Substation equipment state monitoring and intelligent fault early warning method based on deep learning

Abstract

The invention discloses a substation equipment state monitoring and intelligent fault early warning method based on deep learning. The method comprises the following steps: S1, obtaining a preprocessed multi-source state data set; s2, generating a high-dimensional equipment state feature matrix; s3, a fault sensitive deep belief network model is adopted to form a preliminary fault state recognition result; s4, obtaining an optimized sensitive depth belief network model; s5, performing online analysis on the multi-source state data acquired in real time by using the optimized sensitive deep belief network model, generating a real-time fault prediction result of the equipment state, and classifying and grading fault risks; and S6, according to a real-time fault prediction result, triggering a remote fault early warning mechanism, and sending fault early warning information including a fault risk level, an early warning signal and an emergency processing suggestion to a substation operation and maintenance center. According to the invention, intelligent alarm linkage and hierarchical control of the scheduling system are effectively supported.

Description

Technical Field

[0001] The present invention relates to the technical field of substations, and particularly to a method for substation equipment status monitoring and intelligent fault early warning based on deep learning. Background Art

[0002] With the continuous improvement of the intelligent level of the power system, as the core node in the power transmission and distribution network, the operation safety of the substation and the health status of the equipment have an important impact on the overall stability of the power grid. Substation remote fault early warning, as a key link in the intelligent operation and maintenance system, undertakes important responsibilities such as equipment status monitoring, potential fault identification, and risk trend prediction.

[0003] Currently, the mainstream substation fault early warning methods are mostly based on static threshold setting or empirical model establishment. The core is to set fixed upper and lower limits for monitoring parameters such as voltage, current, and temperature, and once the value exceeds the limit, an early warning is triggered. Although such methods are simple to implement and have a certain early fault response ability, they have obvious defects: on the one hand, it is difficult for static thresholds to take into account the normal fluctuation ranges of different devices in various operating states, resulting in frequent false alarms and missed alarms; on the other hand, empirical models cannot capture the non-linear fault evolution process, especially when facing complex system disturbances and hidden fault precursors, the early warning effect is significantly insufficient.

[0004] In addition, some academic and engineering fields have tried to introduce classification and recognition methods based on machine learning to predict and analyze substation faults. However, such methods generally rely on artificial feature construction and are difficult to adapt to the data characteristics of high-dimensional, multi-source, and strong time-series coupling in the actual power system. There are problems such as weak model generalization ability, complex parameter adjustment, and insufficient sensitivity to abnormal states, which limit their popularization and application in engineering practice.

[0005] The current mainstream deep learning applications mostly focus on traditional convolutional neural networks or recurrent neural networks. Although they perform well in image recognition or sequence prediction, when facing the multi-dimensional, low signal-to-noise ratio, and dynamically evolving data characteristics in the operation of substation equipment, they often fall into local optima due to improper initial model parameters or rigid training structures and are difficult to dynamically adjust to adapt to the operation environment differences of substations in different regions. In addition, the parameter tuning process in existing research still mainly relies on manual trial and error methods or random initialization, lacking an intelligent optimization mechanism linked to the real-time operation feedback of the system.

[0006] In summary, there is an urgent need for an innovative intelligent early warning method that can integrate multi-source state information, adaptively optimize the learning structure, and take into account the global search ability and fault sensitivity characteristics to achieve early and accurate identification and hierarchical response to equipment faults in complex power systems, thereby ensuring the safe and stable operation of substations and the overall power grid. Summary of the Invention

[0007] An object of the present invention is to provide a substation equipment status monitoring and intelligent fault warning method based on deep learning. The present invention can not only improve the response speed of fault handling, but also support the dynamic evolution tracking of risks across time periods, effectively supporting the intelligent alarm linkage and hierarchical control of the dispatching system.

[0008] A substation equipment status monitoring and intelligent fault warning method based on deep learning according to an embodiment of the present invention includes the following steps: S1. Real-time collect multi-source status data sets and perform preprocessing to obtain preprocessed multi-source status data sets; S2. Use the preprocessed multi-source status data sets to construct equipment status feature vectors, and adopt the time series data segmentation and multi-scale sliding window method to extract key features related to fault evolution, generating a high-dimensional equipment status feature matrix; S3. A fault-sensitive deep belief network model performs multi-level unsupervised feature learning on the high-dimensional equipment status feature matrix to form a preliminary fault status recognition result; S4. Apply the jellyfish swarm optimization algorithm to globally search and dynamically optimize the weight parameters and bias parameters of the deep belief network model. The deviation between the preliminary fault status recognition result and the predetermined recognition performance index is used as feedback information during the adjustment process to obtain an optimized sensitive deep belief network model; S5. Use the optimized sensitive deep belief network model to perform online analysis on the real-time collected multi-source status data, generate real-time fault prediction results of the equipment status, and classify and grade the fault risks; S6. According to the real-time fault prediction result, trigger a remote fault warning mechanism and send a fault warning message including the fault risk level, warning signal and emergency treatment suggestions to the substation operation and maintenance center.

[0009] Optionally, the S1 includes the following steps: S11. Set a data collection period. In each collection period, voltage sensors, current sensors, temperature sensors and switch status collection devices deployed at key parts of the substation are used to collect voltage data , current data , temperature data and switch status data respectively to construct an original multi-source status data set : ; Wherein, represents the i-th acquisition record, represents the time stamp of the i-th data acquisition, is the number of samples within the set time window; S12. Perform noise removal processing on the original multi-source state dataset. Use the moving average method to smooth the continuous variables. Calculate the average of the current value of each data point and its previous adjacent data points as the denoising result of this data point to eliminate the influence of short-term spike noise on the data; S13. Perform outlier detection and removal on the original multi-source state dataset after preliminary smoothing. Judge the deviation degree of a single point in the overall data distribution. Divide the difference between the value of each data point and the mean of its dimension by the standard deviation of this dimension to obtain the standardized value. When the standardized value is greater than the preset outlier determination threshold, it is regarded as an outlier and removed; S14. Perform missing value filling on the multi-source state dataset after outlier removal. For continuous variables such as voltage data, current data, temperature data, and switch status data, use the average of the previous and next valid data points of this data point as the interpolation result to maintain the continuity and change trend of the data sequence in the multi-source state dataset; S15. Perform normalization processing on the multi-source state dataset after filling missing values. Subtract the minimum value in each dimension from each dimension of data and then divide by the difference between the maximum value and the minimum value in this dimension to map the data of each dimension to the interval , and eliminate the dimensional differences between different physical quantities; S16. Define the multi-source state dataset after noise removal, outlier removal, missing value filling, and normalization processing as the preprocessed multi-source state dataset .

[0010] Optionally, S2 includes the following steps: S21. For the preprocessed multi-source state dataset Perform equally spaced segmentation according to the acquisition time series , set the segmentation length to , and divide the preprocessed multi-source state dataset into several time series data sub-segments : ; Among them, , is the total time span of the preprocessing dataset, represents the th data sub-segment within the time window; S22. For each time series data sub-segment Construct a multi-scale feature extraction structure based on a sliding window. Set the sliding window length set, perform a sliding traversal operation for each window length , and extract the statistical features of voltage data, current data, temperature data, and switch status data within each sliding window, including the mean , Standard Deviation , Skewness and Range ; Mean: ; Standard Deviation: ; Skewness: ; Range: ; Wherein, represents the single-dimensional data at the th segment, the nd sliding window, and the th moment, is the mean of the sliding window index; S23. Integrate the statistical features extracted from all time-series data sub-segments and window length to generate a set of feature vectors with the time period index and scale index as the combined index ; Each feature vector is represented as: ; Wherein, , , , respectively represent the mean, standard deviation, skewness, and range features of the voltage at the th segment and the th scale. The current, temperature, and switch status data of other dimensions are extracted in the same way; S24. Concatenate all the sets of feature vectors to form a high-dimensional device status feature matrix , where is the number of time windows, is the number of sliding window scales, and is the number of single-segment feature dimensions extracted at each scale.

[0011] Optionally, S3 includes the following steps: S31. Calculate the fault sensitivity coefficients of each feature component for the safety boundaries of each key parameter during the operation of the substation, and form a fault sensitivity coefficient matrix : ; Wherein, represents in the th time period, the The eigenvector under the scale is a preset threshold for reflecting the safe operation state of the substation is the fault risk amplification factor; The calculated fault sensitivity coefficient matrix and the high-dimensional equipment state feature matrix are multiplied element by element to form a fault sensitivity input matrix ; S32. Using the fault sensitivity input matrix Construct a fault sensitivity deep belief network model, and the fault sensitivity deep belief network model is represented by a stacked structure of multiple improved restricted Boltzmann machines as: ; Among them, represents the th layer of improved restricted Boltzmann machine, is the total number of layers of the fault sensitivity deep belief network model. In each layer of the fault sensitivity deep belief network model, the input vector is denoted as For the first layer, , and the hidden layer is represented as ; S33. In each layer of the improved restricted Boltzmann machine inside, the input vector and the hidden vector are modeled using an energy function that fuses the fault sensitivity coefficient: ; Among them, represents the th input node in the th layer, is the connection weight between the th input node and the th hidden node in the th layer, and are the biases of the input nodes and hidden nodes in this layer respectively, is the fault sensitivity coefficient corresponding to the th input node, taken from the fault sensitivity input matrix obtained after corresponding mapping, is the number of neurons in the visible layer of the lth layer, is the number of neurons in the hidden layer of the lth layer, represents the hidden layer vector of the lth layer; S34. Using the contrastive divergence algorithm, on the premise of ensuring that the fault abnormal features in the multi-source state data of the substation are fully strengthened, for each layer of the improved restricted Boltzmann machine parameters Perform layer-by-layer unsupervised pre-training and use the output of each hidden layer as the input of the next layer. During the training process, the fault sensitivity coefficient adjusts the weights of each input node in the gradient update, so that the fault-sensitive deep belief network model focuses on the features that deviate abnormally from the safety threshold; S35. After completing the unsupervised pre-training of all improved restricted Boltzmann machine layers, perform supervised fine-tuning on the entire fault-sensitive deep belief network model, and use the fault label vector generated from historical fault records , where is the number of fault categories, and the cross-entropy loss function is used to jointly optimize the parameters of the entire network to generate a fault feature expression vector ; S36. Transfer the fault feature expression vector to the output layer of the fault-sensitive deep belief network model, and use the fault mode classifier for forward inference to output the preliminary fault status recognition result .

[0012] Optionally, the S4 includes the following steps: S41. Unify and expand the connection weight parameters and the bias parameters of each layer in the trained fault-sensitive deep belief network model into a parameter vector , where represents the jellyfish individual number, is the jellyfish population size, and each jellyfish individual represents a set of potential optimal configurations of the fault-sensitive deep belief network model under the current working conditions of the substation; S42. Perform forward inference on the multi-source state data set using the fault-sensitive deep belief network model corresponding to the jellyfish individual position to obtain the preliminary fault recognition output , and compare the preliminary fault recognition output with the historical label , introduce the key equipment weight coefficient of the substation to form a device weighted cross-entropy loss function that combines prediction performance and equipment importance : ; where Q represents the number of devices participating in the evaluation, represents the running criticality weight of the qth device, c represents the total number of fault types that the model can predict, represents the true ith type of fault label of the qth device in the training sample, represents the jellyfish individual The predicted probability output of the deep belief network model represented for the i-th type of fault on the q-th device; Device weighted cross-entropy loss function It not only reflects the accuracy of the model in identifying various types of faults, but also strengthens the attention to key devices by operating critical weights, ensuring that the optimization process does not ignore the components crucial for system stability in the substation while improving the overall accuracy. The device weighted cross-entropy loss function constructed in this way is more in line with the requirements of power engineering for safety redundancy, priority response of device levels, and fault criticality.

[0013] S43. Initialize the jellyfish population And set the maximum number of iterations , in each round t, combined with the dynamic working condition weight of the current substation operation status , use the ocean current drive update formula to achieve global optimization: ; Among them, represents the position vector of the s-th jellyfish individual in the t-th round of iteration, represents the updated position vector of the s-th jellyfish individual in the (t + 1)-th round of iteration represents the dynamic working condition weight introduced for the substation operation status in the t-th round of iteration, which is used to adjust the response ability of the global search behavior of the jellyfish group to environmental disturbances at different times, represents the ocean current drift factor, which is used to control the amplitude of the jellyfish individual approaching the current optimal individual, reflecting the cluster movement speed of the jellyfish group, represents the position vector of the individual with the optimal fitness in the entire jellyfish group in the t-th round of iteration; S44. Introduce an active foraging behavior update mechanism that combines recognition error and fault level response to accelerate the adjustment speed of the fault-sensitive deep belief network model body with an error greater than the preset value: ; Among them, is the active search coefficient, is a randomly selected individual, is the response weight of the s-th jellyfish individual in the current round, which is used to guide the convergence direction to correct the fault misjudgment area; S45. In each round of iteration, dynamically switch between the ocean current drive and the active foraging mechanism according to the change trend of individual fitness and the difference between the global optimal model performance, and record the current optimal model parameter vector ; S46. If the maximum number of iterations is reached or continuously The global optimal fitness improvement in the iteration is less than the convergence threshold , the iteration is terminated and the optimal parameter vector finally obtained is Decoded into a set of fault-sensitive deep belief network model parameters , forming an optimized fault-sensitive deep belief network model .

[0014] Optionally, S5 includes the following steps: S51. Using the optimized fault-sensitive deep belief network model , the pre-processed multi-source state data set collected in real time Input into the model, perform forward reasoning calculations, and obtain each time point The equipment status prediction results under ,in, Indicates Class failure at time point predicted probability of occurrence; S52. Prediction results The predicted probabilities of various types of faults are sorted and the maximum value is determined, and the fault type corresponding to the maximum probability is extracted , the fault type As the main failure mode of the current equipment operation status, and record the predicted probability of the main failure mode ; S53. Combine the typical operating conditions of the substation, the risk level of the equipment and the historical fault impact data to build a fault risk classification rule; S54. According to the fault type With predicted probability , determine the current time point The equipment failure risk level under the condition is calculated, and the real-time fault prediction result label triples including prediction category, prediction probability and corresponding risk level are output. ,in, ; S55. Sequence the prediction result labels in multiple consecutive time periods Perform time series fusion analysis to generate fault trend evolution curve.

[0015] Optionally, the fault risk classification rule is established according to the following fault risk level classification standards: Low risk level: predicted probability , and the corresponding fault type It is a slight fluctuation type, including short-term current slight abnormality, low-amplitude temperature rise fluctuation, and single-point transient interference; Medium risk level: predicted probability , and the corresponding fault type Belongs to the medium warning category, including continuous overheating of equipment, unplanned switching signals, and long-term voltage deviation from the steady state; High risk level: Prediction probability , and the corresponding fault type Involves critical abnormal categories, including breaker operation failure, severe current fluctuation, and continuous temperature overlimit.

[0016] The beneficial effects of the present invention are as follows: (1) In the feature learning stage of the present invention, a fault sensitivity coefficient matrix is introduced. By constructing a fault risk amplification factor proportional to the deviation degree of operating parameters such as voltage, current, temperature, and switch state, weight enhancement processing of key features is realized, which is embedded in the energy function of the improved restricted Boltzmann machine. During the unsupervised pre-training process, it guides the deep network structure to focus on the abnormal signal areas with the most potential for faults during the operation of substation equipment. Compared with the traditional deep belief network structure, it performs better in the extraction ability of low-amplitude fluctuations, multi-dimensional data coupling, and time-series mixed faults, and greatly improves the accuracy of fault mode recognition and the ability to extract precursors under the condition of multi-source information fusion.

[0017] (2) The present invention introduces the jellyfish swarm optimization algorithm into the deep model parameter tuning process, and in view of the characteristics of variable substation working conditions and large differences in operating areas, a dynamic working condition weight function and a behavior selection function based on prediction error response are introduced to construct a hybrid optimization update strategy that combines ocean current driving and active foraging mechanisms. It can dynamically adjust the optimization direction and amplitude according to the real-time state, prediction deviation, and response speed of the equipment, making the model parameters continuously converge to the optimal solution under the current operating state, significantly reducing the overfitting risk, and improving the model stability and generalization ability.

[0018] (3) In the real-time inference stage of the present invention, combined with the output probability of the optimized deep model and the critical degree of equipment operation, a risk classification standard based on equipment importance weighting is constructed, realizing multi-level fault risk division of low risk, medium risk, and high risk. By serializing the prediction results of multiple consecutive moments, the fault evolution trend and the curve of grade change trend are further output, providing a more forward-looking and executable decision-making basis for power operation and maintenance personnel. It can not only improve the response speed of fault handling, but also support cross-time risk dynamic evolution tracking, effectively supporting the intelligent alarm linkage and hierarchical control of the dispatching system. Description of the Drawings

[0019] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1Flow chart of a substation equipment status monitoring and intelligent fault warning method based on deep learning proposed by the present invention. Detailed implementation manners

[0020] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0021] Refer to Figure 1 , a substation equipment status monitoring and intelligent fault warning method based on deep learning, comprising the following steps: S1. Collect multi-source status data sets in real time and perform preprocessing to obtain preprocessed multi-source status data sets; S2. Use the preprocessed multi-source status data sets to construct equipment status feature vectors, adopt the time series data segmentation and multi-scale sliding window method to extract key features related to fault evolution, and generate a high-dimensional equipment status feature matrix; S3. A fault-sensitive deep belief network model is used to perform multi-level unsupervised feature learning on the high-dimensional equipment status feature matrix to form a preliminary fault status recognition result; S4. Apply the jellyfish swarm optimization algorithm to globally search and dynamically optimize the weight parameters and bias parameters of the deep belief network model. The deviation between the preliminary fault status recognition result and the predetermined recognition performance index is used as feedback information during the adjustment process to obtain an optimized sensitive deep belief network model; S5. Use the optimized sensitive deep belief network model to perform online analysis on the multi-source status data collected in real time, generate real-time fault prediction results of the equipment status, and classify and grade the fault risks; S6. According to the real-time fault prediction results, trigger a remote fault warning mechanism and send fault warning information including fault risk levels, warning signals, and emergency handling suggestions to the substation operation and maintenance center.

[0022] In this embodiment, S1 includes the following steps: S11. Set a data collection period. In each collection period, voltage sensors, current sensors, temperature sensors, and switch status collection devices deployed at key parts of the substation are used to collect voltage data , current data , temperature data and switch status data respectively to construct an original multi-source status data set :

[0023] Among them, represents the i-th acquisition record, represents the timestamp of the i-th data acquisition, is the number of samples within the set time window; S12. Perform noise removal processing on the original multi-source status dataset. Use the moving average method to smooth the continuous variables. For each data point, average its current value with its previous adjacent data points as the denoising result of this data point to eliminate the influence of short-term spike noise on the data; S13. Perform outlier detection and removal on the original multi-source status dataset after preliminary smoothing processing. Judge the deviation degree of a single point in the overall data distribution. Divide the difference between the value of each data point and the mean of its dimension by the standard deviation of this dimension to obtain the standardized value. When the standardized value is greater than the preset outlier determination threshold, it is regarded as an outlier and removed; S14. Perform missing value completion on the multi-source status dataset after outlier removal. For continuous variables such as voltage data, current data, temperature data, and switch status data, use the average of the previous and next valid data points of this data point as the interpolation result to maintain the continuity and change trend of the data sequence in the multi-source status dataset; S15. Perform normalization processing on the multi-source status dataset after completing missing value completion. Subtract the minimum value in each dimension from each dimension of data and then divide by the difference between the maximum value and the minimum value in this dimension to map the data of each dimension to the interval to eliminate the dimensional difference between different physical quantities; S16. Define the multi-source status dataset after noise removal, outlier removal, missing value completion, and normalization processing as the preprocessed multi-source status dataset .

[0024] In this embodiment, S2 includes the following steps: S21. For the preprocessed multi-source status dataset perform equally spaced segmentation according to the acquisition time series , set the segmentation length as , and divide the preprocessed multi-source status dataset into several time series data sub-segments : ; wherein, , is the total time span of the preprocessing dataset, represents the th data sub-segment within the time window; S22. For each time series data sub-segment construct a multi-scale feature extraction structure based on a sliding window, set the sliding window length set, and for each window length Perform a sliding traversal operation to extract the statistical features of voltage data, current data, temperature data, and switch status data within each sliding window, including the mean value , standard deviation , skewness and range ; Mean value: ; Standard deviation: ; Skewness: ; Range: ; Among them, represents the single-dimensional data at the th segment, the th moment within the th sliding window, is the mean value of the sliding window index; S23. Integrate the statistical features extracted from all time series data segments and the window length to generate a set of feature vectors with the time period index and the scale index as the combined index ; Each feature vector is expressed as: ; Among them, , , , respectively represent the mean value, standard deviation, skewness, and range features of the voltage at the th segment and the th scale. The current, temperature, and switch status data of other dimensions are extracted in the same way; S24. Concatenate all the sets of feature vectors to form a high-dimensional device status feature matrix , where is the number of time windows, is the number of sliding window scales, is the number of single-segment feature dimensions extracted at each scale.

[0025] In this embodiment, S3 includes the following steps: S31. Calculate the fault sensitivity coefficients of each feature component for the safety boundaries of each key parameter during the operation of the substation to form a fault sensitivity coefficient matrix : ; Among them, denotes the eigenvector at the th time period and the th scale, is a preset threshold reflecting the safe operation state of the substation, is the fault risk amplification factor; The calculated fault sensitivity coefficient matrix and the high-dimensional equipment state feature matrix are multiplied element by element to form the fault sensitivity input matrix ; S32. Using the fault sensitivity input matrix to construct a fault sensitivity deep belief network model, the fault sensitivity deep belief network model is represented by a stacked structure of multiple improved restricted Boltzmann machines as: ; Among them, denotes the th layer of improved restricted Boltzmann machine, is the total number of layers of the fault sensitivity deep belief network model. In each layer of the fault sensitivity deep belief network model, the input vector is denoted as . For the first layer, , and the hidden layer is represented as ; The construction process of the improved restricted Boltzmann machine includes: At the structural level, the improved restricted Boltzmann machine retains the basic structure of the traditional restricted Boltzmann machine, that is, a two-layer symmetric network composed of a visible layer and a hidden layer, with a fully bidirectional connection between the two layers and no connection between nodes within the layer; At the input feature processing level, the improved restricted Boltzmann machine introduces a fault sensitivity coefficient regulation mechanism at the visible layer nodes, that is, weights are assigned to each dimension of the input vector, and the weights are derived from the fault sensitivity matrix to amplify the feature signals significantly related to the abnormal state of the equipment and suppress the weak response inputs under background noise or normal fluctuations; At the energy function modeling level, the improved restricted Boltzmann machine adjusts the energy function of the traditional restricted Boltzmann machine and embeds the fault sensitivity coefficient to form a fault-guided energy expression, enabling the model to automatically focus on the data components deviating from the normal operation state during the learning process; At the training mechanism level, the improved restricted Boltzmann machine retains contrastive divergence as the training method, but during the gradient update process, the fault sensitivity coefficient is introduced into the gradient expression of error backpropagation, enabling the parameter adjustment of the model to give priority to responding to the feature dimensions amplified in the potential fault state of the equipment; At the inter-layer transmission mechanism level, the improved restricted Boltzmann machine supports its output implicit vector As the next layer Input ,Ensure that the layer by layer feature expression is multi-level abstracted and aggregated towards the fault related direction; The improved restricted Boltzmann machine introduces a fault-sensitive control mechanism at the input end, embeds feature weighted expression in energy modeling, and guides parameters to focus on optimizing key feature dimensions during training. This enables the entire network to have stronger perception and representation capabilities for fault precursor data in substations, forming a core feature learning unit for remote fault warning tasks.

[0026] S33. Improved restricted Boltzmann machine at each layer Inside, for the input vector and the implicit vector The energy function integrating the fault sensitivity coefficient is used for modeling: ; in, Indicates Tier Input nodes, For the Layer The input node and The connection weights between hidden nodes, and are the input node and hidden node bias of this layer respectively, For the The fault sensitivity coefficients corresponding to the input nodes are taken from the fault sensitive input matrix After corresponding mapping, we get: is the number of neurons in the lth visible layer, is the number of neurons in the lth hidden layer, Represents the hidden layer vector of the lth layer; S34. Using the contrast divergence algorithm, under the premise of ensuring that the fault anomaly characteristics in the multi-source state data of the substation are fully enhanced, each layer of the improved restricted Boltzmann machine Parameters Perform unsupervised pre-training layer by layer and output each hidden layer As the input of the next layer, during the training process, the fault sensitivity coefficient Adjust the weight of each input node in the gradient update so that the fault-sensitive deep belief network model focuses on the features of abnormal deviation from the safety threshold; After completing the unsupervised pre-training of all improved restricted Boltzmann machine layers, perform supervised fine-tuning on the entire fault-sensitive deep belief network model, and use the fault label vector generated from historical fault records , where is the number of fault categories, and use the cross-entropy loss function to jointly optimize the parameters of the entire network to generate a fault feature expression vector ; S36. Transfer the fault feature expression vector to the output layer of the fault-sensitive deep belief network model, and use the fault mode classifier for forward inference to output the preliminary fault status recognition result .

[0027] In this embodiment, S4 includes the following steps: S41. Unify and expand the connection weight parameters and the bias parameters in the trained fault-sensitive deep belief network model into a parameter vector , where represents the jellyfish individual number, is the jellyfish population size, and each jellyfish individual represents a set of potential optimal configurations of the fault-sensitive deep belief network model under the current working conditions of the substation; S42. Perform forward inference on the multi-source state data set using the fault-sensitive deep belief network model corresponding to the jellyfish individual position to obtain the preliminary fault recognition output , and compare the preliminary fault recognition output with the historical label , introduce the key equipment weight coefficient of the substation to form an equipment weighted cross-entropy loss function that combines prediction performance and equipment importance: ; Among them, Q represents the number of equipment participating in the evaluation, represents the operating criticality weight of the qth equipment, c represents the total number of fault types that the model can predict, represents the true ith fault label of the qth equipment in the training sample, represents the prediction probability output of the deep belief network model represented by the jellyfish individual on the qth equipment for the ith fault; Equipment weighted cross-entropy loss function It not only reflects the accuracy of the model in identifying various types of faults, but also enhances the attention to key equipment by running critical weights, ensuring that the optimization process does not ignore the components crucial for system stability in the substation while improving the overall accuracy. The device weighted cross-entropy loss function constructed in this way better meets the requirements of power engineering for safety redundancy, priority response of equipment levels, and fault criticality.

[0028] S43. Initialize the jellyfish population and set the maximum number of iterations , in each round t, combined with the dynamic working condition weight of the current substation operation status , use the ocean current drive update formula to achieve global optimization: ; where, represents the position vector of the s-th jellyfish individual in the t-th round of iteration, represents the updated position vector of the s-th jellyfish individual in the (t + 1)-th round of iteration represents the dynamic working condition weight introduced for the substation operation status in the t-th round of iteration, which is used to adjust the response ability of the global search behavior of the jellyfish group to environmental disturbances at different times, represents the ocean current drift factor, which is used to control the amplitude of the jellyfish individual approaching the current optimal individual and reflects the cluster movement speed of the jellyfish group, represents the position vector of the individual with the optimal fitness in the entire jellyfish group in the t-th round of iteration; S44. Introduce an active foraging behavior update mechanism that combines recognition error and fault level response to accelerate the adjustment speed of the fault-sensitive deep belief network model body with an error greater than the preset value: ; where, is the active search coefficient, is a randomly selected individual, is the -th jellyfish individual's response weight in the current round, which is used to guide the convergence direction to correct the fault misjudgment area; S45. In each round of iteration, dynamically switch between the ocean current drive and the active foraging mechanism according to the change trend of individual fitness and the difference between the global optimal model performance, and record the current optimal model parameter vector ; S46. If the maximum number of iterations is reached or the improvement of the global optimal fitness is less than the convergence threshold in consecutive iterations, then terminate the iteration and take the finally obtained optimal parameter vector Decode into a set of parameters of the fault-sensitive deep belief network model , and form an optimized fault-sensitive deep belief network model .

[0029] In this embodiment, S5 includes the following steps: S51. Utilize the optimized fault-sensitive deep belief network model , and input the preprocessed multi-source state data set collected in real time into the model for forward inference calculation to obtain the device state prediction results at each time point , where represents the predicted probability of the th type of fault occurring at time point ; S52. Sort and determine the maximum value of the predicted probabilities of various types of faults in the prediction results , extract the fault type corresponding to the maximum probability , and use the fault type as the main fault mode of the current device operation state, and record the predicted probability of this main fault mode ; S53. Combine the typical operation conditions of the substation, the equipment risk level, and the historical fault impact data to construct a fault risk grading and classification rule; S54. According to the fault type and the predicted probability , determine the equipment fault risk level at the current time point , and output a real-time fault prediction result label triple including the prediction category, the predicted probability, and the corresponding risk level , where ; S55. Perform time-series fusion analysis on the prediction result label sequences in multiple consecutive time periods to generate a fault trend evolution curve.

[0030] In this embodiment, the fault risk grading and classification rule is divided according to the following fault risk level criteria: Low risk level: The predicted probability , and the corresponding fault type is a slightly fluctuating type, including slight abnormal short-term current, low-amplitude temperature rise fluctuation, and single-point transient interference; Medium risk level: The predicted probability , and the corresponding fault type belongs to the medium warning type, including continuous overheating of the equipment, unplanned switching signal, and long-term voltage deviation from the steady state; High risk level: The predicted probability , and corresponding fault types involve key exception classes, including breaker operation failure, severe current fluctuations, and continuous temperature overlimit.

[0031] Example 1: In June 2024, a 220 kV substation is located in the western industrial zone. The substation is an important hub of the regional power transmission and distribution network, undertaking the power supply dispatching tasks for 8 large enterprise users in the surrounding area and the urban power grid. The substation includes 3 groups of main transformers, 6 groups of circuit breakers, 16 groups of disconnectors, and a large number of current, voltage, and temperature monitoring devices. With the arrival of the peak industrial electricity consumption period in summer, the probability of equipment failure increases significantly due to long-term high-load operation. To ensure the safe and stable operation of the substation, the operation and maintenance unit decides to deploy and conduct actual verification based on the present invention at this station, and conduct a performance comparative analysis with the current traditional static threshold warning system and the intelligent identification method based on the SVM model.

[0032] The project test lasts for 30 days, from June 10, 2024 to July 9, 2024. During the deployment process, synchronous sensing devices are installed at key areas such as the outgoing side of the main transformer, the temperature rise position of the breaker contact, the primary side current bus, and the ambient temperature and humidity monitoring points. The system collects the following four types of real-time status data: Voltage data : The sampling frequency is 10 Hz, and the monitoring range is 0 - 250 kV; Current data : The sampling frequency is 10 Hz, and the monitoring range is 0 - 3000 A; Temperature data : The sampling frequency is 1 Hz, and the monitoring range is -20°C - 150°C; Switch status data : Event-triggered acquisition, recording the operation time, duration, and feedback status code of the circuit breaker or disconnector.

[0033] In the data processing link, first, the original data is preprocessed standardly, including noise filtering, outlier removal, missing value interpolation, and normalization mapping, to generate a cleaned status data set . Subsequently, the data is segmented into 10-minute time periods, and statistical features (mean, standard deviation, skewness, range) of the equipment status are extracted using sliding windows (three scales of 30 seconds, 60 seconds, and 120 seconds), and the corresponding status feature vectors for each segment are calculated.

[0034] By introducing prior experience rules for equipment failures, reference thresholds are set for voltage fluctuations exceeding ±5%, current continuously deviating from the rated value by more than 10%, abnormal temperature rise rate exceeding 5°C / min, and abnormal switch tripping frequency higher than 3 times per hour. A fault sensitivity coefficient matrix is constructed in combination with the degree of sample deviation and embedded into the training structure of a multi-layer deep belief network.

[0035] During the model training stage, the actual labeled operation data occurring from March to May 2024 is collected as the training set, including: Table 1 Training set parameter data

[0036] After initially training the deep belief network using the data in Table 1 above, the jellyfish swarm optimization algorithm is introduced to globally and dynamically optimize the network parameters. The optimization process uses the cross-entropy loss function and the weighted error of the equipment risk weight as evaluation indicators, with 100 rounds of iteration and a swarm size of 30 jellyfish individuals. Finally, a deep belief network model with optimal parameter configuration is obtained. 。

[0037] During the 30-day deployment and operation period, approximately 1.224 million monitoring data were collected in total. Table 2 shows the comparison results of the output performance of the early warning system: Table 2 Comparison results of the performance of the present invention and traditional methods

[0038] From the operation feedback, at 02:47 on the early morning of June 28, 2024, this early warning system successfully identified the unplanned opening abnormality in Area A3 of the circuit breaker for the first time and classified it as a "high-risk" event. The system completed the reasoning within only 0.47 seconds after identifying the abnormal state and automatically sent a remote warning message to the Jining power grid dispatching system. After on-site investigation and confirmation, the fault reason was that the arc discharge triggered the protection switch action, avoiding the equipment burnout accident caused by excessive temperature rise at the contact point. The traditional system did not generate an alarm during this period and only recorded the over-temperature event after the equipment stopped running 1 hour later.

[0039] In another 4 on-site equipment abnormal scenarios, this system completed the risk early warning more than 30 seconds in advance and accurately classified it as the "medium-risk" or "low-risk" level, providing a scientific basis for the duty operation and maintenance personnel to formulate the inspection sequence and dispatching plan.

[0040] In addition, during the operation of the model, the jellyfish swarm optimization algorithm automatically adjusts the weight and bias parameters of the network according to the current sampling data distribution. When the equipment load showed a continuous growth trend from June 15 to June 18, the model achieved faster adaptation and convergence through optimization, effectively avoiding the occurrence of fault identification lag and misjudgment phenomena.

[0041] In summary, the remote fault warning method for substations proposed by the present invention, which combines the jellyfish swarm optimization algorithm and the deep belief network, has the advantages of high precision, low latency, and strong adaptability. It can achieve real-time warning and dynamic risk assessment under complex working conditions in the scenario of multi-source status data, effectively improving the automation and security guarantee capabilities of the substation intelligent operation and maintenance system. Example 1 verifies the feasibility and effectiveness of the present invention in practical engineering applications.

[0042] In the feature learning stage of the present invention, a fault sensitivity coefficient matrix is introduced. By constructing a fault risk amplification factor that is proportional to the deviation degree of operating parameters such as voltage, current, temperature, and switch status, the weight enhancement processing of key features is realized, which is embedded in the energy function of the improved restricted Boltzmann machine. During the unsupervised pre-training process, the deep network structure is guided to focus on the abnormal signal areas with the most potential for faults in the operation of substation equipment. Compared with the traditional deep belief network structure, it performs better in the extraction ability of low-amplitude fluctuations, multi-dimensional data coupling, and time-series mixed faults, and greatly improves the accuracy of fault pattern recognition and the ability to extract precursors under the condition of multi-source information fusion.

[0043] The present invention introduces the jellyfish swarm optimization algorithm into the deep model parameter tuning process. Considering the characteristics of variable working conditions and large differences in operating areas of substations, a dynamic working condition weight function and a behavior selection function based on prediction error response are introduced to construct a hybrid optimization update strategy that combines ocean current driving and active foraging mechanisms. It can dynamically adjust the optimization direction and amplitude according to the real-time state, prediction deviation, and response speed of the equipment, making the model parameters continuously converge to the optimal solution under the current operating state, significantly reducing the overfitting risk, and improving the model stability and generalization ability.

[0044] In the real-time inference stage of the present invention, by combining the output probability of the optimized deep model and the critical degree of equipment operation, a risk classification standard based on equipment importance weighting is constructed to achieve multi-level fault risk classification of low risk, medium risk, and high risk. By serializing the prediction results of multiple consecutive moments, the fault evolution trend and the grade change trend curve are further output, providing a more forward-looking and executable decision-making basis for power operation and maintenance personnel. It can not only improve the response speed of fault handling but also support the dynamic evolution tracking of risks across time periods, effectively supporting the intelligent alarm linkage and hierarchical control of the dispatching system.

[0045] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A substation equipment status monitoring and intelligent fault warning method based on deep learning, characterized in that: The steps include: S1. Real-time acquisition of multi-source state data sets and preprocessing to obtain preprocessed multi-source state data sets; S2. Use the preprocessed multi-source state data set to construct the equipment state feature vector, use the time series data segmentation and multi-scale sliding window method to extract the key features related to the fault evolution, and generate a high-dimensional equipment state feature matrix; S3. Fault-sensitive deep belief network model, which performs multi-level unsupervised feature learning on high-dimensional equipment state feature matrix to form preliminary fault state recognition results; S4. Apply the jellyfish swarm optimization algorithm to perform global search and dynamic tuning on the weight parameters and bias parameters of the deep belief network model. The adjustment process uses the deviation between the preliminary fault state recognition result and the predetermined recognition performance index as feedback information to obtain an optimized sensitive deep belief network model. S5. Use the optimized sensitive deep belief network model to perform online analysis on multi-source status data collected in real time, generate real-time fault prediction results of equipment status, and classify and grade fault risks; S6. Based on the real-time fault prediction results, the remote fault warning mechanism is triggered to send fault warning information including fault risk level, warning signal and emergency handling suggestions to the substation operation and maintenance center.

2. According to a deep learning-based substation equipment status monitoring and intelligent fault early warning method according to claim 1, it is characterized in that: The S1 comprises the following steps: S11. Set the data collection cycle. In each collection cycle, voltage data is collected by using voltage sensors, current sensors, temperature sensors and switch status collection devices deployed at key locations of the substation. , current data , Temperature data And switch status data , construct the original multi-source state dataset ; S12. Remove noise from the original multi-source state data set, use the moving average method to smooth the continuous variables, and compare the current value of each data point with its previous value. The adjacent data points are averaged as the denoising result of the data point to eliminate the influence of short-term spike noise on the data. S13. Perform outlier detection and elimination on the original multi-source state data set after the preliminary smoothing process, determine the degree of deviation of a single point in the overall data distribution, divide the difference between the value of each data point and the mean of its dimension by the standard deviation of the dimension to obtain a standardized value, and when the standardized value is greater than the preset abnormal judgment threshold, it is regarded as an abnormal point and eliminated; S14. Fill missing values ​​in the multi-source state data set after the anomaly is eliminated. For the continuous variables of voltage data, current data, temperature data, and switch state data, the average value of the valid data point before and after the data point is used as the interpolation result to maintain the continuity and change trend of the data sequence in the multi-source state data set; S15. Normalize the multi-source state data set with missing values, subtract the minimum value in each dimension from the data, and divide it by the difference between the maximum and minimum values ​​of the dimension, and map the data of each dimension to the interval , eliminating the dimensional differences between different physical quantities; S16. The multi-source state dataset after noise removal, outlier removal, missing value filling and normalization is defined as the preprocessed multi-source state dataset .

3. According to a method for substation equipment status monitoring and intelligent fault early warning based on deep learning in claim 1, it is characterized in that: The S2 comprises the following steps: S21. Preprocessed multi-source state dataset According to the collection time series Divide the segment into equal intervals and set the segment length to , divide the preprocessed multi-source state data set into several time series data sub-segments ,in, , is the total time span of the preprocessed dataset, Indicates Data subsegments within a time window; S22. For each time series data segment Construct a multi-scale feature extraction structure based on sliding windows, set the sliding window length set, and for each window length Perform sliding traversal operations to extract statistical features of voltage data, current data, temperature data, and switch status data in each sliding window, including the mean , Standard Deviation , slope and range ; S23. For all time series data sub-segments and window length The extracted statistical features are integrated to generate a time period index and scale index A set of feature vectors for combined indexes ; S24. Collect all feature vectors Splicing to form a high-dimensional device state feature matrix ,in is the number of time windows, is the sliding window scale, The number of single-segment feature dimensions extracted at each scale.

4. According to a method for substation equipment status monitoring and intelligent fault early warning based on deep learning according to claim 1, it is characterized in that: The S3 comprises the following steps: S31. According to the safety boundary of each key parameter in the operation process of the substation, the fault sensitivity coefficient of each characteristic component is calculated to form a fault sensitivity coefficient matrix : ； in, express Middle Period, The feature vector at scale, To reflect the preset threshold value of the safe operation status of the substation, is the failure risk amplification factor; Calculated fault sensitivity coefficient matrix and high-dimensional equipment state characteristic matrix Multiply element by element to form a fault-sensitive input matrix ; S32. Using the Fault Sensitive Input Matrix Building a Fault-Sensitive Deep Belief Network Model , the fault-sensitive deep belief network model adopts a multi-layer improved restricted Boltzmann machine stacking structure and is expressed as: ； in, Indicates Layer-modified restricted Boltzmann machine, is the total number of layers of the fault-sensitive deep belief network model. In each layer of the fault-sensitive deep belief network model, the input vector is recorded as , for layer 1, , the hidden layer is represented as ; S33. Improved restricted Boltzmann machine at each layer Inside, for the input vector and the implicit vector The energy function integrating the fault sensitivity coefficient is used for modeling: ； in, Indicates Tier Input nodes, For the Layer The input node and The connection weights between hidden nodes, and are the input node and hidden node bias of this layer respectively, For the The fault sensitivity coefficients corresponding to the input nodes are taken from the fault sensitive input matrix After corresponding mapping, we get: is the number of neurons in the lth visible layer, is the number of neurons in the lth hidden layer, Represents the hidden layer vector of the lth layer; S34. Using the contrast divergence algorithm, under the premise of ensuring that the fault anomaly characteristics in the multi-source state data of the substation are fully enhanced, each layer of the improved restricted Boltzmann machine Parameters Perform unsupervised pre-training layer by layer and output each hidden layer As the input of the next layer, during the training process, the fault sensitivity coefficient Adjust the weight of each input node in the gradient update so that the fault-sensitive deep belief network model focuses on the features of abnormal deviation from the safety threshold; S35. After completing the unsupervised pre-training of all modified restricted Boltzmann machine layers, supervised fine-tuning of the entire fault-sensitive deep belief network model is performed, using the fault label vector generated by historical fault records ,in, is the number of fault categories, and the cross entropy loss function is used to jointly optimize the parameters of the entire network to generate a fault feature expression vector ; S36. Express the fault feature vector Passed to the output layer of the fault-sensitive deep belief network model, the fault mode classifier is used for forward reasoning to output the preliminary fault state recognition result .

5. A substation equipment status monitoring and intelligent fault early warning method based on deep learning according to claim 4, characterized in that: The S4 comprises the following steps: S41. Set the connection weight parameters of each layer in the trained fault-sensitive deep belief network model With bias parameters , uniformly expanded into parameter vectors ,in, Indicates the jellyfish individual number, is the size of the jellyfish group, each jellyfish individual represents a set of potential optimal configurations of the fault-sensitive deep belief network model under the current working conditions of the substation; S42. Perform forward reasoning on the multi-source state data set using the fault-sensitive deep belief network model corresponding to the individual jellyfish positions to obtain preliminary fault identification output , and compare the initial fault identification output with the historical label Comparison, introducing the weight coefficient of key equipment in substation Forming a device-weighted cross entropy loss function that combines prediction performance and device importance : ； Among them, Q represents the number of devices participating in the evaluation, represents the critical weight of the operation of the qth device, c represents the total number of fault types that can be predicted by the model, represents the true i-th fault label of the q-th device in the training sample, Indicates individual jellyfish The predicted probability output of the deep belief network model represented by the representation for the i-th type of fault on the q-th device; S43. Initialize the jellyfish population And set the maximum number of iterations , in each round t, combined with the dynamic operating weight of the current substation operating status , use the ocean current driven update formula to achieve global optimization: ； in, represents the position vector of the sth jellyfish individual in the tth iteration, Represents the updated position vector of the sth jellyfish individual in the t+1th iteration It represents the dynamic operating condition weight introduced for the substation operation status in the tth iteration, which is used to adjust the responsiveness of the jellyfish swarm global search behavior to environmental disturbances at different times. Represents the ocean current drift factor, which is used to control the extent to which the jellyfish individual approaches the current optimal individual and reflects the cluster movement speed of the jellyfish group. represents the position vector of the individual with the best fitness in the entire jellyfish population in the tth iteration; S44. Introduce an active foraging behavior update mechanism that combines recognition error and fault level response to speed up the adjustment of fault-sensitive deep belief network models whose errors are greater than a preset value: ； in, is the active search coefficient, is a randomly selected individual, For the The response weight of each jellyfish in the current round is used to guide the convergence direction to correct the fault misjudgment area; S45. In each iteration, the current-driven and active foraging mechanisms are dynamically switched according to the individual fitness change trend and the performance difference of the global optimal model, and the current optimal model parameter vector is recorded. ; S46. If the maximum number of iterations is reached or continuous The global optimal fitness improvement in the iteration is less than the convergence threshold , the iteration is terminated and the optimal parameter vector finally obtained is Decoded into a set of fault-sensitive deep belief network model parameters , forming an optimized fault-sensitive deep belief network model .

6. A substation equipment status monitoring and intelligent fault early warning method based on deep learning according to claim 5, characterized in that: The S5 comprises the following steps: S51. Using the optimized fault-sensitive deep belief network model , the pre-processed multi-source state data set collected in real time Input into the model, perform forward reasoning calculations, and obtain each time point The equipment status prediction results under ; S52. Prediction results The predicted probabilities of various types of faults are sorted and the maximum value is determined, and the fault type corresponding to the maximum probability is extracted , the fault type As the main failure mode of the current equipment operation status, and record the predicted probability of the main failure mode ; S53. Combine the typical operating conditions of the substation, the risk level of the equipment and the historical fault impact data to build a fault risk classification rule; S54. According to the fault type With predicted probability , determine the current time point The following equipment failure risk level.

7. A method for substation equipment status monitoring and intelligent fault early warning based on deep learning according to claim 6, characterized in that: The fault risk classification rules are constructed according to the following fault risk level classification standards: Low risk level: predicted probability , and the corresponding fault type It is a slight fluctuation type, including short-term current slight abnormality, low-amplitude temperature rise fluctuation, and single-point transient interference; Medium risk level: predicted probability , and the corresponding fault type Belongs to the medium warning category, including continuous over-temperature of equipment, unplanned switching signals, and long-term deviation of voltage from steady state; High risk level: predicted probability , and the corresponding fault type Key anomalies involved include failure of circuit breaker operation, sharp current fluctuations, and continuous temperature exceeding the limit.

8. A method for substation equipment status monitoring and intelligent fault early warning based on deep learning according to claim 6, characterized in that: S5 also includes outputting a real-time fault prediction result label triple including a prediction category, a prediction probability and a corresponding equipment failure risk level. ,in, .

Images