can use the fault protection trigger graph structure and the fault type vector representation corresponding to the graph vertices as input features, and use machine learning algorithms (such as support vector machines, decision trees, etc.) for classification analysis. Taking the support vector machine as an example, it separates different categories of data (in this case, the correct triggering and false triggering of the protection action) by finding an optimal hyperplane. The computer system converts the current fault protection trigger graph structure and the fault type vector representation into a feature vector form that the support vector machine can handle, and then classifies it through the trained support vector machine model to obtain the result of whether the protection action is falsely triggered. In this process, some preprocessing of the input features may be required, such as normalization, to ensure that features of different dimensions have the same weights. The formula can be expressed as: For the feature vector , the normalized eigenvector ,in are the minimum and maximum values ​​of the i-th feature dimension respectively.

[0040] Through the above implementation, the computer system can comprehensively consider the various relationships in the fault protection trigger graph structure and the various features in the fault type vector representation, and comprehensively analyze and identify the false triggering of the protection action of the target photovoltaic module power generation system during the data acquisition cycle. This identification result is of great significance for timely discovery and correction of false triggering of protection actions in the system, and can improve the operating efficiency and stability of the photovoltaic module power generation system. If the false triggering of the protection action is not discovered in time, it may cause unnecessary system shutdown or unnecessary operation of the normally operating equipment, thereby affecting the power generation of the power generation system and the life of the equipment. Accurate protection false triggering identification can ensure that the system takes correct protection measures when a fault occurs, while avoiding the negative impact of false operation.

[0041] As an implementation mode, step S300, based on the fault protection trigger graph structure and the fault type vector representations corresponding to the X graph vertices, performs protection false triggering identification on the target photovoltaic module power generation system based on the target false triggering identification model, including:

[0042] Step S310: Determine a relation array corresponding to the fault protection trigger graph structure, the array size of the relation array is X×X, a and b are used to indicate different graph vertices, a and b are positive integers less than or equal to X, in the relation array, the array cell located in the ath column and the bth row is used to indicate the influence coefficient of the connection line from the graph vertex represented by a to the graph vertex represented by b;

[0043] Step S320: According to the relation array and the fault type vector representations corresponding to the X graph vertices, protection false triggering identification is performed on the target photovoltaic module power generation system based on the target false triggering identification model.

[0044] For step S310, the computer system determines the relationship array corresponding to the fault protection trigger graph structure. This relationship array is a two-dimensional array, which can be understood as an adjacency matrix, and its array size is X×X, where X is the number of graph vertices in the fault protection trigger graph structure, and each graph vertex represents a type of fault. In this relationship array, let a and b be used to indicate different graph vertices, and a and b are both positive integers less than or equal to X. The array unit located in the ath column and the bth row is used to indicate the influence coefficient of the connection line between the graph vertex represented by a and the graph vertex represented by b.

[0045] For example, suppose that in the target photovoltaic module power generation system, the fault protection trigger graph structure has 4 graph vertices (i.e., X=4), corresponding to the four types of faults: photovoltaic module short circuit fault, inverter efficiency reduction fault, junction box overcurrent fault, and photovoltaic module open circuit fault. For a connection line between the graph vertex representing the photovoltaic module short circuit fault (assuming that the graph vertex a=1) and the graph vertex representing the inverter efficiency reduction fault (assuming that the graph vertex b=2), the influence coefficient is 0.6, then in the relationship array, the value of the array cell located in the first column and the second row is 0.6. This means that in the fault association relationship of this system, when the photovoltaic module short circuit fault occurs, the trigger association degree (expressed by the influence coefficient) for the inverter efficiency reduction fault is 0.6. The computer system can construct this relationship array by traversing and parsing the fault protection trigger graph structure. For each pair of connection lines between graph vertices, the computer system queries its influence coefficient and fills it into the corresponding position of the relationship array.

[0046] In step S320, the computer system identifies protection false triggering of the target photovoltaic module power generation system based on the target false triggering identification model according to the relationship array determined in step S310 and the fault type vector representations corresponding to the X graph vertices.

[0047] Take the graph neural network model as an example to illustrate the target false trigger identification model. Graph neural networks can effectively process graph structure data (such as the fault protection trigger graph structure here) and node features (such as the fault type vector representation). For the influence coefficients in the relationship array, they can be regarded as the weights of the edges between the graph vertices in the graph neural network. These weights play a key role in the information propagation and feature learning process of the model.

[0048] Assume that the fault type vector representation contains multiple quantitative features related to the fault. For example, for the graph vertex corresponding to the inverter efficiency reduction fault, one dimension in the fault type vector representation may be the ratio of the inverter input power to the normal input power when the fault occurs, and another dimension may be the difference between the inverter internal temperature and the normal operating temperature. The values ​​of these dimensions constitute the vector representation of the fault type.

[0049] When a computer system performs protection false triggering identification based on a graph neural network model, one implementation method that can be adopted is based on a message passing mechanism. In each layer of the graph neural network, each graph vertex will update its own state based on the relationship with its adjacent graph vertices (represented by the influence coefficient in the relationship array) and the characteristics of the adjacent graph vertices (represented by the fault type vector). For example, for a graph vertex v, its new state It can be calculated by the following formula (this is just a simple example, the actual situation may be more complicated): , where N(v) represents the set of neighboring graph vertices of graph vertex v, is the influence coefficient of the edge from graph vertex u to graph vertex v (i.e. the value in the relationship array), is the current state of the vertex u (which can be initialized to the corresponding fault type vector representation), is an activation function (such as ReLU function, etc.), and b is a bias term. Through such message passing and state update mechanism, the graph neural network can gradually learn the complex relationship between different fault types in the graph structure and their impact on false triggering of protection.

[0050] Another implementation method is based on feature fusion. The computer system fuses the relationship information in the relationship array with the feature information in the fault type vector representation to generate a more representative feature vector for protection against false triggering identification. For example, the influence coefficient in the relationship array can be weighted summed with the corresponding dimension in the fault type vector representation. Assume that the fault type vector representation , the influence coefficient vector related to the vertex of the graph in the relationship array is , then the fused feature vector The formula can be (It can be understood that only a simple fusion method is shown here, and a more complex design can be made as needed) and then the fused feature vector is passed as input to the target false triggering recognition model (such as a classifier, etc.) to determine whether there is a false triggering of the protective action.

[0051] In actual photovoltaic module power generation systems, the relationships between different fault types are complex and diverse. For example, an overcurrent fault in a combiner box may not only be related to its own internal circuit problems, but may also be affected by the output power fluctuations of the upstream photovoltaic modules and the load capacity of the downstream inverter. In the fault protection trigger graph structure, the influence coefficients of the connecting lines from the photovoltaic module to the combiner box and from the combiner box to the inverter will reflect this complex correlation. By accurately constructing a relationship array and using a fault type vector representation, the computer system enables the target false trigger identification model to fully understand these relationships, thereby more accurately identifying protection false triggers.

[0052] At the same time, as the scale of photovoltaic module power generation systems continues to expand and the operating environment continues to change, the mode of fault occurrence and the protection triggering situation will also change. The relational array and fault type vector representation can dynamically reflect these changes, and the target false triggering recognition model can adapt to these changes and continuously learn new patterns. For example, when new photovoltaic modules are added to the system or the inverter is upgraded, new fault modes may appear, and the corresponding graph vertex relationships (values ​​in the relational array) and fault type vector representations in the fault protection trigger graph structure will also change. By continuously collecting data, updating the relational array and fault type vector representation, and retraining the target false triggering recognition model, the computer system can always maintain the ability to accurately identify the protection false triggering situation, ensuring the stable and efficient operation of the photovoltaic module power generation system.

[0053] Steps S310 and S320 can deeply explore the relationship between fault protection triggering events and accurately determine whether the protection action is falsely triggered by reasonably constructing a relationship array, effectively utilizing the fault type vector representation, and using the target false triggering identification model, thereby providing a strong guarantee for the reliable operation of the photovoltaic module power generation system.

[0054] As an implementation mode, step S320, based on the relationship array and the fault type vector representations corresponding to the X graph vertices, performs protection false triggering identification on the target photovoltaic module power generation system based on the target false triggering identification model, including:

[0055] Step S321: Determine a graph vertex array according to the fault type vector representations corresponding to X graph vertices, the graph vertex array includes X columns of array elements, and c is set to be a positive integer less than or equal to X, c is used to indicate a graph vertex, and the c-th column array element is used to indicate the fault type vector representation corresponding to the graph vertex represented by c;

[0056] Step S322: According to the relationship array and the graph vertex array, based on the target false triggering identification model, the protection false triggering identification of the target photovoltaic module power generation system is performed.

[0057] For step S321, the computer system determines a graph vertex array according to the fault type vector representations corresponding to the X graph vertices. This graph vertex array is also a two-dimensional array, including X columns of array elements. Let c be a positive integer less than or equal to X, c is used to indicate a graph vertex, and the cth column array element is used to indicate the fault type vector representation corresponding to the graph vertex represented by c.

[0058] For example, suppose there are three graph vertices (X=3) in the fault protection trigger graph structure, corresponding to the hot spot fault of the photovoltaic module, the DC component too high fault of the inverter, and the grounding fault of the junction box. For the graph vertex corresponding to the hot spot fault of the photovoltaic module (assuming c=1), its fault type vector representation may contain quantitative information such as the hot spot area ratio, the temperature difference between the hot spot and the normal temperature, and the computer system fills this vector representation into the first column array unit of the graph vertex array. Similarly, for the graph vertices corresponding to the DC component too high fault of the inverter (assuming c=2) and the grounding fault of the junction box (assuming c=3), their respective fault type vector representations are also filled into the second and third columns of the graph vertex array. The computer system can construct the graph vertex array by sequentially reading the fault type vector representation corresponding to each graph vertex, and filling it into the corresponding column of the array in the order of the index of the graph vertex.

[0059] In step S322, the computer system performs protection false triggering identification on the target photovoltaic module power generation system based on the target false triggering identification model according to the graph vertex array determined in step S321 and the relationship array determined in the previous step S310.

[0060] Take a simple target false trigger identification model based on matrix operations as an example. The data in the relationship array (adjacency matrix) and the graph vertex array can be regarded as important components of the model input. For the relationship array, it reflects the connection relationship between the graph vertices and the influence coefficient of the connection, which is similar to defining a connection weight of a network structure in the model. For example, the value of an element in the relationship array represents the connection strength or influence probability from the graph vertex corresponding to a specific fault type to the graph vertex corresponding to another fault type.

[0061] The graph vertex array is a concentrated reflection of the characteristics of each graph vertex (fault type vector representation). When a computer system performs protection false triggering identification, one implementation method that can be adopted is to perform a joint matrix operation on the graph vertex array and the relationship array. Assuming that the graph vertex array is A and the relationship array is R, the computer system can calculate M=A×R (here × represents matrix multiplication operation). The result M obtained by calculation is a new matrix that combines the characteristics of the graph vertices themselves and the relationship information between the graph vertices. Each element in this new matrix contains information related to protection false triggering identification. For example, an element in the matrix M may represent the comprehensive influence of a certain fault type on another fault type in terms of protection false triggering after considering the connection relationship and characteristics of the graph vertices.

[0062] Another implementation method is based on feature mapping. The computer system can define a mapping function f to map the elements in the graph vertex array and the relationship array to a new feature space. For example, for an element in the fault type vector representation in the graph vertex array, (representing the jth element in the i-th graph vertex array) and the influence coefficient in the relationship array (representing the influence coefficient from the jth graph vertex to the kth graph vertex), the mapping function f can be defined as (It can be understood that this application only shows a schematic mapping function, and the actual situation can be designed according to specific needs.) Through such mapping, the computer system converts the original data into a representation in a new feature space. These new features can better reflect the complex relationship between fault types and the comprehensive impact on the false triggering of protection. Then, the computer system can use these new features for subsequent analysis, such as inputting the new features into a classifier (as part of the target false triggering identification model) to determine whether the target photovoltaic module power generation system has false triggering of the protection action.

[0063] In the actual operation scenario of photovoltaic module power generation system, the data in the graph vertex array and relationship array are dynamically changing. With the increase of system operation time, changes in environmental conditions, and aging of equipment, the fault type vector representation may change, which in turn affects the graph vertex array; at the same time, the correlation between faults (reflected in the relationship array) may also change. For example, in a high temperature environment, the hot spot fault of the photovoltaic module may more frequently cause the DC component over-high fault of the inverter, which requires the computer system to update the relevant data in the graph vertex array and relationship array in a timely manner. In addition, due to differences in scale, module model, installation environment and other factors of different photovoltaic module power generation systems, the specific form and value of the graph vertex array and relationship array will also be different. The computer system accurately constructs and updates these data structures according to the characteristics of the specific system, and uses a suitable target false trigger identification model to perform protection false trigger identification to ensure the stable operation of the system and the effectiveness of the fault protection mechanism.

[0064] Through steps S321 and S322, the computer system can make full use of the fault type vector representation corresponding to the graph vertex and the relationship information between the graph vertices, integrate and convert this information into an input form suitable for the target false trigger identification model through a reasonable implementation method, so as to accurately analyze and identify the protection false triggering of the target photovoltaic module power generation system. This is of vital importance to ensure the efficient and stable operation of the photovoltaic module power generation system, and can timely discover and correct possible false triggering problems of protection actions, and reduce the occurrence of unnecessary system shutdowns or equipment losses caused by false triggering.

[0065] As an implementation mode, step S322, based on the relationship array and the graph vertex array, performs protection false triggering identification on the target photovoltaic module power generation system based on the target false triggering identification model, including:

[0066] Step S3221: Generate an initial state representation for each graph vertex according to the fault type vector representation corresponding to each graph vertex in the graph vertex array, and obtain a state set consisting of the initial state representations of all graph vertices, wherein the initial state representation includes the fault type information of the graph vertex itself and the influence of other graph vertices adjacent to it;

[0067] Step S3222: using a multi-layer perceptron in the target false trigger recognition model to update the initial state representation of each graph vertex in the state set to obtain an intermediate state set, and processing the intermediate state set through a nonlinear transformation enhancement network to obtain an updated state set;

[0068] Step S3223: Based on the connection relationship in the relationship array, aggregate the state representation of each graph vertex in the update state set with the state representation of its neighboring graph vertices to obtain an aggregated state set;

[0069] Step S3224: input the aggregated state set into the recurrent neural network layer in the target false trigger recognition model, mine the dependency relationship of the fault protection triggering event in the time series, and obtain the dependency state set;

[0070] Step S3225: performing a global pooling operation on the dependency state set to obtain a pooling representation vector;

[0071] Step S3226: passing the pooled representation vector to the classifier in the target false triggering identification model, classifying the pooled representation vector based on the classifier, and obtaining an identification result of whether there is false triggering of the protective action.

[0072] In step S3221, the computer system generates an initial state representation for each graph vertex according to the fault type vector representation corresponding to each graph vertex in the graph vertex array, and obtains a state set consisting of the initial state representations of all graph vertices. The initial state representation here includes the fault type information of the graph vertex itself and the influence of other graph vertices adjacent to it.

[0073] For example, suppose there are three graph vertices in the graph vertex array corresponding to PV module short circuit fault, inverter overtemperature fault and combiner box overcurrent fault. For the graph vertex corresponding to the PV module short circuit fault, its fault type vector representation contains information such as the short circuit location and the amount of current change during the short circuit. When generating the initial state representation, the computer system not only considers the information in these own fault type vector representations, but also considers the influence of the graph vertices adjacent to it (for example, there may be an association with the inverter overtemperature fault graph vertex, and if the short circuit occurs, it may cause the inverter to overheat). Assume that there is a simple calculation method to consider the influence of adjacent graph vertices. For each adjacent graph vertex, calculate the inner product of its fault type vector representation and a predefined association weight vector, and then add these results to its own fault type vector representation to obtain the initial state representation. Suppose the fault type vector representation of the PV module short circuit fault graph vertex itself is , and the associated weight vector with the vertices of the inverter over-temperature fault graph is , assuming that the fault type vector of the inverter overtemperature fault graph vertex is expressed as , then the part after considering the influence of the adjacent graph vertices is , then the initial state of the vertices of the PV module short-circuit fault graph is represented by The computer system performs similar operations on each vertex of the graph to obtain the entire state set.

[0074] In step S3222, the computer system uses the multi-layer perceptron in the target false trigger recognition model to update the initial state representation of each graph vertex in the state set to obtain an intermediate state set, and processes the intermediate state set through a nonlinear transformation enhancement network to obtain an updated state set.

[0075] A multilayer perceptron (MLP) is a feedforward neural network that consists of an input layer, a hidden layer, and an output layer. The computer system takes the initial state representation of each vertex as input to the input layer. For example, for a multilayer perceptron with m input neurons (corresponding to the dimension of the initial state representation) and h hidden neurons, the output H of its hidden layer can be expressed by the formula Calculate, where is the value of the input neuron (i.e., the element in the initial state representation), is the connection weight from the input neuron to the hidden neuron, b is the bias term, and f is the activation function (such as the ReLU function f(x)=max(0,x)). Through multiple layers of such calculations, the multi-layer perceptron extracts and transforms the initial state representation to obtain the intermediate state representation, thereby forming an intermediate state set.

[0076] The intermediate state set is then further processed by a nonlinear transformation enhancement network. This network can use a variety of nonlinear transformations to enhance the representation of the data. For example, it can use functions such as the hyperbolic tangent function. The nonlinear function transforms each element in the intermediate state set. Assume that an intermediate state in the intermediate state set is represented by , the updated state representation obtained after the nonlinear transformation enhancement network processing , where (this is just a simple example) each intermediate state representation in the intermediate state set is processed in this way to obtain the updated state set.

[0077] In step S3223, based on the connection relationship in the relationship array, the computer system aggregates each graph vertex state representation in the update state set with the state representation of its neighboring graph vertices to obtain an aggregated state set.

[0078] The relationship array describes the connection relationship between the vertices of the graph. For example, an element in the relationship array Represents the connection relationship between graph vertex i and graph vertex j (such as influence coefficient, etc.). For a graph vertex state in the update state set, , the computer system finds its neighbor relationship in the relationship array and then aggregates it with the state representation of the neighbor graph vertex. A simple aggregation operation can be a weighted sum. Assume that graph vertex i has k neighbor graph vertices, and its neighbor graph vertex state is represented as , the corresponding influence coefficient in the relationship array is , then the aggregated state is represented by The computer system performs such an aggregation operation on each graph vertex in the update state set, thereby obtaining an aggregated state set.

[0079] In step S3224, the computer system inputs the aggregated state set into the recurrent neural network layer in the target false trigger identification model, mines the dependency relationship of the fault protection triggering events in the time series, and obtains the dependency state set.

[0080] Recurrent Neural Networks (RNNs) have the ability to process sequential data and capture temporal dependencies in the data. When a set of aggregate states is input to a recurrent neural network layer, such as a simple basic recurrent unit (such as the Elman unit), it updates its hidden state at each time step. Let an element in the aggregate state set be (representing the aggregate state at time t), the hidden state of the recurrent neural network layer The formula can be Calculate, where is the connection weight from hidden state to hidden state, is the connection weight from the input (aggregate state) to the hidden state, b is the bias term, and f is the activation function. By performing such calculations on the entire time series, the recurrent neural network layer can mine the dependencies of fault protection triggering events on the time series, such as the order relationship between a certain fault protection triggering event and other events at different time points and the impact of this order relationship on the overall protection mis-triggering. The hidden state of each time step constitutes a set of dependency states.

[0081] In step S3225, the computer system performs a global pooling operation on the dependency state set to obtain a pooling representation vector.

[0082] The global pooling operation is a way to compress and extract features from data. For example, for the elements in the dependency state set (Assuming the length of the time series is T), the maximum pooling operation can be used, that is, the elements in the pooling representation vector P (Here we assume Indicates h t ), or an average pooling operation This pooled representation vector is a compact representation that incorporates the information of all fail-safe triggering events and can reflect the interactions between fail-safe triggering events.

[0083] Finally, in step S3226, the computer system transfers the pooled representation vector to the classifier in the target false triggering recognition model, classifies the pooled representation vector based on the classifier, and obtains the recognition result of whether there is a false triggering of the protective action.

[0084] Classifiers can be of many types, such as support vector machines (SVMs), decision trees, etc. For example, support vector machines separate data of different categories by finding an optimal hyperplane. The computer system takes the pooled representation vector as input, and the support vector machine selects the optimal hyperplane based on its internal decision function (e.g., the decision function of a linear support vector machine). , where w is the weight vector, x is the input vector (i.e., the pooled representation vector), and b is the bias term) to determine which category the input vector belongs to. If the output result indicates that the input vector belongs to the protection action false triggering category, it means that the target PV module power generation system has a protection action false triggering during the data collection cycle; otherwise, it means that there is no protection action false triggering.

[0085] During the operation of the entire photovoltaic module power generation system, this process from the initial state representation of the graph vertex to the final protection action false triggering identification result is a complex and interrelated process. Each step depends on the data processing results of the previous step, and the operations in each step are aimed at deeply exploring the relationship between fault protection triggering events, whether it is a comprehensive analysis from the characteristics of the graph vertex itself, the relationship between the graph vertices, or the dependency relationship in the time series. This helps the computer system to accurately identify the false triggering of the protection action in the target photovoltaic module power generation system, thereby ensuring the stable operation of the system, improving the power generation efficiency, and reducing unnecessary losses caused by false triggering of protection actions, such as equipment downtime and increased maintenance costs. At the same time, with the continuous development of photovoltaic module power generation systems and changes in the operating environment, this identification process can also be continuously optimized and improved, such as by updating the parameters in the target false triggering identification model and adjusting the calculation method in each step to adapt to new fault modes and system requirements.

[0086] As an implementation method, the trigger parameter is obtained based on one or both of the following two acquisition strategies:

[0087] Acquisition strategy 1: Based on Y fault protection triggering events and the order of being triggered, determine the total number of triggering times of the fault protection triggering event binary group as the triggering parameter;

[0088] Acquisition strategy two: Based on Y fault protection triggering events and the order in which they are triggered, determine the fault cumulative effect value corresponding to the fault protection triggering event that triggers the second fault type in response to the fault protection triggering event binary as the trigger parameter.

[0089] For acquisition strategy 1, based on Y fault protection trigger events and the order of being triggered, the total number of triggering times of the fault protection trigger event binary is determined as the trigger parameter. During the operation of the target photovoltaic module power generation system, the fault protection trigger events will occur in a certain order. The computer system determines the order between these events by recording the occurrence time of each fault protection trigger event and the corresponding fault type. For example, suppose that in a data collection cycle, the computer system monitors a series of fault protection trigger events such as photovoltaic module short circuit fault, inverter overtemperature fault, and junction box overcurrent fault. For a specific fault protection trigger event binary, such as the combination of photovoltaic module short circuit fault and inverter overtemperature fault, the computer system will count this situation every time it detects that the photovoltaic module short circuit fault occurs and then the inverter overtemperature fault occurs during the entire data collection cycle.

[0090] Assume that this situation occurs a total of 5 times during this data collection cycle. The 5 times here is the total number of triggers of this fault protection trigger event binary group (PV module short circuit fault-inverter overtemperature fault), and this total number of triggers is determined as the trigger parameter of this binary group. The computer system can implement this counting by setting a counter. Whenever the first fault (such as a PV module short circuit fault) is detected, a counter for subsequent specific faults (such as an inverter overtemperature fault) is started. Once a subsequent fault is detected, the counter is incremented by 1. When the data collection cycle ends, the value of this counter is the total number of triggers of the fault protection trigger event binary group.

[0091] The second acquisition strategy is based on Y fault protection trigger events and the order of being triggered, and determines the fault cumulative effect value corresponding to the fault protection trigger event that triggers the second fault type with the fault protection trigger event binary as the trigger parameter. The fault cumulative effect value reflects the degree of cumulative impact of the fault protection trigger event on the system.

[0092] For example, consider the fault protection trigger event binary of the aging fault of the photovoltaic module and the output power drop fault of the inverter. For the aging fault of the photovoltaic module, it may have multiple effects on the entire power generation system, and these effects vary with the accumulation of time and the number of faults. The computer system can quantify the cumulative effect value of this fault by monitoring the key performance indicators of the photovoltaic module. Assume that the key performance indicators monitored include the photovoltaic module's photoelectric conversion efficiency, output current stability, etc. For the photoelectric conversion efficiency indicator, the computer system can calculate the cumulative effect value of the fault based on its decline ratio relative to the initial value.

[0093] Assume that the initial photoelectric conversion efficiency is , the photoelectric conversion efficiency detected at a certain moment is , then the cumulative effect value of the fault based only on the photoelectric conversion efficiency is The formula can be If multiple performance indicators are considered at the same time, the total fault cumulative effect value can be calculated by weighted summation. Assume that the fault cumulative effect value corresponding to the output current stability indicator is , whose weight is The cumulative effect of failure corresponding to the photoelectric conversion efficiency is , whose weight is , then the total cumulative effect of failure is .

[0094] When a photovoltaic module aging fault occurs, the computer system calculates its fault cumulative effect value, and when this fault cumulative effect value reaches a certain level, it may trigger the inverter output power reduction fault. The computer system uses the fault cumulative effect value of the photovoltaic module aging fault related to the inverter output power reduction fault as the trigger parameter of the fault protection trigger event binary group (photovoltaic module aging fault-inverter output power reduction fault).

[0095] These two strategies for obtaining trigger parameters reflect the relationship between fault protection trigger events from different perspectives. The total trigger count of strategy 1 mainly focuses on the frequency relationship between the occurrence of two fault protection trigger events. This frequency information can reflect the probability trend of one fault occurring after another fault occurs in the target photovoltaic module power generation system. For example, if the total trigger count of a fault protection trigger event binary is high, it means that there may be a strong correlation or causal relationship between the two faults, which may be caused by factors such as the physical structure, electrical connection or control logic of the system.

[0096] The fault cumulative effect value of strategy 2 focuses more on measuring the relationship between fault protection triggering events from the perspective of the cumulative impact of faults on the system. It takes into account the gradual impact of faults on system performance during the development process, and only when this cumulative impact reaches a certain level will another fault be triggered. This strategy is more accurate in describing those fault relationships with progressive and cumulative characteristics. For example, in a photovoltaic module power generation system, the failure of certain components may be caused by the gradual accumulation of long-term minor damage or performance degradation. The fault cumulative effect value can better capture the inherent connection in the development process of this fault.

[0097] In the actual target photovoltaic module power generation system fault analysis and protection false triggering identification, the computer system can choose to use one of the two strategies or both at the same time according to the specific needs and system characteristics. If the system pays more attention to the frequent correlation patterns between faults, strategy one may be given priority; if the system failure is more manifested as a chain reaction caused by cumulative effects, then strategy two is more appropriate. In addition, the combination of the two strategies can provide a more comprehensive and accurate description of the relationship between fault protection triggering events, thereby providing more reliable basic data for constructing the fault protection triggering graph structure and subsequent protection false triggering identification.

[0098] At the same time, as the operating environment, operating time and system structure of the target photovoltaic module power generation system change, these two strategies for obtaining trigger parameters also need to be continuously adjusted and optimized. For example, as the service life of photovoltaic modules increases, their aging rate may accelerate, which requires re-evaluation of the calculation method and weight distribution of the fault cumulative effect value; or when the system is upgraded, new equipment is added or the electrical connection method is changed, the order and frequency relationship between fault protection trigger events may change, and at this time, the statistical method of the total number of triggers in strategy one needs to be adjusted accordingly. This ensures that the computer system can accurately obtain the trigger parameters, and then effectively construct the fault protection trigger graph structure and identify the protection false trigger, so as to ensure the stable operation of the target photovoltaic module power generation system.

[0099] As an implementation mode, the method further includes a process of determining the fault type:

[0100] Step S10: performing a similarity comparison between the protection event vector representation corresponding to the fault protection trigger event and the fault type vector representation corresponding to the preset fault type, wherein the protection event vector representation corresponding to the fault protection trigger event is used to characterize the fault protection trigger event, and the fault type vector representation corresponding to the preset fault type is used to characterize the preset fault type;

[0101] Step S20: determining the fault type corresponding to the fault protection triggering event based on the preset fault types that meet the similarity condition.

[0102] In the fault analysis of the target photovoltaic module power generation system, the fault type determination process is implemented through steps S10-S20, and the computer system executes these steps to accurately determine the fault type corresponding to the fault protection triggering event.

[0103] In step S10, the computer system compares the protection event vector representation corresponding to the fault protection trigger event with the fault type vector representation corresponding to the preset fault type for similarity. The protection event vector representation is a quantitative form used to characterize the fault protection trigger event, which contains multiple information elements related to the event. For example, for a fault protection trigger event of a photovoltaic component, its protection event vector representation may contain information elements such as the ambient temperature when the fault occurs, the output voltage of the component, the current change, and the time when the fault occurs. Assume that in a specific photovoltaic component fault protection trigger event, the ambient temperature is 30°C when the fault occurs, the component output voltage drops by 10% compared to the normal voltage, the current change is reduced by 8%, and the fault occurs at 10 am, then the protection event vector representation of this fault protection trigger event may be (30, -10%, -8%, 10:00) (this is just a simple illustration, the actual vector representation may contain more complex parameters and more precise quantization methods).

[0104] The fault type vector representation corresponding to the preset fault type is a quantitative description of the characteristics of each preset fault type. Taking the short-circuit fault of a photovoltaic module as an example, its fault type vector representation may include relevant information elements such as the typical voltage close to zero during a short circuit, the current suddenly increases to above a certain threshold, and the short circuit may occur in a certain area inside the module. Assume that the fault type vector representation of the short-circuit fault is (close to 0, increases to above the threshold, and the internal area of ​​the module) (also a simple illustration).

[0105] When a computer system performs similarity comparison, it can adopt a variety of implementation methods. One feasible method is based on vector distance measurement. For example, the Euclidean distance formula ,in is an element in the protection event vector representation of the fault protection triggering event, is an element in the fault type vector representation corresponding to the preset fault type, and n is the dimension of the vector. The computer system calculates the Euclidean distance between the two vectors. The smaller the distance, the higher the similarity between the two vectors.

[0106] In step S20, the computer system determines the fault type corresponding to the fault protection triggering event based on the preset fault types that meet the similarity condition. The similarity condition is a preset threshold or judgment standard. For example, if the Euclidean distance is used for similarity comparison in step S10, the computer system sets a distance threshold T. When the calculated Euclidean distance between the two vectors is less than T, it is considered that the similarity condition is met.

[0107] Continuing with the above example of photovoltaic module failure, if the Euclidean distance calculated between the protection event vector representation of a fault protection trigger event and the preset short-circuit fault type vector representation is less than the set threshold T, then the computer system determines that the fault type corresponding to the fault protection trigger event is a short-circuit fault.

[0108] In the actual target photovoltaic module power generation system, the manifestations of faults may be diverse and complex. Different faults may have similarities in certain parameters. For example, the output power drop of the module may be caused by a variety of reasons, which may be the fault of the module itself, or it may be indirectly caused by external environmental factors or other equipment failures. By comparing the similarity between the protection event vector representation and the fault type vector representation, the fault type that best matches the fault protection trigger event can be accurately determined from many possible fault types.

[0109] At the same time, this method helps the system's automated fault diagnosis and management. The computer system can quickly analyze and process a large number of fault protection trigger events, without the need for manual judgment of the fault type one by one. As the scale of photovoltaic module power generation systems continues to expand, this automated fault type determination method can improve the efficiency and accuracy of fault diagnosis, thereby ensuring the stable operation of the system. For example, in a large photovoltaic power station, there may be tens of thousands of photovoltaic modules, which will generate a large number of fault protection trigger events every day. If the fault type of each event is determined manually, the workload is huge and prone to errors. The computer system can use this vector-based similarity comparison method to complete this task efficiently and accurately.

[0110] In addition, in order to ensure the accuracy of this method of determining the type of fault, the computer system continuously updates and optimizes the fault type vector representation corresponding to the preset fault type. With the in-depth study of the fault mechanism of the photovoltaic module power generation system and the discovery of new fault phenomena, it may be necessary to adjust the element content or value range in the fault type vector representation. For example, with the application of new photovoltaic module materials and technologies, some new failure modes may appear, and the computer system needs to update the preset fault type vector representation to adapt to new fault diagnosis needs. In addition, similarity conditions (such as distance threshold T) may also need to be adjusted based on actual operating experience and feedback on the accuracy of fault diagnosis, so as to continuously improve the accuracy and reliability of fault type determination.

[0111] As an implementation method, the process of determining the preset fault type includes:

[0112] Step S101: obtaining T fault dimensions corresponding to a fault protection triggering event that triggers a response in the operating state of a target photovoltaic module, wherein the fault dimension includes one or more fault sub-dimensions, and the fault protection triggering event is a triggering response of a target photovoltaic module power generation system in the operating state of the target photovoltaic module;

[0113] Step S102: Integrate the fault sub-dimensions corresponding to the T fault dimensions to obtain a preset fault type.

[0114] In step S101, the computer system obtains T fault dimensions corresponding to the trigger response fault protection trigger event in the operating state of the target photovoltaic module, where the fault dimension covers one or more fault sub-dimensions. The fault dimension is a classification method for a comprehensive description of the fault, which describes the characteristics of the fault protection trigger event in the operating state of the target photovoltaic module from multiple angles.

[0115] Taking the operating status of photovoltaic modules as an example, one of the fault dimensions can be the electrical characteristics dimension. This dimension contains multiple fault sub-dimensions, such as voltage fluctuations, current anomalies, etc. Assume that at a certain moment, the computer system detects that the output voltage of the photovoltaic module has frequent ups and down fluctuations, exceeding the normal fluctuation range, and the current is also unstable. These phenomena are all manifestations of fault sub-dimensions under the fault dimension of electrical characteristics.

[0116] Another fault dimension may be the environmental impact dimension. Fault sub-dimensions under this dimension may include excessively high or low temperatures, abnormal light intensity, etc. For example, when the ambient temperature of a photovoltaic module is too high, it may affect the photovoltaic conversion efficiency of the module, thereby triggering a series of possible fault protection triggering events. The computer system obtains information related to these fault dimensions through various sensors and monitoring devices. For monitoring of the electrical characteristics dimension, the computer system can obtain voltage and current data through voltage sensors and current sensors installed in the photovoltaic module circuit; for the environmental impact dimension, the temperature sensor and light intensity sensor can provide temperature and light intensity information respectively.

[0117] In step S102, the computer system integrates the fault sub-dimensions corresponding to the T fault dimensions, thereby obtaining a preset fault type. This integration process is a complex process that comprehensively considers the relationship between each fault dimension and its sub-dimension.

[0118] Continuing with the previous example, the computer system begins to integrate the relevant information of the electrical characteristics dimension (including sub-dimensions such as voltage fluctuation and current anomaly) and the environmental impact dimension (including sub-dimensions such as excessive temperature and abnormal light intensity). Assuming that in a specific operating scenario, it is often the case that when the temperature is too high, the electrical characteristics of the photovoltaic module will change significantly, such as increased voltage fluctuation and abnormal current. The computer system will integrate these related fault sub-dimensions into a preset fault type, such as "abnormal electrical characteristics fault caused by high temperature".

[0119] The integration of computer systems can be achieved through data analysis and pattern recognition. One possible approach is to construct a correlation matrix. Assume that there are three failure dimensions: , each fault dimension has several fault sub-dimensions. For each pair of fault sub-dimensions (representing the jth fault sub-dimension under the i-th fault dimension) and (representing the lth fault sub-dimension under the kth fault dimension), the computer system determines the correlation between them based on historical data or pre-set rules , thereby constructing the correlation matrix A. For example, if it is found based on historical data that when the temperature is too high (a sub-dimension under the environmental impact dimension), the probability of voltage fluctuation of the photovoltaic module (a sub-dimension under the electrical characteristics dimension) is high, then the correlation between these two fault sub-dimensions is will be set to a higher value, such as 0.8.

[0120] Then, the computer system can integrate the fault sub-dimensions by clustering the association matrix. The clustering algorithm can cluster related fault sub-dimensions into one category according to the degree of association, thereby forming a preset fault category. For example, using the K-Means clustering algorithm, the computer system regards the fault sub-dimensions as data points and divides them into different clusters according to the degree of association in the association matrix. Each cluster can be defined as a preset fault category.

[0121] Determining the preset fault types in this way helps to more comprehensively and accurately describe the possible fault conditions in the target photovoltaic module power generation system. In the actual operation of photovoltaic module power generation systems, faults are often the result of the interaction of multiple factors, and a single-dimensional fault description may not accurately cover all fault characteristics. By integrating multiple fault dimensions and their sub-dimensions, a preset fault type that is more in line with the actual situation can be constructed, thereby providing a more effective basis for subsequent fault protection trigger event analysis, fault type determination, and protection false trigger identification.

[0122] At the same time, with the continuous development of photovoltaic module power generation systems and changes in the operating environment, the computer system continuously updates and optimizes the process of determining the preset fault types. For example, with the application of new photovoltaic module materials and technologies, new fault dimensions or fault sub-dimensions may appear, and the computer system promptly incorporates these new factors into the process of determining the preset fault types. In addition, with the in-depth study of the fault mechanism, the correlation degree in the original correlation matrix may need to be re-evaluated and adjusted, and the parameters of the clustering algorithm may also need to be optimized to ensure that the preset fault types can accurately reflect the actual fault conditions and improve the accuracy and effectiveness of the entire fault management system.

[0123] As an implementation manner, step S101, obtaining T fault dimensions corresponding to the trigger response fault protection trigger event in the operating state of the target photovoltaic component, includes:

[0124] Step S1011: based on past fault protection triggering events that triggered responses in the operating state of the target photovoltaic component, T fault dimensions are determined; wherein the fault dimension includes a fault cumulative impact dimension, a fault sub-dimension corresponding to the fault protection triggering event under the fault cumulative impact dimension is used to indicate the fault cumulative effect value corresponding to the fault protection triggering event, a fault sub-dimension corresponding to the preset fault type under the fault cumulative impact dimension is used to indicate the fault cumulative effect value interval corresponding to the preset fault type, and the fault cumulative effect value corresponding to the fault protection triggering event is located in the fault cumulative effect value interval corresponding to the fault type to which the fault protection triggering event belongs.

[0125] In the fault analysis process of the target photovoltaic module power generation system, step S1011 in step S101 is executed by a computer system, aiming to determine T fault dimensions based on past fault protection triggering events that triggered responses in the operating state of the target photovoltaic module. Among them, the fault dimension includes the fault cumulative impact dimension, etc. This process involves in-depth mining and analysis of data related to past fault protection triggering events.

[0126] The fault cumulative impact dimension is a dimension that reflects the impact of faults on the system as time or the number of events accumulates. Taking photovoltaic modules as an example, during their operation, some problems may accumulate over time, which may eventually trigger fault protection triggering events. For example, the accumulation of dust on the surface of photovoltaic modules is a gradual accumulation process. As dust continues to accumulate, its impact on the photovoltaic module's photoelectric conversion efficiency will gradually increase. In this process, the computer system can use relevant factors such as the amount of dust accumulation and the degree of decrease in photoelectric conversion efficiency as parameters under the fault cumulative impact dimension.

[0127] Assume that the computer system monitors the photovoltaic conversion efficiency of the photovoltaic module over a period of time. Initial photoelectric conversion efficiency There is the following relationship: the decrease ratio of photoelectric conversion efficiency When p reaches a certain threshold, it may trigger a corresponding fault protection trigger event, such as an alarm when the output power is lower than the normal range. The computer system uses the photoelectric conversion efficiency reduction ratio p as an important parameter in the fault cumulative impact dimension to measure the degree of fault impact in this dimension.

[0128] The computer system can determine the fault dimension based on data mining and analysis algorithms. For the data of past fault protection triggering events, the computer system first collects and organizes the data. These data may come from multiple sensors, such as sensors that monitor the voltage, current, temperature, etc. of photovoltaic modules, as well as log files that record fault alarm information.

[0129] For example, the computer system collects a series of data about the fluctuation of the output power of photovoltaic panels, which includes the power values ​​at different time points. And the corresponding ambient temperature and light intensity Computer systems can process this data using time series analysis methods.

[0130] A simple time series analysis method is the moving average method. (Here we take the output power of photovoltaic modules as an example. ), calculate the moving average of length m By observing the changing trend of the moving average, the computer system can find out whether there are regular changes in power fluctuations, such as whether there is a gradual downward or unstable trend over time. If it is found that the power is gradually decreasing, and this downward trend has no obvious direct correlation with the ambient temperature or light intensity, the computer system may attribute it to the influence of a certain fault dimension, such as internal aging of components, which becomes a basis for determining the fault dimension.

[0131] In addition to time series analysis, computer systems can also use association rule mining algorithms. For example, for different parameters in fault protection triggering events, such as voltage V, current I, and temperature T, the computer system analyzes a large amount of past data to find association rules between them. If it is found that when the voltage V drops to a certain level, the current I will also decrease accordingly, and this situation is more obvious when the temperature T is in a certain range, the computer system can attribute this relationship between these three parameters to a fault dimension. For example, it is defined as the "voltage-current-temperature associated fault dimension", and the fault sub-dimensions under this dimension can be the voltage drop amplitude, current reduction ratio, and temperature range, etc.

[0132] The computer system continuously mines various fault dimensions by continuously monitoring past fault protection triggering events, collecting data, and using a variety of data analysis algorithms. These fault dimensions cover multiple aspects, from simple physical quantity changes to complex multi-factor correlations, and can fully reflect the possible fault conditions in the target photovoltaic module power generation system. This provides important basic data for subsequent operations such as accurately determining the preset fault type, building the fault protection trigger graph structure, and identifying false protection triggers, which helps to improve the accuracy and effectiveness of fault management of the entire photovoltaic module power generation system.

[0133] In the actual operation of photovoltaic power generation systems, due to the complexity of the system and the variability of the environment, faults manifest in various forms. The computer system can adapt to such complex and changing situations by determining the fault dimension based on past fault protection trigger events, timely discover potential fault factors, and take corresponding measures to prevent and repair them, thereby ensuring the stable operation and efficient power generation of photovoltaic power generation systems.

[0134] As an implementation method, if the target photovoltaic component operating state is the linkage device energy exchange state, the fault dimension includes the linkage device involvement relationship dimension, and the fault sub-dimension corresponding to the fault protection trigger event under the linkage device involvement relationship dimension is used to indicate the device characteristics of the linkage device involved in the fault protection trigger event in the linkage device energy exchange state, and the fault sub-dimension corresponding to the preset fault type under the linkage device involvement relationship dimension is used to indicate the device characteristics of the linkage device involved in the fault protection trigger event belonging to the preset fault type in the linkage device energy exchange state.

[0135] When the target PV module is in the state of linked device energy exchange, the operation of the entire system involves the coordinated work of multiple devices and the interaction of energy. This is a more complex operating state, and the computer system performs a detailed analysis of various situations to ensure the normal operation of the system.

[0136] The dimension of the relationship between linkage equipment is an important dimension used to describe the relationship between various equipment and their association with fault protection triggering events in the state of linkage equipment energy exchange. In this state, the photovoltaic module does not operate independently, but is connected to other equipment such as inverters, energy storage devices, monitoring systems, etc. to exchange energy.

[0137] For example, in a small distributed energy system that includes photovoltaic panels, inverters and energy storage batteries, when in the state of linked equipment energy exchange, the inverter plays a key role in converting the DC power generated by the photovoltaic panels into AC power so that it can be connected to the power grid or used by local loads. The energy storage battery is used to store energy when photovoltaic power generation is in excess, and release energy when photovoltaic power generation is insufficient to maintain a stable power supply for the system.

[0138] Suppose a fault protection trigger event occurs in this system, such as a sudden drop in the charging efficiency of the energy storage battery. Under the dimension of linkage device involvement, the fault sub-dimension corresponding to this fault protection trigger event will indicate the device characteristics of the energy storage battery (as an involved linkage device) in the linkage device energy exchange state. These device characteristics may include parameters such as charging current, charging voltage, battery temperature, and battery internal resistance. For example, the computer system detects that when a fault occurs, the charging current of the energy storage battery is lower than the normal level, the charging voltage fluctuates abnormally, and the battery temperature is too high. These parameter values ​​constitute the fault sub-dimension information for the fault protection trigger event under the dimension of linkage device involvement.

[0139] For preset fault types, the corresponding fault sub-dimension under the linkage device involvement relationship dimension also indicates the equipment characteristics of the linkage equipment involved in the fault protection trigger event belonging to the preset fault type in the linkage equipment energy exchange state. Continuing with the example of the decrease in the charging efficiency of the energy storage battery, if it is defined as a preset fault type, then under this preset fault type, the fault sub-dimension corresponding to the linkage device involvement relationship dimension may contain some typical equipment characteristic ranges. For example, during normal charging, the charging current should be between I1 and I2, the charging voltage should be stable at around V1, and the battery temperature should be maintained between T1 and T2. If it exceeds these ranges, it may be judged to meet the situation of this preset fault type.

[0140] When the computer system monitors and analyzes the energy exchange status of such linkage devices, there are many ways to implement it. The monitoring of device characteristics can be achieved by installing sensors on each linkage device. For example, current sensors, voltage sensors, and temperature sensors are installed on energy storage batteries to obtain data on the battery's charging current I, charging voltage V, and battery temperature T in real time.

[0141] When analyzing data to determine the relationship between device characteristics and fault protection triggering events, the computer system can use a threshold-based judgment method. Assume that the lower limit threshold of normal charging current is , the upper threshold is , when the monitored charging current I meets When the battery temperature is lower than the charging voltage V, the computer system will regard it as an abnormal situation that may be related to the fault. Similar thresholds can also be set for the charging voltage V and battery temperature T. Make a judgment.

[0142] In addition, the computer system can also use data trend analysis methods. For example, by analyzing the charging current data of the energy storage battery over a period of time, if it is found that the charging current shows a continuous downward trend, even if it has not reached the set lower limit threshold, the computer system can use it as a potential fault warning signal. Simple linear regression analysis can be used to determine the trend of the data. For a set of charging current data The corresponding time point , let the linear regression equation be I=mt+b, where m is the slope and b is the intercept. If m<0, it means that the charging current is on a downward trend.

[0143] In the state of energy exchange of linked devices, the mutual influence relationship between different devices is complex and diverse. For example, the failure of the inverter may affect the output power adjustment strategy of the photovoltaic module, and then affect the charging and discharging process of the energy storage battery. If the inverter fails to output unstable power, it may cause the output power of the photovoltaic module to fluctuate frequently. Such fluctuations may cause the charging controller of the energy storage battery to frequently adjust the charging parameters, thereby accelerating the aging of the battery or causing a decrease in charging efficiency.

[0144] In this case, the computer system comprehensively considers the equipment characteristics of each linkage device and the mutual influence relationship between them to accurately determine the nature of the fault protection trigger event and the possible preset fault types. For example, when it is detected that the charging efficiency of the energy storage battery decreases, the computer system must not only analyze the equipment characteristic parameters of the energy storage battery itself, but also consider relevant factors such as whether the output power of the inverter is stable and whether the output power of the photovoltaic module is normal. Only by comprehensively considering these factors can the fault sub-dimension corresponding to the fault protection trigger event and the fault sub-dimension corresponding to the preset fault type be accurately determined under the dimension of the linkage device involvement relationship, thereby providing a reliable basis for the fault diagnosis, prevention and repair of the entire linkage device energy exchange system, and ensuring the stable operation of the system and the high efficiency of energy exchange.

[0145] In summary, when the target PV module is in the energy exchange state of the linkage equipment, the linkage equipment involvement relationship dimension provides a framework for the computer system to comprehensively analyze the relationship between fault protection triggering events and equipment characteristics. Monitoring and analyzing various parameters under this dimension in a reasonable implementation method helps to improve the reliability and stability of the entire system.

[0146] As an implementation mode, the method further includes a training process of a target false triggering recognition model, including the following steps:

[0147] Step S1: obtaining a debugging learning sample with a priori mark, the priori mark is used to indicate whether the photovoltaic module power generation system sample has a protection action false triggering in the sample monitoring period, and the debugging learning sample is used to indicate the sample fault protection triggering graph structure corresponding to the photovoltaic module power generation system sample in the sample monitoring period and the fault type vector representation of the graph vertex included in the sample fault protection triggering graph structure;

[0148] Step S2: according to the debugging learning sample, based on the initialization false trigger identification model, determining the reasoning information indicating whether the photovoltaic component power generation system sample has a protection action false trigger in the sample monitoring period;

[0149] Step S3: According to the loss between the inference information and the priori mark, the initialization false trigger recognition model is debugged to obtain the target false trigger recognition model.

[0150] In step S1, the computer system obtains a debugging learning sample with a priori mark. The priori mark is used to indicate whether the photovoltaic module power generation system sample has a protection action false triggering in the sample monitoring cycle, and the debugging learning sample is used to indicate the sample fault protection trigger graph structure corresponding to the photovoltaic module power generation system sample in the sample monitoring cycle and the fault type vector representation of the graph vertex included in the sample fault protection trigger graph structure.

[0151] For example, suppose there are multiple photovoltaic module power generation system samples. For one of the samples, within a specific sample monitoring period (such as one month), the computer system records all fault protection trigger events of the system and their related information, thereby constructing a sample fault protection trigger graph structure. Assuming that there are three fault protection trigger events in this sample system, namely, photovoltaic module short circuit fault, inverter overtemperature fault, and combiner box overcurrent fault, the computer system constructs a graph structure based on their triggering sequence, fault types, and mutual relationships. If the photovoltaic module short circuit fault occurs first, then it may cause the inverter overtemperature fault, and finally the combiner box overcurrent fault, then there will be corresponding graph vertices in the graph structure to represent these three fault types, and there will be connecting lines to represent their triggering sequence relationship.

[0152] At the same time, for each vertex of the graph, there is a corresponding fault type vector representation. Taking the photovoltaic module short circuit fault as an example, its fault type vector representation may contain information such as the short circuit location and the current change during the short circuit. The computer system integrates this information as a debugging learning example. In addition, this example has a priori mark. If it is determined that there is a false triggering of the protection action in this example through manual or other accurate detection means, then the priori mark is "yes"; if there is no false triggering of the protection action, then the priori mark is "no".

[0153] The computer system can obtain these debugging learning samples by collecting and sorting data from a large number of photovoltaic module power generation systems that are actually in operation. Data collection points can be set up in different photovoltaic power stations to collect relevant data such as fault protection trigger events and equipment operating parameters. Then, data mining technology is used to filter out debugging learning samples that meet the requirements from these massive data. For example, it can be filtered according to a specific time window (sample monitoring cycle) and fault-related keywords (such as various fault names).

[0154] In step S2, the computer system determines, according to the debugging learning sample and based on the initialization false trigger identification model, reasoning information indicating whether the photovoltaic assembly power generation system sample has a false triggering of a protective action during the sample monitoring period.

[0155] The initialization false trigger identification model is a pre-built model, which may have some initial parameter settings. This model can be built based on a neural network, such as a multi-layer perceptron (MLP) model. Taking the MLP model as an example, its input layer receives data related to the sample fault protection trigger graph structure and the fault type vector representation in the debugging learning sample. Assume that the connection relationship and influence coefficient in the sample fault protection trigger graph structure can be converted into a matrix form of data, and the fault type vector is represented as a vector data. After appropriate preprocessing (such as normalization), these data are used as input to the input layer.

[0156] In the MLP model, feature extraction and nonlinear transformation are performed through the hidden layer. For example, suppose there are n neurons in the input layer, h neurons in the hidden layer, and the input vector is , the output of the hidden layer The formula can be Calculate, where is the connection weight from the ith neuron in the input layer to the jth neuron in the hidden layer, is the bias term of the jth neuron in the hidden layer, and f is the activation function (such as the ReLU function). Through such calculations, the hidden layer extracts features from the input data, and then the output layer makes the final classification judgment based on the output of the hidden layer.

[0157] For the output layer, if it is a binary classification problem (i.e., judging whether there is a false triggering of the protection action), the logistic regression function can be used as the activation function of the output layer. Suppose the connection weight from the hidden layer to the output layer is (j represents the hidden layer neuron, k represents the output layer neuron, where k=1), the bias term of the output layer is , then the output z of the output layer can be expressed by the formula Calculate, where g is the logistic regression function The value of z indicates the probability of false triggering of the protection action in the PV module power generation system sample. If z>0.5, the reasoning information is judged as "yes", indicating that there is false triggering of the protection action; if z 0.5, the reasoning information is judged as “no”, indicating that there is no false triggering of the protective action.

[0158] In step S3, the computer system debugs the initialized false trigger recognition model according to the loss between the inference information and the priori mark to obtain the target false trigger recognition model.

[0159] Loss is an indicator used to measure the difference between the inference information and the prior label. In the binary classification problem, the commonly used loss function is the cross entropy loss function. Let the prior label be y (y=1 means there is a false triggering of the protective action, y=0 means there is no false triggering of the protective action), and the probability obtained by the inference information be z, then the cross entropy loss function .

[0160] The computer system debugs the initialized false trigger recognition model according to the value of this loss function. The debugging method can adopt an optimization algorithm based on gradient descent, such as the stochastic gradient descent (SGD) algorithm. In the SGD algorithm, the parameters of the model (such as the connection weight w and the bias term b) are updated according to the gradient of the loss function. Assume that the learning rate is , for a parameter (can be w or b), the update formula is .

[0161] By continuously inputting debugging learning samples, the computer system calculates the loss between the reasoning information and the prior mark, and then uses the gradient descent algorithm to update the model parameters so that the model's prediction results are closer and closer to the prior mark. After multiple iterations, when the value of the loss function converges to a smaller value, the target false trigger identification model is obtained. This target false trigger identification model can more accurately identify whether there is a false triggering of the protective action in the photovoltaic module power generation system.

[0162] Throughout the training process, the computer system processes a large number of debugging learning examples, and it may be necessary to adjust the model's structure and parameter settings to improve the model's performance. For example, if it is found that the model performs poorly on certain types of fail-safe trigger graph structures, you can consider increasing the number of neurons in the hidden layer or adjusting the type of activation function. At the same time, the learning rate The choice is also very important. Too large may cause the model to skip the optimal solution during training; if If it is too small, the model training speed will be very slow. Therefore, the computer system is adjusted according to the actual situation to ensure an accurate and efficient target false trigger identification model, thereby providing reliable protection against false trigger identification for the stable operation of the photovoltaic module power generation system.

[0163] In addition, with the continuous development of photovoltaic module power generation systems and the emergence of new fault conditions, the computer system may need to regularly update and debug learning samples and retrain the target false trigger recognition model to adapt to new situations and improve the generalization ability and accuracy of the model. For example, when new photovoltaic module technology or new equipment is introduced into the power generation system, the pattern of fault protection trigger events may change, and the original model may no longer be applicable. At this time, the above training process needs to be repeated.

[0164] As an embodiment, the method further includes:

[0165] Step S400: if the identification information indicates that the target photovoltaic assembly power generation system has a protection action false triggering during the data collection cycle, then determining a target fault protection triggering event corresponding to the protection action false triggering;

[0166] Step S500: marking photovoltaic equipment related to a target fault protection triggering event.

[0167] In step S400, if the identification information indicates that the target photovoltaic module power generation system has a protection action false triggering during the data collection cycle, the computer system will determine the target fault protection triggering event corresponding to the protection action false triggering. When the identification information obtained by the computer system in step S300 indicates that there is a protection action false triggering, this means that in the previously constructed fault protection triggering graph structure and the analysis process based on the target false triggering identification model, it was found that the actual triggered protection action was inconsistent with the expectation.

[0168] For example, assume that in the target photovoltaic module power generation system, the fault protection trigger graph structure includes relevant graph vertices and their connection relationships such as the overtemperature fault of the photovoltaic module, the overload fault of the inverter, and the overcurrent fault of the junction box. Under normal circumstances, when the temperature of the photovoltaic module is too high, subsequent protection actions may be triggered in a certain logical order, such as reducing the working power of the module or starting the heat dissipation device. However, if the identification information shows that there is a false triggering of the protection action, the computer system determines which fault protection triggering event caused this false triggering.

[0169] The computer system can determine the target fault protection trigger event by tracing back the previously constructed fault protection trigger graph structure and analyzing the data related to each fault protection trigger event. For each connection line and its influence coefficient in the fault protection trigger graph structure, the computer system can re-evaluate whether these relationships are abnormally triggered during the data collection cycle. For example, if it is found that the influence coefficient of a certain connection line should be low under normal circumstances, but it is abnormally high during the data collection cycle, this may indicate that there is a problem with the fault protection trigger event corresponding to both ends of the connection line. At the same time, the computer system can also view the detailed data corresponding to each fault protection trigger event, such as the actual temperature value when the photovoltaic module temperature over-high fault occurs, the rate of temperature rise, etc., as well as other parameter changes related to it. Through comprehensive analysis of these data, the computer system can determine which (or which) fault protection trigger event is the source of the false triggering of the protection action, and determine it as the target fault protection trigger event.

[0170] Next, in step S500, the computer system marks the photovoltaic equipment related to the target fault protection triggering event. The purpose of marking is to mark the equipment related to the false triggering so as to perform targeted inspection, maintenance or further analysis later.

[0171] For example, if it is determined that the overtemperature fault of a photovoltaic module is a target fault protection trigger event, the computer system will mark the photovoltaic module. This marking can contain a variety of information, such as creating a special marking field for the photovoltaic module in the database to mark it as related to the false triggering of the protection action. The marking content can be a simple identifier or more detailed information, such as the time when the false trigger occurred, the operating environment data at that time, etc.

[0172] The computer system can implement labeling by modifying the relevant records in the device management database. Assuming that there is a table about photovoltaic modules in the device management database, which contains information such as the module number, location, performance parameters, etc., the computer system can add a column to this table for labeling, or modify an existing column to add labeling information. If a new column is named "false trigger labeling", when a photovoltaic module is determined to be a device related to the target fault protection trigger event, relevant information can be written in the "false trigger labeling" column corresponding to the component, such as "false triggering of the protection action occurred at [specific time], and the temperature was [specific temperature value] at that time."

[0173] This kind of labeling is of great significance for the maintenance and management of photovoltaic module power generation systems. On the one hand, when technicians conduct equipment inspections, they can quickly identify the equipment related to the false triggering of protective actions, so as to focus on the operating status of these equipment and check whether there are potential faults or abnormal equipment performance. For example, for the labeled photovoltaic modules, technicians can check in detail whether their temperature sensors are working properly and whether there are faults in the heat dissipation device. On the other hand, these labeled data can serve as an important basis for subsequent analysis to improve the fault protection triggering mechanism or optimize the target false triggering identification model. If a device is found to be frequently labeled as related to the false triggering of protective actions, this may mean that there is a problem in the fault judgment logic or protective action triggering logic of the device, which requires further in-depth research and adjustment.

[0174] In the actual operation of the photovoltaic module power generation system, steps S400-S500 help improve the reliability and stability of the system. As the scale of the system continues to expand and the operating environment becomes increasingly complex, the false triggering of the protection action may have a significant impact on the normal operation of the system. By accurately determining the target fault protection trigger event and marking the relevant equipment, the problem can be discovered in time, and effective measures can be taken to correct and prevent it, thereby ensuring the efficient operation of the photovoltaic module power generation system and reducing unnecessary equipment losses and power generation losses caused by false triggering. At the same time, this also provides strong support for the intelligent management of the entire power generation system, allowing the system to continuously optimize its own fault protection and operation management mechanisms.

[0175] The present application embodiment provides a computer system, such as Figure 2 As shown, the computer system 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present application.

[0176] An embodiment of the present application provides a computer system, and the computer system in the embodiment of the present application includes: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processor, the above-mentioned photovoltaic module power generation system data acquisition control method is implemented.

Claims

1. A data acquisition and control method for a photovoltaic module power generation system, characterized in that: The method comprises: Obtain Y fault protection trigger events corresponding to the target photovoltaic module power generation system in the data collection cycle; According to the fault types corresponding to the Y fault protection trigger events and the order in which the Y fault protection trigger events are triggered in the data collection cycle, a fault protection trigger graph structure including X graph vertices is constructed, and different graph vertices are used to indicate different fault types. According to the order in which the fault protection trigger events are triggered, for a fault protection trigger event tuple consisting of an m-th fault protection trigger event and an n-th fault protection trigger event, where n=m+1; if the m-th fault protection trigger event is a first fault type, and the n-th fault protection trigger event is a second fault type, then there is a target connection line from the first graph vertex to the second graph vertex in the fault protection trigger graph structure, the first graph vertex is used to indicate the first fault type, the second graph vertex is used to indicate the second fault type, and the influence coefficient of the target connection line is used to indicate the trigger parameter of the fault protection trigger event tuple; According to the fault protection trigger graph structure and the fault type vector representations corresponding to the X graph vertices, protection false triggering identification is performed on the target photovoltaic assembly power generation system based on the target false triggering identification model to obtain identification information indicating whether the target photovoltaic assembly power generation system has a protection action false triggering in the data collection period; Among them, the method of identifying the protection false triggering of the target photovoltaic module power generation system based on the target false triggering identification model according to the fault protection triggering graph structure and the fault type vector representations corresponding to the X graph vertices respectively includes: Determine a relation array corresponding to the fault protection trigger graph structure, the array size of the relation array is X×X, let a and b be used to indicate different graph vertices, a and b are both positive integers less than or equal to X, in the relation array, the array unit located in the ath column and the bth row is used to indicate the influence coefficient of the connection line from the graph vertex represented by a to the graph vertex represented by b; According to the relationship array and the fault type vector representations respectively corresponding to the X graph vertices, protection false triggering identification is performed on the target photovoltaic assembly power generation system based on the target false triggering identification model.

2. The method according to claim 1, characterized in that The method of identifying the protection false triggering of the target photovoltaic assembly power generation system based on the target false triggering identification model according to the relationship array and the fault type vector representations corresponding to the X graph vertices respectively includes: Determine a graph vertex array according to the fault type vector representations respectively corresponding to the X graph vertices, the graph vertex array comprising X columns of array units, and let c be a positive integer less than or equal to X, c is used to indicate a graph vertex, and the c-th column of array units is used to indicate the fault type vector representation corresponding to the graph vertex represented by c; According to the relationship array and the graph vertex array, protection false triggering identification is performed on the target photovoltaic assembly power generation system based on the target false triggering identification model.

3. The method according to claim 2, characterized in that The step of performing protection false triggering identification on the target photovoltaic assembly power generation system based on the target false triggering identification model according to the relationship array and the graph vertex array includes: According to the fault type vector representation corresponding to each graph vertex in the graph vertex array, an initial state representation is generated for each graph vertex, and a state set consisting of the initial state representations of all graph vertices is obtained, wherein the initial state representation includes the fault type information of the graph vertex itself and the influence of other graph vertices adjacent to it; The multi-layer perceptron in the target false trigger recognition model is used to update the initial state representation of each graph vertex in the state set to obtain an intermediate state set, and the intermediate state set is processed by a nonlinear transformation enhancement network to obtain an updated state set; Based on the connection relationship in the relationship array, aggregating each graph vertex state representation in the update state set with the state representation of its neighboring graph vertices to obtain an aggregated state set; Inputting the aggregated state set into a recurrent neural network layer in a target false trigger identification model, mining the dependency relationship of the fault protection triggering event in the time series, and obtaining a dependency state set; Perform a global pooling operation on the dependency state set to obtain a pooling representation vector; The pooled characterization vector is transferred to a classifier in a target false triggering identification model, and the pooled characterization vector is classified based on the classifier to obtain an identification result of whether a protective action is falsely triggered.

4. The method according to claim 1, characterized in that: The trigger parameter is obtained based on one or both of the following two acquisition strategies: Acquisition strategy 1: based on the Y fault protection triggering events and the triggered order, determine the total triggering times of the fault protection triggering event binary group as the triggering parameter; Acquisition strategy two: Based on the Y fault protection trigger events and the triggering order, determine the fault cumulative effect value corresponding to the fault protection trigger event that triggers the second fault type in response to the fault protection trigger event binary as the trigger parameter.

5. The method according to claim 1, characterized in that The method further comprises: Comparing the protection event vector representation corresponding to the fault protection trigger event with the fault type vector representation corresponding to the preset fault type for similarity, the protection event vector representation corresponding to the fault protection trigger event is used to characterize the fault protection trigger event, and the fault type vector representation corresponding to the preset fault type is used to characterize the preset fault type; Determine the fault type corresponding to the fault protection triggering event based on preset fault types that meet similarity conditions; The process of determining the preset fault type includes: Acquire T fault dimensions corresponding to a trigger response fault protection trigger event in a target photovoltaic assembly operation state, wherein the fault dimension includes one or more fault sub-dimensions, and the fault protection trigger event is a trigger response of the target photovoltaic assembly power generation system in the target photovoltaic assembly operation state; The fault sub-dimensions corresponding to the T fault dimensions are integrated to obtain the preset fault type.

6. The method according to claim 5, characterized in that The step of obtaining T fault dimensions corresponding to the trigger response fault protection trigger event in the target photovoltaic assembly operation state includes: Determining T fault dimensions based on past fault protection triggering events that triggered responses in the target photovoltaic component operating state; The fault dimension includes a fault cumulative impact dimension, and the fault sub-dimension corresponding to the fault protection trigger event under the fault cumulative impact dimension is used to indicate the fault cumulative effect value corresponding to the fault protection trigger event, and the fault sub-dimension corresponding to the preset fault type under the fault cumulative impact dimension is used to indicate the fault cumulative effect value interval corresponding to the preset fault type, and the fault cumulative effect value corresponding to the fault protection trigger event is located in the fault cumulative effect value interval corresponding to the fault type to which the fault protection trigger event belongs.

7. The method according to claim 5, characterized in that If the target photovoltaic component operating state is the linkage device energy exchange state, the fault dimension includes the linkage device involvement relationship dimension, and the fault sub-dimension corresponding to the fault protection trigger event under the linkage device involvement relationship dimension is used to indicate the device characteristics of the linkage device involved in the fault protection trigger event in the linkage device energy exchange state, and the fault sub-dimension corresponding to the preset fault type under the linkage device involvement relationship dimension is used to indicate the device characteristics of the linkage device involved in the fault protection trigger event belonging to the preset fault type in the linkage device energy exchange state.

8. The method according to claim 1, characterized in that The method also includes a training process of a target false triggering recognition model, comprising the following steps: Acquire a debugging learning sample with a priori mark, wherein the priori mark is used to indicate whether the photovoltaic component power generation system sample has a protection action false triggering in the sample monitoring period, and the debugging learning sample is used to indicate the sample fault protection triggering graph structure corresponding to the photovoltaic component power generation system sample in the sample monitoring period and the fault type vector representation of the graph vertex included in the sample fault protection triggering graph structure; According to the debugging learning sample, based on the initialization false trigger identification model, determining the reasoning information indicating whether the photovoltaic assembly power generation system sample has a protection action false trigger in the sample monitoring period; According to the loss between the inference information and the priori mark, the initialization false trigger recognition model is debugged to obtain the target false trigger recognition model; The method further comprises: If the identification information indicates that the target photovoltaic assembly power generation system has a protection action false triggering during the data collection period, determining a target fault protection triggering event corresponding to the protection action false triggering; The photovoltaic equipment related to the target fault protection triggering event is marked.

9. A computer system, characterized in that: include: one or more processors; Memory; one or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 8 is implemented.

Images