Risk assessment method and risk assessment system for distributed photovoltaic system

Abstract

The invention discloses a risk assessment method and system for a distributed photovoltaic system, and the method comprises the steps: carrying out the risk assessment of a pre-trained distributed photovoltaic system risk assessment model based on real-time system data collected from a to-be-assessed distributed photovoltaic system, and obtaining the real-time risk information; the training process of the distributed photovoltaic system risk assessment model comprises the following steps: generating a logic diagram matched with a sample distributed photovoltaic system; based on the logic diagram and sample system data, generating a training sample, the sample system data being data acquired in advance from the sample distributed photovoltaic system; and using the training sample to train a neural network to obtain a distributed photovoltaic system risk assessment model, so that the reliability and timeliness of risk assessment can be improved, and a user can accurately, reliably and timely check out the risk in the distributed photovoltaic system so as to facilitate further risk response.

Description

Technical Field

[0001] The present application relates to the field of power system safety technology, and in particular to a risk assessment method and a risk assessment system for a distributed photovoltaic system. Background Art

[0002] With the increasing maturity of photovoltaic power generation technology and the widespread application of communication technology, distributed photovoltaics, as a clean and flexible new energy solution, are increasingly attracting widespread attention. Different from traditional centralized photovoltaic power generation, distributed photovoltaics are usually deployed directly on the user side. This power generation method can not only utilize local resources, but also connect many sub-nodes in the region through the microgrid system to form a highly flexible energy network. In this network, each node can exchange and monitor data in real time through communication technology to achieve accurate management and scheduling of power generation and power consumption of each node. This distributed structure and deployment mode on the user side enable photovoltaic power generation to achieve efficient utilization of "generated and used", reducing the loss of electricity that may be generated during long-distance transmission. In addition, the modular design of the distributed photovoltaic system allows more users to directly participate in energy production, which not only improves energy utilization, but also greatly reduces dependence on traditional power grids. With the support of communication technology, these distributed photovoltaic nodes can interact flexibly with the central power grid, and can also sell excess electricity to the grid to achieve two-way energy exchange with the grid.

[0003] However, the high efficiency and flexibility of distributed photovoltaic systems and user-side photovoltaic grid-connected power transactions are highly dependent on the massive and diverse heterogeneous sensor monitoring equipment (i.e., some nodes in the system) and the corresponding communication technology support within the distributed photovoltaic system, which can easily become a potential entry point for malicious attacks, thereby destroying the stable operation of the distributed photovoltaic system and interfering with the normal operation of power transactions. At the same time, the distributed structure determines the large number of internal sub-nodes, resulting in a large number of attack entrances. The sub-nodes are interconnected and tightly coupled, which allows risks to be further propagated through the correlation and coupling between the sub-nodes, thereby increasing the system complexity of distributed photovoltaics.

[0004] In related technologies, the risk assessment method of distributed photovoltaics is mainly based on inherent assessment rules and system models to conduct qualitative assessments on specific fault types. It is difficult to adapt to the complex coupling relationships within distributed photovoltaic systems and has limitations under various new types of malicious attacks, resulting in insufficient reliability and timeliness of risk assessment of distributed photovoltaic systems. Summary of the invention

[0005] In order to solve the above technical problems, the embodiments of the present application propose a risk assessment method and a risk assessment system for a distributed photovoltaic system, which can improve the reliability and timeliness of risk assessment, so that users can accurately, reliably and timely identify risks in the distributed photovoltaic system to facilitate further risk response.

[0006] In a first aspect, an embodiment of the present application provides a risk assessment method for a distributed photovoltaic system, comprising:

[0007] Based on the real-time system data collected from the distributed photovoltaic system to be evaluated, a pre-trained distributed photovoltaic system risk assessment model is used to perform risk assessment and obtain real-time risk information;

[0008] The training process of the distributed photovoltaic system risk assessment model includes:

[0009] Generate a logic diagram matching a sample distributed photovoltaic system;

[0010] Generate training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system;

[0011] The training samples are used to train a neural network to obtain a distributed photovoltaic system risk assessment model.

[0012] Optionally, the logic diagram is a directed logic diagram, and generating a logic diagram matching the sample distributed photovoltaic system includes:

[0013] According to the data interaction logic between each sample sub-node in the sample distributed photovoltaic system, a tree-like logic diagram of each sample sub-node is constructed;

[0014] The directed logic graph is determined based on the tree-like logic graph and the topological structure of each sample sub-node.

[0015] Optionally, the sample system data includes power generation data and / or power consumption data of each of the sample sub-nodes;

[0016] The data interaction logic between the sample sub-nodes is obtained by performing dependency analysis on the power generation data and / or power consumption data of the sample sub-nodes.

[0017] Optionally, the tree-like logic graph includes a plurality of tree-like nodes, the plurality of tree-like nodes include N root nodes and at least one leaf node associated with each root node, the plurality of tree-like nodes correspond one-to-one to the respective sample sub-nodes, and the topological structure includes nodes corresponding one-to-one to the respective sample sub-nodes, N≥1, and determining the directed logic graph based on the tree-like logic graph and the topological structure of the respective sample sub-nodes includes:

[0018] According to the N root nodes and their respective associated leaf nodes, connecting the nodes in the topological structure to form the directed logical graph;

[0019] The connection modes of the topological structure include:

[0020] For each root node, the node corresponding to the root node in the topological structure is connected to the node corresponding to the leaf node associated with the root node in the topological structure.

[0021] Optionally, generating a training sample based on the logic diagram and the sample system data includes:

[0022] Performing a two-dimensional structural construction on each node in the logic diagram, so that each node in the logic diagram is configured to have two data channels, wherein one of the two data channels is used to carry power generation data in the sample system data, and the other is used to carry power consumption data in the sample system data;

[0023] Performing zero-fill processing on each configured node in the logic graph;

[0024] Based on the sample system data and the logic graph after zero-filling processing, construct a tensor for training as the training sample;

[0025] The real-time system data is data obtained through the two-dimensional structured construction.

[0026] Optionally, using the training samples to train a neural network to obtain a distributed photovoltaic system risk assessment model includes:

[0027] Inputting the training sample into the neural network for prediction to obtain a prediction result;

[0028] Based on the error between the prediction result and the training sample, the neural network is trained to obtain a distributed photovoltaic system risk assessment model after training.

[0029] Optionally, the neural network is an autoencoder model based on a deep convolutional neural network, the autoencoder model includes an encoder and a decoder, and the sample system data is data generated when the sample distributed photovoltaic system operates normally;

[0030] The step of inputting the training sample into the neural network for prediction to obtain a prediction result includes:

[0031] Inputting the training samples into the encoder, so that the encoder extracts low-dimensional potential features from the training samples that match the normal operation of the sample distributed photovoltaic system;

[0032] Inputting the low-dimensional latent features into the decoder, so that the decoder reconstructs the low-dimensional latent features to generate reconstructed data, and uses the reconstructed data as the prediction result;

[0033] The step of training the neural network based on the error between the prediction result and the training sample to obtain a distributed photovoltaic system risk assessment model after training includes:

[0034] Determine a reconstruction error between the reconstructed data and the training sample, train the autoencoder model with the goal of minimizing the reconstruction error, and obtain a distributed photovoltaic system risk assessment model after training;

[0035] The autoencoder model is trained using forward propagation, and the reconstruction error is determined by calculating the mean square error loss.

[0036] Optionally, the risk assessment is performed based on the real-time system data collected from the distributed photovoltaic system to be assessed using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information, including:

[0037] Input the real-time system data into the distributed photovoltaic system risk assessment model for prediction, obtain a real-time prediction result, and determine a real-time error between the real-time prediction result and the real-time system data, wherein the real-time error includes a real-time node error corresponding to at least one sub-node to be evaluated in the distributed photovoltaic system to be evaluated;

[0038] Based on the comparison result between the real-time node error and the set threshold, determining the risk assessment index value corresponding to each of the at least one sub-node to be assessed;

[0039] The real-time risk information is generated according to the risk assessment indicator value.

[0040] Optionally, determining the risk assessment index value corresponding to each of the at least one to-be-assessed sub-nodes based on the comparison result between the real-time node error and a set threshold value includes:

[0041] Based on the comparison result between the real-time node error and the set threshold, the risk assessment index value corresponding to each of the at least one sub-node to be assessed is generated by a maximum normalization MMN function.

[0042] In a second aspect, an embodiment of the present application provides a risk assessment system for a distributed photovoltaic system, including:

[0043] An assessment module, used to perform risk assessment based on real-time system data collected from the distributed photovoltaic system to be assessed using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information;

[0044] The training process of the distributed photovoltaic system risk assessment model includes:

[0045] Generate a logic diagram matching a sample distributed photovoltaic system;

[0046] Generate training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system;

[0047] The training samples are used to train a neural network to obtain a distributed photovoltaic system risk assessment model.

[0048] In summary, the embodiments of the present application have at least the following beneficial effects:

[0049] According to an embodiment of the present application, risk assessment is performed using a pre-trained distributed photovoltaic system risk assessment model based on real-time system data collected from the distributed photovoltaic system to be evaluated to obtain real-time risk information; wherein, the training process of the distributed photovoltaic system risk assessment model includes: generating a logic diagram matching a sample distributed photovoltaic system; generating training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system; using the training samples, training a neural network to obtain a distributed photovoltaic system risk assessment model, thereby improving the reliability and timeliness of risk assessment, allowing users to accurately, reliably and timely identify risks in distributed photovoltaic systems to facilitate further risk response. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of a process of a risk assessment method for a distributed photovoltaic system provided in an embodiment of the present application;

[0051] Figure 2It is a flowchart of the training process of the distributed photovoltaic system risk assessment model provided in the embodiment of the present application;

[0052] Figure 3 is a schematic diagram of a directed logic graph provided in an embodiment of the present application;

[0053] Figure 4 It is a schematic diagram of two-dimensional dual-channel data construction and local feature extraction provided by an embodiment of the present application;

[0054] Figure 5 is a schematic diagram of a neural network architecture provided in an embodiment of the present application;

[0055] Figure 6 is a schematic diagram of the distribution of distributed photovoltaic sub-nodes under malicious attacks provided in an embodiment of the present application;

[0056] Figure 7 It is a schematic diagram of the structure of a risk assessment system for a distributed photovoltaic system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] In the description of the present application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "multiple" is two or more. In the description of the present application, the term "including" and its variations are open inclusions, i.e., "including but not limited to". The term "based on" means "at least partially based on". The term "according to" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments".

[0059] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0060] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0061] The following is an explanation of some terminology concepts involved in the embodiments of the present application:

[0062] Min-Max Normalization (MMN), also often referred to as minimum-maximum scaling, is a commonly used feature scaling technique used to scale data to a specific range, usually between [0,1].

[0063] First, see Figure 1 , shows a schematic flow chart of a risk assessment method for a distributed photovoltaic system provided in an embodiment of the present application, the method comprising step S101, which is as follows:

[0064] S101, based on real-time system data collected from the distributed photovoltaic system to be evaluated, using a pre-trained distributed photovoltaic system risk assessment model to perform risk assessment to obtain real-time risk information;

[0065] Among them, see Figure 2 The training process of the distributed photovoltaic system risk assessment model includes:

[0066] S201, generating a logic diagram matching a sample distributed photovoltaic system;

[0067] In one example, the data interaction logic of the distributed system sub-nodes under the distributed photovoltaic system (e.g., distributed photovoltaic microgrid) can be analyzed, and the power generation data and / or power consumption data of the grid-connected inverter, power convergence point, and user-side smart meter in the system can be collected to construct a tree-like logic diagram of data interaction between sub-nodes, thereby combing the internal interaction logic of distributed photovoltaics. In the distributed photovoltaic microgrid, the sub-node data interaction logic involves data interaction between the grid-connected inverter, the power convergence point, and the user-side smart meter. The grid-connected inverter is suitable for converting the direct current of photovoltaic power generation into alternating current and collecting power generation data. The power convergence point is suitable for integrating and distributing the power generation of different inverters. The user-side smart meter is suitable for real-time monitoring of power consumption data. Each node exchanges data in real time through the communication network to ensure the balance of power generation and power consumption and efficient operation of the system. According to the logical connection and topological structure of distributed photovoltaics, the dependency relationship between power generation and power consumption data is determined, wherein all data in this embodiment represent the normal situation of the system.

[0068] S202, generating training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system;

[0069] In one example, the sample system data is specifically data collected in advance from the sample distributed photovoltaic system and preprocessed, wherein the preprocessing includes at least one of the following: data screening, data cleaning, and data normalization. In this embodiment, first, records containing missing values ​​and abnormal values ​​in the collected data can be deleted. Subsequently, the data is subjected to the minimum-maximum normalization included in the data normalization to eliminate the scale differences between different features, which can improve the stability and effect of model training. This embodiment can ensure the accuracy and consistency of the data.

[0070] S203, using the training samples to train a neural network to obtain a distributed photovoltaic system risk assessment model.

[0071] In an optional implementation, the logic diagram is a directed logic diagram, and generating a logic diagram matching the sample distributed photovoltaic system includes:

[0072] According to the data interaction logic between each sample sub-node in the sample distributed photovoltaic system, a tree-like logic diagram of each sample sub-node is constructed;

[0073] The directed logic graph is determined based on the tree-like logic graph and the topological structure of each sample sub-node.

[0074] In an optional implementation, the sample system data includes power generation data and / or power consumption data of each of the sample sub-nodes;

[0075] The data interaction logic between the sample sub-nodes is obtained by performing dependency analysis on the power generation data and / or power consumption data of the sample sub-nodes.

[0076] In an optional implementation, the tree-like logic graph includes a plurality of tree-like nodes, the plurality of tree-like nodes include N root nodes and at least one leaf node associated with each root node, the plurality of tree-like nodes correspond one-to-one to the respective sample sub-nodes, and the topological structure includes nodes corresponding one-to-one to the respective sample sub-nodes, N≥1, and determining the directed logic graph based on the tree-like logic graph and the topological structure of the respective sample sub-nodes includes:

[0077] According to the N root nodes and their respective associated leaf nodes, connecting the nodes in the topological structure to form the directed logical graph;

[0078] The connection modes of the topological structure include:

[0079] For each root node, the node corresponding to the root node in the topological structure is connected to the node corresponding to the leaf node associated with the root node in the topological structure.

[0080] Understandably, see Figure 3 In this embodiment, after the corresponding nodes of the root node and its associated leaf nodes in the topological structure are connected together, it is equivalent to making the root node and its associated leaf nodes be able to be arranged adjacently after being mapped to the directed logical graph.

[0081] In an optional implementation, generating a training sample based on the logic diagram and the sample system data includes:

[0082] Performing a two-dimensional structural construction on each node in the logic diagram, so that each node in the logic diagram is configured to have two data channels, wherein one of the two data channels is used to carry power generation data in the sample system data, and the other is used to carry power consumption data in the sample system data;

[0083] Performing zero-fill processing on each configured node in the logic graph;

[0084] Based on the sample system data and the logic graph after zero-filling processing, construct a tensor for training as the training sample;

[0085] The real-time system data is data obtained through the two-dimensional structured construction.

[0086] In this embodiment, by performing zero-filling processing on the edge of two-dimensional data for each configured node in the logic graph, the data size can be effectively controlled.

[0087] In one example, the trained model can be deployed to a distributed photovoltaic microgrid scenario to collect real-time power generation and consumption data from grid-connected inverters, power aggregation points, and user sides in distributed photovoltaic microgrid components. The model is constructed according to the dimensions and channels defined by the network architecture, and a two-channel two-dimensional data structure is input into the model to obtain the reconstruction error of each sub-node.

[0088] In an optional implementation, the using the training samples to train a neural network to obtain a distributed photovoltaic system risk assessment model includes:

[0089] Inputting the training sample into the neural network for prediction to obtain a prediction result;

[0090] Based on the error between the prediction result and the training sample, the neural network is trained to obtain a distributed photovoltaic system risk assessment model after training.

[0091] In an optional implementation, the neural network is an autoencoder model based on a deep convolutional neural network, the autoencoder model includes an encoder and a decoder, and the sample system data is data generated when the sample distributed photovoltaic system operates normally;

[0092] The step of inputting the training sample into the neural network for prediction to obtain a prediction result includes:

[0093] Inputting the training samples into the encoder, so that the encoder extracts low-dimensional potential features from the training samples that match the normal operation of the sample distributed photovoltaic system;

[0094] The low-dimensional latent features are input into the decoder, so that the decoder reconstructs the low-dimensional latent features to generate reconstructed data, and uses the reconstructed data as the prediction result.

[0095] In this embodiment, the intrinsic patterns and implicit features (represented by low-dimensional latent features) under normal operating conditions and the local features of business interactions between sub-nodes in a distributed system can be extracted.

[0096] In one example, see Figure 4 and Figure 5 The autoencoder model can include convolutional layers, pooling layers, and deconvolutional layers, all of which can be built using Python language based on pytorch platform tools and related libraries.

[0097] In one example, the encoder can be composed of a stack of convolutional layers and pooling layers, wherein the first convolutional layer has an input channel number of 2 and an output channel number of 16, and is used to preliminarily extract local information to obtain a feature map; the maximum pooling layer is used to downsample the feature map, thereby simplifying data representation, reducing the amount of computation, and improving the generalization ability of the model; the second convolutional layer has an input channel number of 16, and is used to output an 8-channel feature map based on the data obtained by downsampling, and passes through the pooling layer again to capture low-dimensional potential features under normal operating conditions of the system.

[0098] In one example, the decoder can be composed of deconvolution layers, which are symmetrical with the encoder structure. The first deconvolution layer has 8 input channels and 16 output channels, and the second deconvolution layer has 16 input channels and 2 output channels, thereby restoring the low-dimensional latent features to data in the same format as the data in the input encoder, thereby completing the reconstruction of the data in the input encoder.

[0099] The step of training the neural network based on the error between the prediction result and the training sample to obtain a distributed photovoltaic system risk assessment model after training includes:

[0100] Determine a reconstruction error between the reconstructed data and the training sample, train the autoencoder model with the goal of minimizing the reconstruction error, and obtain a distributed photovoltaic system risk assessment model after training;

[0101] The autoencoder model is trained using forward propagation, and the reconstruction error is determined by calculating the mean squared error (MSE) loss.

[0102] In one example, the input vector (input data) input into the autoencoder model can be represented as x={x1,x1,…,x N}, the forward propagation can be expressed by the following formula:

[0103]

[0104] Among them, T en Represents the encoding process of the encoder, that is, mapping from the input vector x to a low-dimensional latent feature representing the normal operation characteristics of the system The process of de Represents the decoding process of the decoder, that is, reconstructing the low-dimensional potential features into an output x with the same dimension as the input vector R The process of en represents the weight matrix of the encoder, B en Represents the encoder bias parameter, W de represents the weight matrix of the decoder, Bde Represents the bias parameters of the decoder.

[0105] In one example, the mean square error loss can be expressed as follows:

[0106]

[0107] Among them, x i,j represents the input data of node (i, j) (determined by the training sample), Represents the output data of node (i, j) (determined by the reconstructed data).

[0108] In this embodiment, the Adam optimizer can be selected, the learning rate can be set to 0.005, and the stochastic gradient descent method can be used to train the autoencoder model. During the training process, the weight matrix W of the encoder is adjusted. en and the bias parameter B en , and the decoder weight matrix W de and the bias parameter B de Through these adjustments, the model is able to reconstruct the PV microgrid data under normal operating conditions, thereby achieving the goal of unsupervised learning.

[0109] In an optional implementation, the risk assessment is performed based on the real-time system data collected from the distributed photovoltaic system to be assessed using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information, including:

[0110] Input the real-time system data into the distributed photovoltaic system risk assessment model for prediction, obtain a real-time prediction result, and determine a real-time error between the real-time prediction result and the real-time system data, wherein the real-time error includes a real-time node error corresponding to at least one sub-node to be evaluated in the distributed photovoltaic system to be evaluated;

[0111] Based on the comparison result between the real-time node error and the set threshold, determining the risk assessment index value corresponding to each of the at least one sub-node to be assessed;

[0112] The real-time risk information is generated according to the risk assessment indicator value.

[0113] In an optional implementation, determining the risk assessment index value corresponding to each of the at least one to-be-assessed sub-nodes based on the comparison result between the real-time node error and a set threshold value includes:

[0114] Based on the comparison result between the real-time node error and the set threshold, the risk assessment index value corresponding to each of the at least one sub-node to be assessed is generated by a maximum normalization MMN function.

[0115] It can be understood that this embodiment reconstructs the error and sets the threshold based on the real-time data collection, and performs threshold judgment and normalization processing through the evaluation function (maximum normalization MMN function, which is used to map the value to the range of 0-1), and quantifies the risk assessment index of each sub-node under malicious attack, thereby evaluating the risk level and position of each sub-node under malicious attack of distributed photovoltaic.

[0116] In one example, the risk assessment index value can be determined by the following formula:

[0117]

[0118] Among them, σ i,j Represents the risk assessment index value corresponding to node (i, j), Represents the real-time system data x of node (i, j) i,j Real-time prediction results of node (i, j) The real-time error between ε and max is the set threshold. i,j It represents the maximum real-time error among all nodes traversed. The real-time error in this embodiment is preferably determined by the mean squared error (MSE) loss.

[0119] In this embodiment, see Figure 6, this embodiment can be applied to the security analysis of distributed photovoltaic in complex dynamic environment, which is helpful to enhance the reliability of distributed photovoltaic operation. Compared with traditional centralized photovoltaic power station, distributed photovoltaic realizes the instant use of photovoltaic power generation by placing photovoltaic power generation near multiple users, and connects multiple user nodes, power aggregation nodes and grid-connected inverters through microgrids, forming local consumption of clean energy in local areas, and through the deployment of a large number of IoT terminal devices and the application of intelligent communication technology, it realizes the implementation of monitoring of power generation and electricity consumption data, laying the foundation for power market economic scenarios such as photovoltaic access to the Internet. However, the massive terminals and communication technologies inside distributed photovoltaics expose vulnerabilities under malicious attacks, and a risk assessment method is urgently needed. Combined with the business scenarios of distributed photovoltaics and the structural characteristics of distributed systems, real-time quantitative risk assessment is performed for multiple malicious attacks, thereby revealing the risk level of each internal sub-node and the risk position of distributed photovoltaics. Based on the characteristics of highly coupled physical information within distributed photovoltaics, this embodiment considers the interconnected relationships such as data consistency of power generation and power consumption of sub-nodes in distributed systems. First, dual-channel two-dimensional structured data is constructed, and the interconnected relationships between sub-nodes are represented by the local features of the two-dimensional data graph; then, an autoencoder model based on a convolutional neural network is constructed, and the neural network is trained using normal business data without attacks, extracting the local features of the two-dimensional data, and refining the business data characteristics with distributed system characteristics into a low-dimensional coding space, and reconstructing the two-dimensional data through a deconvolutional neural network, thereby training the neural network with the purpose of minimizing the reconstruction error. The reconstruction error of each node is obtained according to the input data, and the risk index of each sub-node in the normalized distributed photovoltaic is obtained through the risk assessment function, thereby indicating the risk level and position of each node in the distributed photovoltaic, thereby realizing the risk assessment of the distributed photovoltaic under malicious attacks.

[0120] Accordingly, an embodiment of the present application also provides a risk assessment system for a distributed photovoltaic system, which can implement all processes of the risk assessment method for a distributed photovoltaic system provided in the above embodiment.

[0121] See also Figure 7 , shows a schematic diagram of the structure of a risk assessment system for a distributed photovoltaic system provided in an embodiment of the present application, the risk assessment system 700 includes:

[0122] An evaluation module 701 is used to perform risk evaluation based on real-time system data collected from the distributed photovoltaic system to be evaluated using a pre-trained distributed photovoltaic system risk evaluation model to obtain real-time risk information;

[0123] The training process of the distributed photovoltaic system risk assessment model includes:

[0124] Generate a logic diagram matching a sample distributed photovoltaic system;

[0125] Generate training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system;

[0126] The training samples are used to train a neural network to obtain a distributed photovoltaic system risk assessment model.

[0127] In an optional implementation, the logic diagram is a directed logic diagram, and generating a logic diagram matching the sample distributed photovoltaic system includes:

[0128] According to the data interaction logic between each sample sub-node in the sample distributed photovoltaic system, a tree-like logic diagram of each sample sub-node is constructed;

[0129] The directed logic graph is determined based on the tree-like logic graph and the topological structure of each sample sub-node.

[0130] In an optional implementation, the sample system data includes power generation data and / or power consumption data of each of the sample sub-nodes;

[0131] The data interaction logic between the sample sub-nodes is obtained by performing dependency analysis on the power generation data and / or power consumption data of the sample sub-nodes.

[0132] In an optional implementation, the tree-like logic graph includes a plurality of tree-like nodes, the plurality of tree-like nodes include N root nodes and at least one leaf node associated with each root node, the plurality of tree-like nodes correspond one-to-one to the respective sample sub-nodes, and the topological structure includes nodes corresponding one-to-one to the respective sample sub-nodes, N≥1, and determining the directed logic graph based on the tree-like logic graph and the topological structure of the respective sample sub-nodes includes:

[0133] According to the N root nodes and their respective associated leaf nodes, connecting the nodes in the topological structure to form the directed logical graph;

[0134] The connection modes of the topological structure include:

[0135] For each root node, the node corresponding to the root node in the topological structure is connected to the node corresponding to the leaf node associated with the root node in the topological structure.

[0136] In an optional implementation, generating a training sample based on the logic diagram and the sample system data includes:

[0137] Performing a two-dimensional structural construction on each node in the logic diagram, so that each node in the logic diagram is configured to have two data channels, wherein one of the two data channels is used to carry power generation data in the sample system data, and the other is used to carry power consumption data in the sample system data;

[0138] Performing zero-fill processing on each configured node in the logic graph;

[0139] Based on the sample system data and the logic graph after zero-filling processing, construct a tensor for training as the training sample;

[0140] The real-time system data is data obtained through the two-dimensional structured construction.

[0141] In an optional implementation, the using the training samples to train a neural network to obtain a distributed photovoltaic system risk assessment model includes:

[0142] Inputting the training sample into the neural network for prediction to obtain a prediction result;

[0143] Based on the error between the prediction result and the training sample, the neural network is trained to obtain a distributed photovoltaic system risk assessment model after training.

[0144] In an optional implementation, the neural network is an autoencoder model based on a deep convolutional neural network, the autoencoder model includes an encoder and a decoder, and the sample system data is data generated when the sample distributed photovoltaic system operates normally;

[0145] The step of inputting the training sample into the neural network for prediction to obtain a prediction result includes:

[0146] Inputting the training samples into the encoder, so that the encoder extracts low-dimensional potential features from the training samples that match the normal operation of the sample distributed photovoltaic system;

[0147] Inputting the low-dimensional latent features into the decoder, so that the decoder reconstructs the low-dimensional latent features to generate reconstructed data, and uses the reconstructed data as the prediction result;

[0148] The step of training the neural network based on the error between the prediction result and the training sample to obtain a distributed photovoltaic system risk assessment model after training includes:

[0149] Determine a reconstruction error between the reconstructed data and the training sample, and train the autoencoder model with the goal of minimizing the reconstruction error to obtain a distributed photovoltaic system risk assessment model after training;

[0150] The autoencoder model is trained using forward propagation, and the reconstruction error is determined by calculating the mean square error loss.

[0151] In an optional implementation, the risk assessment is performed based on the real-time system data collected from the distributed photovoltaic system to be assessed using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information, including:

[0152] Input the real-time system data into the distributed photovoltaic system risk assessment model for prediction, obtain a real-time prediction result, and determine a real-time error between the real-time prediction result and the real-time system data, wherein the real-time error includes a real-time node error corresponding to at least one sub-node to be evaluated in the distributed photovoltaic system to be evaluated;

[0153] Based on the comparison result between the real-time node error and the set threshold, determining the risk assessment index value corresponding to each of the at least one sub-node to be assessed;

[0154] The real-time risk information is generated according to the risk assessment indicator value.

[0155] In an optional implementation, determining the risk assessment index value corresponding to each of the at least one to-be-assessed sub-nodes based on the comparison result between the real-time node error and a set threshold value includes:

[0156] Based on the comparison result between the real-time node error and the set threshold, the risk assessment index value corresponding to each of the at least one sub-node to be assessed is generated by a maximum normalization MMN function.

[0157] In summary, the embodiments of the present application have at least the following beneficial effects:

[0158] According to an embodiment of the present application, risk assessment is performed using a pre-trained distributed photovoltaic system risk assessment model based on real-time system data collected from the distributed photovoltaic system to be evaluated to obtain real-time risk information; wherein, the training process of the distributed photovoltaic system risk assessment model includes: generating a logic diagram matching a sample distributed photovoltaic system; generating training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system; using the training samples, training a neural network to obtain a distributed photovoltaic system risk assessment model, thereby improving the reliability and timeliness of risk assessment, allowing users to accurately, reliably and timely identify risks in distributed photovoltaic systems to facilitate further risk response.

[0159] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary hardware platform, and of course, it can also be implemented entirely by hardware. Based on such an understanding, all or part of the contribution of the technical solution of the present application to the background technology can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0160] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications are also considered to be within the scope of protection of the present application.

Claims

1. A risk assessment method for a distributed photovoltaic system, characterized in that: include: Based on the real-time system data collected from the distributed photovoltaic system to be evaluated, a pre-trained distributed photovoltaic system risk assessment model is used to perform risk assessment and obtain real-time risk information; The training process of the distributed photovoltaic system risk assessment model includes: Generate a logic diagram matching a sample distributed photovoltaic system; Generate training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system; The training samples are used to train a neural network to obtain a distributed photovoltaic system risk assessment model.

2. The method according to claim 1, characterized in that The logic diagram is a directed logic diagram, and the generating of the logic diagram matching the sample distributed photovoltaic system includes: According to the data interaction logic between each sample sub-node in the sample distributed photovoltaic system, a tree-like logic diagram of each sample sub-node is constructed; The directed logic graph is determined based on the tree-like logic graph and the topological structure of each sample sub-node.

3. The method according to claim 2, characterized in that The sample system data includes power generation data and / or power consumption data of each of the sample sub-nodes; The data interaction logic between the sample sub-nodes is obtained by performing dependency analysis on the power generation data and / or power consumption data of the sample sub-nodes.

4. The method according to claim 2, characterized in that The tree-like logic graph includes a plurality of tree-like nodes, the plurality of tree-like nodes include N root nodes and at least one leaf node associated with each root node, the plurality of tree-like nodes correspond to the respective sample sub-nodes one-to-one, and the topological structure includes nodes corresponding to the respective sample sub-nodes one-to-one, N≥1, and determining the directed logic graph based on the tree-like logic graph and the topological structure of the respective sample sub-nodes includes: According to the N root nodes and their respective associated leaf nodes, connecting the nodes in the topological structure to form the directed logical graph; The connection modes of the topological structure include: For each root node, the node corresponding to the root node in the topological structure is connected to the node corresponding to the leaf node associated with the root node in the topological structure.

5. The method according to claim 1, characterized in that The generating of training samples based on the logic diagram and the sample system data comprises: Performing a two-dimensional structural construction on each node in the logic diagram, so that each node in the logic diagram is configured to have two data channels, wherein one of the two data channels is used to carry power generation data in the sample system data, and the other is used to carry power consumption data in the sample system data; Performing zero-fill processing on each configured node in the logic graph; Based on the sample system data and the logic graph after zero-filling processing, construct a tensor for training as the training sample; The real-time system data is data obtained through the two-dimensional structured construction.

6. The method according to claim 1, characterized in that The method of using the training samples to train a neural network to obtain a distributed photovoltaic system risk assessment model includes: Inputting the training sample into the neural network for prediction to obtain a prediction result; Based on the error between the prediction result and the training sample, the neural network is trained to obtain a distributed photovoltaic system risk assessment model after training.

7. The method according to claim 6, characterized in that The neural network is an autoencoder model based on a deep convolutional neural network, the autoencoder model includes an encoder and a decoder, and the sample system data is data generated when the sample distributed photovoltaic system operates normally; The step of inputting the training sample into the neural network for prediction to obtain a prediction result includes: Inputting the training samples into the encoder, so that the encoder extracts low-dimensional potential features from the training samples that match the normal operation of the sample distributed photovoltaic system; Inputting the low-dimensional latent features into the decoder, so that the decoder reconstructs the low-dimensional latent features to generate reconstructed data, and uses the reconstructed data as the prediction result; The step of training the neural network based on the error between the prediction result and the training sample to obtain a distributed photovoltaic system risk assessment model after training includes: Determine a reconstruction error between the reconstructed data and the training sample, train the autoencoder model with the goal of minimizing the reconstruction error, and obtain a distributed photovoltaic system risk assessment model after training; The autoencoder model is trained using forward propagation, and the reconstruction error is determined by calculating the mean square error loss.

8. The method according to claim 1, characterized in that The method of performing risk assessment based on real-time system data collected from the distributed photovoltaic system to be assessed and using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information includes: Input the real-time system data into the distributed photovoltaic system risk assessment model for prediction, obtain a real-time prediction result, and determine a real-time error between the real-time prediction result and the real-time system data, wherein the real-time error includes a real-time node error corresponding to at least one sub-node to be evaluated in the distributed photovoltaic system to be evaluated; Based on the comparison result between the real-time node error and the set threshold, determining the risk assessment index value corresponding to each of the at least one sub-node to be assessed; The real-time risk information is generated according to the risk assessment indicator value.

9. The method according to claim 8, characterized in that The step of determining the risk assessment index value corresponding to each of the at least one sub-node to be assessed based on the comparison result between the real-time node error and the set threshold value comprises: Based on the comparison result between the real-time node error and the set threshold, the risk assessment index value corresponding to each of the at least one sub-node to be assessed is generated by a maximum normalization MMN function.

10. A risk assessment system for a distributed photovoltaic system, characterized in that: include: An assessment module, used to perform risk assessment based on real-time system data collected from the distributed photovoltaic system to be assessed using a pre-trained distributed photovoltaic system risk assessment model to obtain real-time risk information; The training process of the distributed photovoltaic system risk assessment model includes: Generate a logic diagram matching a sample distributed photovoltaic system; Generate training samples based on the logic diagram and sample system data, wherein the sample system data is data pre-collected from the sample distributed photovoltaic system; The training samples are used to train a neural network to obtain a distributed photovoltaic system risk assessment model.

Images