A method and system for predicting faults in substations based on fusion terminals

Abstract

The present invention discloses a fault prediction method and prediction system for a substation based on a fusion terminal, which belongs to the technical field of fault prediction and diagnosis of power systems. The improved AdaGrad optimizer optimizes the LSTM model, and combines the graph convolutional neural network model to achieve high-precision fault prediction of a single device and the entire substation, thereby improving the prediction accuracy and reliability. The warning information is generated by using the equipment failure probability and the credibility function value, so that the warning substation and faulty equipment can be identified in time, ensuring rapid response and processing of the fault, and reducing the power outage time and impact range. By accumulating the instantaneous failure rate and operating years of all fault modes, the comprehensive failure rate and credibility function value of the substation are calculated, which provides a comprehensive evaluation basis for the fault prediction of the substation based on the fusion terminal, and improves the scientificity and comprehensiveness of the prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault prediction and diagnosis, and in particular to a station area fault prediction method and prediction system based on a fusion terminal. Background Art

[0002] As the scale and complexity of power grids continue to increase, the frequency of power grid equipment failures and accidents continues to rise, which poses a great challenge to the safe and stable operation of the power grid. Especially in continuous high temperature weather, power grid equipment faces additional pressure and challenges, because high temperature will increase the load of equipment, reduce insulation performance and make heat dissipation difficult, thereby accelerating the aging of equipment and the possibility of failure. Fault prediction technology can detect potential faults in time before they occur by analyzing and mining the massive data generated by power grid equipment, providing strong support for equipment maintenance and management. The long-time series data generated by power grid equipment has significant topological relationships, which are closely related to factors such as the location, layout, and connection method of the equipment. Due to the topological relationship of the equipment, there are certain similarities in the occurrence of its failures, and this hidden correlation is often overlooked by researchers.

[0003] Under the influence of continuous high temperature weather, the fault prediction of power grid equipment is more difficult, the amount of data increases, the data type is diversified, the data quality decreases, and the topological relationship of the equipment has a more complex impact on the fault. These factors make the traditional fault prediction method have obvious shortcomings in terms of accuracy, false alarm rate and detection range. Summary of the invention

[0004] In view of the above problems, the present invention proposes a substation fault prediction method and prediction system based on a fusion terminal to improve the accuracy and reliability of power grid equipment fault prediction, especially under extreme climatic conditions, to ensure the safe and stable operation of the power grid.

[0005] A technical solution of the present invention is:

[0006] A method for predicting faults in a substation area based on a fusion terminal, the method comprising:

[0007] Using the historical fault data of a single device in the substation obtained based on the fusion terminal as training data, training the long short-term memory network model LSTM, updating the optimization model parameters based on the improved AdaGrad optimizer to obtain a direct prediction model of the device failure probability, and predicting the future device failure probability based on the direct prediction model of the device failure probability obtained by the training;

[0008] Based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by the graph convolutional neural network model;

[0009] The failure of the substation area is predicted based on the failure probability and the credibility function value of each substation area.

[0010] Furthermore, the acquisition of historical fault data of a single device in the substation includes:

[0011] Acquire the substation data, which is acquired based on the fusion terminal, including: equipment operation data, historical fault data, equipment maintenance data and environmental data;

[0012] Based on the substation data, obtain historical fault data of a single device in the substation;

[0013] The historical failure data of a single device includes: the failure mode, instantaneous failure rate and failure rate under operating conditions of the single device.

[0014] Furthermore, the method for obtaining the instantaneous failure rate of a single device includes:

[0015] Obtaining the instantaneous failure rate of the single device based on the device operation data and historical fault data;

[0016] The calculation formula is:

[0017] ;

[0018] in, is the instantaneous failure rate, indicating that the device The rate at which time failure occurs, is the total number of times the device runs. For the device to work from the initial time to The number of failures at time is the number of normal operation of the equipment, For a very short period of time, For the device at time The number of failures that occurred within

[0019] The method for obtaining the failure rate under the operating conditions includes:

[0020] Obtaining the failure rate under the operating condition based on the instantaneous failure rate of the single device;

[0021] The calculation formula of the failure rate under the operating conditions is:

[0022] ;

[0023] in, is the failure rate under operating conditions, indicating that the instantaneous failure rate is between 0 and The cumulative effect over the time period of time, is the instantaneous failure rate;

[0024] The method for obtaining the failure mode of a single device includes:

[0025] The random forest algorithm is used to extract the same features of similar faults from the historical fault data to obtain a historical fault feature set. ;

[0026] Extract feature vectors based on the instantaneous failure rate of a single device, the failure rate under operating conditions, and the equipment operation data , the extracted feature vector and the historical fault feature set The vector features in are compared using the Pearson correlation coefficient formula to find the most similar fault types and determine the possible fault modes of the equipment. The similarity calculation formula is:

[0027] ;

[0028] in, is the historical fault feature set The eigenvectors in is the feature vector Middle eigenvalues, is the feature vector Middle eigenvalues, and The eigenvectors are and The mean of all eigenvalues ​​of .

[0029] Furthermore, the equipment operation data includes: equipment historical operation data and equipment real-time operation data;

[0030] The real-time operation data of the equipment includes: load, temperature, pressure, current, voltage and timestamp;

[0031] The historical fault data includes: the time when the historical fault occurred, the fault type and the maintenance record.

[0032] Furthermore, the updating of optimization model parameters based on the improved AdaGrad optimizer includes:

[0033] At each weight update, the learning rate decay factor is applied to the iteration of the learning rate of the AdaGrad optimizer, and the iteration formula is:

[0034] ;

[0035] Learning rate for AdaGrad optimizer The calculation formula is:

[0036] ;

[0037] in, is the global learning rate, is a non-zero decimal number, is the attenuation factor, is the number of iteration steps, For the The learning rate in iterations, For the The learning rate in iterations, is the accumulated squared gradient.

[0038] Furthermore, the updating of optimization model parameters based on the improved AdaGrad optimizer further includes:

[0039] In the gradient descent iteration of the AdaGrad optimizer, a momentum term is added to the current gradient. The iteration formula is:

[0040] ;

[0041] The momentum term iteration formula is:

[0042] ;

[0043] in, For the The momentum term in the iteration, is the momentum coefficient, is the current gradient, is the global learning rate, For the The parameters in the iterations, For the Parameters in iterations.

[0044] Furthermore, based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by a graph convolutional neural network model, including:

[0045] Extract the current state characteristics of the device and construct a feature vector;

[0046] The graph convolutional neural network model is constructed, including: taking each device as a node, setting the feature vector of the device as the feature of the node, setting the connection relationship between the devices as the edge of the graph, setting the degree of mutual influence between the devices as the weight of the edge, and constructing an adjacency matrix Sum degree matrix ;

[0047] Construct a graph convolutional neural network model. The graph convolution operation is expressed as:

[0048] ;

[0049] in, For Node exist The feature vector of the layer, For Node The set of neighbor nodes of For the The weight matrix of the layer, For the The bias term of the layer, is the activation function;

[0050] Multiple layers of graph convolutional layers are stacked, and the output of each layer is used as the input of the next layer. The calculation formula is:

[0051] ;

[0052] in, For the The feature matrix of the layer is the feature matrix of the layer, is the adjacency matrix, is the degree matrix, and , Representation Node The degree, Representation Node With Node Connection relationship, if there is a connection relationship ,otherwise , and Respectively The weight matrix and bias term of the layer;

[0053] Using the output of the graph convolutional neural network model, the credibility values ​​of the nodes are aggregated and the credibility function value of each node is calculated. The calculation formula is:

[0054] ;

[0055] in, For Node The credibility function value of For Node The instantaneous failure rate, Nodes output by the graph convolutional neural network model The credibility function value of and is the weight coefficient, which is used to adjust the impact of the graph convolutional neural network model output and failure rate. is a nonlinear mapping function;

[0056] The total failure rate of the area is obtained by adding up the instantaneous failure rates of all devices. The calculation formula is:

[0057] ;

[0058] in, is the total failure rate of the area, For Node The instantaneous failure rate, is a set of graph nodes;

[0059] Accumulate the credibility function values ​​of all devices to obtain the overall credibility function value of the substation. The calculation formula is:

[0060] ;

[0061] in, is the overall credibility function value of the station area, For Node The instantaneous failure rate, A collection of graph nodes.

[0062] Furthermore, the predicting of the substation fault based on the failure probability and the credibility function value of each substation includes:

[0063] According to the credibility function values ​​of each area, the area label with the largest credibility function value is selected as the fault label of the area, and it is determined as an early warning area; in the early warning area, the credibility function values ​​of all devices are compared, and the device with the largest credibility function value is selected and marked as the early warning device;

[0064] According to the failure probability of faulty equipment in the early warning area , calculate the equipment failure rate , equipment failure rate , the maximum confidence function value in the warning area And the operating period , calculate the time period from the current moment to the warning moment , predict future fault warning time ; According to the calculated fault warning time, the warning time point from the current time is obtained, and the warning information containing the equipment label, fault warning time and related description is generated, including the following calculation formula:

[0065] Calculating Equipment Failure Rate , the calculation formula is:

[0066] ;

[0067] in, for Equipment failure rate at all times, The failure probability From 0 to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment;

[0068] The time period from the current time to the warning time is calculated by the following equation group: :

[0069] ;

[0070] ;

[0071] in, for Equipment failure rate at all times, The service life is is the time period from the current time to the warning time, is the maximum confidence function value in the warning area, The failure probability From 0 to The cumulative effect of time, The failure probability from to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment;

[0072] The warning time is: ;

[0073] in, is the fault warning time, The service life is It is the time period from the current time to the warning time.

[0074] Furthermore, the connection relationship between the devices includes: a physical connection relationship, an electrical connection relationship or a collaborative relationship between the devices.

[0075] Based on the above-mentioned area fault prediction method based on the fusion terminal, the present invention also provides an area fault prediction system based on the fusion terminal, the system comprising:

[0076] The equipment failure probability prediction module is used to use the historical fault data of a single device in the substation obtained based on the fusion terminal as training data, train the long short-term memory network model LSTM, update the optimization model parameters based on the improved AdaGrad optimizer to obtain the equipment failure probability direct prediction model, and predict the future equipment failure probability based on the equipment failure probability direct prediction model obtained by the training;

[0077] An equipment credibility value calculation module is used to calculate the failure probability and credibility function value of each substation through a graph convolutional neural network model based on the future equipment failure probability;

[0078] The early warning module is used to predict the failure of the substation based on the failure probability and the credibility function value of each substation.

[0079] The substation fault prediction method and prediction system based on the fusion terminal provided in the embodiment of the present invention optimizes the LSTM model through the improved AdaGrad optimizer, and combines the graph convolutional neural network model to achieve high-precision fault prediction of single equipment and the entire substation, thereby improving the prediction accuracy and reliability. By using the equipment failure probability and the credibility function value to generate early warning information, it is possible to timely identify the early warning substation and faulty equipment, ensure rapid response and processing of the fault, and reduce the power outage time and impact range. By accumulating the instantaneous failure rate and years of operation of all fault modes, the comprehensive failure rate and credibility function value of the substation are calculated, which provides a comprehensive evaluation basis for substation fault prediction and improves the scientificity and comprehensiveness of the prediction.

[0080] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0082] Figure 1 A flow chart of a method for predicting substation faults based on a fusion terminal provided in an embodiment of the present invention is shown.

[0083] Figure 2 A block diagram of a substation fault prediction system based on a fusion terminal provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0084] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the invention and to fully convey the scope of the present invention to those skilled in the art.

[0085] Fault prediction technology can detect potential faults in time before they occur by analyzing and mining the massive data generated by power grid equipment, providing strong support for equipment maintenance and management. The long-time series data generated by power grid equipment has significant topological relationships, which are closely related to factors such as the location, layout, and connection method of the equipment. Due to the topological relationship of the equipment, there are certain similarities in the occurrence of their faults, and this hidden correlation is often overlooked by researchers.

[0086] Under the influence of continuous high temperature weather, the fault prediction of power grid equipment becomes more difficult, the amount of data increases, the data type diversifies, the data quality decreases, and the topological relationship of the equipment has a more complex impact on the fault. These factors make the traditional fault prediction method have obvious deficiencies in accuracy, false alarm rate and detection range. Therefore, it is urgent to introduce new theories and technical means to improve the accuracy and reliability of power grid equipment fault prediction, especially under extreme climate conditions, to ensure the safe and stable operation of the power grid.

[0087] The specific implementation of the technical solution of the present invention includes the following contents: Example

[0088] refer to Figure 1 , a method for predicting faults in a substation area based on a fusion terminal, the method comprising:

[0089] S1, using the historical fault data of a single device in the substation obtained based on the fusion terminal as training data to train the long short-term memory network model LSTM, updating the optimization model parameters based on the improved AdaGrad optimizer to obtain a direct prediction model of device failure probability, and predicting the future device failure probability based on the direct prediction model of device failure probability obtained by the training;

[0090] S2, based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by a graph convolutional neural network;

[0091] S3, predicting the substation fault based on the failure probability and the credibility function value of each substation.

[0092] In step S1, the acquisition of historical fault data of a single device in the substation includes:

[0093] Acquire the substation data, which is acquired based on the fusion terminal, including: equipment operation data, historical fault data, equipment maintenance data and environmental data;

[0094] Based on the substation data, obtain historical fault data of a single device in the substation;

[0095] The historical failure data of a single device includes: the failure mode, instantaneous failure rate and failure rate under operating conditions of the single device.

[0096] Among them, the equipment operation data includes: equipment historical operation data and equipment real-time operation data; the equipment real-time operation data includes: load, temperature, pressure, current, voltage and timestamp; the historical fault data includes: the time when the historical fault occurred, the fault type and the maintenance record.

[0097] The fusion terminal of the prior art is a device that combines multiple functions, integrates multiple sensors, data processors, memory, communication interface and other modules, and interacts with the server data through a data link. It has multiple functions such as data acquisition, monitoring, communication, and control, and can automatically process data and perform control tasks. It has interoperability and supports multiple communication protocols to ensure compatibility with other power grid equipment and systems. It can maintain stable operation under various environmental conditions. It also has security and can protect data and systems from unauthorized access and attacks.

[0098] Applying the fusion terminal to the power grid, especially the power grid area in the technical solution of the present invention, can effectively improve the intelligence and automation level of the power grid. Data collection and monitoring can be realized, and power parameters such as voltage, current, and power can be monitored in real time. The communication interface supports a variety of communication protocols to achieve interconnection with other devices and systems. Equipment control and parameter adjustment can also be performed through remote operation. The fusion terminal can be used for fault detection, prediction and alarm, identify and locate faults, and send alarm information in a timely manner. It can further strengthen power quality management, monitor and improve power quality, and reduce losses.

[0099] In the above-mentioned single device historical fault data of the substation, the method for obtaining the instantaneous failure rate of the single device includes: obtaining the instantaneous failure rate of the single device based on the device operation data and the historical fault data;

[0100] The calculation formula is:

[0101] ;

[0102] in, is the instantaneous failure rate, indicating that the device The rate at which time failure occurs, is the total number of times the device runs. For the device to work from the initial time to The number of failures at time To work from the initial time to The number of failures at time It is a very short period of time, generally the interval between data sampling. is the number of normal operation of the equipment, For a very short period of time, For the device at time The number of failures that occurred within.

[0103] The method for obtaining the failure rate under operating conditions in the above data is to obtain the failure rate under operating conditions based on the instantaneous failure rate of the single device, and the calculation formula is:

[0104] ;

[0105] in, is the failure rate under operating conditions, indicating that the instantaneous failure rate is between 0 and The cumulative effect over the time period of time, is the instantaneous failure rate.

[0106] The method for obtaining the failure mode of a single device in the above data includes:

[0107] The random forest algorithm is used to extract the same features of similar faults from the historical fault data to obtain a historical fault feature set. The specific method is: extract historical fault features from the historical fault data, classify the historical faults, including electrical faults (including short circuit faults, open circuit faults, overload faults), communication faults (including offline faults, reading failures), and other faults (including equipment aging faults, environmental factor faults, and human factor faults), and use the random forest algorithm to extract the same features of the same type of faults from various types of faults to obtain a historical fault feature set. .

[0108] Extract equipment feature vectors based on the instantaneous failure rate of a single device, the failure rate under operating conditions, and equipment operation data , the extracted device feature vector and the historical fault feature set The vector features in are compared using the Pearson correlation coefficient formula to find the most similar fault types and determine the possible fault modes of the equipment. The similarity calculation formula is:

[0109] ;

[0110] in, is the historical fault feature set The eigenvectors in is the feature vector Middle eigenvalues, is the feature vector Middle eigenvalues, and The eigenvectors are and The mean of all eigenvalues ​​of .

[0111] After obtaining the above-mentioned area data, the data must be cleaned and normalized to obtain a variety of data to be analyzed. Data cleaning includes operations such as removing noise and filling missing values. Normalization makes the format of each data unified to facilitate model input.

[0112] In this step, the optimization model parameters are updated based on the improved AdaGrad optimizer. The AdaGrad optimizer is improved by adjusting the learning rate decay of AdaGrad and combining momentum, including:

[0113] At each weight update, the learning rate decay factor is applied to the iteration of the learning rate of the AdaGrad optimizer, and the iteration formula is:

[0114] ;

[0115] The AdaGrad optimizer may reduce the learning rate to a too low value too early during each iteration. To reduce this possibility, the AdaGrad optimizer applies a learning rate decay factor to the learning rate after each iteration. Each time the weights are updated, the above learning rate decay formula (4) is used for iteration.

[0116] Specifically, the learning rate of the AdaGrad optimizer The calculation formula is:

[0117] ;

[0118] in, is the global learning rate, is a non-zero decimal number, is the attenuation factor, is the number of iteration steps, For the The learning rate in iterations, For the The learning rate in iterations, is the accumulated squared gradient.

[0119] The AdaGrad optimizer is improved by adjusting the combined momentum of AdaGrad, including: adding a momentum term to the current gradient in the gradient descent iteration of the AdaGrad optimizer. The iteration formula is:

[0120] ;

[0121] The momentum term iteration formula is:

[0122] ;

[0123] in, For the The momentum term in the iteration, is the momentum coefficient, is the current gradient, is the global learning rate, For the The parameters in the iterations, For the Parameters in iterations.

[0124] In this step, the long short-term memory network model LSTM is constructed, including: input layer, LSTM layer, first fully connected layer, the neural unit of the previous LSTM layer is fully connected with each neuron in the first fully connected layer, and a Dropout1 layer is added on top of the first fully connected layer to Some neurons are discarded, and the remaining neurons are The probability of is retained, a fully connected layer is added above the Dropout1 layer, and a Dropout2 layer is added above the second fully connected layer, with probability Some neurons are discarded, and the remaining neurons are The probability of failure is retained, and a third fully connected layer with 10 neurons is added above the Dropout2 layer to ensure that the output result is a continuous prediction value. A regression layer is added above the third fully connected layer. The regression layer is the output layer with only one neuron at the end, which outputs the predicted value of the device failure probability.

[0125] Through the above-mentioned improved optimization, the AdaGrad optimizer updates the model parameters to train the LSTM. The trained LSTM model is the AdaGrad-LSTM model, that is, the direct prediction model of equipment failure probability. After the AdaGrad-LSTM model is obtained through training, the AdaGrad-LSTM model can be used to predict the future equipment failure probability.

[0126] For step S2, based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by the graph convolutional neural network model, including:

[0127] The current state characteristics of the equipment are extracted and a feature vector is constructed. The current state characteristics of the equipment are obtained based on the fusion terminal, including: possible failure modes, equipment type, operating status, instantaneous failure rate of the equipment, failure rate under operating conditions, and feature vector construction.

[0128] The graph convolutional neural network model is constructed, including: taking each device as a node, setting the feature vector of the device as the feature of the node, setting the connection relationship between the devices as the edge of the graph, setting the degree of mutual influence between the devices as the weight of the edge, and constructing an adjacency matrix Sum degree matrix .

[0129] Constructing the adjacency matrix Sum degree matrix Afterwards, the node features are aggregated through the features of its neighboring nodes. The node feature matrix and the edge feature matrix are used as the input of the input layer. The feature representation of each node is updated by aggregating the features of the node and its neighboring nodes. Considering the influence of the adjacency matrix, the weighted sum of the node features is calculated and updated based on the features of the node itself and its neighboring nodes. The cross entropy loss is selected as the loss function, and the Adam optimizer is used for parameter tuning. The training data is used for model training. The features of each device are calculated through the forward propagation of the graph convolutional neural network model to obtain the credibility function value of each device.

[0130] Among them, the graph convolution neural network model is constructed, and the graph convolution operation is expressed as:

[0131] ;

[0132] in, For Node exist The feature vector of the layer, For Node The set of neighbor nodes of For the The weight matrix of the layer, For the The bias term of the layer, is the activation function;

[0133] Multiple layers of graph convolutional layers are stacked, and the output of each layer is used as the input of the next layer, which can capture deeper graph structure information. The calculation formula is:

[0134] ;

[0135] in, For the The feature matrix of the layer, is the adjacency matrix, is the degree matrix, and , Representation Node The degree, Representation Node With Node Connection relationship, if there is a connection relationship ,otherwise , and Respectively The weight matrix and bias term of the layer.

[0136] Among them, the cross entropy loss function is calculated as follows:

[0137] ;

[0138] in, is the cross entropy loss function, which is used to measure the difference between the predicted value and the true value. is the total number of samples, For the True labels of samples, For the The predicted probability of samples;

[0139] Using the output of the graph convolutional neural network model, the credibility values ​​of the nodes are aggregated and the credibility function value of each node is calculated. The calculation formula is:

[0140] ;

[0141] in, For Node The credibility function value of For Node The instantaneous failure rate, Nodes output by the graph convolutional neural network model The credibility function value of and is the weight coefficient, which is used to adjust the impact of the graph convolutional neural network model output and failure rate. is a nonlinear mapping function.

[0142] The total failure rate of the area is obtained by adding up the instantaneous failure rates of all devices. The calculation formula is:

[0143] ;

[0144] in, is the total failure rate of the area, For Node The instantaneous failure rate, A collection of graph nodes.

[0145] Accumulate the credibility function values ​​of all devices to obtain the overall credibility function value of the substation. The calculation formula is:

[0146] ;

[0147] in, is the overall credibility function value of the station area, For Node The instantaneous failure rate, A collection of graph nodes.

[0148] For step S3, predicting the failure of the substation area based on the failure probability and the credibility function value of each substation area includes:

[0149] According to the credibility function values ​​of each substation, the substation label with the largest credibility function value is selected as the fault label of the substation, and it is determined as a warning substation; within the warning substation, the credibility function values ​​of all devices are compared, and the device with the largest credibility function value is selected to be marked as the warning device. The device with the largest credibility function value indicates that it has the greatest risk of failure; thereby constructing detailed warning information, including the labels of the warning substation and warning equipment and their credibility function values.

[0150] The above-mentioned warning information includes:

[0151] The early warning area includes the name or logo of the area and the credibility function value of the area.

[0152] Early warning equipment, including the name or identification of the equipment and the credibility function value of the equipment.

[0153] Risk Assessment: Provide an overview of the failure risk of the area and equipment to help relevant personnel make decisions.

[0154] This embodiment predicts the fault warning time based on the failure probability of the faulty equipment in the warning area and the equipment operation life obtained from the AdaGrad-LSTM model, where the failure probability of the faulty equipment indicates the possibility of the equipment failing in a certain time period in the future.

[0155] The specific prediction method is: based on the failure probability of faulty equipment in the early warning area , calculate the equipment failure rate , equipment failure rate , the maximum confidence function value in the warning area And the operating period , calculate the time period from the current moment to the warning moment , predict future fault warning time ; According to the calculated fault warning time, the warning time point from the current time is obtained, and the warning information containing the equipment label, fault warning time and related description is generated, including the following calculation formula:

[0156] Calculating Equipment Failure Rate :

[0157] ;

[0158] in, for Equipment failure rate at all times, The failure probability From 0 to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment;

[0159] The time period from the current moment to the warning moment is calculated by the following equations (15) and (16): :

[0160] ;

[0161] ;

[0162] in, for Equipment failure rate at all times, The service life is is the time period from the current time to the warning time, is the maximum confidence function value in the warning area, The failure probability From 0 to The cumulative effect of time, The failure probability from to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment;

[0163] The warning time is: ;

[0164] in, is the fault warning time, The service life is It is the time period from the current time to the warning time.

[0165] In this embodiment, the connection relationship between devices includes: a physical connection relationship, an electrical connection relationship or a collaborative relationship between the devices.

[0166] Embodiment 2:

[0167] refer to Figure 2 Based on the above-mentioned method for predicting faults in a substation area based on a fusion terminal, the present invention further provides a system for predicting faults in a substation area based on a fusion terminal, the system comprising:

[0168] The equipment failure probability prediction module is used to use the historical fault data of a single device in the substation obtained based on the fusion terminal as training data, train the long short-term memory network model LSTM, update the optimization model parameters based on the improved AdaGrad optimizer to obtain a direct prediction model of equipment failure probability, and predict the future equipment failure probability based on the direct prediction model of equipment failure probability obtained by the training;

[0169] An equipment credibility value calculation module is used to calculate the failure probability and credibility function value of each substation through a graph convolutional neural network model based on the future equipment failure probability;

[0170] The early warning module is used to predict the failure of the substation based on the failure probability and the credibility function value of each substation.

[0171] The specific implementation method of this embodiment is the same as that of Embodiment 1, which will not be repeated here. Please refer to the description of Embodiment 1 for details.

[0172] The method and system of the above embodiment are adopted, based on the intelligent fusion terminal, to meet the needs of fault prediction of the substation based on the fusion terminal under continuous high temperature weather, and solve the following technical problems:

[0173] 1. High-precision fault prediction: Based on the substation data obtained by the fusion terminal, the LSTM model is optimized by the improved AdaGrad optimizer, combined with the graph convolutional neural network model, to achieve high-precision fault prediction for a single device and the entire substation, improving the prediction accuracy and reliability.

[0174] 2. Real-time early warning and response: Using the equipment failure probability and the credibility function value to generate early warning information, it is possible to timely identify the early warning area and faulty equipment, ensure rapid response and handling of the fault, and reduce the power outage time and impact range.

[0175] 3. Comprehensive failure rate evaluation: By accumulating the instantaneous failure rate and operating years of all failure modes, the comprehensive failure rate and reliability function value of the substation are calculated, which provides a comprehensive evaluation basis for substation fault prediction and improves the scientificity and comprehensiveness of the prediction.

[0176] Those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, the combination of features from different embodiments is meant to be within the scope of the present invention and to form different embodiments.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting substation faults based on a fusion terminal, characterized in that: The method comprises: The historical fault data of a single device in the substation obtained based on the fusion terminal is used as training data to train the long short-term memory network model LSTM, and the parameters of the optimized model are updated based on the improved AdaGrad optimizer to obtain a direct prediction model of the device failure probability, and the future device failure probability is predicted based on the direct prediction model of the device failure probability obtained by the training; Based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by the graph convolutional neural network model; Predicting the failure of the substation based on the failure probability and the credibility function value of each substation; The acquisition of historical fault data of a single device in the substation includes: Acquire the substation data, which is acquired based on the fusion terminal, including: equipment operation data, historical fault data, equipment maintenance data and environmental data; Based on the substation data, obtain historical fault data of a single device in the substation; The historical failure data of a single device includes: failure mode, instantaneous failure rate and failure rate under operating conditions of a single device; The method for obtaining the instantaneous failure rate of a single device includes: Obtaining the instantaneous failure rate of the single device based on the device operation data and historical fault data; The calculation formula is: ； in, is the instantaneous failure rate, indicating that the device The rate at which time failure occurs, is the total number of times the device runs. To work from the initial time to The number of failures at time is the number of normal operation of the equipment, For a very short period of time, For in time The number of failures that occurred within The method for obtaining the failure rate under the operating conditions includes: Obtaining the failure rate under the operating condition based on the instantaneous failure rate of the single device; The calculation formula of the failure rate under the operating conditions is: ； in, is the failure rate under operating conditions, indicating that the instantaneous failure rate is between 0 and The cumulative effect over the time period of time, is the instantaneous failure rate; The method for obtaining the failure mode of a single device includes: The random forest algorithm is used to extract the same features of similar faults from the historical fault data to obtain a historical fault feature set. ; Extract feature vectors based on the instantaneous failure rate of a single device, the failure rate under operating conditions, and the equipment operation data , the extracted feature vector and the historical fault feature set The eigenvectors in are compared using the Pearson correlation coefficient formula to find the most similar fault types and determine the possible fault modes of the equipment. The similarity calculation formula is: ； in, is the historical fault feature set The eigenvectors in is the feature vector Middle eigenvalues, is the feature vector Middle eigenvalues, and The eigenvectors are and The mean of all eigenvalues ​​of .

2. According to claim 1, a method for predicting substation faults based on a fusion terminal is characterized in that: The equipment operation data includes: equipment historical operation data and equipment real-time operation data; The real-time operation data of the equipment includes: load, temperature, pressure, current, voltage and timestamp; The historical fault data includes: the time when the historical fault occurred, the fault type and the maintenance record.

3. The method for predicting substation faults based on a fusion terminal according to claim 1, characterized in that: The updating of optimization model parameters based on the improved AdaGrad optimizer includes: At each weight update, the learning rate decay factor is applied to the iteration of the learning rate of the AdaGrad optimizer, and the iteration formula is: ； Learning rate for AdaGrad optimizer The calculation formula is: ； in, is the global learning rate, is a non-zero decimal number, is the attenuation factor, is the number of iteration steps, For the The learning rate in iterations, For the The learning rate in iterations, is the accumulated squared gradient.

4. The method for predicting substation faults based on a fusion terminal according to claim 3 is characterized in that: The method of updating the optimization model parameters based on the improved AdaGrad optimizer further includes: In the gradient descent iteration of the AdaGrad optimizer, a momentum term is added to the current gradient. The iteration formula is: ； The momentum term iteration formula is: ； in, For the The momentum term in the iteration, is the momentum coefficient, is the current gradient, is the global learning rate, For the The parameters in the iterations, For the Parameters in iterations.

5. The method for predicting substation faults based on a fusion terminal according to claim 1, characterized in that: Based on the future equipment failure probability, the failure probability and the credibility function value of each substation are calculated by the graph convolutional neural network model, including: Extract the current state characteristics of the device and construct a feature vector; The graph convolutional neural network model is constructed, including: taking each device as a node, setting the feature vector of the device as the feature of the node, setting the connection relationship between the devices as the edge of the graph, setting the degree of mutual influence between the devices as the weight of the edge, and constructing an adjacency matrix Sum degree matrix ; Construct a graph convolutional neural network model. The graph convolution operation is expressed as: ； in, For Node exist The feature vector of the layer, For Node The set of neighbor nodes of For the The weight matrix of the layer, For the The bias term of the layer, is the activation function; Multiple layers of graph convolutional layers are stacked, and the output of each layer is used as the input of the next layer. The calculation formula is: ； in, For the The feature matrix of the layer, is the adjacency matrix, is the degree matrix, and , Representation Node The degree, Representation Node With Node Connection relationship, if there is a connection relationship ,otherwise , and Respectively The weight matrix and bias term of the layer; Using the output of the graph convolutional neural network model, the credibility values ​​of the nodes are aggregated and the credibility function value of each node is calculated. The calculation formula is: ； in, For Node The credibility function value of For Node The instantaneous failure rate, Nodes output by the graph convolutional neural network model The credibility function value of and is the weight coefficient, which is used to adjust the impact of the graph convolutional neural network model output and failure rate. is a nonlinear mapping function; The total failure rate of the area is obtained by adding up the instantaneous failure rates of all devices. The calculation formula is: ； in, is the total failure rate of the area, For Node The instantaneous failure rate, is a set of graph nodes; Accumulate the credibility function values ​​of all devices to obtain the overall credibility function value of the substation. The calculation formula is: ； in, is the overall credibility function value of the station area, For Node The credibility function value of A collection of graph nodes.

6. The method for predicting substation faults based on a fusion terminal according to claim 1, characterized in that: The predicting of the substation fault based on the failure probability and the credibility function value of each substation includes: According to the credibility function values ​​of each area, the area label with the largest credibility function value is selected as the fault label of the area, and it is determined as an early warning area; in the early warning area, the credibility function values ​​of all devices are compared, and the device with the largest credibility function value is selected and marked as the early warning device; According to the failure probability of faulty equipment in the early warning area , calculate the equipment failure rate , equipment failure rate , the maximum confidence function value in the warning area And the operating period , calculate the time period from the current moment to the warning moment , predict future fault warning time ; According to the calculated fault warning time, the warning time point from the current time is obtained, and the warning information containing the equipment label, fault warning time and related description is generated, including the following calculation formula: Calculating Equipment Failure Rate , the calculation formula is: ； in, for Equipment failure rate at all times, The failure probability From 0 to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment; The time period from the current time to the warning time is calculated by the following equation group: : ； ； in, for Equipment failure rate at all times, The service life is is the time period from the current time to the warning time, is the maximum confidence function value in the warning area, The failure probability From 0 to The cumulative effect of time, The failure probability from to The cumulative effect of time, for Always warn of the failure probability of faulty equipment in the area, From 0 to A time parameter within a moment; The warning time is: ; in, is the fault warning time, The service life is It is the time period from the current time to the warning time.

7. The method for predicting substation faults based on a fusion terminal according to claim 5 is characterized in that: The connection relationship between the devices includes: a physical connection relationship, an electrical connection relationship or a cooperative relationship between the devices.

8. A fault prediction system for a substation based on a fusion terminal, characterized in that: The system comprises: The equipment failure probability prediction module is used to use the historical fault data of a single device in the substation obtained based on the fusion terminal as training data, train the long short-term memory network model LSTM, update the optimization model parameters based on the improved AdaGrad optimizer to obtain a direct prediction model of equipment failure probability, and predict the future equipment failure probability based on the direct prediction model of equipment failure probability obtained by the training; An equipment credibility value calculation module is used to calculate the failure probability and credibility function value of each substation through a graph convolutional neural network model based on the future equipment failure probability; An early warning module, used for predicting the failure of the substation based on the failure probability and the credibility function value of each substation; The acquisition of historical fault data of a single device in the substation includes: Acquire the substation data, which is acquired based on the fusion terminal, including: equipment operation data, historical fault data, equipment maintenance data and environmental data; Based on the substation data, obtain historical fault data of a single device in the substation; The historical failure data of a single device includes: failure mode, instantaneous failure rate and failure rate under operating conditions of a single device; The method for obtaining the instantaneous failure rate of a single device includes: Obtaining the instantaneous failure rate of the single device based on the device operation data and historical fault data; The calculation formula is: ； in, is the instantaneous failure rate, indicating that the device The rate at which time failure occurs, is the total number of times the device runs. To work from the initial time to The number of failures at time is the number of normal operation of the equipment, For a very short period of time, For in time The number of failures that occurred within The method for obtaining the failure rate under the operating conditions includes: Obtaining the failure rate under the operating condition based on the instantaneous failure rate of the single device; The calculation formula of the failure rate under the operating conditions is: ； in, is the failure rate under operating conditions, indicating that the instantaneous failure rate is between 0 and The cumulative effect over the time period of time, is the instantaneous failure rate; The method for obtaining the failure mode of a single device includes: The random forest algorithm is used to extract the same features of similar faults from the historical fault data to obtain a historical fault feature set. ; Extract feature vectors based on the instantaneous failure rate of a single device, the failure rate under operating conditions, and the equipment operation data , the extracted feature vector and the historical fault feature set The eigenvectors in are compared using the Pearson correlation coefficient formula to find the most similar fault types and determine the possible fault modes of the equipment. The similarity calculation formula is: ； in, is the historical fault feature set The eigenvectors in is the feature vector Middle eigenvalues, is the feature vector Middle eigenvalues, and The eigenvectors are and The mean of all eigenvalues ​​of .

Images