An intelligent monitoring system for power supply and distribution in a distribution network

Abstract

The present invention belongs to the technical field of intelligent power monitoring for distribution networks. The present invention discloses an intelligent power supply and distribution monitoring system for distribution networks, which includes: collecting power data and status data, strengthening the power data to obtain enhanced power data; performing waveform analysis on the enhanced power data to obtain a load curve graph, performing status influence on the load curve graph to obtain a load correction graph, performing feature extraction on the load correction graph to obtain load features; predicting the load features based on the constructed fault prediction model to obtain a fault message; performing load prediction on the load correction graph to obtain future load demands, constructing a demand topology graph based on the future load demands, performing fault cutting on the demand topology graph to obtain a local demand graph, and performing power optimization scheduling on the local demand graph to obtain an optimal scheduling plan, which greatly improves the stability of the power supply and distribution system.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent power monitoring for distribution networks. More specifically, the present invention relates to an intelligent power supply and distribution monitoring system for distribution networks. Background Art

[0002] Under the background of industrialization, the electricity demand is complex and diverse. The power supply and distribution system is the core facility to ensure the stable operation of production. With the growth of the number of devices and the advancement of industrial production automation, the existing power supply and distribution monitoring systems for distribution networks have shown significant deficiencies in functionality and adaptability, and it is difficult to meet the requirements of modern intelligent manufacturing. For example: The diverse production tasks have made the dynamic changes in electricity demand more and more significant, resulting in the need for frequent adjustments to the power supply and distribution system. The existing systems lack flexible dynamic regulation capabilities. The power supply and distribution strategies usually only target local optimization, ignoring the overall situation of the power supply and distribution network, and cannot be quickly updated according to real-time data. There are often significant delays in scheduling adjustments, lacking real-time adaptability and a global perspective, resulting in frequent equipment failures, and a significant decline in production efficiency and safety; In the power supply and distribution system, many faults lurk without causing an immediate power outage, but these hidden faults are usually weak fluctuations or intermittent abnormalities of electrical signals, which cannot be effectively captured in normal monitoring systems. Once a hidden fault evolves into an obvious fault, it usually causes irreversible damage to the equipment, which often leads to production stoppages, equipment damage, and expensive repair costs.

[0003] In view of this, the present invention proposes an intelligent power supply and distribution monitoring system for distribution networks to solve the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An intelligent power supply and distribution monitoring system for distribution networks, comprising:

[0005] A data acquisition and processing module: collecting power data and status data of distribution network equipment in the same time series, and performing data enhancement on the power data to obtain enhanced power data;

[0006] A waveform analysis module: performing waveform analysis on the enhanced power data to obtain a load curve graph, performing a state impact on the load curve graph based on the status data to obtain a load correction graph, and performing feature extraction on the load correction graph to obtain load features;

[0007] A fault message module: predicting the load features based on a constructed fault prediction model to obtain a fault message, and issuing a fault alarm based on the fault message;

[0008] Optimization Scheduling Module: It includes a demand forecasting unit and a scheduling unit. The demand forecasting unit conducts load forecasting on the load correction graph to obtain future load demands, constructs a demand topology graph based on the future load demands, performs fault cutting on the demand topology graph based on fault messages to obtain a local demand graph, and the scheduling unit uses an optimization algorithm to perform power optimization scheduling on the local demand graph to obtain the best scheduling plan.

[0009] Furthermore, the power data includes: device voltage and device current; the status data includes: device temperature.

[0010] Furthermore, the method for data enhancement of the power data includes:

[0011] Taking each type of data in the power data as the original data, performing waveform transformation on the original data to obtain a waveform function, presetting a noise threshold, performing waveform smoothing on the waveform function based on the noise threshold to obtain a smoothed value; recording the maximum and minimum values in the smoothed value, taking the smoothed value at each moment as a neuron data, performing enhanced weight learning on the smoothed value based on the neuron data to obtain an enhanced weight; performing value enhancement on the smoothed value at each moment, multiplying the smoothed value at each moment by the corresponding enhanced weight to obtain enhanced data, and all the enhanced data constitutes enhanced device data.

[0012] Furthermore, the formula for performing enhanced weight learning is:

[0013] ; where represents the enhanced weight at the th moment, represents the enhanced weight at the th moment, represents the enhancement balance factor, represents the smoothed value at the th moment, represents the maximum value in the smoothed value, represents the minimum value in the smoothed value, represents the th moment of the original data.

[0014] Furthermore, the method for influencing the status of the load curve graph includes:

[0015] Using the spline interpolation method to depict curves for each type of data in the enhanced power data to obtain a device voltage curve graph and a device current curve graph; performing load fitting on the device voltage curve graph and the device current curve graph to obtain the power load, and the power loads at different moments constitute the load curve graph;

[0016] ​Predict the state data based on the constructed load status influence model to obtain the load influence coefficient. Preset the frequency determination function, and perform frequency determination on the load influence coefficient based on the frequency determination function to obtain the corrected frequency. Perform load correction on the load curve graph based on the load influence coefficient. The formula for load correction is as follows: ; where represents the corrected load at time, represents the load offset factor, represents the power load at time, represents the load influence coefficient at time, represents the corrected frequency at time, represents the th correction. The corrected loads at different times constitute the load correction graph.

[0017] Furthermore, the method for feature extraction of the load correction graph includes:

[0018] Integrate the load correction graph over the time series span to obtain the total load value, and divide the total load value by the time series span to obtain the average load; calculate the difference between the corrected loads at consecutive times of the load correction graph, subtract the corrected load at the previous time from the corrected load at the next time to obtain the load change over time, calculate the standard deviation of the load change over time, and divide the standard deviation of the load change over time by the average load to obtain the load volatility; calculate the load change rate at each time in the load correction graph, and average the load change rates to obtain the average change rate; preset a sliding time window, capture the short-term fluctuations of the load correction graph, slide the sliding time window starting from the start time in the load correction graph, and calculate the standard deviation of the corrected loads within each sliding time window to obtain the window standard deviation; form the load features with the mean values of the load volatility, average load, average change rate, and window standard deviation.

[0019] Furthermore, the construction method of the fault prediction model includes:

[0020] Collect historical fault message data and historical equipment load characteristic data. Use the historical fault message data and historical equipment load characteristic data as the input of the fault prediction model, use the predicted fault message as the output of the fault prediction model, and use the historical fault message as the prediction target of the fault prediction model for model training to construct the fault prediction model. The fault prediction model is a linear regression model, and the target loss function of the fault prediction model is the recall rate function. Take minimizing the value of the target loss function as the training target to obtain the fault prediction model with the minimum value of the target loss function.

[0021] Further, the construction method of the local demand graph includes:

[0022] Predict the load correction graph based on the trained load prediction model to obtain the future load demand. Use the future load demand as the load nodes, preset the connection threshold, calculate the correlation between different load characteristics based on the cosine similarity formula, use the correlation as the weight of the connected edge, and connect the load nodes with a correlation greater than or equal to the connection threshold to obtain the demand topology graph; Remove the fault nodes and the edges directly connected to the fault nodes in the demand topology graph based on the fault message to obtain the local demand graph.

[0023] Further, the method for performing power optimization scheduling on the local demand graph includes:

[0024] Preset the total load, evenly distribute each load node in the local demand graph based on the total load to obtain the initial load, and preset the load disturbance factor; Initialize B ants, randomly assign each ant to different graph nodes in the local demand graph, initialize the initial pheromone concentration of each edge in the local demand graph as the reciprocal of the edge weight, and initialize the global fitness as infinitesimal;

[0025] Each ant conducts distribution exploration in the local demand graph, selects the next position for the ant to move based on the initial pheromone concentration of each edge. The ant uses the edge with a high initial pheromone concentration as the moving direction. When the graph node information of each ant changes, record the current position and the previous position of the ant, update the initial load corresponding to the current position and the previous position based on the load disturbance factor. The load node corresponding to the previous position will reduce the load by the magnitude of the load disturbance factor and allocate it to the load node corresponding to the current position, and use the allocated value as the new initial load for the current position and the previous position. Perform load adaptation calculation on the local demand graph at this time to obtain the load fitness; When the load fitness is less than the global fitness, the ant stops moving and uses the load fitness at this time as the new global fitness. When the load fitness is greater than or equal to the global fitness, the ant returns to the previous position and takes back the load allocation. When all ants stop moving, use the current position of the ant as the initial position to repeat the distribution exploration until the number of ants on each load node in the local demand graph no longer changes, and output the initial load of each load node in the local demand graph at this time as the optimal scheduling plan.

[0026] Further, the formula for performing load adaptation calculation is:

[0027] ; where represents the load fitness, represents the number of load nodes in the local demand graph, represents the balance weight, represents the The initial load of a load node Representing the average load of all load nodes Representing the adaptation weight Representing the future load demand of the nth load node

[0028] The technical effects and advantages of an intelligent monitoring system for power supply and distribution in a distribution network according to the present invention

[0029] By collecting various data, the present invention fully considers the key factors in power supply and distribution, which helps to comprehensively monitor the operation status of equipment; by strengthening the power data, the mutation data is effectively removed, the accuracy of the data is improved, the time series characteristics are strengthened, and the system's perception ability of instantaneous fluctuations is optimized; by affecting the state of the load curve graph, real-time load correction based on equipment status and load fluctuations is realized, effectively coping with complex load change scenarios and improving the accuracy of load monitoring; by quickly issuing an alarm through the fault message module, potential equipment failures are timely responded to, reducing the operation risk of the power supply and distribution system of the distribution network; by constructing a local demand graph, the reliability during subsequent optimal dispatching is ensured; by performing power optimal dispatching on the local demand graph, the allocation of each load node is dynamically adjusted, ensuring the balance and adaptability of load distribution, reducing equipment failure rate and energy waste, improving the utilization rate of power resources, and reducing operating costs Description of the Drawings

[0030] Figure 1 Schematic diagram of an intelligent monitoring system for power supply and distribution in a distribution network according to the present invention

[0031] Figure 2 Schematic diagram of an intelligent monitoring method for power supply and distribution in a distribution network according to the present invention Detailed Embodiments

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention

[0033] Embodiment 1

[0034] Please refer to Figure 1 as shown. An intelligent monitoring system for power supply and distribution in a distribution network in this embodiment includes

[0035] Data acquisition and processing module: Collecting power data and status data of distribution network equipment in the same time series, and strengthening the power data to obtain strengthened power data

[0036] Waveform analysis module: It performs waveform analysis on the enhanced power data to obtain a load curve graph, conducts state influence on the load curve graph based on the state data to obtain a load correction graph, and extracts features from the load correction graph to obtain load characteristics;

[0037] Fault message module: It predicts the load characteristics based on the constructed fault prediction model to obtain a fault message, and issues a fault alarm based on the fault message;

[0038] Optimized scheduling module: It includes a demand prediction unit and a scheduling unit. The demand prediction unit conducts load prediction on the load correction graph to obtain future load demands, constructs a demand topology graph based on the future load demands, performs fault cutting on the demand topology graph based on the fault message to obtain a local demand graph, and the scheduling unit uses an optimization algorithm to perform power optimized scheduling on the local demand graph to obtain an optimal scheduling plan;

[0039] Each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0040] Power data includes: equipment voltage, equipment current, and equipment rated power; state data includes: equipment temperature and equipment vibration; the equipment voltage is obtained through a voltage sensor to monitor the voltage fluctuation of the equipment; the equipment current is obtained through a current sensor to monitor the current value of the equipment; the equipment temperature is obtained through a temperature sensor to monitor the temperature of the equipment, and too high equipment temperature is likely to cause equipment failure; the equipment vibration is obtained through a vibration sensor, and different vibration frequencies of the equipment indicate different states of the equipment.

[0041] The methods for data enhancement of power data include:

[0042] Taking each type of data in the power data (referring to equipment voltage and equipment current) as the original data, performing waveform transformation on the original data to obtain a waveform function. Common waveform transformations include wavelet transformation and Fourier transformation. A noise threshold is preset, and waveform smoothing is performed on the waveform function based on the noise threshold. The formula for waveform smoothing is: ; where represents the smoothed value of the waveform function at the th moment, represents the function value of the waveform function at the th moment, represents the sign function. When the value of is greater than zero, the value of the sign function is 1. When the value of is less than zero, the value of the sign function is -1. When the value of is equal to zero, the value of the sign function is 0, represents the noise threshold. By waveform smoothing, the localization characteristics of the original data can be optimized, the instantaneous changes of the original data can be accurately determined, and the mutation data in the original data can be removed; record the maximum and minimum values of the smoothed values. Take the smoothed value at each moment as a neuron data, and perform enhanced weight learning on the smoothed values based on the neuron data. The formula for performing enhanced weight learning is:

[0043] ; where represents the enhanced weight at the moment, represents the enhanced weight at the moment, represents the enhanced balance factor, which is used to balance the change of the enhanced weight, represents the moment of the smoothed value, represents the maximum value in the smoothed values, represents the minimum value in the smoothed values, represents the moment of the original data; perform value enhancement on the smoothed value at each moment based on the enhanced weight. Multiply the enhanced weight at each moment by the corresponding smoothed value to obtain the enhanced data. All enhanced data constitutes the enhanced device data; the value enhancement method based on the enhanced weight can adaptively adjust the influence of the smoothed value, so as to dynamically enhance the smoothed data in different time periods, thereby improving the accuracy of data representation.

[0044] The ways to perform waveform analysis on the enhanced power data include:

[0045] Use the spline interpolation method to depict the curve of each type of data in the enhanced power data to obtain the device voltage curve and the device current curve; perform load fitting on the device voltage curve and the device current curve. The formula for performing load fitting is: ; where represents the moment of the power load, represents the voltage of the device voltage curve at the moment, represents the current of the device current curve at the moment, represents the phase difference between the voltage and the current. The power loads at different moments constitute the load curve.

[0046] The ways to affect the state of the load curve include:

[0047] Collect historical device status data, historical device power data, and historical device power. The historical device status data is of the same type as the device status data, and the historical device power data and the power data are of the same data type. Use the spline interpolation method to depict the curve of each type of data in the historical device power data, obtain the historical voltage curve and the historical current curve, perform load fitting on the historical voltage curve and the historical current curve, obtain the historical power load, and use the ratio of the historical device power to the historical power load as the historical load impact coefficient;

[0048] Use the historical device status data and the historical load impact coefficient as the training sample set, use the GAN model as the initial model, train the GAN model using the training sample set, use the historical device status data as the input data of the load status impact model, and use the predicted load impact coefficient as the output data of the load status impact model; use minimizing the error between the actual historical load impact coefficient and the predicted load impact coefficient as the training objective, use the mean square error function as the loss function of the load status impact model, and stop training to obtain the load status impact model when the loss function converges, and record the value of the loss function at this time as the load offset factor;

[0049] Based on the constructed load status impact model, predict the status data to obtain the load impact coefficient, preset a frequency determination function, perform frequency determination on the load impact coefficient based on the frequency determination function to obtain the correction frequency, and perform load correction on the load curve based on the load impact coefficient. The formula for load correction is: ; where represents the corrected load at time, represents the load offset factor, represents the power load at time, represents the load impact coefficient at time, represents the correction frequency at time, represents the th correction. The corrected loads at different times form a load correction graph; by introducing the load offset factor and using the load correction formula, the system can be adjusted in real time according to the device status and load fluctuations, effectively cope with the load fluctuations and device status changes in the power supply and distribution system of the distribution network, and perform precise power monitoring on the distribution network.

[0050] The ways to extract features from the load correction graph include:

[0051] Integrate the load correction diagram over the time series span to obtain the total load value. Divide the total load value by the time series span to obtain the average load. Calculate the difference between the corrected loads at consecutive moments of the load correction diagram, subtract the corrected load at the previous moment from the corrected load at the next moment to obtain the load change over time. Calculate the standard deviation of the load change over time. Divide the standard deviation of the load change over time by the average load to obtain the load volatility. Calculate the load change rate at each moment in the load correction diagram and take the average of the load change rates to obtain the average change rate. Preset a sliding time window to capture short-term fluctuations in the load correction diagram. Slide the sliding time window starting from the start time in the load correction diagram and calculate the standard deviation of the corrected loads within each sliding time window to obtain the window standard deviation. The load characteristics are composed of the mean values of the load volatility, average load, average change rate, and window standard deviation. It should be noted that the time series span is the acquisition duration during data collection.

[0052] The construction method of the fault prediction model includes:

[0053] Collect historical fault message data and historical equipment load characteristic data. Use the historical fault message data and historical equipment load characteristic data as the input of the fault prediction model, use the predicted fault message as the output of the fault prediction model, and use the historical fault message as the prediction target of the fault prediction model for model training to construct the fault prediction model. The fault prediction model is a linear regression model, and the target loss function of the fault prediction model is the recall rate function. Take minimizing the value of the target loss function as the training target to obtain the fault prediction model with the minimum value of the target loss function. Based on the constructed fault prediction model, predict the load characteristics to obtain the fault message. The historical fault message data is the fault report data that has occurred in the equipment, including the equipment number, fault type, fault occurrence time, and fault description. The equipment load characteristic data is of the same type as the load characteristic data.

[0054] The method for load prediction of the load correction diagram includes:

[0055] A preset sample window is used to perform window cutting on the load correction graph to obtain a sample load sequence. Each sample load data is normalized to obtain standard sample data. A classification algorithm is used to perform sample clustering on the training samples to obtain sample clusters. Commonly used classification algorithms include the K-Means clustering algorithm and the hierarchical clustering algorithm. The preset load prediction model is an LSTM model. The standard sample data is used as the input and time scale of the load prediction model, and the predicted future load demand is used as the output of the load prediction model. A preset output probability is used to select the output of each prediction. The Adam optimizer is used as the model optimizer of the load prediction model, and the cross-entropy function is used as the loss function of the load prediction model. The load prediction model is cyclically trained based on the sample clusters, the parameters of the load prediction model are optimized by the Adam optimizer, and the mean value of the loss function values of each round of cyclic training is calculated as the loss mean. When the loss mean converges, the load prediction model at this time is output as the optimal load prediction model. Based on the trained load prediction model, the load correction graph is predicted to obtain the future load demand.

[0056] The sample window is a window with a fixed time span size. Assuming the time span size is three hours, then the window cutting is to divide the load correction graph in terms of time with a time span of every three hours.

[0057] The ways to perform fault cutting on the demand topology graph include:

[0058] Using the future load demand as the load nodes, a preset connection threshold is set. The correlation between different load characteristics is calculated based on the cosine similarity formula, and the correlation is used as the weight of the connected edge. The load nodes with a correlation greater than or equal to the connection threshold are connected by edges to obtain the demand topology graph. Based on the fault message, the fault nodes and the edges directly connected to the fault nodes in the demand topology graph are removed (and the faulty equipment is shut down by the system) to obtain the local demand graph.

[0059] The ways to perform power optimization scheduling on the local demand graph include:

[0060] A preset total load is set. Based on the total load, an initial load is obtained by evenly distributing to each load node in the local demand graph. A preset load perturbation factor is set. Initialize B ants, where B is greater than the number of load nodes in the local demand graph. Each ant is randomly assigned to different graph nodes in the local demand graph. The initial pheromone concentration of each edge in the local demand graph is set to the reciprocal of the edge weight, and the global fitness is initialized to infinitesimal.

[0061] Each ant conducts distributed exploration in the local demand graph, selects the position where the ant will move next based on the initial pheromone concentration of each edge. The ant uses the edge with a high initial pheromone concentration as the moving direction. When the graph node information of each ant changes, record the current position and the previous position of the ant, and update the initial load corresponding to the current position and the previous position based on the load perturbation factor. The load node corresponding to the previous position will reduce the load by the magnitude of the load perturbation factor and allocate it to the load node corresponding to the current position. Take the allocated value as the new initial load of the current position and the previous position, and perform load adaptation calculation on the local demand graph at this time. The formula for performing load adaptation calculation is:

[0062] ; where represents the load fitness, represents the number of load nodes in the local demand graph, represents the balance weight, which is used to balance the proportion of load balance in the load fitness, represents the th initial load of the load node, represents the average load of all load nodes, represents the adaptation weight, which is used to balance the proportion of future load demand in the load fitness. The larger the adaptation weight, the more likely the initial load allocated to this node is to be close to the future load demand, represents the th future load demand of the load node; when the load fitness is less than the global fitness, the ant stops moving and takes the current load fitness as the new global fitness. When the load fitness is greater than or equal to the global fitness, the ant returns to the previous position and withdraws the load allocation. When all ants stop moving, repeat the distributed exploration with the current position of the ant as the initial position until the number of ants on each load node in the local demand graph no longer changes. Output the initial load of each load node in the local demand graph at this time as the optimal scheduling plan.

[0063] In this embodiment, by collecting various data and fully considering the key factors in power supply and distribution, it helps to comprehensively monitor the operation status of equipment; by strengthening the power data, mutation data is effectively removed, the accuracy of the data is improved, the time series characteristics are strengthened, and the system's perception ability of instantaneous fluctuations is optimized; by affecting the state of the load curve graph, real-time load correction based on equipment status and load fluctuations is achieved, effectively coping with complex load change scenarios and improving the accuracy of load monitoring; by quickly issuing an alarm through the fault message module, potential equipment failures are timely addressed, reducing the operation risks of the power supply and distribution system of the distribution network; by constructing a local demand graph, the reliability during subsequent optimized dispatching is ensured; by performing power optimized dispatching on the local demand graph and dynamically adjusting the allocation of each load node, the balance and adaptability of load distribution are ensured, reducing equipment failure rates and energy waste, improving the utilization rate of power resources, and reducing operating costs.

[0064] Embodiment 2

[0065] Please refer to Figure 2 as shown. For the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A power supply and distribution intelligent monitoring method for a distribution network is provided, including:

[0066] S1. Collect power data and status data of distribution network equipment in the same time series, and strengthen the power data to obtain enhanced power data;

[0067] S2. Perform waveform analysis on the enhanced power data to obtain a load curve graph, affect the state of the load curve graph based on the status data to obtain a load correction graph, and extract features from the load correction graph to obtain load characteristics;

[0068] S3. Predict the load characteristics based on the constructed fault prediction model to obtain a fault message, and issue a fault alarm based on the fault message;

[0069] S4. Perform load prediction on the load correction graph to obtain future load demands, construct a demand topology graph based on the future load demands, perform fault cutting on the demand topology graph based on the fault message to obtain a local demand graph, and use an optimization algorithm to perform power optimized dispatching on the local demand graph to obtain an optimal dispatching plan.

[0070] Embodiment 3

[0071] This embodiment publicly provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided power supply and distribution intelligent monitoring method for a distribution network.

[0072] Since the electronic device introduced in this embodiment is the electronic device adopted for implementing an intelligent monitoring method for power supply and distribution in a distribution network in the embodiments of the present application, based on the intelligent monitoring method for power supply and distribution in a distribution network introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail herein. As long as those skilled in the art implement the electronic device adopted for an intelligent monitoring method for power supply and distribution in a distribution network in the embodiments of the present application, it falls within the scope of protection of the present application.

[0073] The above formulas are all calculated by taking the numerical values after dimensionless processing. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0074] The above description is only a preferred embodiment of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An intelligent monitoring system for power supply and distribution in a distribution network, characterized in that: include: Data acquisition and processing module: collects power data and status data of distribution network equipment in the same time series, enhances the power data, and obtains enhanced power data; Waveform analysis module: perform waveform analysis on enhanced power data to obtain a load curve diagram, perform state impact on the load curve diagram based on state data to obtain a load correction diagram, perform feature extraction on the load correction diagram to obtain load characteristics; Fault message module: predicts load characteristics based on the constructed fault prediction model, obtains fault messages, and issues fault alarms based on the fault messages; Optimization and dispatching module: including demand forecasting unit and dispatching unit. The demand forecasting unit performs load forecasting on the load correction diagram to obtain future load demand, builds a demand topology diagram based on future load demand, performs fault cutting on the demand topology diagram based on fault messages to obtain a local demand diagram, and the dispatching unit uses an optimization algorithm to perform power optimization dispatching on the local demand diagram to obtain the best dispatching solution; The method of enhancing the power data includes: Each type of data in the power data is taken as the original data, and the original data is subjected to waveform transformation to obtain a waveform function. A noise threshold is preset, and the waveform function is subjected to waveform smoothing based on the noise threshold to obtain a smoothed value; the maximum and minimum values ​​in the smoothed value are recorded, and the smoothed value at each moment is taken as a neuron data, and enhanced weight learning is performed on the smoothed value based on the neuron data to obtain an enhanced weight; the smoothed value at each moment is enhanced based on the enhanced weight, and the enhanced weight at each moment is multiplied by the smoothed value at the corresponding moment to obtain enhanced data, and all enhanced data constitute enhanced equipment data; The formula for enhancing weight learning is: Among them, w t+1 represents the enhanced weight at time t+1, w t represents the enhanced weight at time t, γ represents the enhanced balance factor, and Org t represents the smoothed value at the tth moment, Lar represents the maximum value of the smoothed value, Min represents the minimum value of the smoothed value, Cha t Represents the original data at time t.

2. The power distribution network intelligent monitoring system according to claim 1, characterized in that: The power data includes: device voltage and device current; the status data includes: device temperature.

3. The power distribution network intelligent monitoring system according to claim 2 is characterized in that: The method of influencing the state of the load curve diagram includes: The spline difference method is used to draw a curve for each type of data in the enhanced power data to obtain the equipment voltage curve and the equipment current curve; the equipment voltage curve and the equipment current curve are load fitted to obtain the power load, and the power load at different times constitutes a load curve; Based on the constructed load state influence model, the state data is predicted to obtain the load influence coefficient, the frequency determination function is preset, the frequency determination of the load influence coefficient is performed based on the frequency determination function, the correction frequency is obtained, and the load correction is performed on the load curve based on the load influence coefficient. The formula for load correction is: Among them, COP t represents the corrected load at time t, σ represents the load offset factor, P t represents the power load at time t, Represents the load influence factor at time t, BD t represents the correction frequency at time t, i represents the i-th correction, and the correction loads at different times constitute the load correction diagram.

4. The power distribution network intelligent monitoring system according to claim 3 is characterized in that: The method of extracting features from the load correction diagram includes: Integrate the load correction graph over the time series span to obtain the total load value, and divide the total load value by the time series span to obtain the average load; perform difference calculation on the corrected loads at consecutive moments of the load correction graph, subtract the corrected load at the previous moment from the corrected load at the next moment to obtain the load momentary change, calculate the standard deviation of the load momentary change, and divide the standard deviation of the load momentary change by the average load to obtain the load fluctuation rate; calculate the load change rate at each moment in the load correction graph, average the load change rates, and obtain the average change rate; preset a sliding time window to capture short-term fluctuations of the load correction graph, slide the sliding time window at the start time in the load correction graph, calculate the standard deviation of the corrected load in each sliding time window, and obtain the window standard deviation; the load characteristic is composed of the mean of the load fluctuation rate, average load, average change rate and window standard deviation.

5. The power distribution network intelligent monitoring system according to claim 4 is characterized in that: The fault prediction model is constructed in the following manner: Collect historical fault message data and historical equipment load characteristic data, use the historical fault message data and historical equipment load characteristic data as the input of the fault prediction model, use the predicted fault message as the output of the fault prediction model, use the historical fault message as the prediction target of the fault prediction model for model training, and build a fault prediction model. The fault prediction model is a linear regression model. The objective loss function of the fault prediction model is the recall rate function. Minimizing the value of the objective loss function is used as the training goal, and a fault prediction model with the smallest value of the objective loss function is obtained.

6. The power distribution network intelligent monitoring system according to claim 5, characterized in that: The construction method of the local demand graph includes: The load correction graph is predicted based on the trained load forecasting model to obtain the future load demand. The future load demand is used as the load node, and the connection threshold is preset. The correlation between different load characteristics is calculated based on the cosine similarity formula. The correlation is used as the weight of the connected edge, and the load nodes with a correlation greater than or equal to the connection threshold are connected to obtain the demand topology graph. Based on the fault message, the faulty nodes and the edges directly connected to the faulty nodes in the demand topology graph are removed to obtain the local demand graph.

7. The intelligent monitoring system for power supply and distribution of distribution network according to claim 6, characterized in that: The method of performing power optimization dispatch on the local demand graph includes: Preset the total load, and evenly distribute each load node in the local demand graph based on the total load to obtain the initial load and preset the load disturbance factor; initialize B ants, and randomly assign each ant to a different graph node in the local demand graph. Initialize the initial pheromone concentration of each edge in the local demand graph to the inverse of the edge weight, and initialize the global fitness to infinitesimal; Each ant performs allocation exploration in the local demand graph, and selects the position to which the ant will move next based on the initial pheromone concentration of each edge. The ant takes the edge with high initial pheromone concentration as the moving direction. When the graph node information of each ant changes, the current position and the previous position of the ant are recorded, and the initial load corresponding to the current position and the previous position is updated based on the load disturbance factor. The load node corresponding to the previous position will reduce the load by the size of the load disturbance factor and allocate it to the load node corresponding to the current position. The allocated value is used as the new initial load of the current position and the previous position, and the load adaptation calculation is performed on the local demand graph at this time to obtain the load fitness; when the load fitness is less than the global fitness, the ant stops moving, and the load fitness at this time is used as the new global fitness. When the load fitness is greater than or equal to the global fitness, the ant returns to the previous position and retracts the load allocation. When all ants stop moving, the current position of the ant is used as the initial position allocation and the allocation exploration is repeated until the number of ants on each load node in the local demand graph no longer changes, and the initial load of each load node in the local demand graph at this time is output as the optimal scheduling solution.

8. The power distribution network intelligent monitoring system according to claim 7, characterized in that: The formula for load adaptation calculation is: Among them, Fit represents the load fitness, All represents the number of load nodes in the local demand graph, b1 represents the balance weight, and L a represents the initial load of the ath load node, Lavg represents the average load of all load nodes, b2 represents the adaptation weight, Lned a Represents the future load demand of the ath load node.

Images