Power transformation equipment state evaluation probability graph network method and system considering operation condition
The substation equipment status assessment method that combines deep learning and graph networks solves the problem that the existing technology fails to consider operating conditions and dynamic changes, achieves efficient and accurate status assessment, and improves the adaptability and interpretability of the assessment.
Patent Information
- Application Number
- CN202510738968.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-04
- Publication Date
- 2025-10-17
AI Technical Summary
Existing substation equipment status assessment methods fail to fully consider operating conditions, have difficulty adapting to dynamic changes, are inefficient in processing complex data, and have poor model interpretability.
A deep learning algorithm is used to establish a state transition network. Combining working condition adaptive feature coding and graph network reasoning, a probabilistic graph network system for substation equipment state assessment is constructed. By obtaining equipment state parameters and environmental data, dynamic adaptive feature coding and state assessment are performed.
It achieves a comprehensive, accurate and real-time assessment of the status of substation equipment, improves the reliability and computational efficiency of the assessment results, maintains the interpretability of the model, and supports practical decision-making.
Smart Images
Figure CN120806181A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of power transformation equipment, and particularly to a power transformation equipment state evaluation probabilistic graph network method and system considering operating conditions. BACKGROUND
[0002] As a key component of the power system, the operation state of power transformation equipment directly affects the safety and reliability of the entire power grid.
[0003] With the rapid development of smart grid technology, higher requirements are placed on the accuracy and real-time performance of power transformation equipment state evaluation.
[0004] Traditional power transformation equipment state evaluation methods mainly rely on periodic inspections and simple threshold judgments, which are difficult to adapt to complex and changing operating environments and equipment states.
[0005] In recent years, with the progress of artificial intelligence and big data technology, data-driven power transformation equipment state evaluation methods have gradually become a research hotspot.
[0006] These methods collect and analyze a large amount of equipment operation data, and establish complex mathematical models to evaluate equipment states.
[0007] Among them, probabilistic graph models have attracted widespread attention due to their ability to effectively handle uncertainty and complex dependency relationships.
[0008] However, existing power transformation equipment state evaluation methods based on probabilistic graph models still have some significant limitations.
[0009] Firstly, most methods fail to adequately consider the operating conditions of power transformation equipment, resulting in a lack of accuracy and reliability in evaluation results under different environmental conditions.
[0010] Secondly, existing methods often use static model structures, which are difficult to adapt to dynamic changes in equipment states and environments.
[0011] In addition, when dealing with high-dimensional, multi-source heterogeneous data, the computational efficiency and scalability of existing methods also face challenges.
[0012] The closest prior art typically employs simple probabilistic graph models, such as Bayesian networks or Markov random fields, to describe the state transition process of power transformation equipment.
[0013] While these methods can capture the probabilistic dependency relationships of equipment states to some extent, they are not well-equipped to handle complex time-series data and multi-scale features.
[0014] At the same time, they often ignore the impact of operating conditions on equipment states, resulting in significant deviations between evaluation results in actual applications and the true situation.
[0015] On the other hand, some studies have attempted to introduce deep learning techniques to improve the accuracy of state assessment.
[0016] However, these methods usually use deep learning as an independent feature extraction tool and fail to organically combine it with probabilistic graphical models, resulting in poor model interpretability and difficulty in providing reliable support for actual decision-making. Summary of the Invention
[0017] In view of the shortcomings of existing technologies, there is an urgent need for a substation equipment status assessment method that can comprehensively consider operating conditions, adapt to dynamic changes, process complex data, and has high efficiency and scalability.
[0018] The present invention is an innovative solution proposed in response to this urgent need.
[0019] The present invention proposes a probability graph network method and system for substation equipment status assessment considering operating conditions, including: The acquisition steps include: Obtain the original state parameter sequence and operating environment data of the substation equipment; Processing steps include: Based on the original state parameter sequence, a state transition network between the state parameters of the substation equipment and the equipment state is established; Based on the state transition network, a deep learning algorithm is used to train and solve the prior probabilities of the initial state parameters and state transition model parameters of the substation equipment; Based on the operating environment data, construct a working condition adaptive feature coding module; Using the working condition adaptive feature coding module, the online state parameters of the substation equipment are subjected to working condition adaptive feature coding to obtain a priori probability of the working condition; Output steps include: Based on the prior probability of the operating condition, a statistical evaluation of the status of the substation equipment is performed through a graph network reasoning algorithm, and an evaluation result is output.
[0020] Preferably, the establishing of a state transition network between the state parameters of the power transformation equipment and the equipment state specifically includes: Define the initial state parameter sequence as ,in For the substation equipment Initial state parameter values; Define the initial state variable sequence of the substation equipment as ,in For the substation equipment Initial state variable values; Building a multi-layer state transfer network wherein is the number of network layers, is the first layer state transition network; state parameter nodes and state variable nodes are defined in each layer state transition network, and the connection relationship between nodes is established.
[0021] As a preferred, the prior probability of the initial state parameter and the state transition model parameter of the power transformation equipment trained and solved by the deep learning algorithm specifically includes: An online estimator algorithm including ridge regression and neural network is used to estimate the prior probability distribution and transition probability distribution of the node variable in the state transition network; An offline trainer algorithm including multilayer perceptron and convolutional neural network is used to estimate the prior probability distribution and transition probability distribution of the node variable in the state transition network; The probability distribution of the initial state parameter sequence of the power transformation equipment belonging to different working conditions is calculated wherein represents the first working condition.
[0022] As a preferred, the working condition adaptive feature encoding module specifically includes: A working condition perception feature extraction module of the power transformation equipment is constructed; The single online state parameter of the power transformation equipment is processed by using the LIN network to obtain the prior probability distribution of the state parameter belonging to different working conditions; The prior probability distribution of the online state sequence parameter of the power transformation equipment belonging to different working conditions is calculated.
[0023] As a preferred, the calculation formula of the prior probability distribution of the state parameter belonging to different working conditions is: , is the prior probability distribution of the first state parameter of the power transformation equipment belonging to the first working condition, is the prior probability of the first working condition, is the conditional probability of the first state parameter under the first working condition, is the marginal probability of the first state parameter.
[0024] As a preferred, the calculation formula of the prior probability distribution of the online state sequence parameter of the power transformation equipment belonging to different working conditions is: , in, For substation equipment No. 1 to No. The state parameter sequence of the online phase belongs to The prior probability distribution of the working conditions, For 1st to A sequence of state parameters for an online phase.
[0025] Preferably, the statistical evaluation of the status of the substation equipment by using a graph network reasoning algorithm specifically includes: Calculate the posterior probability distribution of the state transition network of the substation equipment; Calculate the probability of the state variables of the substation state variables: , For the Tier The probability distribution of state variable nodes, is the probability calculation function; Based on the posterior probability distribution, the status of the substation equipment is statistically evaluated.
[0026] Preferably, the statistical evaluation of the status of the power transformation equipment further comprises: Calculation of state parameters of substation equipment belongs to the The probability of the following working conditions: , in, is the state parameter sequence of the substation equipment, is the state variable sequence; Calculate the reliability evaluation index of substation equipment: , in, is the reliability index, For the The state variable sequence under different working conditions.
[0027] As an advantage, it also includes: Set environmental characteristic parameters based on working conditions, including ambient temperature, ambient humidity, current and voltage parameters; A convolution feature extraction module is used at the input end of the substation equipment status assessment to extract data features; Use the fully connected layer to perform probabilistic reasoning on state features; Factor graphs are introduced and message-based variational inference methods are used for probabilistic graph network reasoning.
[0028] The network system for probabilistic graphs of substation equipment status assessment considering operating conditions includes: Data acquisition module, used to obtain the original state parameter sequence and operating environment data of the substation equipment; Processing module, including: A network establishing unit, configured to establish a state transition network between the state parameters of the power transformation equipment and the equipment state based on the original state parameter sequence; A deep learning unit, configured to train and solve the prior probabilities of the initial state parameters and state transition model parameters of the substation equipment through a deep learning algorithm according to the state transition network; A feature coding unit is used to construct a working condition adaptive feature coding module based on the operating environment data, and use the module to perform working condition adaptive feature coding on the online state parameters of the substation equipment to obtain a priori probability of the working condition; an inference evaluation unit, configured to perform a statistical evaluation of the state of the substation equipment by using a graph network inference algorithm based on the prior probability of the operating condition; An output module is used to output the result of the statistical evaluation.
[0029] The present invention proposes a probability graph network method and system for substation status assessment considering operating conditions, which aims to solve key problems in the existing technology, such as failure to fully consider operating conditions, difficulty in adapting to dynamic changes, and low efficiency in processing complex data.
[0030] By innovatively integrating advanced technologies such as deep learning, condition-adaptive feature coding, and graph network reasoning, the present invention achieves a comprehensive, accurate, and real-time assessment of the status of substation equipment.
[0031] Specifically, the method of the present invention first establishes a dynamic relationship model between the state parameters of the substation equipment and the equipment state through a deep learning algorithm, overcoming the limitations of the traditional static model.
[0032] Secondly, the innovative introduction of the working condition adaptive feature coding module enables the evaluation process to fully consider the impact of different operating environments, greatly improving the accuracy and reliability of the evaluation results.
[0033] In addition, the use of graph network reasoning algorithms for state assessment can not only effectively process high-dimensional, multi-source heterogeneous data, but also significantly improve computing efficiency and model scalability.
[0034] The method and system of the present invention have shown significant advantages and beneficial effects in practical applications. First, by considering operating conditions, the present invention can maintain high-precision state assessment in various complex environments, greatly improving the reliability and practicality of the assessment results.
[0035] Secondly, the dynamic self-adaptive feature enables the present invention to quickly respond to changes in device status and environment, making it possible to discover potential faults in a timely manner.
[0036] Furthermore, the efficient graph network inference algorithm enables the invention to handle large-scale substation state evaluation tasks in real-time, meeting the stringent real-time requirements of smart grids.
[0037] More importantly, the method of the invention combines deep learning and probabilistic graph models organically, improving evaluation accuracy while maintaining good interpretability.
[0038] This feature is crucial for actual decision support, helping operation and maintenance personnel better understand evaluation results and develop more reasonable maintenance strategies.
[0039] In summary, the substation state evaluation probabilistic graph network method and system considering operating conditions proposed by the invention not only solves the key problems in the prior art, but also makes significant progress in accuracy, adaptability, efficiency and interpretability.
[0040] This innovative solution provides strong technical support for improving the safety and reliability of power grids, and has important theoretical value and broad application prospects. BRIEF DESCRIPTION OF DRAWINGS
[0041] Fig. 1 The network system module diagram of the invention; Fig. 2 The internal structure diagram of the processing module of the invention; Fig. 3 The probabilistic graph network system structure diagram of the invention. DETAILED DESCRIPTION
[0042] Please refer to the accompanying Figs. 1-3 The invention provides a substation state evaluation probabilistic graph network method and system considering operating conditions.
[0043] This method fully considers the state characteristics of substation equipment under different operating conditions, and realizes more accurate and reliable state evaluation through an innovative probabilistic graph network model.
[0044] Firstly, the method includes an acquisition step, a processing step and an output step.
[0045] In the acquisition step, the original state parameter sequence and operating environment data of the substation equipment are acquired.
[0046] These data are the basis for subsequent processing, among which the original state parameter sequence may include key indicators such as voltage, current and temperature of the equipment, while the operating environment data may involve external factors such as ambient temperature and humidity.
[0047] In the processing step, the method of the present application first establishes a state transition network between the state parameters of the power transformation equipment and the equipment state based on the obtained original state parameter sequence.
[0048] This network is the core of the entire evaluation method, which describes how the equipment state transitions over time and environmental changes.
[0049] Preferably, the present application adopts a multi-layer network structure, each layer representing a different time scale or state granularity, so as to more comprehensively capture the dynamic characteristics of the equipment state.
[0050] Next, the method trains and solves the prior probability of the initial state parameters and state transition model parameters of the power transformation equipment according to the established state transition network through a deep learning algorithm.
[0051] This step fully utilizes the advantages of deep learning in complex pattern recognition and nonlinear mapping, and can learn the internal rules of state transition from a large amount of historical data.
[0052] In one embodiment of the present application, a deep learning model suitable for processing sequence data such as recurrent neural network (RNN) or long short-term memory network (LSTM) can be used.
[0053] On the basis of considering the operating conditions, the method further constructs a condition-adaptive feature encoding module.
[0054] Based on the obtained operating environment data, this module can adaptively extract and encode features under different operating conditions.
[0055] This is an important innovation of the present application, because it enables the evaluation model to dynamically adjust according to the current operating environment, thereby improving the accuracy and adaptability of the evaluation.
[0056] Using the constructed condition-adaptive feature encoding module, the method performs condition-adaptive feature encoding on the online state parameters of the power transformation equipment, obtaining a condition prior probability.
[0057] This step combines real-time state parameters with the current operating conditions to generate a prior probability distribution that comprehensively considers the equipment state and operating environment.
[0058] Finally, in the output step, the method performs statistical evaluation of the state of the power transformation equipment based on the obtained condition prior probability through a graph network inference algorithm, and outputs the evaluation result.
[0059] The graph network inference algorithm used here can effectively handle complex probability dependencies, resulting in more reliable evaluation results.
[0060] Further, the application adopts a detailed mathematical model when establishing the state transition network between the state parameters of the power transformation equipment and the equipment state.
[0061] Specifically, the initial state parameter sequence is defined as , where is the initial state parameter value of the i-th power transformation equipment.
[0062] These parameters may include voltage, current, power factor, and other key indicators.
[0063] At the same time, the initial state variable sequence of the power transformation equipment is defined as , where is the initial state variable value of the i-th power transformation equipment, which may represent the working state and health degree of the equipment.
[0064] The application constructs a multi-layer state transition network , where is the number of network layers, is the i-th layer state transition network.
[0065] This multi-layer structure design enables the model to capture state transition characteristics at different time scales and different abstraction levels.
[0066] In each layer of the state transition network, state parameter nodes and state variable nodes are defined, and the connection relationship between the nodes is established.
[0067] This detailed network structure design provides a solid foundation for subsequent probabilistic reasoning.
[0068] When solving the prior probability of the initial state parameters and state transition model parameters of the power transformation equipment, the application adopts multiple deep learning algorithms.
[0069] Among them, the online estimator algorithm includes ridge regression and neural network. Ridge regression is suitable for handling multicollinearity problems, while neural network can handle complex nonlinear relationships.
[0070] For example, for ridge regression, the following formula can be used: , where is the estimated parameter vector, is the input matrix, is the output vector, is the regularization parameter, is the identity matrix.
[0071] The selection is usually between 0.1 and 10, and the specific value can be determined by cross-validation.
[0072] Meanwhile, the method also adopts an offline trainer algorithm, including a multilayer perceptron and a convolutional neural network.
[0073] These algorithms can learn complex patterns and features from a large amount of historical data. For example, for a multilayer perceptron, the following activation function can be used: , This is a commonly used ReLU activation function, which can effectively alleviate the gradient vanishing problem and speed up the training process of the network.
[0074] In addition, the application also calculates the probability distribution of the initial state parameter sequence of the power transformation equipment belonging to different working conditions , where represents the th working condition. The calculation of this probability distribution fully considers the initial state of the equipment and the possible operating conditions, providing important prior information for subsequent state assessment.
[0075] Through the above steps, the method of the application can comprehensively consider the operating conditions of the power transformation equipment and establish an accurate state assessment model.
[0076] This method not only improves the accuracy of the assessment, but also enhances the adaptability of the model to different operating environments, providing reliable decision support for the maintenance and management of power transformation equipment.
[0077] The method of the application further proposes an innovative working condition adaptive feature encoding module.
[0078] The construction of this module is the key to considering the operating conditions in the method, which enables the state assessment to dynamically adjust according to the actual operating environment, thereby improving the accuracy and adaptability of the assessment.
[0079] Specifically, the working condition adaptive feature encoding module first constructs a working condition perception feature extraction module for the power transformation equipment.
[0080] This module can extract key working condition features from complex operating environment data, laying the foundation for subsequent adaptive encoding.
[0081] Preferably, the application uses a multilayer perceptron structure to realize this function, where each layer focuses on extracting working condition features of different scales or abstraction levels.
[0082] Based on the working condition perception feature extraction, the method innovatively introduces a LIN network to process individual online state parameters of the power transformation equipment.
[0083] LIN network is a lightweight neural network structure that can quickly process input data under limited computing resources.
[0084] Through the processing of the LIN network, the method can obtain the prior probability distribution of the state parameters belonging to different working conditions. The introduction of this step greatly improves the recognition ability and adaptability of the method to different working conditions.
[0085] In order to more accurately describe the relationship between the state parameters and the working conditions, the present application proposes an innovative calculation formula: , represents the prior probability distribution of the first state parameter belonging to the first working condition, is the prior probability of the first working condition, is the conditional probability of the first state parameter under the first working condition, is the marginal probability of the first state parameter.
[0086] This formula ingeniously combines Bayesian theory and working condition characteristics, enabling the model to more accurately capture the complex relationship between state parameters and working conditions.
[0087] In practical applications, can be obtained by statistical analysis of historical data, such as the frequency of occurrence of a certain working condition; can be obtained by probability density estimation of historical data, preferably using kernel density estimation method. can be obtained by marginalization calculation of all working conditions.
[0088] The method of the present application not only considers a single state parameter, but also further calculates the prior probability distribution of the online state sequence parameter of the power transformation equipment belonging to different working conditions.
[0089] This innovation enables the method to fully utilize the time series information of the state parameters, thereby obtaining more reliable working condition judgment results.
[0090] The specific calculation formula is as follows: , represents the prior probability distribution of the state parameter sequence of the first to the tth online stage belonging to the first working condition, is the state parameter sequence of the first to the tth online stage.
[0091] This formula considers the time sequence characteristics of state parameters, and can better capture the change rule of working conditions over time.
[0092] In the preferred embodiment of the present application, It can be modeled by a Hidden Markov Model (HMM).
[0093] HMM can effectively process sequence data and capture the time sequence dependence of state parameter sequences.
[0094] Specifically, an HMM model can be trained for each working condition, and then the forward algorithm is used to calculate the likelihood probability of the sequence.
[0095] Based on the above working condition adaptive feature coding, the method further statistically evaluates the state of the power transformation equipment through a graph network inference algorithm.
[0096] This step makes full use of the working condition prior probability obtained in the previous step, and combines the powerful inference ability of the graph network to achieve accurate evaluation of the state of the power transformation equipment.
[0097] Specifically, the method first calculates the posterior probability distribution of the state transition network of the power transformation equipment.
[0098] This step considers the prior probability, observation data and network structure, and obtains the approximate posterior distribution through variational inference and other methods.
[0099] Then, the method calculates the probability of the state variable node state variable of the power transformation equipment, and the calculation formula is: , denotes the probability distribution of the state variable node at the layer and the th state variable node, is a probability calculation function, is the state transition network at the layer, is the corresponding state parameter node, is the relevant state variable node of the previous layer.
[0100] This formula reflects the hierarchical dependence between state variables, so that the model can capture complex state transition patterns.
[0101] In practical applications, the function can be designed as a neural network, with the state of the relevant nodes as input and the probability distribution of the target node as output.
[0102] Preferably, a graph convolutional network (GCN) can be used to implement this function, as GCN can effectively handle graph-structured data and capture the topological relationship between nodes.
[0103] Finally, based on the calculated posterior probability distribution, the method statistically evaluates the state of the power transformation equipment.
[0104] This step comprehensively considers various possible states and their probabilities, thereby obtaining a comprehensive and reliable evaluation result.
[0105] In this way, the method of the present application not only gives the current state evaluation of the power transformation equipment, but also provides the possible future state change trend, which provides an important decision basis for the preventive maintenance of the equipment.
[0106] The method of the present application further improves the statistical evaluation process of the state of the power transformation equipment.
[0107] When evaluating the state of the power transformation equipment, the method not only considers the single state probability, but also introduces a more comprehensive evaluation index, thereby providing a more reliable and meaningful evaluation result.
[0108] Specifically, the method calculates the probability that the state parameter of the power transformation equipment belongs to the i-th working condition.
[0109] This calculation process can be represented as: , represents the probability that the power transformation equipment belongs to the i-th working condition under the condition that the given state parameter sequence ; represents the state variable sequence.
[0110] This formula ingeniously combines the state parameter and the state variable, and obtains a more robust working condition probability estimate through marginalization.
[0111] In a preferred embodiment of the present application, can be obtained by the graph network reasoning algorithm described above, and can be approximately calculated by variational inference and the like.
[0112] This calculation method fully utilizes the advantages of the probabilistic graph model and can effectively handle complex conditional dependencies.
[0113] Further, the method innovatively introduces a reliability evaluation index of the power transformation equipment.
[0114] The calculation formula of this index is: , wherein, R represents a reliability index, P represents the prior probability of the th working condition, X represents the state variable sequence under the th working condition.
[0115] This index comprehensively considers the prior probability of the working condition, the working condition probability under the current state parameter, and the conditional probability of the state variable, providing a comprehensive reliability measurement.
[0116] In practical applications, the value range of the reliability index R is usually between 0 and 1.
[0117] Preferably, a threshold value, such as 0.8, can be set, and when the R value is lower than this threshold value, the system will issue a warning, prompting the need for equipment inspection or maintenance.
[0118] The selection of this threshold value can be adjusted according to specific application scenarios and requirements.
[0119] The method of the present invention further optimizes the overall process of substation equipment state assessment.
[0120] Firstly, the method sets environmental characteristic parameters based on working conditions, including environmental temperature, environmental humidity, current and voltage parameters.
[0121] The selection of these parameters is based on the actual situation of substation equipment operation, which can fully reflect the operating environment of the equipment.
[0122] For example, the value range of the environmental temperature may be between -20℃ and 50℃, and the range of the environmental humidity may be between 20% and 90%.
[0123] The range of current and voltage parameters needs to be determined according to the specific specifications of substation equipment.
[0124] Accurate measurement and recording of these parameters are crucial for accurate assessment of equipment state.
[0125] Secondly, the method uses a convolutional feature extraction module at the input end of the substation equipment state assessment.
[0126] The use of the convolutional feature extraction module enables the method to automatically learn and extract key features from raw data, reducing the workload of manual feature engineering and improving the efficiency and effectiveness of feature extraction.
[0127] In an embodiment of the present invention, the convolutional feature extraction module can adopt a multi-layer convolutional neural network structure.
[0128] For example, 3 layers of convolutional layers can be used, each followed by a max-pooling layer.
[0129] The first layer can be set to a kernel size of 3x3, the second layer to 5x5, and the third layer to 7x7. This gradually increasing kernel design can capture features of different scales.
[0130] Next, the method uses a fully connected layer to perform probabilistic inference on the state features.
[0131] The use of a fully connected layer allows the model to learn complex nonlinear relationships between features, resulting in a more accurate probability distribution.
[0132] Preferably, a multi-layer fully connected network can be used, such as a 3-layer structure, with the number of neurons decreasing layer by layer, such as 1024, 512, 256, and the output dimension of the last layer being the same as the number of state categories.
[0133] Finally, the method of the present application introduces a factor graph and uses a message-based variational inference method for probabilistic graph network inference.
[0134] The introduction of the factor graph allows complex probabilistic dependencies to be clearly represented, while the message-based variational inference method provides an efficient approximate inference algorithm.
[0135] This inference method can greatly reduce the computational complexity while ensuring the accuracy of the inference, allowing the method to be applied to large-scale real-time state evaluation tasks.
[0136] During variational inference, mean-field approximation or structured mean-field approximation can be used to simplify calculations.
[0137] The number of iterations can be set to 50 to 100, or until convergence (e.g., when the parameter changes between two consecutive iterations are less than a certain threshold, such as 0.001).
[0138] To implement the above method, the present application also proposes a power transformation device state evaluation probabilistic graph network system considering operating conditions.
[0139] The system includes a data acquisition module 1, a processing module 2, and an output module 3.
[0140] The data acquisition module 1 is responsible for acquiring the original state parameter sequence and operating environment data of the power transformation device.
[0141] This module can include various sensors and data acquisition devices to ensure that the required data can be acquired in real time and accurately.
[0142] The processing module 2 is the core of the system, which includes multiple functional units.
[0143] First is the network establishment unit 21, which establishes the state transition network between the state parameters of the power transformation equipment and the equipment state based on the original state parameter sequence.
[0144] This unit implements the network construction step in the method of the present application, laying the foundation for subsequent processing.
[0145] The deep learning unit 22 is responsible for training and solving the prior probability of the initial state parameters and state transition model parameters of the power transformation equipment according to the state transition network through a deep learning algorithm.
[0146] This unit implements the deep learning training step in the method of the present application, and is the key to the learning ability of the system.
[0147] The feature encoding unit 23 constructs a working condition adaptive feature encoding module based on the operating environment data, and uses the module to perform working condition adaptive feature encoding on the online state parameters of the power transformation equipment to obtain the working condition prior probability. This unit implements the working condition adaptive feature encoding step in the method of the present application, and is the core of the system considering the operating conditions.
[0148] The reasoning and evaluation unit 24 performs statistical evaluation on the state of the power transformation equipment based on the working condition prior probability through a graph network reasoning algorithm.
[0149] This unit implements the state evaluation step in the method of the present application, and is the key to outputting the final evaluation result of the system.
[0150] Finally, the output module 3 is responsible for outputting the results of the statistical evaluation.
[0151] This module can include display devices, alarm devices, etc., for presenting the evaluation results in an appropriate manner.
[0152] Through the collaborative work of the above modules and units, the system of the present application can comprehensively and accurately evaluate the state of the power transformation equipment, providing reliable decision support for the operation and maintenance management of the equipment.
[0153] It should be noted that: the above only describes the preferred embodiments of the present application, and does not limit the present application, any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.
Claims
1. A probability graph network method for substation equipment status assessment considering operating conditions is characterized by: include: The acquisition steps include: Obtain the original state parameter sequence and operating environment data of the substation equipment; Processing steps include: Based on the original state parameter sequence, a state transition network between the state parameters of the substation equipment and the equipment state is established; Based on the state transition network, a deep learning algorithm is used to train and solve the prior probabilities of the initial state parameters and state transition model parameters of the substation equipment; Based on the operating environment data, construct a working condition adaptive feature coding module; Using the working condition adaptive feature coding module, the online state parameters of the substation equipment are subjected to working condition adaptive feature coding to obtain a priori probability of the working condition; Output steps include: Based on the prior probability of the operating condition, a statistical evaluation of the status of the substation equipment is performed through a graph network reasoning algorithm, and an evaluation result is output.
2. The method according to claim 1, characterized in that The establishment of a state transition network between the state parameters of the power transformation equipment and the equipment state specifically includes: Define the initial state parameter sequence as ,in For the substation equipment Initial state parameter values; Define the initial state variable sequence of the substation equipment as ,in For the substation equipment Initial state variable values; Building a multi-layer state transfer network ,in is the number of network layers, For the Layer state transfer network; Define state parameter nodes in each layer of the state transfer network and state variable nodes , and establish connections between nodes.
3. The method according to claim 1, characterized in that The method of training and solving the priori probabilities of the initial state parameters and state transition model parameters of the substation equipment through the deep learning algorithm specifically includes: Online estimator algorithms, including ridge regression and neural networks, are used to estimate the prior probability distribution and transition probability distribution of node variables in state transition networks. An offline trainer algorithm, including multilayer perceptron and convolutional neural network, is used to estimate the prior probability distribution of node variables and the transition probability distribution in the state transition network. Calculate the probability distribution of the initial state parameter sequence of substation equipment belonging to different working conditions ,in Indicates the Working conditions.
4. The method according to claim 1, wherein The construction of the working condition adaptive feature coding module specifically includes: Construct a working condition perception feature extraction module for substation equipment; The LIN network is used to process the single online status parameter of the substation equipment and obtain the prior probability distribution of the status parameter belonging to different working conditions. Calculate the prior probability distribution of the online state sequence parameters of substation equipment belonging to different working conditions.
5. The method according to claim 4, characterized in that The calculation formula for obtaining the prior probability distribution of the state parameters belonging to different working conditions is: , For the substation equipment The state parameters belong to The prior probability distribution of the working conditions, For the The prior probability of the working condition, For the Under the working condition The conditional probability of the state parameters, For the The marginal probability of a state parameter.
6. The method according to claim 4, characterized in that The calculation formula for calculating the prior probability distribution of the online state sequence parameters of the substation equipment belonging to different working conditions is: , in, For substation equipment No. 1 to No. The state parameter sequence of the online phase belongs to The prior probability distribution of the working conditions, For 1st to A sequence of state parameters for an online phase.
7. The method according to claim 1, characterized in that The statistical evaluation of the status of the substation equipment by using the graph network reasoning algorithm specifically includes: Calculate the posterior probability distribution of the state transition network of the substation equipment; Calculate the probability of the state variables of the substation state variables: , For the Tier The probability distribution of state variable nodes, is the probability calculation function; Based on the posterior probability distribution, the status of the substation equipment is statistically evaluated.
8. The method according to claim 7, characterized in that The statistical evaluation of the status of the power substation equipment further includes: Calculation of state parameters of substation equipment belongs to the The probability of the following working conditions: , in, is the state parameter sequence of the substation equipment, is the state variable sequence; Calculate the reliability evaluation index of substation equipment: , in, is the reliability index, For the The state variable sequence under different working conditions.
9. The method according to claim 1, characterized in that Also includes: Set environmental characteristic parameters based on working conditions, including ambient temperature, ambient humidity, current and voltage parameters; A convolution feature extraction module is used at the input end of the substation equipment status assessment to extract data features; Use the fully connected layer to perform probabilistic reasoning on state features; Factor graphs are introduced and message-based variational inference methods are used for probabilistic graph network reasoning.
10. A network system for evaluating the state of substation equipment considering operating conditions, for executing the method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to obtain the original state parameter sequence and operating environment data of the substation equipment; Processing module, including: A network establishing unit, configured to establish a state transition network between the state parameters of the power transformation equipment and the equipment state based on the original state parameter sequence; A deep learning unit, configured to train and solve the prior probabilities of the initial state parameters and state transition model parameters of the substation equipment through a deep learning algorithm according to the state transition network; A feature coding unit is used to construct a working condition adaptive feature coding module based on the operating environment data, and use the module to perform working condition adaptive feature coding on the online state parameters of the substation equipment to obtain a priori probability of the working condition; an inference evaluation unit, configured to perform a statistical evaluation of the state of the substation equipment by using a graph network inference algorithm based on the prior probability of the operating condition; An output module is used to output the result of the statistical evaluation.