A real-time monitoring system and method for a smart substation based on big data

Abstract

The present invention discloses a real-time monitoring system and method for an intelligent substation based on big data, specifically relating to the technical field of power equipment monitoring, including a multi-source acquisition and processing module, a dynamic topology update module, a device status monitoring module, a fault diagnosis and prediction module, a device life prediction module, an energy efficiency analysis and optimization module, and a remote operation and maintenance scheduling module; the present invention can achieve high-precision fault detection and latent association mining, improve the abnormal detection ability of the substation, and effectively optimize the life management of various power equipment in the substation; it can ensure reasonable allocation of maintenance resources, improve equipment reliability, enhance comprehensive energy efficiency, reduce carbon emissions, and achieve intelligent operation of the substation without manual intervention to adjust the operation mode and improve the optimization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment monitoring, and specifically relates to a real-time monitoring system and method for a smart substation based on big data. Background Art

[0002] With the rapid development of the energy Internet, the power grid has put forward higher requirements for intelligence, reliability, and real-time performance. Traditional substation monitoring systems mainly rely on single data sources, suffering from problems such as data islands, insufficient real-time performance, and low fault diagnosis accuracy. In recent years, with the breakthroughs in technologies such as big data, artificial intelligence, and the Internet of Things, smart substations have gradually become an important part of the smart grid. As the core hub of the power grid, the stable operation of substations is directly related to the safety and economy of the power system. With the increase in the proportion of new energy grid connection, the uncertainty of power grid operation has increased, and traditional monitoring means are difficult to meet the requirements of modern smart grids for substation state perception, fault early warning, and autonomous optimization. Therefore, how to build an efficient, accurate, and intelligent substation real-time monitoring system has become a hot research issue.

[0003] Existing real-time monitoring systems and methods for smart substations cannot perform high-precision fault detection and latent association mining, reducing the abnormal detection ability of substations; in addition, existing real-time monitoring systems and methods for smart substations cannot reasonably allocate maintenance resources, reducing equipment reliability, and require manual intervention to adjust the operation mode, reducing the optimization efficiency. For this reason, we propose a real-time monitoring system and method for a smart substation based on big data. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects existing in the prior art; therefore, a real-time monitoring system and method for a smart substation based on big data are proposed.

[0005] In the first aspect of the implementation of the present invention, a real-time monitoring system for a smart substation based on big data is first proposed. The system includes:

[0006] A multi-source acquisition and processing module, a dynamic topology update module, an equipment status monitoring module, a fault diagnosis and prediction module, an equipment life prediction module, an energy efficiency analysis and optimization module, and a remote operation and maintenance scheduling module;

[0007] The multi-source acquisition and processing module is used to collect real-time detection data collected by different sensors deployed on each power equipment in the substation and preprocess each group of collected detection data;

[0008] The dynamic topology update module is used to establish an association map of power equipment and dynamically update the association map of power equipment based on each preprocessed detection data;

[0009] The device status monitoring module monitors and analyzes the operating status of each power device in the substation based on the preprocessed detection data received in real time and the power device association map;

[0010] The fault diagnosis and prediction module predicts the potential faults of each power device in the substation based on the real-time analysis results of the operating status of each power device in the substation and the equipment aging situation;

[0011] The equipment life prediction module is used to construct a virtual model of the equipment, and based on the real-time analysis results of the operating status of each power device in the substation and the fault prediction results, analyze the aging situation of each device;

[0012] The energy efficiency analysis and optimization module optimizes the operating mode of each power device in the substation based on the real-time analysis results of the operating status of each power device in the substation and the power device association map;

[0013] The remote operation and maintenance scheduling module is used to optimize the maintenance plan of the substation by combining the fault detection data, operating mode and life prediction results of each power device in the substation.

[0014] As a further solution of the present invention, the specific steps for the multi-source acquisition and processing module to preprocess the collected groups of detection data are as follows:

[0015] S1.1: The multi-source acquisition and processing module receives the detection data collected by different sensors, including temperature data, voltage data, humidity data and ultrasonic data. The detection data of the same type are divided into a group, and then the sampling time step of each group of detection data is synchronized by the time window resampling method. After the synchronization is completed, through unit standardization processing, the units of each group of detection data are unified;

[0016] S1.2: Arrange each group of detection data in ascending order of data value, and respectively take the values at the 25% position and 75% position of each group of detection data as the first quartile Q1 and the third quartile Q3 of each group of data, and obtain the interquartile range IQR of each group of detection data through Q3 - Q1. Then, based on the IQR, take Q1 - 1.5×IQR as the lower boundary and Q3 + 1.5×IQR as the upper boundary to draw the corresponding box plot. Traverse each group of detection data. If there is detection data below the lower boundary or above the upper boundary, mark the detection data as an outlier;

[0017] S1.3: Traverse each group of detection data, and count the missing values in each group of detection data. Then, fill in each missing value by linear interpolation, and replace each outlier by mean filling processing. After the outlier processing is completed, recalculate Q1, Q3 and IQR, and draw a new box plot to perform outlier detection again to verify the quality of each group of detection data. After the verification is passed, normalize each group of detection data to the range of [0, 1].

[0018] As a further solution of the present invention, the dynamic topology update module collects each group of power equipment in the substation and uses them as nodes. According to the electrical connection relationship between the equipment, edges are established between each node to connect each node. An associated map of power equipment is established based on each node and the edges between each node, and the characteristics of each node are initialized. The characteristics include detection data, historical operating status of the equipment, and fault records, and the associated map of power equipment is dynamically updated based on the detection data of each group collected in real time.

[0019] As a further solution of the present invention, the specific steps for the equipment status monitoring module to monitor and analyze the operating status of each power equipment in the substation are as follows:

[0020] S2.1: The status monitoring module establishes a monitoring and analysis model based on the ST-GCN architecture. The monitoring and analysis model includes an input layer, a GCN layer, a TCN layer, and an output layer. The data of each group recorded in the associated map of power equipment established by the dynamic topology update module and the preprocessed real-time detection data of each group are used as input data and transmitted to the monitoring and analysis model. Among them, the data of each group recorded in the associated map of power equipment include historical detection data, connection relationships of each power equipment, historical operating status of the equipment, and fault records;

[0021] S2.2: The monitoring and analysis model receives the input data and performs forward propagation. The GCN layer establishes an adjacency matrix between each power equipment based on the input data, normalizes the adjacency matrix, and then performs graph convolution processing on the normalized adjacency matrix to extract the spatial characteristics of the status of each power equipment. Then, the 1D convolution kernel of the TCN layer is used to extract the time characteristics of each power equipment;

[0022] S2.3: The spatial characteristics and time characteristics of each power equipment are extracted layer by layer through multiple groups of GCN layers and TCN layers. The output layer receives the finally extracted spatial characteristics and time characteristics, and performs nonlinear activation processing on each characteristic data through an activation function, and outputs the fault association between each power equipment and predicts the occurrence probability of each type of fault. At the same time, based on the fault association result and the occurrence probability of each type of fault, the existing fault propagation path is warned in advance, and the parameters of the monitoring and analysis model are updated according to the actual power equipment fault situation.

[0023] As a further solution of the present invention, the specific calculation formula of the graph convolution processing in S2.2 is as follows:

[0024]

[0025] In the formula, represents the normalized adjacency matrix; represents the degree matrix; represents the adjacency matrix, where, if device is electrically connected to device , then , otherwise it is 0; represents the node space feature of the th layer; represents the activation function; represents the node space feature of the th layer; represents the training weight of the th layer;

[0026] The specific calculation formula for the 1D convolution kernel to extract the time features of each power device as described in S2.2 is as follows:

[0027]

[0028] In the formula, represents the time feature at time; represents the size of the time convolution kernel; represents the th time convolution kernel weight; represents the dilation rate, which controls the sampling interval.

[0029] As a further solution of the present invention, the specific steps for the device life prediction module to analyze the aging conditions of each device are as follows:

[0030] S3.1: The device life prediction module collects the physical parameters of each electrical device through actual measurement by the staff or referring to the user manual of the device manufacturer, constructs a device simulation model based on the collected physical parameters, establishes physical equations for the corresponding components according to the internal structure, components and operating mechanism of the electrical device, and uses the finite element method to perform mesh division on the electrical device. Then, combined with the real-time sensor data, the physical equations of each device simulation model are solved to obtain the internal state distribution of the electrical device;

[0031] S3.2: Based on each device simulation model, a corresponding aging prediction model is constructed, and the historical operation data of each power device is divided into a training set and a test set. The training set data is input into the aging prediction model, and through the forward propagation algorithm, each piece of input training set data is processed layer by layer, and the final power device aging prediction value is output. Then, the aging prediction value is compared with the historical true aging value, and the corresponding model error value is calculated through the mean square error;

[0032] S3.3: If the model error value is higher than the preset threshold, input the error value into the aging prediction model, and adjust the parameters of the aging prediction model through the backpropagation algorithm. After each round of training, input the test set data into the aging prediction model, and calculate the error value of the trained aging prediction model. If the error value converges within the preset range, stop the training; otherwise, continue to train the model.

[0033] S3.4: Input the current operating states of each power device and the fault prediction results into the corresponding aging prediction models respectively. Predict the aging trend residuals of each power device through the forward propagation algorithm, and correct the remaining life of each power device based on the aging trend residuals. At the same time, output the corrected remaining life of each power device. Compare the latest life prediction result with the previous round of life prediction result. If the difference in the life prediction values of the power devices exceeds the preset threshold, generate an early warning in advance and formulate a maintenance plan.

[0034] As a further solution of the present invention, the specific calculation formula of the model error value in S3.2 is as follows:

[0035]

[0036] In the formula, represents the total number of power devices in the substation; represents the true state value of the th power device measured by the sensor; represents the predicted state value of the th power device.

[0037] In the second aspect of the implementation of the present invention, a real-time monitoring method for a smart substation based on big data is proposed. The method includes the following steps:

[0038] Ⅰ. Collect and preprocess the operation data and environmental data of each power device in the substation;

[0039] Ⅱ. According to the collected data sets, construct a power device association graph, analyze the implicit fault associations among multiple devices, and optimize the maintenance strategy of the substation;

[0040] Ⅲ. Real-time manage the data access permissions through the zero-trust access control mechanism, and predict the life of each power device in the substation based on the real-time working condition data;

[0041] Ⅳ. Real-time analyze the factors of the loss of each power device and the output of renewable energy, and dynamically adjust the operation mode.

[0042] As a further solution of the present invention, the specific steps of optimizing the maintenance strategy of the substation in step Ⅱ are as follows:

[0043] S4.1: Based on the operating status, detection data, and maintenance resources of each power equipment, construct corresponding equipment state spaces respectively, and collect different maintenance operations and the corresponding costs, risks, and benefits of each operation to construct the corresponding action space. Set the reward function based on the maintenance cost, reliability benefit, and economic loss;

[0044] S4.2: Take the current state of the power equipment as the root node. Based on each maintenance decision in the action space, generate multiple sets of child nodes corresponding to the root node to establish an initial policy tree. Then, starting from the root node, calculate the upper confidence bound values of each child node, and based on the UCB selection strategy, layer by layer select the child node with the highest upper confidence bound value until a set of unexpanded child nodes is selected;

[0045] S4.3: Select the maintenance operation not used by this child node in the action space and expand new child nodes. Then, randomly select any new child node based on the current child node, simulate the evolution of the health state of the corresponding power equipment under the current detection operation, calculate the policy benefit through the reward function, and then transmit the simulation result back to the root node and update the cumulative reward and access times of each node;

[0046] S4.4: Repeat the processes of selection, expansion, simulation, and backtracking until the change value of the cumulative reward of each node converges to a preset threshold after multiple rounds of iteration. Then, stop the iteration. After that, select the node path with the highest cumulative reward, construct a complete maintenance strategy based on the detection operations corresponding to each node in this node path, adjust the original maintenance strategy, and dynamically update the maintenance strategy according to the changes in the operating status, detection data, and maintenance resources of subsequent power equipment.

[0047] As a further solution of the present invention, the specific steps of dynamically adjusting the operation mode in step IV are as follows:

[0048] S5.1: Collect the input carbon emissions, output carbon emissions, carbon emission reduction, and carbon emission loss of the substation, construct a carbon flow matrix to generate the corresponding substation carbon flow balance equation, and establish fitness functions respectively with three optimization objectives of minimizing the total carbon emissions, maximizing the comprehensive energy efficiency of the substation, and reducing the operating cost;

[0049] S5.2: Initialize a group of populations, where each individual in the population represents a set of substation operation parameters. Then, initialize the position and velocity of each individual, calculate the fitness value of each group of individuals through the fitness function respectively, and then take the position of the individual with the highest fitness value as the global optimal position, and select the position of the individual with the highest historical fitness value of each individual itself as the local optimal position;

[0050] S5.3: Update the velocity of each individual according to the global optimal position and the local optimal position. Then, based on the updated velocity values, update the positions of each individual. After the positions of each individual are updated, select the optimal individual through the NSGA-II algorithm, and re-perform the global optimal position update, local optimal position update, individual velocity update, and individual position update;

[0051] S5.4: Repeatedly update and iterate until the individual fitness value of the global optimal position converges to a preset threshold, then stop the optimization, select the substation operation parameters corresponding to the optimal individual, and use them as the optimal operation mode. Adjust the substation load distribution, energy storage charge and discharge, and equipment operation based on the selected optimal operation mode.

[0052] As a further solution of the present invention, the specific formula of the carbon flow matrix in S5.1 is as follows:

[0053]

[0054] Based on the constructed carbon flow matrix Construct the following substation carbon flow balance equation:

[0055]

[0056] In the above formulas, represents the input carbon emission; represents the carbon emission per unit of electricity generation; represents the electricity generated by fossil fuel power generation; represents the output carbon emission; represents the carbon emission factor; represents the user load supplied by the substation; represents the carbon emission reduction; represents photovoltaic power generation; represents the discharge of the energy storage system; represents equipment loss; represents the copper loss of the transformer; represents the iron loss of the transformer; represents the substation carbon flow balance value;

[0057] The specific calculation formula for updating the positions of each individual in S5.3 is as follows:

[0058]

[0059] In the formula, represents the velocity of the th individual at ; represents the inertia weight; represents the th individual at The speed at and represent the acceleration factors respectively; and represent the random numbers respectively; represents the individual optimal position of the th individual; represents the global optimal position; represents the current position of the th individual; represents the position of the th individual at ; represents the position of the th individual at .

[0060] Advantages of the present invention:

[0061] 1. Based on the electrical connection relationship between devices, the present invention establishes an associated map of power devices, initializes node features at the same time, constructs a monitoring and analysis model using the ST-GCN architecture, normalizes the adjacency matrix between devices through the GCN layer of the model, and extracts spatial features using graph convolution. The TCN layer extracts temporal features through 1D convolutional kernels. After multi-layer feature extraction, the implicit fault associations between power devices are finally output, and the fault propagation path is predicted to achieve early warning. Physical parameters of electrical devices are collected, a device simulation model is constructed, and physical equations are established based on the internal structure and components of the device. The finite element method is used for mesh generation, combined with real-time sensor data, the internal state distribution of the device is solved, and an aging prediction model is constructed. Historical operation data is divided into a training set and a test set, the parameters of the aging prediction model are trained and optimized, and the aging rate and crack growth rate are corrected by combining the Arrhenius equation and the Paris formula. Real-time device data is input, the aging trend of the device is predicted and the remaining life is calculated. If the difference between the two-round prediction results exceeds the threshold, an early warning is generated in advance and a maintenance plan is formulated, which can achieve high-precision fault detection and implicit association mining, improve the abnormal detection ability of the substation, and effectively optimize the life management of each power device in the substation.

[0062] 2. According to the operating status, environmental conditions, and maintenance resources of power equipment, the present invention constructs a maintenance decision-making space, collects the costs, risks, and benefits of different decisions, establishes a corresponding action space, uses the current equipment status as the root node, expands child nodes according to different maintenance decisions, and selects the optimal child nodes layer by layer through the upper confidence bound value selection strategy until the incompletely expanded child nodes. Subsequently, simulate the impact of the maintenance strategy on the future health status of the equipment, calculate the benefits based on the reward function, backtrack and update the cumulative rewards and visit counts of each node, and continuously iterate this process until the reward value converges. Finally, select the maintenance strategy corresponding to the path with the highest cumulative reward to optimize the maintenance plan. Then, construct a carbon flow matrix, establish a fitness function with the goals of minimizing carbon emissions, maximizing comprehensive energy efficiency, and reducing operating costs, initialize the population, where each individual represents a set of substation operating parameters, calculate the fitness value, determine the global and local optimal positions. Subsequently, adjust the individual speed and position according to the optimal positions, and use the NSGA-II algorithm to screen the optimal individuals, continuously optimize until the fitness value converges. Finally, select the substation operating parameters corresponding to the optimal individuals to optimize the load distribution, energy storage charging and discharging strategies, and equipment operating parameters, which can ensure reasonable allocation of maintenance resources, improve equipment reliability, enhance comprehensive energy efficiency, reduce carbon emissions, eliminate the need for manual intervention to adjust the operation mode, improve the optimization efficiency, and achieve the intelligent operation of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The present invention will be further described below with reference to the accompanying drawings.

[0064] Figure 1 It is a system block diagram of a real-time monitoring system for a smart substation based on big data provided by an embodiment of the present invention;

[0065] Figure 2 It is a flowchart of a real-time monitoring method for a smart substation based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] An embodiment of the present invention provides a real-time monitoring system for a smart substation based on big data. Refer to Figure 1 ,Figure 1 This is a system block diagram of a real-time monitoring system for a smart substation based on big data provided by an embodiment of the present invention. The system includes the following modules: a multi-source acquisition and processing module, a dynamic topology update module, a device status monitoring module, a fault diagnosis and prediction module, a device life prediction module, an energy efficiency analysis and optimization module, and a remote operation and maintenance scheduling module.

[0069] The multi-source acquisition and processing module is used to collect real-time detection data collected by different sensors deployed on various power equipment in the substation and preprocess each group of collected detection data.

[0070] Specifically, the multi-source acquisition and processing module receives each detection data collected by different sensors, including temperature data, voltage data, humidity data, and ultrasonic data. It divides the detection data of the same type into a group, and then synchronizes the sampling time steps of each group of detection data through the time window resampling method. After synchronization, through unit standardization processing, it unifies the units of each group of detection data, arranges each group of detection data in ascending order of data values, and respectively takes the values at the 25% position and 75% position of each group of detection data as the first quartile Q1 and the third quartile Q3 of each group of data, and obtains the interquartile range IQR of each group of detection data through Q3 - Q1. Then, based on the IQR, it takes Q1 - 1.5×IQR as the lower boundary and Q3 + 1.5×IQR as the upper boundary to draw the corresponding box plot. It traverses each group of detection data. If there is detection data lower than the lower boundary or higher than the upper boundary, it marks the detection data as an outlier. It traverses each group of detection data and counts the missing values in each group of detection data. Then, it fills each missing value through linear interpolation and replaces each outlier through mean filling processing. After the outlier processing is completed, it recalculates Q1, Q3, and IQR, and draws a new box plot to perform outlier detection again to verify the quality of each group of detection data. After verification, it normalizes each group of detection data to the range [0, 1].

[0071] The dynamic topology update module is used to establish a power equipment association graph and dynamically update the power equipment association graph based on the preprocessed detection data.

[0072] It should be further noted that the dynamic topology update module collects each group of power equipment in the substation and uses them as nodes. According to the electrical connection relationship between the equipment, it establishes edges between each node to connect each node. It establishes a power equipment association graph based on each node and the edges between each node, and initializes each node feature, where the features include detection data, the historical operation status of the equipment, and fault records, and dynamically updates the power equipment association graph based on the real-time collected detection data.

[0073] The device status monitoring module monitors and analyzes the operating status of each power device in the substation based on the preprocessed detection data received in real time and the power device association map.

[0074] Specifically, the status monitoring module establishes a monitoring and analysis model based on the ST-GCN architecture. This monitoring and analysis model includes an input layer, a GCN layer, a TCN layer, and an output layer. The groups of data recorded in the power device association map established by the dynamic topology update module and the preprocessed groups of real-time detection data are used as input data and transmitted to the monitoring and analysis model. Among them, the groups of data recorded in the power device association map include historical detection data, the connection relationships of each power device, the historical operating status of the device, and fault records. The monitoring and analysis model receives the input data and performs forward propagation. The GCN layer establishes an adjacency matrix between each power device based on the input data, normalizes the adjacency matrix, and then performs graph convolution processing on the normalized adjacency matrix to extract the spatial features of the status of each power device. Then, the 1D convolution kernel of the TCN layer is used to extract the time features of each power device. The spatial features and time features of each power device are extracted layer by layer through multiple groups of the GCN layer and the TCN layer. The output layer receives the finally extracted spatial features and time features, performs nonlinear activation processing on each feature data through an activation function, and outputs the fault associations between each power device and predicts the occurrence probability of each type of fault. At the same time, based on the fault association results and the occurrence probability of each type of fault, the existing fault propagation paths are warned in advance, and the parameters of the monitoring and analysis model are updated according to the actual power device fault situation.

[0075] It should be further noted that the specific calculation formula for graph convolution processing is as follows:

[0076]

[0077] In the formula, represents the normalized adjacency matrix; represents the degree matrix; represents the adjacency matrix, where if device is electrically connected to device , then , otherwise it is 0; represents the node spatial feature of the th layer; represents the activation function; represents the node spatial feature of the th layer; represents the training weight of the th layer;

[0078] The specific calculation formula for the 1D convolution kernel to extract the time features of each power device is as follows:

[0079]

[0080] In the formula, represents the time feature at a moment; represents the size of the time convolution kernel; represents the weight of the nth time convolution kernel; represents the dilation rate, which controls the sampling interval.

[0081] Based on the real-time analysis results of the operating states of various power equipment in the substation and the equipment aging situation, the fault diagnosis and prediction module predicts the potential faults of various power equipment in the substation;

[0082] The equipment life prediction module is used to construct a virtual model of the equipment, and based on the real-time analysis results of the operating states of various power equipment in the substation and the fault prediction results, analyze the aging situation of each equipment.

[0083] Specifically, the equipment life prediction module collects the physical parameters of each electrical equipment through actual measurement by staff or referring to the operation manuals of equipment manufacturers, constructs an equipment simulation model based on the collected physical parameters, establishes physical equations for the corresponding components according to the internal structure, components and operation mechanism of the electrical equipment, and uses the finite element method to perform mesh division on the electrical equipment. Then, combined with real-time sensor data, solve the physical equations of each equipment simulation model to obtain the internal state distribution of the electrical equipment, construct a corresponding aging prediction model based on each equipment simulation model, divide the historical operation data of each power equipment into a training set and a test set, input the training set data into the aging prediction model, and through the forward propagation algorithm, process the input training set data layer by layer and output the final aging prediction value of the power equipment. Then, compare the aging prediction value with the historical true aging value, calculate the corresponding model error value through the mean square error. If the model error value is higher than the preset threshold, input the error value into the aging prediction model, and through the backpropagation algorithm, adjust the parameters of the aging prediction model. After each round of training, input the test set data into the aging prediction model and calculate the error value of the trained aging prediction model. If the error value converges within the preset range, stop training; otherwise, continue to train the model. Input the current operating states and fault prediction results of each power equipment into the corresponding aging prediction models respectively, predict the aging trend residuals of each power equipment through the forward propagation algorithm, and correct the remaining life of each power equipment based on the aging trend residuals, and at the same time output the corrected remaining life of each power equipment. Compare the latest life prediction result with the previous round of life prediction result. If the difference in the life prediction value of the power equipment exceeds the preset threshold, generate an early warning in advance and formulate a maintenance plan.

[0084] In this embodiment, the specific calculation formula of the model error value is as follows:

[0085]

[0086] In the formula, represents the total number of power equipment in the substation; represents the th true state value of the power equipment measured by the sensor; represents the predicted th state value of the power equipment.

[0087] The energy efficiency analysis and optimization module optimizes the operation modes of the power equipment in the substation according to the real-time analysis results of the operation states of the power equipment in the substation and the power equipment association graph.

[0088] The remote operation and maintenance scheduling module is used to optimize the maintenance plan of the substation by combining the fault detection data, operation modes and life prediction results of the power equipment in the substation.

[0089] The embodiment of the present invention also provides a real-time monitoring method for a smart substation based on big data. As Figure 2 shown, the method includes the following steps:

[0090] Collect and preprocess the operation data and environmental data of the power equipment in the substation.

[0091] According to the collected data groups, construct a power equipment association graph, analyze the hidden fault associations among multiple equipment, and optimize the maintenance strategy of the substation.

[0092] Specifically, according to the operating states, detection data, and maintenance resources of each power equipment, corresponding equipment state spaces are constructed respectively, and different maintenance operations and the corresponding costs, risks, and benefits of each operation are collected to construct the corresponding action space. A reward function is set based on the maintenance cost, reliability benefit, and economic loss. Taking the current state of the power equipment as the root node, multiple groups of child nodes corresponding to the root node are generated based on each maintenance decision in the action space to establish an initial policy tree. Then, starting from the root node, the upper confidence bound values of each child node are calculated, and based on the UCB selection strategy, the child node with the highest upper confidence bound value is selected layer by layer until a group of unexpanded child nodes is selected. The maintenance operation not used by this child node in the action space is selected and new child nodes are expanded. Then, any new child node is randomly selected based on the current child node, and the evolution of the health state of the corresponding power equipment under the current detection operation is simulated. The policy benefit is calculated through the reward function, and the simulation results are then passed back to the root node, and the cumulative rewards and visit times of each node are updated. The selection, expansion, simulation, and backtracking are repeated until the change value of the cumulative rewards of each node converges to a preset threshold in multiple rounds of iteration, and then the iteration stops. After that, the node path with the highest cumulative reward is selected, and a complete maintenance strategy is constructed based on the detection operations corresponding to each node in this node path, and the original maintenance strategy is adjusted. At the same time, according to the changes in the operating states, detection data, and maintenance resources of subsequent power equipment, the maintenance strategy is dynamically updated.

[0093] Real-time manage data access permissions through a zero-trust access control mechanism and predict the service life of power equipment in each substation based on real-time operating condition data.

[0094] Real-time analyze various factors of power equipment loss and renewable energy output, and dynamically adjust the operation mode.

[0095] Specifically, collect the input carbon emissions, output carbon emissions, carbon emission reduction, and carbon emission losses of the substation, and construct a carbon flow matrix to generate the corresponding substation carbon flow balance equation. Then, establish fitness functions with three optimization objectives: minimizing the total carbon emissions, maximizing the comprehensive energy efficiency of the substation, and reducing the operating cost. Initialize a population, where each individual represents a set of substation operating parameters. Then, initialize the position and velocity of each individual, and calculate the fitness value of each group of individuals through the fitness function. After that, take the position of the individual with the highest fitness value as the global optimal position, and select the position of the individual with the highest historical fitness value of each individual as the local optimal position. Update the velocity of each individual according to the global optimal position and the local optimal position. Then, based on the updated velocity value, update the position of each individual. After the position of each individual is updated, select the optimal individual through the NSGA-II algorithm, and re-update the global optimal position, local optimal position, individual velocity, and individual position. Repeat the update and iteration until the individual fitness value of the global optimal position converges to the preset threshold, then stop the optimization, and select the substation operating parameters corresponding to the optimal individual, and use them as the optimal operating mode. Based on the selected optimal operating mode, adjust the substation load distribution, energy storage charge and discharge, and equipment operation.

[0096] It should be further noted that the specific formula of the carbon flow matrix is as follows:

[0097]

[0098] Based on the constructed carbon flow matrix To construct the following substation carbon flow balance equation:

[0099]

[0100] In the above formulas, represents the input carbon emissions; represents the carbon emissions per unit of electricity generation; represents the electricity generated by fossil fuel power generation; represents the output carbon emissions; represents the carbon emission factor; represents the user load supplied by the substation; represents the carbon emission reduction; represents photovoltaic power generation; represents the discharge of the energy storage system; represents equipment losses; represents the copper loss of the transformer; represents the iron loss of the transformer; represents the substation carbon flow balance value;

[0101] The specific calculation formula for updating the position of each individual is as follows:

[0102]

[0103] wherein, represents the speed of the th individual at ; represents the inertia weight; represents the speed of the th individual at ; and respectively represent the acceleration factors; and respectively represent random numbers; represents the individual optimal position of the th individual; represents the global optimal position; represents the current position of the th individual; represents the position of the th individual at ; represents the position of the th individual at .

[0104] The above has described in detail an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention application shall still fall within the scope covered by the patent of the present invention.

Claims

1. A real-time monitoring system for smart substations based on big data, characterized in that: include: Multi-source acquisition and processing module, dynamic topology update module, equipment status monitoring module, fault diagnosis and prediction module, equipment life prediction module, energy efficiency analysis and optimization module, and remote operation and maintenance scheduling module; The multi-source acquisition and processing module is used to collect real-time detection data collected by different sensors deployed in various power equipment in the substation, and pre-process each group of collected detection data; The dynamic topology update module is used to establish a power equipment association map and dynamically update the power equipment association map based on the pre-processed detection data; The dynamic topology update module collects each group of power equipment in the substation and takes it as a node. According to the electrical connection relationship between the equipment, the edges between the nodes are established to connect the nodes. The power equipment association map is established according to the nodes and the edges between the nodes, and each node feature is initialized. The features include detection data, equipment historical operation status and fault records, and the power equipment association map is dynamically updated according to the detection data of each group collected in real time. The equipment status monitoring module monitors and analyzes the operating status of each power equipment in the substation based on the pre-processed detection data received in real time and the power equipment association map; The fault diagnosis and prediction module predicts potential faults of each power equipment in the substation based on the real-time analysis results of the operating status of each power equipment in the substation and the aging of the equipment; The equipment life prediction module is used to build a virtual model of the equipment, and analyze the aging of each device based on the real-time analysis results of the operating status of each power equipment in the substation and the fault prediction results; The energy efficiency analysis and optimization module optimizes the operation mode of each power equipment in the substation based on the real-time analysis results of the operation status of each power equipment in the substation and the power equipment association map; The remote operation and maintenance scheduling module is used to optimize the substation maintenance plan by combining the fault detection data, operation mode and life prediction results of each power equipment in the substation.

2. According to the big data-based smart substation real-time monitoring system of claim 1, it is characterized in that: The specific steps of preprocessing each group of collected detection data by the multi-source acquisition processing module are as follows: S1.1: The multi-source acquisition processing module receives various detection data collected by different sensors, including temperature data, voltage data, humidity data and ultrasonic data, and groups the detection data of the same type into one group. Then, the sampling time step of each group of detection data is synchronized through the time window resampling method. After the synchronization is completed, the unit of each group of detection data is unified through unit standardization processing; S1.2: Arrange each group of test data from small to large according to the data value, and use the values ​​of the 25% and 75% positions of each group of test data as the first quartile Q1 and the third quartile Q3 of each group of data, and obtain the interquartile range IQR of each group of test data through Q3-Q1. Then, based on IQR, use Q1-1.5×IQR as the lower boundary and Q3+1.5×IQR as the upper boundary to draw the corresponding box plot, traverse each group of test data, and if there is a test data below the lower boundary or above the upper boundary, mark the test data as an outlier; S1.3: Traverse each group of test data and count the missing values ​​in each group of test data. Then fill the missing values ​​through linear interpolation, and then replace the outliers through mean filling. After the outlier processing is completed, recalculate Q1, Q3 and IQR, and draw a new box plot to perform outlier detection again to verify the quality of each group of test data. After the verification, normalize the test data of each group to the range of [0, 1].

3. A real-time monitoring system for smart substations based on big data according to claim 2, characterized in that: The specific steps of the equipment status monitoring module to monitor and analyze the operating status of each power equipment in the substation are as follows: S2.1: The state monitoring module establishes a monitoring and analysis model based on the ST-GCN architecture. The monitoring and analysis model includes an input layer, a GCN layer, a TCN layer, and an output layer. Each group of data recorded in the power equipment association map established by the dynamic topology update module and each group of pre-processed real-time detection data are transmitted to the monitoring and analysis model as input data. Among them, each group of data recorded in the power equipment association map includes historical detection data, connection relationships of various power equipment, historical operation status of equipment, and fault records; S2.2: The monitoring and analysis model receives input data and performs forward propagation. The GCN layer establishes the adjacency matrix between each power device based on the input data, normalizes the adjacency matrix, and then performs graph convolution on the normalized adjacency matrix to extract the spatial characteristics of the state of each power device. Then, the 1D convolution kernel of the TCN layer is used to extract the temporal characteristics of each power device. S2.3: The spatial and temporal features of each power equipment are extracted layer by layer through multiple groups of GCN layers and TCN layers. The output layer receives the final extracted spatial and temporal features, and performs nonlinear activation processing on each feature data through the activation function, and outputs the fault correlation between each power equipment and predicts the probability of occurrence of each type of fault. At the same time, based on the fault correlation results and the probability of occurrence of each type of fault, the existing fault propagation path is warned in advance, and the parameters of the monitoring and analysis model are updated according to the actual power equipment fault situation.

4. A real-time monitoring system for smart substations based on big data according to claim 3, characterized in that: The specific steps for the equipment life prediction module to analyze the aging of each device are as follows: S3.1: The equipment life prediction module collects the physical parameters of each electrical equipment through actual measurements by staff or equipment manufacturer's manuals, and builds an equipment simulation model based on the collected physical parameters. According to the internal structure, components and operation mechanism of the electrical equipment, the physical equations of the corresponding components are established, and the finite element method is used to mesh the electrical equipment. Then, combined with real-time sensor data, the physical equations of the simulation models of each equipment are solved to obtain the internal state distribution of the electrical equipment; S3.2: Based on the simulation model of each device, a corresponding aging prediction model is constructed, and the historical operation data of each power device is divided into a training set and a test set. The training set data is input into the aging prediction model. Through the forward propagation algorithm, each input training set data is processed layer by layer, and the final power equipment aging prediction value is output. After that, the aging prediction value is compared with the historical actual aging value, and the corresponding model error value is calculated by the mean square error; S3.3: If the model error value is higher than the preset threshold, the error value is input into the aging prediction model, and the aging prediction model parameters are adjusted through the back propagation algorithm. After each round of training, the test set data is input into the aging prediction model, and the error value of the trained aging prediction model is calculated. If the error value converges to the preset range, the training is stopped, otherwise, the model training continues; S3.4: Input the current operating status of each power equipment and the fault prediction results into the corresponding aging prediction model respectively, predict the aging trend residual of each power equipment through the forward propagation algorithm, and correct the remaining life of each power equipment based on the aging trend residual. At the same time, output the corrected remaining life of each power equipment, compare the latest life prediction result with the previous round of life prediction results, and if the difference between the power equipment life prediction values ​​exceeds the preset threshold, generate an early warning and formulate a maintenance plan.

5. A real-time monitoring method for a smart substation based on big data, used to implement the function of a real-time monitoring system for a smart substation based on big data as described in any one of claims 1-4, characterized in that: The following steps are involved: Ⅰ. Collect and pre-process the operation data and environmental data of each power equipment in the substation; Ⅱ. Based on the collected data, a power equipment correlation map is constructed to analyze the hidden fault correlation between multiple devices and optimize the maintenance strategy of the substation; III. Manage data access rights in real time through zero-trust access control mechanism, and predict the life of power equipment in each substation based on real-time operating data; IV. Real-time analysis of various power equipment losses and renewable energy output factors, and dynamic adjustment of operating modes.

6. A real-time monitoring method for smart substations based on big data according to claim 5, characterized in that: The specific steps of optimizing the maintenance strategy of the substation described in step II are as follows: S4.1: According to the operating status, detection data and maintenance resources of each power equipment, the corresponding equipment state space is constructed respectively, and different maintenance operations and the corresponding costs, risks and benefits of each operation are collected to construct the corresponding action space, and the reward function is set based on maintenance cost, reliability benefit and economic loss; S4.2: The current state of the power equipment is taken as the root node, and multiple groups of child nodes corresponding to the root node are generated based on each maintenance decision in the action space to establish an initial strategy tree. Then, starting from the root node, the upper confidence bound value of each child node is calculated, and based on the UCB selection strategy, the child node with the highest upper confidence bound value is selected layer by layer until a group of incompletely expanded child nodes is selected; S4.3: Select the unused maintenance operation of the child node in the action space and expand the new child node. Then randomly select any new child node based on the current child node, simulate the evolution of the health state of the corresponding power equipment under the current detection operation, calculate the strategy benefit through the reward function, and then send the simulation result back to the root node, and update the cumulative reward and visit count of each node; S4.4: Repeat the selection, expansion, simulation and backtracking until the cumulative reward of each node converges to the preset threshold in multiple rounds of iterations, then stop the iteration, select the node path with the highest cumulative reward, and build a complete maintenance strategy based on the detection operations corresponding to each node in the node path, and adjust the original maintenance strategy. At the same time, dynamically update the maintenance strategy based on the subsequent changes in the operating status of each power equipment, detection data and maintenance resources.

7. A real-time monitoring method for smart substations based on big data according to claim 5, characterized in that: The specific steps of dynamically adjusting the operation mode described in step IV are as follows: S5.1: Collect the input carbon emissions, output carbon emissions, carbon emission reductions and carbon emission losses of the substation, and construct a carbon flow matrix to generate the corresponding substation carbon flow balance equation, and then establish a fitness function with three sets of optimization goals: minimizing the total carbon emissions, maximizing the comprehensive energy efficiency of the substation and reducing the operating cost; S5.2: Initialize a group of populations, each individual in the population represents a group of substation operating parameters, then initialize the position and speed of each individual, and calculate the fitness value of each group of individuals through the fitness function, then take the individual position with the highest fitness value as the global optimal position, and select the individual position with the highest historical fitness value of each individual as the local optimal position; S5.3: Update the speed of each individual according to the global optimal position and the local optimal position, and then update the position of each individual based on the updated speed value. After the position of each individual is updated, the best individual is selected through the NSGA-II algorithm, and the global optimal position update, local optimal position update, individual speed update and individual position update are performed again; S5.4: Repeatedly update and iterate until the individual fitness value of the global optimal position converges to the preset threshold, then stop the optimization, select the substation operation parameters corresponding to the optimal individual, and use it as the optimal operation mode. Based on the selected optimal operation mode, adjust the substation load distribution, energy storage charging and discharging, and equipment operation.

Images