Intelligent substation auxiliary control monitoring system based on gateway machine

Abstract

The invention discloses a substation intelligent auxiliary control monitoring system based on a gateway machine. The substation intelligent auxiliary control monitoring system comprises a data acquisition module, a protocol adaptive adjustment module, an anomaly detection module, a fault early warning module and an intelligent operation control module. The invention relates to the technical field of transformer substation intelligent management, in particular to a transformer substation intelligent auxiliary control monitoring system based on a gateway machine. According to the scheme, protocol features are automatically extracted and predicted through self-supervised comparative learning and a second-order Markov chain, and label-free rapid switching and high-reliability communication are achieved in combination with confidence and delay decision; fusing a sliding window transfer entropy and a kernel PCA reconstruction residual error, performing unsupervised detection and positioning electrical, network and environment anomalies; by means of multi-scale Morlet wavelet entropy and DBSCAN rare cluster recognition, unknown fault precursor spontaneous early warning is achieved; based on real-time apparent power autoregression prediction and mixed integer convex optimization, peak suppression, load smoothing and temperature constraint are considered, and a scheduling strategy is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent management of substations, and in particular to an intelligent auxiliary control and monitoring system for substations based on a gateway machine. Background Art

[0002] As power systems evolve toward digitalization and intelligence, substation automation and monitoring technologies have become critical to ensuring the safe and stable operation of power grids. With the continuous improvement of substation networking, various intelligent terminals and communication protocols coexist between on-site equipment and monitoring centers. Traditional approaches that rely on manual pre-configuration or a single protocol stack are no longer able to adapt to the demands of mixed operation using multiple vendors and heterogeneous protocols. Communication handover delays and misjudgments are frequent, significantly increasing operational and maintenance costs and risks.

[0003] At the same time, existing anomaly detection is often based on simple thresholds or single statistical models, which cannot effectively integrate the complex coupling relationship between electrical conditions, network traffic and environmental parameters. It lacks the ability to identify unknown fault precursors and weak abnormal signals, and is prone to missed reports and false alarms. In addition, with the intensified fluctuations in substation load curves and frequent equipment switching operations, traditional empirical scheduling is difficult to take into account peak suppression, load smoothing and equipment life management, and lacks the ability to support real-time prediction and optimization decisions.

[0004] Therefore, it is necessary to build an intelligent auxiliary control and monitoring system at the gateway level to achieve deep integration and collaborative optimization of the communication layer, monitoring layer and operation control layer through adaptive adjustment of label-free protocols, joint multi-source unsupervised anomaly detection and precise early warning, and closed-loop optimization control based on short-term load forecasting. Summary of the Invention

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a substation intelligent auxiliary control monitoring system based on a gateway machine. In view of the problems that traditional protocol identification relies on manual configuration and labeled training, protocol switching is delayed and prone to misjudgment, this solution uses self-supervised comparative learning to deeply extract protocol features, second-order Markov chain to accurately predict protocol evolution, and combines prediction confidence and communication delay for joint decision-making to achieve label-free automatic identification of new protocols; in view of the problems that traditional anomaly detection mostly relies on a single statistical threshold or reconstructed residual method, it is difficult to take into account the causal coupling relationship between signals and the mutation characteristics of each dimension, the positioning accuracy is low and it is easy to generate missed reports and false alarms, this solution introduces a dynamic causal network based on transfer entropy in the sliding window to construct a causal drift score, and combines the kernel principal component analysis to reconstruct the residual to extract the mutation signal score, normalize it according to the historical mean and then perform weighted fusion, and finally realize the joint anomaly detection and precise positioning of electrical working conditions, network traffic and environmental monitoring signals under unsupervised conditions. ; To address the problem that traditional fault warning systems are difficult to adaptively capture the multi-scale precursor patterns of complex signals and are prone to ignoring rare fault signals, this solution extracts the energy distribution of signals of each dimension at different wavelet scales and calculates the wavelet entropy to construct a multi-scale precursor feature vector, combines DBSCAN density clustering to automatically identify rare cluster patterns, and triggers an early warning when rare clusters are maintained in multiple consecutive windows. At the same time, the fault dimension is located based on the median entropy value, realizing unsupervised spontaneous detection and accurate early warning of unknown fault precursors; To address the problem that traditional operation control is difficult to balance peak suppression, load smoothing and equipment switch life and lacks dynamic response to environmental constraints, this solution obtains future load trends through autoregressive load forecasting based on real-time apparent power, and uses peak load minimization, overall load smoothing and switch action penalty as optimization criteria in a mixed integer convex optimization framework, combined with ambient temperature constraints to solve the optimal switching strategy, realizing refined scheduling of load curves and intelligent control of equipment operation.

[0006] The technical solution adopted by the present invention is as follows: The present invention provides a gateway-based substation intelligent auxiliary control and monitoring system, which includes a data acquisition module, a protocol adaptive adjustment module, an anomaly detection module, a fault warning module and an intelligent operation control module;

[0007] The data acquisition module collects communication flow data, substation electrical data, and substation environmental data;

[0008] The protocol adaptive adjustment module first trains the encoder on the frame payload subsequence through self-supervised contrastive learning to learn the deep representation of the protocol. Then, based on a second-order Markov chain, it updates the transition matrix using the previous two protocol states to predict the protocol type at the next moment. Finally, it combines the prediction confidence threshold and real-time communication delay to make a joint decision and dynamically select and switch to the optimal protocol.

[0009] The anomaly detection module first calculates the transfer entropy between each dimension pair of the time series signal in each sliding window, constructs a dynamic causal network, and obtains the inter-window difference matrix. Then, based on kernel principal component analysis, it projects and reverse maps the multidimensional data in the same window to obtain the reconstructed residual and extract the maximum residual score. Finally, the causal drift score and the residual score are fused according to the weight, and the window is determined to be abnormal based on the threshold. The abnormal dimension is located by the maximum difference or the maximum residual position.

[0010] The fault warning module first calculates the energy distribution of each dimensional signal on a multi-scale Morlet wavelet and normalizes it to probability, then obtains the wavelet entropy of each scale to form a precursor feature vector. It then applies DBSCAN clustering to the historical window feature sequence to identify rare clusters with low occurrence frequency. When D consecutive windows fall into the same rare cluster, a warning is triggered, and the fault dimension is determined by comparing it with the median of the historical entropy.

[0011] The intelligent operation control module constructs an autoregressive model based on real-time apparent power to predict future load curves. It then solves the optimal device switching strategy in a mixed integer convex optimization framework that considers peak suppression, load smoothing, switching action penalties, and ambient temperature constraints. Finally, it issues control commands through the gateway and executes them in real time.

[0012] Furthermore, the protocol adaptive adjustment module specifically includes the following units:

[0013] The protocol feature extraction unit uses subsequence comparison of frame payloads for self-supervised learning, separates the deep representation of the protocol, constructs frame sequence slices, and trains the encoder to automatically learn protocol features;

[0014] The Markov chain protocol prediction unit introduces the second-order Markov assumption and combines the weighted prediction of the next protocol based on the previous two protocol types. It also uses online updates to quickly incorporate new protocols into the transfer matrix.

[0015] The protocol switching decision unit adopts a joint decision-making method based on prediction confidence and communication delay.

[0016] Furthermore, the anomaly detection module targets the time series signals from the data acquisition module; introduces a dynamic causal network based on transfer entropy and kernel principal component analysis reconstruction residual fusion detection strategy to achieve joint unsupervised anomaly location of electrical conditions, network traffic and environmental monitoring signals. Specifically, it includes the following units:

[0017] The transfer entropy network constructs a unit that analyzes a multi-dimensional signal set in fixed-length time periods. For any pair of signal dimensions, a sliding window transfer entropy is defined. All transfer entropies are then combined into a transfer entropy matrix. The difference amplitude from the previous window is calculated. Finally, a causal drift score is defined as an anomaly indicator.

[0018] The kernel principal component reconstructs the residual unit, which specifically includes the following:

[0019] Kernel principal component projection: constructs a kernel matrix based on the multi-dimensional data in the window and extracts the principal components, and obtains the reconstructed signal through inverse mapping;

[0020] Residual calculation, defining the reconstruction residual for each dimension;

[0021] Set the residual score, and set the maximum residual in the window as the residual score;

[0022] The fusion decision and positioning units first introduce fusion weights and calculate the comprehensive score;

[0023] Abnormal judgment: When the comprehensive score of a window is greater than or equal to the abnormal judgment threshold, the window is marked as abnormal and the abnormal signal is located.

[0024] Furthermore, the fault warning module uses a method combining multi-time-scale wavelet entropy features with density clustering to utilize the data's own structure to spontaneously identify abnormal precursors in the time-frequency domain and issue warnings. Specifically, it includes the following units:

[0025] The wavelet entropy feature extraction unit specifically includes the following contents:

[0026] Continuous wavelet transformation, calculating the wavelet coefficients on the scale set for each dimensional signal;

[0027] Calculate energy distribution;

[0028] To generate fault precursor features, first normalize the energy distribution to a probability distribution; then calculate the wavelet entropy; and finally concatenate all the wavelet entropies to form the fault precursor features.

[0029] Density cluster warning unit, specifically including the following contents:

[0030] DBSCAN clustering, in the feature space with parameters 、 Run, where represents the neighborhood radius, Indicates the minimum number of points in the cluster within the neighborhood radius; obtain the label set;

[0031] Rare cluster determination, determining rare clusters based on their frequency of occurrence;

[0032] Fault warning, when D consecutive windows fall into the same rare cluster, a warning is triggered, and the fault dimension is determined by comparing it with the median of historical entropy.

[0033] Furthermore, the intelligent operation control module specifically includes the following units:

[0034] The load forecasting unit constructs a B-order autoregressive model based on the power time series output by the real-time power calculation unit to predict the load in the future period;

[0035] The control strategy optimization unit solves a mixed integer convex optimization problem based on the predicted load curve, the equipment switching status and ambient temperature provided by the data acquisition module, while reducing the peak load, smoothing the load curve and minimizing the number of equipment switching operations;

[0036] The execution unit sends control commands to the primary devices through the gateway machine according to the device switching strategy output by the optimization unit, and executes them in real time.

[0037] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0038] (1) To address the problems of traditional protocol recognition relying on manual configuration and labeled training, protocol switching lags and is prone to misjudgment, this solution uses self-supervised comparative learning to deeply extract protocol features, a second-order Markov chain to accurately predict protocol evolution, and a combination of prediction confidence and communication delay for joint decision-making, thereby achieving label-free automatic recognition of new protocols, fast switching, and low-latency reliable communication.

[0039] (2) Traditional anomaly detection relies on a single statistical threshold or reconstructed residual method, which makes it difficult to take into account the causal coupling relationship between signals and the mutation characteristics of each dimension, and has low positioning accuracy and is prone to missed reports and false alarms. This scheme introduces a dynamic causal network based on transfer entropy in the sliding window to construct a causal drift score, and combines the kernel principal component analysis to reconstruct the residual to extract the mutation signal score, which is normalized by the historical mean and then weighted fused. Finally, it realizes the joint anomaly detection and precise positioning of electrical working conditions, network traffic and environmental monitoring signals under unsupervised conditions, effectively improving the detection sensitivity and positioning accuracy.

[0040] (3) To address the problem that traditional fault warning systems have difficulty in adaptively capturing the multi-time-scale precursor patterns of complex signals and are prone to ignoring rare fault signals, this scheme extracts the energy distribution of signals of each dimension at different wavelet scales and calculates the wavelet entropy to construct a multi-time-scale precursor feature vector. This scheme automatically identifies rare cluster patterns by combining DBSCAN density clustering, and triggers an early warning when rare clusters are maintained in multiple consecutive windows. At the same time, it locates the fault dimension based on the median entropy value, thus achieving unsupervised spontaneous detection and accurate early warning of unknown fault precursors.

[0041] (4) In view of the fact that traditional operation control is difficult to balance peak suppression, load smoothing and equipment switch life and lacks dynamic response to environmental constraints, this scheme obtains future load trends through autoregressive load forecasting based on real-time apparent power, and uses peak load minimization, overall load smoothing and switch action penalty as optimization criteria under the mixed integer convex optimization framework, combined with ambient temperature constraints to solve the optimal switching strategy, thereby achieving refined scheduling of load curves and intelligent control of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of a gateway-based substation intelligent auxiliary control and monitoring system provided by the present invention;

[0043] Figure 2 Schematic diagram of the protocol adaptive adjustment module;

[0044] Figure 3 Schematic diagram of the anomaly detection module;

[0045] Figure 4 Schematic diagram of the fault warning module;

[0046] Figure 5 Schematic diagram of the intelligent operation control module.

[0047] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0050] Example 1, see Figure 1The present invention provides a gateway-based intelligent auxiliary control and monitoring system for substations, comprising a data acquisition module, a protocol adaptive adjustment module, an anomaly detection module, a fault warning module, and an intelligent operation control module;

[0051] The data acquisition module collects communication flow data, substation electrical data, and substation environmental data; and sends the data to the protocol adaptive adjustment module, the anomaly detection module, the fault warning module, and the intelligent operation control module;

[0052] The protocol adaptive adjustment module receives data sent by the data acquisition module and trains the encoder on the frame payload subsequence through self-supervised comparative learning to learn the deep representation of the protocol. It then uses the previous two protocol states to update the transition matrix based on a second-order Markov chain to predict the protocol type at the next moment. Finally, it combines the prediction confidence threshold and real-time communication delay to make a joint decision to dynamically select and switch to the optimal protocol.

[0053] The anomaly detection module receives data sent by the data acquisition module, calculates the transfer entropy between each dimension pair of the time series signal in each sliding window, constructs a dynamic causal network and obtains the inter-window difference matrix; then, based on kernel principal component analysis, projects and reverse maps the multidimensional data in the same window to obtain the reconstructed residual and extract the maximum residual score; finally, the causal drift score and the residual score are fused according to the weight, and the window is determined to be abnormal based on the threshold. The abnormal dimension is located by the maximum difference or the maximum residual position, and the data is sent to the fault warning module;

[0054] The fault warning module receives data sent by the data acquisition module and the anomaly detection module, calculates the energy distribution of each dimensional signal on a multi-scale Morlet wavelet and normalizes it into probability, thereby obtaining the wavelet entropy of each scale to form a precursor feature vector; then, DBSCAN clustering is applied to the historical window feature sequence to identify rare clusters with low occurrence frequency; when D consecutive windows fall into the same rare cluster, a warning is triggered, and the fault dimension is determined by comparing it with the median of the historical entropy;

[0055] The intelligent operation control module receives data sent by the data acquisition module, constructs an autoregressive model based on real-time apparent power to predict future load curves, then solves the optimal device switching strategy in a mixed integer convex optimization framework that considers peak suppression, load smoothing, switching action penalties, and ambient temperature constraints. Finally, control commands are issued through the gateway machine and executed in real time.

[0056] Example 2, see Figure 1This embodiment is based on the above embodiment. The data acquisition module collects communication flow data, substation electrical data, and substation environmental data. The communication flow data is specifically a set of original protocol frames on the substation local Ethernet communication bus captured from the gateway Ethernet interface. Each frame contains a complete Ethernet frame header, network layer protocol header, frame payload and packet capture timestamp; the substation electrical data specifically collects electrical quantities of the substation primary equipment, including: bus voltage , line current , device switch status The substation environmental data includes: the substation main control room temperature obtained by the temperature and humidity detector , ambient relative humidity obtained by temperature and humidity probe , the transformer cabinet vibration intensity collected by the acceleration sensor .

[0057] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the protocol adaptive adjustment module specifically includes the following units:

[0058] The protocol feature extraction unit uses the subsequence comparison of the frame payload to perform self-supervised learning, separate the deep representation of the protocol, and construct the frame sequence slice pair. For example, is a negative example, and the encoder is trained to automatically learn the protocol features, which are expressed as follows:

[0059] ;

[0060] in, Represents the deep features of the protocol, i and j represent the index of the protocol frame, represents the payload subsequence of the i-th frame, Represents A sequence of positive examples of the same protocol, Represents Negative example sequences for different protocols, represents an exponential function with a natural constant as the base, Represents a parameter set The feature encoder below, represents the cosine similarity, Represents the temperature hyperparameter, with a value range of ; N and M represent the number of samples of positive and negative examples respectively;

[0061] The Markov chain protocol prediction unit introduces the second-order Markov assumption and combines the weighted prediction of the next protocol based on the previous two protocol types. It also uses online updates to quickly incorporate new protocols into the transition matrix, which is expressed as follows:

[0062] ;

[0063] Where k represents the index of the protocol frame, It indicates the probability that the protocol type at the next moment is exactly k, given that the protocol type at the current moment t is i and the protocol type at the previous moment t-1 is j. Indicates the protocol type index at time t, represents the state transition count matrix entry, which indicates the number of consecutive three steps from protocol j to protocol i and then to protocol k. Indicates the online update rate, Represents an indicator function, which is 1 when the event occurs and 0 otherwise;

[0064] The protocol switching decision unit adopts the joint decision-making of prediction confidence and communication delay, which is expressed as follows:

[0065] ;

[0066] in, Indicates the predicted optimal protocol type for the next moment, Indicates the state switching condition, Indicates the preset confidence threshold, represents the communication time delay calculated from the timestamp, represents the maximum acceptable switching delay, Represents the logical AND symbol.

[0067] By performing the above operations, this solution addresses the problems of traditional protocol identification relying on manual configuration and labeled training, and protocol switching being delayed and prone to misjudgment. It uses self-supervised comparative learning to deeply extract protocol features, a second-order Markov chain to accurately predict protocol evolution, and a combination of prediction confidence and communication latency for joint decision-making, achieving label-free automatic identification of new protocols, fast switching, and low-latency reliable communication.

[0068] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The anomaly detection module is used to detect the time series signal from the data acquisition module. , which includes six dimensions; innovatively introduces a dynamic causal network based on transfer entropy and kernel principal component analysis reconstruction residual fusion detection strategy to achieve joint unsupervised anomaly location of electrical conditions, network traffic and environmental monitoring signals, specifically including the following units:

[0069] Transfer entropy network building unit, multi-dimensional signal set With fixed length Time period Analyze for the unit, where Indicates the length of the sliding window, u indicates the window number, Indicates the end time of the u-th window; for any signal dimension pair , define the sliding window transfer entropy from p to q, expressed as follows:

[0070] ;

[0071] in, and Represents the dimension index, M represents the total number of dimensions, Represents the transfer entropy from dimension p to dimension q on the u-th window; represents the discretized value of dimension q at time t+1, represents the historical embedding vector of dimension q at time t and the past k-1 steps, k represents the embedding dimension, represents the historical embedding vector of dimension p at time t and the past k-1 steps, represents the ternary joint probability, specifically the empirical frequency of all corresponding discrete samples in the u-th window, and represents the conditional probability; then all the transfer entropies are combined into the transfer entropy matrix ; and calculate the difference amplitude with the previous window Finally, we define the causal drift score as an abnormality indicator, which is expressed as follows:

[0072] ;

[0073] in, Indicates that the u-th window ends at the time The average causal drift score of Self-loop, Indicates the window differential amplitude The element at row p and column q in ;

[0074] The kernel principal component reconstructs the residual unit, which specifically includes the following:

[0075] Kernel principal component projection, based on multi-dimensional data within the window Construct the kernel matrix and extract the first r principal components, and obtain the reconstructed signal through inverse mapping ;

[0076] Residual calculation, for each dimension, defines the reconstructed residual at time t, which is expressed as follows:

[0077] ;

[0078] in, represents the reconstructed residual of dimension m at time t, and Represent the signal values ​​of the original and reconstructed time series signals in the mth dimension respectively;

[0079] Set the residual score and set the maximum residual in the window as the residual score, which is expressed as follows:

[0080] ;

[0081] in, represents the reconstructed residual score at the final moment;

[0082] The fusion decision and positioning unit first introduces the fusion weight, calculates the comprehensive score, and then performs anomaly determination, which specifically includes the following:

[0083] Calculate the composite score, expressed as follows:

[0084] ;

[0085] in, represents the fusion weight, and Represent the historical score means of drift score and residual score respectively, Indicates the comprehensive score at the final moment;

[0086] Abnormal judgment, when When the mark window Abnormal, among which Indicates the abnormality judgment threshold, which is specifically set to the 95% quantile of the normal period; and locates it. If Maximum, then dimension To q is considered as the causal coupling instability position; if and , then the dimension The intrinsic signal is considered as an abnormal signal.

[0087] By performing the above operations, traditional anomaly detection relies heavily on a single statistical threshold or reconstructed residual method, which makes it difficult to take into account the causal coupling relationship between signals and the mutation characteristics of each dimension, and has low positioning accuracy and is prone to missed and false alarms. This scheme introduces a dynamic causal network based on transfer entropy in a sliding window to construct a causal drift score, and combines the kernel principal component analysis to reconstruct the residual to extract the mutation signal score, which is normalized by the historical mean and then weightedly fused. Finally, it realizes the joint anomaly detection and precise positioning of electrical working conditions, network traffic and environmental monitoring signals under unsupervised conditions, effectively improving the detection sensitivity and positioning accuracy.

[0088] Example 5, see Figure 1 and Figure 4This embodiment is based on the above embodiment. The fault warning module adopts a method combining multi-time-scale wavelet entropy features with density clustering, and uses the data's own structure to spontaneously identify abnormal precursors in the time-frequency domain and issue warnings. It specifically includes the following units:

[0089] The wavelet entropy feature extraction unit specifically includes the following contents:

[0090] Continuous wavelet transformation, for each dimension signal Computational scale set Wavelet coefficients on :

[0091] ;

[0092] in, represents the wavelet scale set, represents the conjugate Morlet wavelet function, Represents the scale on dimension m Continuous wavelet coefficients of ;

[0093] Calculate the energy distribution, within the window u, and calculate the scale The energy is expressed as follows:

[0094] ;

[0095] in, Indicates the signal of the mth dimension in window u, scale Energy on

[0096] Generate fault precursor features, first distribute the energy Normalized to a probability distribution ; Then calculate the wavelet entropy ; Finally, all wavelet entropies are spliced ​​into fault precursor features ;

[0097] Density cluster warning unit, specifically including the following contents:

[0098] DBSCAN clustering, in the feature space with parameters 、 Run, where Represents the neighborhood radius, and its value range is , Indicates the minimum number of points in the cluster within the neighborhood radius, and its value range is ; Get the label set, which is represented as follows:

[0099] ;

[0100] in, represents the cluster label set of window u, v represents the index of the window, represents the fault precursor feature of window v, and label 0 represents the noise label;

[0101] Rare cluster determination, calculate the frequency of occurrence of clusters cx with labels greater than or equal to 1 ;if , then cluster cx is determined to be a rare cluster, where Indicates the cluster frequency threshold, the value range is ;

[0102] Fault warning, if the current window label , Represents a rare cluster set; and this label is uniformly maintained in D consecutive windows, and the value range of D is , triggering a fault warning and indicating that the fault dimension is: ,in, represents the wavelet entropy set of dimension m on all windows, Represents the median of the historical wavelet entropy of dimension m.

[0103] By performing the above operations, in order to solve the problem that traditional fault warning systems have difficulty in adaptively capturing the multi-time-scale precursor patterns of complex signals and tend to ignore rare fault signals, this solution extracts the energy distribution of signals of each dimension at different wavelet scales and calculates the wavelet entropy to construct a multi-time-scale precursor feature vector. Combined with DBSCAN density clustering, this solution automatically identifies rare cluster patterns and triggers an early warning when rare clusters are maintained in multiple consecutive windows. At the same time, it locates the fault dimension based on the median entropy value, thus achieving unsupervised spontaneous detection and accurate early warning of unknown fault precursors.

[0104] Example 6, see Figure 1 and Figure 5 This embodiment is based on the above embodiment, and the intelligent operation control module specifically includes the following units:

[0105] Load forecasting unit, based on the power time series output by the real-time power calculation unit , construct a B-order autoregressive model to predict the load in the future period , which is expressed as follows:

[0106] ;

[0107] Where d represents the index of the prediction moment, represents the apparent power at time t, Indicates the maximum order of the autoregressive model, b indicates the index of the model order, represents the b-th order autoregressive coefficient, obtained by fitting the historical power series with the least squares method. represents the time lag parameter, represents the zero-mean white noise error term;

[0108] Control strategy optimization unit, based on the predicted load curve , o represents the optimization time index, and the device switch status given by the data acquisition module and ambient temperature , solve the following mixed integer convex optimization problem while reducing peak load, smoothing the load curve, and minimizing the number of equipment switching operations:

[0109] ;

[0110] in, represents the objective function, represents the set of discrete moments in the future optimization period, Indicates the total number of device switches, g indicates the index of the device switch, It represents the optimal control strategy for the switch of the g-th device at time o. When it is 1, it means the switch is closed, and when it is 0, it means the switch is open. Indicates the switch status collected by the data acquisition module. Represents the weight coefficient of the balance between peak suppression and total energy consumption, represents the penalty weight of the control action, Indicates the temperature of the main control room collected by the data acquisition module. Indicates the maximum temperature threshold allowed by the environment;

[0111] Execution unit, based on the device switching strategy output by the optimization unit , control commands are sent to primary devices through the gateway machine and executed in real time.

[0112] By performing the above operations, this solution addresses the problem that traditional operation control is difficult to balance peak suppression, load smoothing, and equipment switch life, and lacks dynamic response to environmental constraints. It obtains future load trends through autoregressive load forecasting based on real-time apparent power, and uses peak load minimization, overall load smoothing, and switch action penalty as optimization criteria within a mixed integer convex optimization framework. Combined with ambient temperature constraints, it solves the optimal switching strategy, achieving refined scheduling of load curves and intelligent control of equipment operation.

[0113] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0114] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0115] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A gateway-based intelligent auxiliary control and monitoring system for substations, characterized by: It includes data acquisition module, protocol adaptive adjustment module, anomaly detection module, fault warning module and intelligent operation control module; The data acquisition module collects communication flow data, substation electrical data, and substation environmental data; The protocol adaptive adjustment module first trains the encoder on the frame payload subsequence through self-supervised contrastive learning to learn the deep representation of the protocol. Then, based on a second-order Markov chain, it updates the transition matrix using the previous two protocol states to predict the protocol type at the next moment. Finally, it combines the prediction confidence threshold and real-time communication delay to make a joint decision and dynamically select and switch to the optimal protocol. The anomaly detection module first calculates the transfer entropy between each dimension pair of the time series signal in each sliding window, constructs a dynamic causal network and obtains the inter-window difference matrix; Then, based on kernel principal component analysis, the multidimensional data within the same window is projected and reversely mapped to obtain the reconstructed residual and extract the maximum residual score. Finally, the causal drift score and the residual score are fused according to the weights, and the window is judged to be abnormal based on the threshold. The abnormal dimension is located by the maximum difference or the maximum residual position. The fault warning module first calculates the energy distribution of each dimensional signal on a multi-scale Morlet wavelet and normalizes it to probability, then obtains the wavelet entropy of each scale to form a precursor feature vector. It then applies DBSCAN clustering to the historical window feature sequence to identify rare clusters with low occurrence frequency. When D consecutive windows fall into the same rare cluster, a warning is triggered, and the fault dimension is determined by comparing it with the median of the historical entropy. The intelligent operation control module constructs an autoregressive model based on real-time apparent power to predict future load curves. It then solves the optimal device switching strategy in a mixed integer convex optimization framework that considers peak suppression, load smoothing, switching action penalties, and ambient temperature constraints. Finally, it issues control commands through the gateway and executes them in real time.

2. The gateway-based intelligent auxiliary control and monitoring system for substations according to claim 1 is characterized by: The protocol adaptive adjustment module specifically includes the following units: The protocol feature extraction unit uses subsequence comparison of frame payloads for self-supervised learning, separates the deep representation of the protocol, constructs frame sequence slices, and trains the encoder to automatically learn protocol features; The Markov chain protocol prediction unit introduces the second-order Markov assumption and combines the weighted prediction of the next protocol based on the previous two protocol types. It also uses online updates to quickly incorporate new protocols into the transfer matrix. The protocol switching decision unit adopts a joint decision-making method based on prediction confidence and communication delay.

3. The gateway-based intelligent auxiliary control and monitoring system for substations according to claim 1 is characterized by: The anomaly detection module targets the time series signals from the data acquisition module; introduces a dynamic causal network based on transfer entropy and kernel principal component analysis reconstruction residual fusion detection strategy to perform joint unsupervised anomaly location on electrical conditions, network traffic, and environmental monitoring signals. Specifically, it includes the following units: The transfer entropy network constructs a unit that analyzes a multi-dimensional signal set in fixed-length time periods. For any pair of signal dimensions, a sliding window transfer entropy is defined. All transfer entropies are then combined into a transfer entropy matrix. The difference amplitude from the previous window is calculated. Finally, a causal drift score is defined as an anomaly indicator. The kernel principal component reconstructs the residual unit, which specifically includes the following: Kernel principal component projection: constructs a kernel matrix based on the multi-dimensional data in the window and extracts the principal components, and obtains the reconstructed signal through inverse mapping; Residual calculation, defining the reconstruction residual for each dimension; Set the residual score, and set the maximum residual in the window as the residual score; The fusion decision and positioning units first introduce fusion weights and calculate the comprehensive score; Abnormal judgment: When the comprehensive score of a window is greater than or equal to the abnormal judgment threshold, the window is marked as abnormal and the abnormal signal is located.

4. The gateway-based intelligent auxiliary control and monitoring system for substations according to claim 1 is characterized by: The fault warning module uses a method that combines multi-time-scale wavelet entropy features with density clustering, and uses the data's own structure to spontaneously identify abnormal precursors in the time-frequency domain and issue warnings. It specifically includes the following units: The wavelet entropy feature extraction unit specifically includes the following contents: Continuous wavelet transformation, calculating the wavelet coefficients on the scale set for each dimensional signal; Calculate energy distribution; To generate fault precursor features, first normalize the energy distribution to a probability distribution; then calculate the wavelet entropy; and finally concatenate all the wavelet entropies to form the fault precursor features. Density cluster warning unit, specifically including the following contents: DBSCAN clustering, in the feature space with parameters 、 Run, where represents the neighborhood radius, Indicates the minimum number of points in the cluster within the neighborhood radius; obtain the label set; Rare cluster determination, determining rare clusters based on their frequency of occurrence; Fault warning, when D consecutive windows fall into the same rare cluster, a warning is triggered, and the fault dimension is determined by comparing it with the median of historical entropy.

5. The gateway-based intelligent auxiliary control and monitoring system for substations according to claim 1 is characterized by: The intelligent operation control module specifically includes the following units: The load forecasting unit constructs a B-order autoregressive model based on the power time series output by the real-time power calculation unit to predict the load in the future period; The control strategy optimization unit solves a mixed integer convex optimization problem based on the predicted load curve, the equipment switching status and ambient temperature provided by the data acquisition module, while reducing the peak load, smoothing the load curve and minimizing the number of equipment switching operations; The execution unit sends control commands to the primary devices through the gateway machine according to the device switching strategy output by the optimization unit, and executes them in real time.

Images