A substation intelligent auxiliary control monitoring system based on a gateway machine

Abstract

The application discloses a kind of based on gateway machine's intelligent auxiliary control monitoring system of substation, including data acquisition module, protocol self-adapting adjustment module, abnormality detection module, fault early warning module and intelligent operation control module.The application relates to the technical field of intelligent management of substation, specifically refers to a kind of based on gateway machine's intelligent auxiliary control monitoring system of substation, the present scheme utilizes self-supervised contrast learning and second-order Markov chain automatically extracts and predicts protocol features, combines confidence and delay decision, realizes no label fast switching, high reliable communication;Fusion sliding window transfer entropy and kernel PCA reconstruction residual, unsupervised detection and positioning electrical, network and environmental anomaly;With the help of multi-scale Morlet wavelet entropy and DBSCAN rare cluster identification, unknown fault precursor spontaneous early warning is realized;Based on real-time apparent power autoregressive prediction and mixed integer convex optimization, peak suppression, load smoothing and temperature constraint are considered, and scheduling strategy is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent management of substations, and particularly to a substation intelligent auxiliary control monitoring system based on a gateway computer. BACKGROUND

[0002] In the process of the development of the power system in the direction of digitization and intelligentization, substation automation and monitoring technology has become a key link to ensure the safe and stable operation of the power grid. With the continuous improvement of the networking level of substations, various intelligent terminals and communication protocols coexist between field devices and monitoring centers. The traditional method of relying on manual pre-configuration or a single protocol stack has been difficult to adapt to the needs of mixed operation of multiple manufacturers and heterogeneous protocols. Communication switching lags or misjudgment problems occur frequently, and the operation and maintenance costs and risks have increased substantially.

[0003] At the same time, existing anomaly detection is often based on simple threshold values or single statistical models, which cannot effectively integrate the complex coupling relationship between electrical working conditions, network traffic and environmental parameters, and has insufficient recognition ability for unknown fault precursors and weak abnormal signals, and is prone to missed reports and false reports. In addition, the fluctuation of substation load curves and the frequent switching of devices make it difficult for traditional experience scheduling to balance peak suppression, load smoothing and device life management, and the ability to support real-time prediction and optimization decision-making is insufficient.

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

[0005] In view of the above, in order to overcome the defects of the prior art, the application provides a transformer substation intelligent auxiliary control monitoring system based on a gateway machine, aiming at the problems that traditional protocol identification relies on manual configuration and labeled training, protocol switching lags and is prone to misjudgment, etc., the scheme extracts protocol features in depth through self-supervised contrast learning, accurately predicts protocol evolution through a second-order Markov chain, and realizes automatic identification of new protocols without labels by combining prediction confidence and communication time delay for joint decision; aiming at the problems that traditional anomaly detection relies on a single statistical threshold or reconstruction residual method, it is difficult to consider the causal coupling relationship between signals and mutation characteristics in each dimension, the positioning accuracy is low, and false positives and false negatives are prone to occur, the scheme introduces a dynamic causal network based on transfer entropy to construct a causal drift score in a sliding window, and combines kernel principal component analysis to extract a mutation signal score, and after standardization according to the historical mean, the scores are weighted and fused, and finally the joint anomaly detection and accurate positioning of electrical working conditions, network traffic and environmental monitoring signals are realized under unsupervised conditions; aiming at the problems that traditional fault warning is difficult to adaptively capture the multi-time scale precursor mode of complex signals and is prone to ignore rare fault signals, the scheme extracts energy distribution and calculates wavelet entropy to construct a multi-time scale precursor feature vector for each dimension signal at different wavelet scales, automatically identifies rare cluster patterns by combining DBSCAN density clustering, and triggers an alarm when the rare cluster is maintained in continuous multiple windows, and locates the fault dimension based on the median of the entropy value, realizing unsupervised spontaneous detection and accurate warning of unknown fault precursors; aiming at the problems that traditional operation control is difficult to consider peak suppression, load smoothing and equipment switch life, and lacks dynamic response to environmental constraints, the scheme obtains future load trend based on autoregressive load prediction of real-time apparent power, and solves the optimal switch strategy under the mixed integer convex optimization framework with the optimization criteria of minimizing peak load, smoothing overall load and switch action penalty, combined with environmental temperature constraints, realizing fine scheduling of load curve and intelligent control of equipment operation.

[0006] The technical scheme adopted by the application is as follows: the transformer substation intelligent auxiliary control monitoring system based on the gateway machine provided by the application comprises 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 acquires communication traffic data, transformer substation electrical data and transformer substation environmental data.

[0008] The protocol adaptive adjustment module first trains an encoder on a frame load subsequence through self-supervised contrast learning to learn deep representations of the protocol, then updates the transition matrix based on a second-order Markov chain to predict the protocol type at the next time, and finally combines the prediction confidence threshold and the real-time communication delay for joint decision to dynamically select and switch to the optimal protocol.

[0009] The abnormality detection module first calculates the transfer entropy between each dimension pair in the time series signal in each sliding window, constructs a dynamic causal network and obtains a window difference matrix; then, based on kernel principal component analysis, projects and inversely maps the multi-dimensional data in the same window to obtain a reconstruction residual and extract a maximum residual score; finally, the causal drift score and the residual score are fused according to the weight, and whether the window is abnormal is determined according to the threshold, and the abnormal dimension is located through the maximum difference or the maximum residual position.

[0010] The fault early warning module first calculates the energy distribution of each dimension signal on the multi-scale Morlet wavelet and normalizes it into a probability, and then obtains 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 the consecutive D windows fall into the same rare cluster, the early warning is triggered, and the fault dimension is determined by comparing 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 the future load curve, then solves the optimal device switching strategy in a mixed integer convex optimization framework considering peak suppression, load smoothing, switch action penalty and environmental temperature constraint, and finally issues control commands through the gateway machine and executes them in real time.

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

[0013] The protocol feature extraction unit uses sub-sequence 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 the protocol features;

[0014] The Markov chain protocol prediction unit introduces a second-order Markov assumption, combines the weighted prediction of the previous two protocol types to predict the next protocol, and uses online updating to quickly incorporate new protocols into the transition matrix;

[0015] The protocol switching decision unit uses a joint decision of prediction confidence and communication delay.

[0016] Further, the abnormality detection module is aimed at time series signals from the data acquisition module; a dynamic causal network based on transfer entropy and a kernel principal component analysis reconstruction residual fusion detection strategy are introduced to realize joint unsupervised abnormal positioning of electrical working conditions, network traffic and environmental monitoring signals, specifically including the following units:

[0017] The transfer entropy network construction unit analyzes the multi-dimensional signal set in fixed length time periods; for any signal dimension pair, define the sliding window transfer entropy; then combine all the transfer entropies into a transfer entropy matrix; and calculate the difference amplitude with the previous window; finally, define the causal drift score as an abnormality indicator;

[0018] Kernel principal component reconstruction residual unit, specifically includes the following contents:

[0019] Kernel principal component projection, based on the window multi-dimensional data to construct kernel matrix and extract principal component, through the inverse mapping to get the reconstruction signal;

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

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

[0022] Fusion decision and positioning unit, first introduce fusion weight, calculate the comprehensive score;

[0023] Abnormality determination, when the comprehensive score of the window is greater than or equal to the abnormality determination threshold, mark the window as abnormal, and carry out abnormal signal positioning.

[0024] Further, the fault warning module adopts a method combining multi-time scale wavelet entropy feature and density clustering, uses the data structure itself to identify abnormal precursors in time-frequency domain and issue a warning, specifically including the following units:

[0025] Wavelet entropy feature extraction unit, specifically includes the following contents:

[0026] Continuous wavelet change, calculate the wavelet coefficient on the scale set for each dimension signal;

[0027] Calculate the energy distribution;

[0028] Generate fault precursor feature, first normalize the energy distribution to probability distribution; then calculate the wavelet entropy; finally, all wavelet entropies are spliced into fault precursor features;

[0029] Density clustering warning unit, specifically includes the following contents:

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

[0031] Rare cluster determination, determine the rare cluster according to the frequency of occurrence;

[0032] Fault warning, when the continuous D windows fall into the same rare cluster, trigger the warning, and determine the fault dimension by comparing with the historical entropy median.

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

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

[0035] The control strategy optimization unit solves a mixed integer convex optimization problem for the predicted load curve and the device switch state and environmental temperature given by the data acquisition module, while reducing the peak load, smoothing the load curve and minimizing the number of device switch actions.

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

[0037] The above scheme has the following beneficial effects:

[0038] (1) To solve the problems of traditional protocol recognition, such as relying on manual configuration and labeled training, lagging in protocol switching and misjudgment, this scheme extracts protocol features through self-supervised contrast learning, accurately predicts protocol evolution through second-order Markov chain, and makes joint decisions based on prediction confidence and communication delay, achieving automatic identification of new protocols, fast switching and low-delay reliable communication.

[0039] (2) To solve the problems of traditional anomaly detection, such as relying on single statistical threshold or reconstruction residual method, difficult to consider causal coupling relationship between signals and mutation characteristics in each dimension, low positioning accuracy and easy to produce false alarm, this scheme introduces a dynamic causal network based on transfer entropy to construct a causal drift score in a sliding window, and combines kernel principal component analysis to extract a mutation signal score, and then normalizes the historical mean and weights the fusion, finally realizes joint anomaly detection and accurate positioning of electrical working conditions, network traffic and environmental monitoring signals under unsupervised conditions, effectively improving detection sensitivity and positioning accuracy.

[0040] (3) To solve the problem of traditional fault warning, which is difficult to adaptively capture multi-time scale precursor patterns of complex signals and easily ignores rare fault signals, this scheme extracts energy distribution and calculates wavelet entropy to construct a multi-time scale precursor feature vector for each dimension signal in different wavelet scales, automatically identifies rare cluster patterns by combining DBSCAN density clustering, and triggers an alarm when the rare cluster is maintained in consecutive multiple windows, and locates the fault dimension based on the median of the entropy value, realizing unsupervised spontaneous detection and accurate warning of unknown fault precursors.

[0041] (4) In view of the problems that the traditional operation control is difficult to balance peak suppression, load smoothing and switch life, and lacks dynamic response to environmental constraints, the scheme obtains the future load trend through autoregressive load prediction based on real-time apparent power, and solves the optimal switch strategy in the mixed integer convex optimization framework with the optimization criteria of minimum peak load, overall load smoothing and switch action penalty, combined with environmental temperature constraints, to realize fine scheduling of the load curve and intelligent control of device operation. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A schematic diagram of a substation intelligent auxiliary control monitoring system based on a gateway machine is provided.

[0043] Figure 2 A schematic diagram of a protocol adaptive adjustment module is provided.

[0044] Figure 3 A schematic diagram of an anomaly detection module is provided.

[0045] Figure 4 A schematic diagram of a fault early warning module is provided.

[0046] Figure 5 A schematic diagram of an intelligent operation control module is provided.

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

[0048] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0049] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0050] Embodiment one, refer to Figure 1The application provides a transformer substation intelligent auxiliary control monitoring system based on a gateway machine, which comprises a data acquisition module, a protocol adaptive adjustment module, an abnormality detection module, a fault early warning module and an intelligent operation control module.

[0051] The data acquisition module acquires communication flow data, transformer substation electrical data and transformer substation environmental data, and sends the data to the protocol adaptive adjustment module, the abnormality detection module, the fault early warning module and the intelligent operation control module.

[0052] The protocol adaptive adjustment module receives the data sent by the data acquisition module, trains an encoder on a frame load subsequence through self-supervised contrast learning to learn deep representation of the protocol, then predicts the protocol type at the next moment based on a second-order Markov chain and the transition matrix updated for the previous two times of protocol states, and finally dynamically selects and switches to the optimal protocol through joint decision-making in combination with a prediction confidence threshold and real-time communication delay.

[0053] The abnormality detection module receives the 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 a window difference matrix, then projects and inversely maps the multi-dimensional data in the same window based on kernel principal component analysis to obtain a reconstruction residual and extract a maximum residual score, and finally fuses the causal drift score and the residual score according to a weight, determines whether the window is abnormal according to a threshold, locates the abnormal dimension through the maximum difference or the maximum residual position, and sends the data to the fault early warning module.

[0054] The fault early warning module receives the data sent by the data acquisition module and the abnormality detection module, calculates the energy distribution on the multi-scale Morlet wavelet of each dimension signal and normalizes it into a probability, then obtains the wavelet entropy of each scale to form a precursor feature vector, applies DBSCAN clustering to the historical window feature sequence to identify rare clusters with low occurrence frequency, triggers early warning when the consecutive D windows fall into the same rare cluster, and determines the fault dimension by comparing with the historical entropy median.

[0055] The intelligent operation control module receives the data sent by the data acquisition module, constructs an autoregressive model based on real-time apparent power to predict the future load curve, then solves the optimal device switch strategy in a mixed integer convex optimization framework considering peak suppression, load smoothing, switch action penalty and environmental temperature constraint, and finally issues a control command through the gateway machine and executes it in real time.

[0056] Embodiment two, refer to Figure 1This embodiment is based on the above embodiment. The data acquisition module collects communication traffic data, substation electrical data, and substation environmental data. Specifically, the communication traffic data is a set of raw protocol frames captured from the substation's local Ethernet communication bus via the gateway's 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 refers to the collected electrical quantities of the substation's primary equipment, including: bus voltage. Line current Equipment on / off status The substation environmental data includes: the temperature of the substation main control room obtained by temperature and humidity detectors. Ambient relative humidity obtained by temperature and humidity detector Vibration intensity of transformer cabinet collected by accelerometer .

[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 subsequence comparison of frame payloads for self-supervised learning to separate the deep representation of the protocol and construct frame sequence slice pairs. As a positive example, for For negative examples, the encoder is trained to automatically learn protocol features, as shown below:

[0059] ;

[0060] in, This represents the deep features of the protocol, where i and j represent the indices of the protocol frames. This represents the payload subsequence of the i-th frame. Indicates and Positive example sequences of the same protocol, Indicates and Negative example sequences of different protocols, This represents an exponential function with the natural constant as its base. Represents parameter set The feature encoder below, Represents cosine similarity. This represents the temperature hyperparameter, with a value range of [value missing]. N and M represent the number of positive and negative samples, respectively.

[0061] The Markov chain protocol prediction unit introduces the second-order Markov assumption and combines the weighted average of the previous two protocol types to predict the next protocol. It also utilizes online updates to quickly incorporate the new protocol into the transition matrix, as shown below:

[0062] ;

[0063] wherein k represents the index of the protocol frame, represents the probability that the protocol type of the next moment is exactly k, given that the protocol type of the current moment t is i and the protocol type of the previous moment t-1 is j, represents the protocol type index of the moment t, represents the state transition count matrix entry, representing the number of times that the continuous three steps of changing from protocol j to protocol i and then to protocol k occur; represents the online update rate, represents an indicator function, which is 1 when the event occurs and 0 otherwise;

[0064] The protocol switching decision unit jointly decides using the prediction confidence and the communication delay, as shown below:

[0065] ;

[0066] wherein, represents the predicted optimal protocol type of the next moment, represents the state switching condition, represents the preset confidence threshold, represents the communication time delay calculated by the timestamp, represents the maximum acceptable switching delay, represents a logical AND symbol.

[0067] By performing the above operations, in view of the problems of traditional protocol recognition, such as relying on manual configuration and labeled training, protocol switching lag, and easy misjudgment, the present scheme deeply extracts protocol features through self-supervised contrast learning, accurately predicts protocol evolution through a second-order Markov chain, and jointly decides using the prediction confidence and the communication delay, to realize automatic identification of new protocols without labels, fast switching, and low-delay reliable communication.

[0068] In the fourth embodiment, referring to Figure 1 and Figure 3 , the abnormality detection module, for the time series signal from the data acquisition module, contains six dimensions; innovatively introduces a dynamic causal network based on transfer entropy and a residual fusion detection strategy based on kernel principal component analysis, to realize joint unsupervised abnormal positioning of electrical working conditions, network traffic, and environmental monitoring signals, specifically including the following units:

[0069] The transfer entropy network construction unit analyzes the multi-dimensional signal set in a fixed length time period as a unit, wherein, denotes the length of the sliding window, u denotes the window index, denotes the ending time of the u-th window; for any signal dimension pair , the sliding window transfer entropy from p to q is defined as follows:

[0070] ;

[0071] wherein, and denote the dimension index, M denotes the total number of dimensions, denotes the transfer entropy from dimension p to dimension q on the u-th window; denotes the discretized value of dimension q at time t+1, denotes the history embedding vector of dimension q at time t and the past k-1 steps, k denotes the embedding dimension, denotes the history embedding vector of dimension p at time t and the past k-1 steps, denotes the ternary joint probability, specifically the empirical frequency of all corresponding discrete samples within the u-th window, and denote the conditional probability; then all the transfer entropies are combined into a transfer entropy matrix ; and the window difference amplitude is calculated with the previous window; finally, the causal drift score is defined as an anomaly indicator, which is represented as follows:

[0072] ;

[0073] wherein, denotes the average causal drift score of the u-th window at the ending time , excluding self-loop, denotes the element in the p-th row and the q-th column of the window difference amplitude ;

[0074] The kernel principal component reconstruction residual unit specifically includes the following contents:

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

[0076] Residual calculation, for each dimension, the reconstruction residual at time t is defined as follows:

[0077] ;

[0078] wherein, denotes the reconstruction residual of dimension m at time t, and respectively represent the signal value of the original and reconstructed time series signal in the mth dimension;

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

[0080]

[0081] wherein, represents the final time point of the reconstructed residual score;

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

[0083] The comprehensive score is calculated and represented as follows:

[0084]

[0085] wherein, represents the fusion weight, and respectively represent the historical score mean of the drift score and the residual score, represents the final time point of the comprehensive score;

[0086] Anomaly determination, when , the window is marked as an anomaly, wherein represents an anomaly determination threshold, which is specifically set as the 95th percentile of the normal period; and positioning is performed, if is the largest, then the dimension is considered as the causal coupling instability position; if and , then the dimension is considered as the anomaly signal.

[0087] By performing the above operation, the traditional anomaly detection method relies on a single statistical threshold or reconstruction residual method, which is difficult to consider the causal coupling relationship between signals and the mutation characteristics of each dimension, and has low positioning accuracy and is prone to false positives and false negatives. The present scheme introduces a dynamic causal network based on transfer entropy to construct a causal drift score in a sliding window, and extracts a mutation signal score by combining a kernel principal component analysis reconstruction residual. After normalization according to the historical mean, the scores are weighted and fused, and finally the joint anomaly detection and accurate positioning of electrical working conditions, network traffic and environmental monitoring signals are realized under unsupervised conditions, effectively improving the detection sensitivity and positioning accuracy.

[0088] Embodiment five, refer to Figure 1 and Figure 4 ​​The embodiment is based on the above-mentioned embodiment, the fault early warning module adopts a method combining multi-time scale wavelet entropy features and density clustering, uses the data structure itself to spontaneously identify abnormal precursors in the time-frequency domain and issue early warnings, and specifically includes the following units:

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

[0090] The continuous wavelet change is performed on each dimension signal The wavelet coefficients on the scale set are calculated:

[0091] ;

[0092] wherein, denotes the wavelet scale set, denotes the conjugate Morlet wavelet function, denotes the continuous wavelet coefficient of the scale on the dimension m;

[0093] The energy distribution is calculated, the energy of the scale is calculated in the window u, and is denoted as follows:

[0094] ;

[0095] wherein, denotes the energy of the signal of the mth dimension in the window u and on the scale ;

[0096] The fault precursor feature is generated, the energy distribution is first normalized into a probability distribution ; then the wavelet entropy is calculated; finally, all the wavelet entropies are spliced into the fault precursor feature ;

[0097] The density clustering early warning unit specifically includes the following content:

[0098] The DBSCAN clustering is performed in the feature space with parameters , , wherein, denotes the neighborhood radius, and the value range is , denotes the minimum number of cluster points in the neighborhood radius, and the value range is ; the label set is obtained and is denoted as follows:

[0099] ;

[0100] wherein, denotes the clustering label set of the window u, and v denotes the index of the window, The fault precursor feature of the window v is represented, and the label 0 represents a noise label;

[0101] Rare cluster determination, the occurrence frequency of the cluster cx with a label greater than or equal to 1 is calculated ; if , the cluster cx is determined as a rare cluster, wherein represents a cluster frequency threshold, and the value range is ;

[0102] Fault warning, if the current window label , represents a rare cluster set; and the label is uniformly maintained in the consecutive D windows, and the value range of D is , a fault warning is triggered, and the fault dimension is indicated as: , wherein represents a set of wavelet entropies of the dimension m on all windows, represents the median of the historical wavelet entropy of the dimension m.

[0103] Through the above operation, in order to solve the problems that the traditional fault warning is difficult to adaptively capture the multi-time scale precursor mode of complex signals and easily ignore rare fault signals, the present scheme extracts the energy distribution of each dimension signal on different wavelet scales and calculates the wavelet entropy to construct a multi-time scale precursor feature vector, automatically identifies the rare cluster mode in combination with DBSCAN density clustering, and triggers the warning when the rare cluster is maintained in the consecutive multiple windows. Meanwhile, based on the median of the entropy value, the fault dimension is located, and the unsupervised spontaneous detection and accurate warning of unknown fault precursors are realized.

[0104] Embodiment six, refer to Figure 1 and Figure 5 , this embodiment is based on the above-mentioned embodiment, and the intelligent operation control module specifically comprises the following units:

[0105] The load prediction unit constructs a B-order autoregressive model to predict the load in the future period based on the power time sequence output by the real-time power calculation unit , which is represented as follows:

[0106] ;

[0107] , wherein d represents the index of the prediction time, represents the apparent power at time t, represents the maximum order of the autoregressive model, b represents the index of the model order, represents the bth-order autoregressive coefficient, which is obtained by least square fitting of the historical power sequence, represents a time delay parameter, represents a zero-mean white noise error term;​

[0108] a control strategy optimization unit, for the predicted load curve , o represents the index of the optimization time, and the device switch state given by the data acquisition module and the ambient temperature , solve the following mixed integer convex optimization problem while reducing peak load, smoothing the load curve and minimizing the number of device switch actions:

[0109] ;

[0110] wherein, represents the objective function, represents the discrete time set of the future optimization period, represents the total number of device switches, g represents the index of the device switch, represents the optimization control strategy of the gth device switch at time o, 1 represents the closing of the switch, and 0 represents the opening of the switch; represents the switch state collected by the data acquisition module, represents the weight coefficient of peak suppression and total energy consumption balance, represents the control action penalty weight, represents the main control room temperature collected by the data acquisition module, represents the maximum temperature threshold allowed by the environment;

[0111] an execution unit, according to the device switch strategy output by the optimization unit , through the gateway machine, the control command is issued to the primary device, and real-time execution is performed.

[0112] By performing the above operation, for the traditional operation control which is difficult to balance peak suppression, load smoothing and device switch life and lacks dynamic response to environmental constraints, the scheme obtains the future load trend based on the autoregressive load prediction of real-time apparent power, and solves the optimal switch strategy under the mixed integer convex optimization framework with the optimization criteria of peak load minimization, overall load smoothing and switch action penalty, combined with the environmental temperature constraint, realizes the fine scheduling of the load curve and the intelligent control of the device operation.

[0113] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other present or future technologies can provide. Specifically, it is contemplated that such technologies will provide for variations on this description, and the scope of the present application is to be understood to include all such variations as can become apparent to those of ordinary skill in the art upon reading the description herein.

[0114] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and alterations can be made therein without departing from the spirit and scope of the application.

[0115] The above description of the application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the application.

Claims

1. A gateway machine-based intelligent auxiliary control and monitoring system for a substation, characterized by: The system comprises a data acquisition module, a protocol adaptive adjustment module, an abnormality detection module, a fault early warning module and an intelligent operation control module. The data acquisition module acquires communication flow data, substation electrical data and substation environmental data. The protocol adaptive adjustment module firstly trains an encoder on a frame load subsequence through self-supervised contrast learning to learn deep representations of the protocol, then updates a transition matrix based on a second-order Markov chain using the previous two protocol states to predict the protocol type at the next time, and finally dynamically selects and switches to the optimal protocol based on a joint decision of a prediction confidence threshold and real-time communication delay. The abnormality detection module firstly calculates transition entropy between each dimension pair in a time series signal in each sliding window, constructs a dynamic causal network and obtains a window difference matrix. Then, the module projects and inversely maps multi-dimensional data in the same window based on kernel principal component analysis to obtain reconstruction residuals and extract maximum residual scores. Finally, the module fuses the causal drift scores and residual scores according to a weight, determines whether the window is abnormal according to a threshold, and locates the abnormal dimension through the maximum difference or the maximum residual position. The fault early warning module firstly calculates the energy distribution of each dimension signal on a multi-scale Morlet wavelet and normalizes it into a probability, and then obtains wavelet entropy of each scale to form a precursor feature vector.

2. The intelligent auxiliary control and monitoring system for a substation based on a gateway computer according to claim 1, characterized in that: Then, the module applies DBSCAN clustering to a historical window feature sequence to identify rare clusters with low occurrence frequency. When the same rare cluster is entered by consecutive D windows, the module triggers an early warning, and determines the fault dimension by comparing with the historical entropy median. The intelligent operation control module constructs an autoregressive model based on real-time apparent power to predict a future load curve, and then solves an optimal device switching strategy in a mixed integer convex optimization framework considering peak suppression, load smoothing, switch action penalty and environmental temperature constraints. The protocol adaptive adjustment module comprises the following units:

3. The intelligent auxiliary control and monitoring system for a substation based on a gateway computer of claim 1, wherein: A protocol feature extraction unit uses subsequence comparison of frame load for self-supervised learning to separate deep representations of the protocol, constructs frame sequence slices, and trains an encoder to automatically learn protocol features. A Markov chain protocol prediction unit introduces a second-order Markov assumption to predict the next protocol type based on the previous two protocol types, and uses online updating to quickly incorporate new protocols into the transition matrix. A protocol switching decision unit uses a joint decision of prediction confidence and communication delay. The abnormality detection module is directed to time series signals from the data acquisition module, and introduces a dynamic causal network based on transition entropy and a kernel principal component analysis reconstruction residual fusion detection strategy to jointly and unsupervisedly locate abnormalities of electrical working conditions, network flow and environmental monitoring signals. A transition entropy network construction unit analyzes a multi-dimensional signal set in a fixed time period, defines a sliding window transition entropy for any signal dimension pair, then combines all transition entropies into a transition entropy matrix, calculates a difference amplitude from the previous window, and finally defines a causal drift score as an abnormality index. A kernel principal component reconstruction residual unit comprises the following contents: Kernel principal component projection, kernel matrix is constructed based on multi-dimensional data in window and principal components are extracted, and the reconstructed signal is obtained by inverse mapping; Residual calculation, define the reconstruction residual for each dimension; Set residual score, set the maximum residual in the window as the residual score; Fusion decision and positioning unit, first introduce fusion weight, calculate comprehensive score; Abnormality judgment, when the comprehensive score of the window is greater than or equal to the abnormality judgment threshold, mark the window as abnormal, and locate the abnormal signal.

4. The intelligent auxiliary control and monitoring system for a substation based on a gateway computer of claim 1, wherein: The fault early warning module adopts a method combining multi-time scale wavelet entropy feature and density clustering, uses the data structure itself to identify abnormal precursors in time-frequency domain and issue early warning, and specifically includes the following units: Wavelet entropy feature extraction unit, specifically including the following contents: Continuous wavelet change, calculate the wavelet coefficient on the scale set for each dimension signal; Calculate energy distribution; Generate fault precursor feature, first normalize the energy distribution into probability distribution, then calculate wavelet entropy, and finally splice all wavelet entropies into fault precursor feature; Density clustering early warning unit, specifically including the following contents: DBSCAN clustering, in a feature space with parameters , running, wherein, denotes a neighborhood radius, denotes a minimum number of points within a cluster within the neighborhood radius; obtaining a set of labels; Rare cluster judgment, judge the rare cluster according to the frequency of occurrence; Fault early warning, trigger early warning when the consecutive D windows fall into the same rare cluster, and determine the fault dimension by comparing with the historical entropy median.

5. The intelligent auxiliary control and monitoring system for a substation based on a gateway computer of claim 1, wherein: The intelligent operation control module specifically includes the following units: Load prediction unit, based on the power time series output by the real-time power calculation unit, build a B-order autoregressive model to predict the load in the future period; Control strategy optimization unit, for the predicted load curve, and the equipment switch state and environmental temperature given by the data acquisition module, solve the mixed integer convex optimization problem, reduce the peak load, smooth the load curve and reduce the number of equipment switch actions as much as possible; Execution unit, according to the equipment switch strategy output by the optimization unit, issue control commands to primary equipment through gateway machine, and execute in real time.

Images