Arc control device under coupling interference condition of multi-source load system and configuration method

Through the method based on TT transformation and deep reinforcement learning, the characteristics of fault arcs in multi-source load systems are extracted and the configuration of arc prevention and control devices are optimized, which solves the problem of fault arc detection in multi-source load systems, improves detection accuracy and coverage, and ensures that The safe and efficient operation of the power system.

CN118920427APending Publication Date: 2024-11-08XIAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411033949.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art cannot effectively detect and locate fault arcs under complex coupling conditions in multi-source load systems, resulting in low accuracy of fault detection and the inability to achieve full coverage arc prevention and control, which poses potential hidden dangers.

Method used

Adaptive detection is performed by using feature extraction methods based on TT transformation and improved sparrow algorithm, combined with the DQN model of deep reinforcement learning, optimize the configuration of arc prevention and control device, and by establishing a simulation model of the multi-source charge system, multi-source arc data are obtained and fault characteristics are extracted. , and the deployment plan of the arc prevention and control device is determined through an optimization algorithm.

Benefits of technology

It improves the accuracy and coverage of fault arc detection, solves the problem of fault arc detection in multi-source charge systems, enhances the coordination and coverage of arc prevention and control systems, and ensures the safety and efficient operation of the power system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118920427A_ABST
    Figure CN118920427A_ABST
Patent Text Reader

Abstract

The invention discloses an electric arc prevention and treatment device under a multi-source load system coupling interference condition and a configuration method, and the method comprises the steps: simulating an evolution process of an electric arc at a multi-source position through constructing a multi-source load system model, and extracting the characteristics of the electric arc through the lossless amplification of an electric arc high-frequency signal through TT transformation; when arc time-frequency characteristics are weakened, an improved sparrow algorithm is used for optimizing a maximum correlation kurtosis deconvolution method to improve the arc signal-to-noise ratio, then deep reinforcement learning is used for improving the generalization ability of a detection model, and finally an optimization algorithm is used for obtaining a simplest configuration scheme considering the input cost of an arc detection device and the protection coverage range. Different system modules are updated and added on the basis of an original single system, the applicability and coverage of a simulation model can be effectively expanded, and the problems of fault arc detection, positioning and protection caused by uncertainty and strong coupling of a multi-source load system are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical fault detection, and in particular relates to an arc prevention device and a configuration method under conditions of coupled interference of a multi-source load system. Background Art

[0002] In order to reduce carbon emissions, new infrastructure such as charging piles, energy storage, and distributed photovoltaic integrated stations have emerged rapidly, focusing on building a clean, low-carbon, safe and efficient energy system, improving the level of clean energy utilization and the efficiency of power system operation, implementing the new development concept, and better playing the role of multi-source load system and multi-energy complementarity in ensuring energy security. Some people have proposed "improving the level of electrification of building terminals, and building 'light storage direct flexible' buildings that integrate photovoltaic power generation, energy storage, DC power distribution, and flexible power consumption", including low-carbon parks, zero-carbon civil buildings, etc. In the case of severe shortage of flexible power regulation resources, effectively improving the level of interaction between power supply and demand is a means to ensure system safety and a high proportion of new energy consumption.

[0003] The construction of a multi-source-load system and the development of multi-energy complementarity are the inherent requirements of the power industry to adhere to the system concept, the objective need to achieve high-quality development of the power system, and the inevitable choice to improve the level of renewable energy development and consumption and the proportion of non-fossil energy consumption. It is of great significance to promote my country's energy transformation and economic and social development. In recent years, multi-source-load systems have continued to develop, and photovoltaic storage and charging systems, diversified photovoltaic storage and direct-flexible systems, etc. are constantly realizing the efficient use of renewable energy in zero-carbon buildings. However, due to the randomness of the time and quantity of these source loads, the morphology and characteristics of the fault arc have changed, which increases the difficulty of arc detection. At the same time, there is a lack of relevant measures for efficient arc disposal, and the safety of electricity use caused by arc-induced fires needs to be urgently addressed.

[0004] In recent years, with the advancement of science and technology, optics, computers, and image processing technologies have begun to be applied to the study of fault arcs, providing powerful tools for the study of fault arcs. Combined with the latest intelligent algorithm analysis and computer real-time information processing, the research on fault arcs has become more and more mature and complete. At present, the main methods used in foreign research on DC fault arcs include time domain analysis, frequency domain analysis, wavelet analysis, neural networks, hidden Markov models, etc. In addition, there are certain studies on DC arc volt-ampere characteristics, three-dimensional models, arc-starting methods and devices. However, most of the construction of fault arc simulation models in China is a single system, which cannot solve the simulation of arc access in complex scenarios of multi-source load systems, so it is impossible to form multi-source arc data. Secondly, facing the complex working conditions of the actual power grid, the original time domain, frequency domain, and STFT features have the general characteristics of low anti-arc capability. Wavelet feature extraction will have the disadvantage of a significant decrease in diagnosis rate in special scenarios, poor adaptability to multiple working conditions, and weakening of features in some scenarios, which brings challenges to the extraction of fault arc features and the improvement of detection accuracy. In terms of detection algorithms, the existing time-frequency domain threshold detection methods set different thresholds in different systems, are easily disturbed by working conditions, and are not suitable for fault location in large areas and open spaces. Pattern recognition detection methods such as artificial neural networks and support vector machines have complex algorithm optimization, interference elimination, and model establishment problems, resulting in the inability to improve the accuracy of fault arc detection. In terms of arc prevention device configuration, at present, power system fault detection mainly relies on traditional monitoring technologies and equipment, such as sensors and relays. These methods can achieve fault detection and location to a certain extent, but traditional power system fault detection methods may not be able to achieve effective detection on all power grid nodes, resulting in some potential fault hazards being ignored, so it is necessary to propose the simplest configuration of the device.

[0005] In order to ensure the good development of multi-source load system, the current fault arc detection technology cannot be applied to the complex coupling system of multi-source load, and the current arc positioning method at home and abroad cannot meet the problem of accurate positioning and efficient disposal of fault arc in long distance range. This patent will establish a fault arc model of a multi-source load system, namely, a photovoltaic storage and charging integrated system, and propose an arc prevention device and its configuration method under the coupling interference condition of the multi-source load system. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides an arc prevention device and a configuration method thereof under the conditions of coupled interference of a multi-source load system, which can improve the DC arc prevention effect in the multi-source load coupled interference system with minimal investment in protection equipment, thereby solving the problems of fault arc detection, positioning and protection caused by the uncertainty and strong coupling of the multi-source load system.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is: a method for configuring an arc prevention device under the condition of coupled interference of a multi-source load system, comprising the following steps:

[0008] Step a, obtain multi-source arc data; according to the frequency f s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is also measured;

[0009] Step b, sampling the original current at equal intervals based on TT transformation, obtaining a 3D graph of the TT sequence obtained by sampling before and after the fault, detecting the fault by judging the current signal amplitude near the main diagonal, extracting the fault arc characteristics, and going to step c after detecting the fault according to the fault arc characteristics, otherwise going to step a;

[0010] Step c: if the fault arc feature is weakened, the improved sparrow algorithm is used to optimize the maximum correlation kurtosis deconvolution algorithm to enhance the feature and go to step d; otherwise, go directly to step d;

[0011] Step d, using the DQN model based on deep reinforcement learning to adaptively detect DC fault arcs, and by testing different working conditions to obtain a fault detection accuracy that meets the requirements, go to step e. If the fault detection accuracy does not meet the requirements, update the model parameters and retest until the fault detection accuracy meets the requirements;

[0012] Step e, taking the target distribution network that needs to be equipped with arc prevention devices as the research object, establishing the optimization objective function and constraint conditions based on the coverage effect and deployment quantity of arc prevention equipment in the power system, using the optimization algorithm to solve the optimization objective function, and obtaining the arc prevention device configuration scheme under the condition of multi-source load system coupling interference;

[0013] Step f, deploy arc prevention devices and analyze the coverage effect; if the coverage effect is achieved, write the DQN model based on deep reinforcement learning into the arc prevention device system to configure the arc prevention device; if the coverage effect is not achieved, determine whether the installation is restricted. If the installation is restricted, appropriately increase the number of nodes and finally select the optimal configuration scheme among the non-unique configuration schemes. Otherwise, go to step e to optimize the algorithm or adjust the expected optimization target.

[0014] Furthermore, obtaining multi-source arc data includes: setting different fault arc occurrence points so that multi-source faults occur simultaneously in a system, thereby forming multi-source arc data.

[0015] Furthermore, based on the improved sparrow algorithm to optimize the maximum correlation kurtosis deconvolution parameters, the three parameters [L, T, M] are optimized to obtain the best maximum correlation kurtosis deconvolution parameters, and the SCSSA-MCKD algorithm is used to enhance and widen the pulse during the fault and amplify the fault characteristics after the fault arc occurs.

[0016] Furthermore, the SCSSA-MCKD algorithm is used to enhance and widen the pulse during the fault, and the fault characteristics after the fault arc occurs are amplified, including:

[0017] Decompose the signal collected in step a by VMD or EMD, calculate the envelope spectrum peak factor of each IMF component, and the IMF component with the largest envelope spectrum peak factor is the optimal IMF component;

[0018] The three parameters of the maximum correlation kurtosis deconvolution are optimized to obtain the enhanced IMF component; finally, the envelope spectrum analysis of the IMF component is performed to obtain the fault characteristic frequency.

[0019] Furthermore, the DQN model based on deep reinforcement learning is used to adaptively detect DC fault arcs. By testing different working conditions, the fault detection accuracy meets the requirements and goes to step e. If the fault detection accuracy does not meet the requirements, the model parameters are updated and retested until the fault detection accuracy meets the requirements, including:

[0020] The fault feature data is normalized to the maximum and minimum values, and the holdout method is used to divide the data set into a training set S and a test set T according to randomness. The training set S is used as the input set of the deep reinforcement learning DQN algorithm. The model is trained by the agent interacting with the environment. The trained model is tested with the test set T to verify the model performance. The accuracy is determined by comparing and matching the output results with the actual label set of the test set. If the accuracy of the test set is low, the parameters in the deep reinforcement learning DQN model are adjusted, and then the features under the new working condition are processed in the same way and input into the model for fault diagnosis to obtain the accuracy under the new working condition. If the accuracy is low, the parameters in the deep reinforcement learning DQN model are adjusted, and the old working condition during training is retested using the updated model until the fault detection accuracy under both the new and old working conditions meets the requirements.

[0021] Furthermore, the structure and key monitoring areas of the power system are determined, and arc prevention devices are deployed on node i. 1 means deployment, and 0 means no deployment. A number of n arc prevention devices are deployed on the nodes of the power system, and the detection range is r. The maximum coverage range is defined as the set of nodes within a circle with a radius of r and centered on node i, including i. The objective function is defined as the coverage rate of the area covered by the formula:

[0022]

[0023] Where: m is the number of nodes that the arc protection device can cover, i is the value of each node when deployed,

[0024] The constraints include: node quantity constraint: the number of arc prevention devices cannot exceed the number of nodes in the power system; detection range constraint: the arc prevention device must cover key nodes within its detection range; resource limitation: consider quantity limitation.

[0025] Furthermore, the present invention can also improve an arc prevention device under multi-source load system coupling interference conditions, which is obtained based on the above configuration method.

[0026] Based on the concept of the above method, the present invention also provides a simulation configuration system for arc prevention devices under multi-source load system coupling interference conditions, including a data acquisition module, a feature extraction module, a feature enhancement module, a fault detection module, a simulation configuration module and an optimization module;

[0027] The data acquisition module is used to acquire multi-source arc data; according to the frequency f s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is also measured;

[0028] The feature extraction module takes equidistant sampling of the original current based on TT transformation, obtains the TT sequence three-dimensional graph by sampling before and after the fault, detects the fault by judging the current signal amplitude near the main diagonal, extracts the fault arc characteristics, and transfers to the feature enhancement module for feature enhancement after the fault is detected according to the fault arc characteristics, otherwise it transfers to the multi-source arc data of the data acquisition module;

[0029] The feature enhancement module is used to optimize the maximum correlation kurtosis deconvolution algorithm using the improved sparrow algorithm to enhance the feature when the fault arc feature is weakened and transfer it to the fault detection module, otherwise it is directly transferred to the fault detection module for detection;

[0030] The fault detection module uses the DQN model based on deep reinforcement learning to perform adaptive detection of DC fault arcs. By testing different working conditions, the fault detection accuracy meets the requirements and goes to step e. If the fault detection accuracy does not meet the requirements, the model parameters are updated and retested until the fault detection accuracy meets the requirements.

[0031] The simulation configuration module takes the target distribution network that needs to be configured with arc protection devices as the object, establishes the optimization objective function and constraint conditions based on the coverage effect and deployment quantity of arc protection equipment in the power system, and uses the optimization algorithm to solve the optimization objective function to obtain the configuration of arc protection devices under the conditions of multi-source load system coupling interference;

[0032] The optimization module is used to deploy arc prevention devices and analyze the coverage effect. If the coverage effect is achieved, the DQN model based on deep reinforcement learning is written into the arc prevention device system to configure the arc prevention device. If the coverage effect is not achieved, it is determined whether the installation is restricted. If the installation is restricted, the number of nodes is appropriately increased, and finally the optimal configuration scheme among the non-unique configuration schemes is selected. Otherwise, the simulation configuration module is turned to optimize the algorithm or adjust the expected optimization target.

[0033] The present invention also provides a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the arc prevention device configuration method under the multi-source load system coupling interference condition described in the present invention.

[0034] A computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, the method for configuring an arc prevention device under the coupling interference condition of a multi-source-load system of the present invention can be implemented.

[0035] Compared with the prior art, the present invention has at least the following beneficial effects:

[0036] (1) The new idea of ​​the multi-source load system simulation model proposed in this invention can effectively expand the applicability and coverage of the simulation model by updating and adding different system modules on the basis of the original single system, and solve the problem that the interaction process between the uncertainty of multi-source load access and the randomness of fault occurrence is difficult to simulate. The simulation can provide data, provide reliable arc feature evolution mode simulation for analyzing the internal mechanism of fault arc, and greatly improve the success rate of fault arc detection.

[0037] (2) The DC fault arc feature extraction method based on TT transform proposed in the present invention has time-frequency atoms with a small inner product with the arc signal. The diagonal elements formed by decomposition can extract the DC fault arc signal features to the maximum extent, greatly improving the fault diagnosis accuracy, solving the problem of decreased fault feature extraction diagnosis rate caused by multi-source load interference, and providing accurate fault arc feature construction technology for complex working conditions existing in actual power grids, thereby protecting people’s lives and property.

[0038] (3) The improved sparrow algorithm proposed in the present invention optimizes the maximum correlation kurtosis deconvolution. By performing multi-parameter optimization on the basis of the original parameters to ensure that the optimal MCKD parameters are found, the time pulse duration scale of the fault characteristics can be widened, and the fault characteristics after the arc occurs can be amplified, thereby further improving the signal-to-noise ratio of the fault arc detection signal and solving the problem that the weak signal of the fault arc is submerged in the strong coupling noise condition of the system.

[0039] (4) The DC fault arc adaptive detection method based on deep reinforcement learning DQN algorithm proposed in the present invention significantly improves the model's adaptability and strong generalization capabilities by combining the perception ability of deep learning with the decision-making ability of reinforcement learning. It solves the problem that the threshold setting of the time-frequency domain threshold detection method is difficult in different systems and is easily affected by multiple factors. At the same time, it makes up for the deficiency that the existing detection algorithm cannot provide complete fault location information in a large range and open space.

[0040] (5) The method for the simplest configuration of arc prevention equipment based on the greedy algorithm proposed in the present invention forms a simplest protection paradigm for the arc prevention system by optimizing the configuration of arc prevention equipment in a multi-source power system. This can effectively enhance the coverage and coordination of the prevention system, thereby achieving full-link prevention of arc hazards, solving the contradiction between the number of prevention equipment and the protection range in the power system, and greatly improving the fault prevention efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the present invention;

[0042] Figure 2a It is a simulation model for arc faults with coupling interference in multi-source load systems;

[0043] Figure 2b This is the current and wavelet characteristic waveform of the energy storage system when the photovoltaic system fails;

[0044] Figure 2c This is the current and wavelet characteristic waveform of the energy storage system when the charging pile system fails;

[0045] Figure 2d This is the current and wavelet characteristic waveform of the energy storage system when the photovoltaic system and the charging pile system fail at the same time;

[0046] Figure 3a This is the arc current waveform diagram under actual working conditions;

[0047] Figure 3b It is the TT sequence three-dimensional diagram of arc current;

[0048] Figure 3c The characteristic output result constructed by the diagonal modulus matrix of the TT transformation;

[0049] Figure 4a It is the current and wavelet characteristic waveform diagram of the coupling interference arc of large current, aluminum cable and inverter multi-source load system;

[0050] Figure 4b The effect diagram of arc feature enhancement for SCSSA-MCKD algorithm;

[0051] Table 1 shows the comparison of characteristic significance before and after arc enhancement under the influence of different source load factors;

[0052] Figure 5a The deep learning model is used to identify the interference fault of the new energy photovoltaic-rice cooker multi-source load system;

[0053] Figure 5b The deep learning model is used to identify the interference fault of the new energy photovoltaic-humidifier multi-source load system;

[0054] Figure 6 Improved fault identification model for deep reinforcement learning;

[0055] Figure 7a The identification results of interference faults in the new energy photovoltaic-rice cooker multi-source load system using deep reinforcement learning;

[0056] Figure 7b The identification results of interference faults in the new energy photovoltaic-humidifier multi-source load system using deep reinforcement learning;

[0057] Figure 8 This is a comparison chart of the effect of deep reinforcement learning on improving the generalization ability of arc identification;

[0058] Fig. 9 Schematic diagram of the IEEE 33-node distribution network connected to the photovoltaic storage and charging system (part);

[0059] Fig.10a The protection coverage of 8 arc protection devices;

[0060] Fig.10b The protection coverage of 23 arc protection devices;

[0061] Fig.10c Description of the system installation locations for the simplest configuration of the 8 selected arc protection devices; DETAILED DESCRIPTION

[0062] The present invention is described in detail below in conjunction with the accompanying drawings and embodiments. The embodiment of an arc prevention device under the condition of multi-source-load system coupling interference and its configuration method are only used to explain the present invention, but not to limit the protection scope of the present invention.

[0063] Step 1: In a multi-source load system, due to multiple factors interfering with the arc characteristics, the arc morphology characteristics change. Different fault arc occurrence points are set in the multi-source load system to make multiple source faults occur simultaneously in one system and generate multi-source arc data. s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is used for comparative analysis;

[0064] Step 2: Use TT transformation to perform signal processing and analysis on the fault current waveforms of different working conditions obtained in step 1. Use the lossless nature of TT transformation and the high-frequency signal amplification to extract features. For this, use the equidistant sampling method (sampling interval is D) to fully cover the sampling points, extract the time range of the fault, perform equidistant sampling within the range, and obtain a three-dimensional graph of the TT sequence obtained by sampling before and after the fault. The current characteristic signal is concentrated near the main diagonal. After a certain time point, the amplitude change is no longer regular and becomes chaotic with a large amplitude increase or decrease. Determine whether a fault arc fault occurs through feature extraction. If no fault arc occurs, return to step 1 to re-collect the signal. If a fault arc occurs, go to step 3.

[0065] Step 3: Optimize the maximum correlation kurtosis deconvolution (MCKD) parameters based on the improved sparrow algorithm; optimize the three parameters [L, T, M] to ensure that the best MCKD parameters are found. The SCSSA-MCKD algorithm can not only enhance and widen the pulse during the fault, but also amplify the fault characteristics after the fault arc occurs. The enhancement effect is obvious and can meet the requirements.

[0066] Step 4: Use the DQN algorithm based on deep reinforcement learning to perform adaptive detection of DC fault arcs. Use the maximum and minimum normalization method to normalize the features extracted by TT transformation, and limit the size of the features extracted by TT transformation to [0,1]. Then use the holdout method to divide the data into training set and test set according to 7:3. Use the training set as the input set to train the deep reinforcement learning network and obtain the fault diagnosis model. After obtaining the fault diagnosis model, use the test set as the input set to perform performance testing on the fault diagnosis model to ensure that the fault detection accuracy under the old working conditions meets the requirements, and then update it to the new working conditions to perform fault arc detection; output the time and location information of the fault arc occurrence for coordinated action.

[0067] Step 5: Taking the IEEE33 node distribution network as the research object, determine the optimization objectives and constraints of the power system, and select the optimization algorithm. Considering factors such as simple calculation and rich usage scenarios, the greedy algorithm is selected to design the solution. Based on the local optimal selection, the current optimal solution is selected at each step in the hope of obtaining the global optimal solution. Based on the optimization algorithm, debugging simulation is carried out, the deployment and coverage effect of arc prevention devices are analyzed for the research object, the optimization objective function is established based on the coverage effect and deployment number of arc prevention equipment in the power system, and the optimization algorithm is applied to solve the multi-objective function. The optimization results are analyzed to determine whether the expected optimization goal has been achieved, that is, the number of arc prevention devices used is the least under the premise of full coverage, the individual cost is low, and the efficient collaborative configuration between devices can be achieved to maximize the operating efficiency. If the expected goal is not achieved, it is determined whether the installation is limited. If the installation is limited, the number of nodes is appropriately increased to obtain a non-unique configuration plan, and the optimal configuration plan is selected from them. Otherwise, the optimization algorithm is reselected or the optimization goal is adjusted.

[0068] Figure 2a It is a simulation model of the photovoltaic storage and charging system built on the Matlab / Simulink simulation platform. The photovoltaic storage and charging system is divided into four parts: photovoltaic system, energy storage system, DC charging pile, and grid-connected system. The photovoltaic system is set as single fault point 1, the charging pile system is set as single fault point 2, and the photovoltaic system and the charging pile system fail at the same time as a double fault point. Figure 2b This is the current waveform of the energy storage system after a single fault arc occurs in the photovoltaic system and the waveform after wavelet transformation. The current of the energy storage system rises sharply 0.5s after the fault occurs, providing energy to the system. It is judged that the photovoltaic system power cannot support the sum of the grid-connected power and the power required by the charging pile system due to the failure of the photovoltaic system. At this time, the energy storage system plays a role. Judging from the perspective of wavelet transformation, the wavelet amplitude changes from decreasing to increasing. It can be seen that the introduction of arcs in the photovoltaic system will transmit high-frequency noise to the energy storage system, which has a significant impact on the energy storage system.

[0069] like Figure 2c The figure shows the current waveform of the energy storage system and the waveform after wavelet transformation after a single fault arc occurs in the charging pile system. After the fault arc occurs 0.5s, the current has a small fluctuation and basically remains at 500A, and the system works normally. According to the interpretation and analysis of wavelet transformation, the amplitude of the wavelet waveform remains stable and basically does not change. It can be seen that the fault arc on the primary side of the charging pile system transformer has no significant impact on the energy storage system, indicating that the fault point cannot be directly determined from the current waveform. It is necessary to extract more significant features of the fault arc to locate and prevent the fault arc.

[0070] like Figure 2dThe figure shows the current waveform of the energy storage system and the waveform after wavelet transformation after the photovoltaic system and the charging pile system have a double fault. The current of the energy storage system rises sharply 0.5s after the fault occurs, providing energy to the system. Judging from the wavelet transformation, the wavelet amplitude changes from decreasing to increasing. It can be seen that the introduction of arc in the photovoltaic system will transfer high-frequency noise to the energy storage system, and the noise is more significant. Figure 2b and Figure 2c By comparison, it is found that the current waveform of the energy storage system is similar to that of the photovoltaic system when a single fault occurs, but the oscillation is more intense. It is speculated that the fault arc in the photovoltaic system is dominant, and the charging pile system has an impact on the system but it is relatively weak. Since the detection point is in the energy storage system, which is close to the photovoltaic system, it is speculated that the impact of the fault on the system is also related to the detection location.

[0071] The results of signal analysis of current data under actual working conditions are as follows: Figure 3a As shown in the figure, it can be seen that the current amplitude suddenly drops at a certain moment. The results show that a DC fault arc occurs during the operation of the electrical equipment under actual working conditions. The commonly used fault feature extraction methods currently have the disadvantages of low accuracy and high cost, and the diagnostic accuracy of some feature extraction methods is significantly reduced under complex working conditions. Therefore, the existing DC fault feature extraction methods still have certain limitations, and the effectiveness, accuracy and stability of DC fault arc diagnosis need to be improved urgently.

[0072] The original current is processed and analyzed by TT transform (Trapezoidal Transform), and its features are extracted by utilizing the lossless nature of TT transform and the high-frequency signal amplification. The sampling points are fully covered by the equidistant sampling method (sampling interval is 100 counting points). Figure 3a The time range of the fault occurrence is extracted from the results, and equidistant sampling is performed within the range. The TT sequence three-dimensional graph obtained by sampling before and after the fault is as follows Figure 3b As shown in the figure, it can be seen that the current characteristic signals are concentrated near the main diagonal. After a certain time point, the amplitude change is no longer regular, and becomes chaotic with large amplitude increases and decreases. This time point is the moment when the fault occurs. The results prove the feasibility of TT transformation in fault arc detection technology.

[0073] The diagonal modulus matrix obtained by extracting the features of the original current using the DC fault extraction method based on TT transformation is as follows: Figure 3cAs shown in the figure, it can be clearly seen that after the DC fault arc occurs in the line, the characteristic signal is significantly enhanced and the amplitude increases sharply. It is calculated that the feature extraction of the fault arc using TT transformation has the characteristics of low cost and high accuracy compared with other feature extraction methods, and its accuracy reaches 99.36%. The results show that when a DC fault arc occurs in the line, the DC fault arc feature extraction method based on TT transformation can accurately obtain the time when the fault arc occurs, so as to timely eliminate electrical accidents caused by the fault arc, provide important support for power safety, and protect the lives and property of the people.

[0074] Figure 4a The figure in the figure shows the characteristics of the fault arc data after wavelet extraction. The red F is the data diagram after wavelet feature extraction, and the blue I is the original fault arc current data diagram. The first stage is the normal circuit stage, and the second stage is the stage where the fault arc occurs. The fault arc under the influence of multiple factors such as large current, aluminum electrode and inverter will not have obvious pulsation when it occurs, which is difficult to detect during manual detection and the detection accuracy is not high. After adding Rbio3.1 wavelet analysis, the characteristics of the DC fault arc have been improved, which shows that Rbio3.1 wavelet analysis extracts more significant characteristics of the fault arc. The average current output in the fault arc section is compared with the normal current average value, which is the improvement ratio. After calculation, T=1.15, which is greater than 1, so it is determined that the fault characteristics of the single inverter are obvious after wavelet extraction, but it is significantly reduced compared with 1.56 for a single inverter, and slightly reduced compared with 1.19 for long lines and inverters.

[0075] Figure 4b In order to use the fault arc data under the influence of large current, aluminum electrode and inverter, SCSSA-MCKD operation is performed after wavelet transformation and compared with Figure 4a By comparison, we can see that the SCSSA-MCKD algorithm can not only enhance and widen the pulse during the fault, but also amplify the fault characteristics after the fault arc occurs. The enhancement effect is obvious and can meet the requirements. Through the T ratio calculation, it can be obtained that the T ratio after wavelet transform is 1.15, and after the SCSSA-MCKD algorithm calculation, the T ratio can reach 1.99, which is significantly improved compared with before. Table 1 shows the comparison of the enhanced feature ratio T under different system influencing factors. It can be seen that the improved sparrow algorithm optimizes the maximum correlation kurtosis deconvolution (SCSSA-MCKD) algorithm, and the enhancement effect is obvious.

[0076] Table 1 Analysis of fault arc feature enhancement effect under multi-system factor coupling interference conditions

[0077] System Influencing Factors Long lines and inverters High current, aluminum cables and inverters Before enhancement 1.19 1.15 After enhancement 1.49 1.89 Significant improvement percentage 25.2% 64.3%

[0078] from Figure 5aIt can be seen from the figure that when a fault arc occurs, the fault arc adaptive detection device can quickly and accurately determine that a fault arc has occurred in the specific working condition of the new energy photovoltaic-rice cooker.

[0079] from Figure 5b It can be seen that the results predicted by the convolutional neural network have serious misjudgments and jumps under fault conditions. If it runs in the equipment, it will cause the circuit breaker to trip incorrectly, affecting the power quality and safe operation of the power system.

[0080] Figure 6 The fault detection model of the DQN algorithm based on deep reinforcement learning is first obtained through sensor acquisition, and then the feature extraction is performed through the transformation method of time domain and frequency domain, and the obtained time-frequency features are normalized by the maximum and minimum standardization processing method, and the processed values ​​are limited to [0,1]. Then, the data volume is divided into training set and test set according to 7:3 by the retention method, and the training set is used as the input set to train the deep reinforcement learning network, and finally the fault diagnosis model is obtained. After obtaining it, the test set is input into the fault diagnosis model for model performance testing to ensure the high accuracy of fault detection under the old working conditions, and then updated to the new working conditions for fault arc detection.

[0081] from Figure 7a It can be seen that when a fault arc occurs, the fault detection model obtained by the DQN algorithm based on deep reinforcement learning can quickly and accurately determine that a fault arc has occurred in the specific working condition of the new energy photovoltaic-rice cooker.

[0082] from Figure 7b It can be seen that the result of deep reinforcement learning prediction determines the fault time with a delay. Compared with the result of CNN fault diagnosis, it improves the power quality of the power system operation and ensures the operation of the load more reliably.

[0083] Figure 8 It means that under three different new working conditions, the fault detection model based on the DQN algorithm of deep reinforcement learning has greatly improved the accuracy compared with the fault detection model based on the CNN convolutional neural network, and has improved the fault detection accuracy. It can be applied to faults under different working conditions and has high adaptability and strong generalization ability.

[0084] Fig. 9The IEEE33 node distribution network simulation model is built based on Simulink, and the photovoltaic storage and charging system simulation model is inserted to verify the feasibility of the photovoltaic storage and charging system in the distribution network. The figure shows some example diagrams after the photovoltaic storage and charging system is inserted into the IEEE33 node distribution network model. I is the DC charging pile system, II is the energy storage system, and the I system and II system act as the load in the IEEE33 node distribution network model. In Simulink modeling, the IEEE33 distribution network is modeled at a ratio of 1:10, and the distance between the lines of each node in the original distribution network is proportionally reduced to 1 / 10 of the original size, which can more accurately reflect the topological structure and characteristics of the original distribution network. Since the arc prevention device has a coverage radius of 80m as required by UL1699B, the coverage radius of the arc prevention device is set to 800m according to the ratio of 1:10 in this modeling.

[0085] The present invention aims at the full coverage problem in the power system, realizes efficient collaborative configuration between devices, takes the IEEE33-node distribution network as the research object, and proposes a solution based on a greedy algorithm. The greedy algorithm is an algorithm based on local optimal selection, which selects the current optimal solution at each step in the hope of obtaining the global optimal solution. In the algorithm described in the present invention, each arc prevention device is selected according to the farthest node that is not currently covered, and gradually covers more nodes until all nodes are covered. By deploying multiple arc prevention devices at the nodes in the distribution network, the nodes are covered to the greatest extent possible with the least number of arc prevention devices, thereby realizing the simplest configuration for full coverage of the power system.

[0086] Fig.10a The number of arc prevention devices is set to 8, and 4 of them are selected, as shown in the figure for the coverage of each node (solid). It can be observed that the full coverage of the nodes can be basically met. The number of arc prevention devices used under the premise of full coverage is the least, and is less than the number of 33 nodes in the IEEE system. However, if the number of arc prevention devices is not enough, the coverage rate will be low, which indirectly reflects the decline in the full coverage rate.

[0087] Fig.10b The number of arc prevention devices is set to 23, and 4 of them are selected. The figure shows the coverage of each node (solid). It can be observed that when the number of arc prevention devices is 23, it can basically meet the full coverage of nodes, and is less than the 33 nodes of the IEEE system. However, this does not meet the requirements of the simplest configuration. Under the premise of full coverage, the number of arc prevention devices used is large, which not only has a high individual cost, but also cannot achieve efficient collaborative configuration between devices, resulting in poor operating efficiency.

[0088] During the implementation process, in order to maximize the economic efficiency and stability of the distribution network operation, we considered key factors such as the number of arc prevention devices and the deployment location of the arc prevention devices. The full coverage requirement is: by deploying multiple arc prevention devices at the nodes in the distribution network, the nodes are covered to the greatest extent with the minimum number of arc prevention devices, thereby achieving the simplest configuration for full coverage of the power system.

[0089] Fig.10c This is the installation position of the equipment full coverage detection device when the number of arc prevention devices is set to 8, and this is the simplest configuration (hollow) to achieve full coverage of the equipment. The full coverage rate changes linearly with the increase in the number of arc prevention devices. When the number of arc prevention devices reaches 23, the system can be basically fully covered. In actual engineering applications, it is necessary to consider the limiting factors brought about by terrain structure, installation area, etc. For example, if the geographical location of a certain area is in a remote mountainous area, an arc prevention device must be installed there. Doing so may not meet the requirements of the simplest configuration and cannot achieve efficient collaborative configuration between equipment. Therefore, it is necessary to consider the situation when the investment is more than the number of devices that achieve the simplest configuration.

[0090] Table 2 Comparison of coverage and full coverage under different individual quantity costs

[0091]

[0092]

[0093] After many experiments, debugging and data analysis, the key results are summarized in Table 2: In the equipment full coverage test, when the number of detection devices was set to 7, the system failed to achieve full coverage. Subsequently, by increasing the number of detection devices to 8, we observed full coverage of the equipment, and this configuration was confirmed to be the simplest solution to achieve full equipment coverage. In addition, through statistical analysis of the full coverage rate under different numbers of detection devices, we found that there is a linear growth relationship between the full coverage rate and the number of detection devices. Specifically, when the number of detection devices increases to 23, regardless of the configuration scheme, the system can basically achieve full coverage.

[0094] The present invention provides a simulation configuration system for arc prevention devices under multi-source load system coupling interference conditions, including a data acquisition module, a feature extraction module, a feature enhancement module, a fault detection module, a simulation configuration module and an optimization module;

[0095] The data acquisition module is used to acquire multi-source arc data; according to the frequency f s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is also measured;

[0096] The feature extraction module takes equidistant sampling of the original current based on TT transformation, obtains the TT sequence three-dimensional graph by sampling before and after the fault, detects the fault by judging the current signal amplitude near the main diagonal, extracts the fault arc characteristics, and transfers to the feature enhancement module for feature enhancement after the fault is detected according to the fault arc characteristics, otherwise it transfers to the multi-source arc data of the data acquisition module;

[0097] The feature enhancement module is used to optimize the maximum correlation kurtosis deconvolution algorithm using the improved sparrow algorithm to enhance the feature when the fault arc feature is weakened and transfer it to the fault detection module, otherwise it is directly transferred to the fault detection module for detection;

[0098] The fault detection module uses the DQN model based on deep reinforcement learning to perform adaptive detection of DC fault arcs. By testing different working conditions, the fault detection accuracy meets the requirements and goes to step e. If the fault detection accuracy does not meet the requirements, the model parameters are updated and retested until the fault detection accuracy meets the requirements.

[0099] The simulation configuration module takes the target distribution network that needs to be configured with arc protection devices as the object, establishes the optimization objective function and constraint conditions based on the coverage effect and deployment quantity of arc protection equipment in the power system, and uses the optimization algorithm to solve the optimization objective function to obtain the configuration scheme of arc protection devices under the conditions of multi-source load system coupling interference;

[0100] The optimization module is used to deploy arc prevention devices and analyze the coverage effect. If the coverage effect is achieved, the DQN model based on deep reinforcement learning is written into the arc prevention device system to configure the arc prevention device. If the coverage effect is not achieved, it is determined whether the installation is restricted. If the installation is restricted, the number of nodes is appropriately increased, and finally the optimal configuration scheme among the non-unique configuration schemes is selected. Otherwise, the simulation configuration module is turned to optimize the algorithm or adjust the expected optimization target.

[0101] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the method for configuring an arc prevention device under the coupling interference conditions of a multi-source load system described in the present invention.

[0102] A computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, the method for configuring an arc prevention device under the coupling interference condition of a multi-source-load system of the present invention can be implemented.

[0103] The computer device may be a laptop computer, a desktop computer or a workstation.

[0104] The processor of the present invention may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC) or a readily available field programmable gate array (FPGA).

[0105] The memory of the present invention may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0106] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0107] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for configuring an arc prevention device under the condition of multi-source load system coupling interference, characterized in that: The following steps are involved: Step a, obtain multi-source arc data; according to the frequency f s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is also measured; Step b, sampling the original current at equal intervals based on TT transformation, obtaining a 3D graph of the TT sequence obtained by sampling before and after the fault, detecting the fault by judging the current signal amplitude near the main diagonal, extracting the fault arc characteristics, and going to step c after detecting the fault according to the fault arc characteristics, otherwise going to step a; Step c: if the fault arc feature is weakened, the improved sparrow algorithm is used to optimize the maximum correlation kurtosis deconvolution algorithm to enhance the feature and go to step d; otherwise, go directly to step d; Step d, using the DQN model based on deep reinforcement learning to adaptively detect DC fault arcs, and by testing different working conditions to obtain a fault detection accuracy that meets the requirements, go to step e. If the fault detection accuracy does not meet the requirements, update the model parameters and retest until the fault detection accuracy meets the requirements; Step e, taking the target distribution network that needs to be equipped with arc prevention devices as the research object, establishing the optimization objective function and constraint conditions based on the coverage effect and deployment quantity of arc prevention equipment in the power system, using the optimization algorithm to solve the optimization objective function, and obtaining the arc prevention device configuration scheme under the condition of multi-source load system coupling interference; Step f, deploy arc prevention devices and analyze the coverage effect; if the coverage effect is achieved, write the DQN model based on deep reinforcement learning into the arc prevention device system to configure the arc prevention device; if the coverage effect is not achieved, determine whether the installation is restricted. If the installation is restricted, appropriately increase the number of nodes and finally select the optimal configuration scheme among the non-unique configuration schemes. Otherwise, go to step e to optimize the algorithm or adjust the expected optimization target.

2. According to the method for configuring an arc prevention device under the condition of multi-source load system coupling interference as described in claim 1, it is characterized in that: Acquiring multi-source arc data includes: setting different fault arc occurrence points so that multi-source faults occur simultaneously in a system, thereby forming multi-source arc data.

3. According to the method for configuring an arc prevention device under the condition of multi-source load system coupling interference as described in claim 1, it is characterized in that: Based on the improved sparrow algorithm to optimize the maximum correlation kurtosis deconvolution parameters, the three parameters [L, T, M] are optimized to obtain the best maximum correlation kurtosis deconvolution parameters. The SCSSA-MCKD algorithm is used to enhance and widen the pulse during the fault and amplify the fault characteristics after the fault arc occurs.

4. According to the method for configuring an arc prevention device under the condition of multi-source load system coupling interference as described in claim 1, it is characterized in that: The SCSSA-MCKD algorithm is used to enhance and widen the pulse during the fault, and the fault characteristics after the fault arc occurs include: Decompose the signal collected in step a by VMD or EMD, calculate the envelope spectrum peak factor of each IMF component, and the IMF component with the largest envelope spectrum peak factor is the optimal IMF component; The three parameters of the maximum correlation kurtosis deconvolution are optimized to obtain the enhanced IMF component; finally, the envelope spectrum analysis of the IMF component is performed to obtain the fault characteristic frequency.

5. According to claim 1, a method for configuring an arc prevention device under multi-source load system coupling interference conditions, characterized in that: Use the DQN model based on deep reinforcement learning to adaptively detect DC fault arcs. By testing different working conditions, the fault detection accuracy meets the requirements and goes to step e. If the fault detection accuracy does not meet the requirements, update the model parameters and retest until the fault detection accuracy meets the requirements, including: The fault feature data is normalized to the maximum and minimum values, and the holdout method is used to divide the data set into a training set S and a test set T according to randomness. The training set S is used as the input set of the deep reinforcement learning DQN algorithm. The model is trained by the agent interacting with the environment. The trained model is tested with the test set T to verify the model performance. The accuracy is determined by comparing and matching the output results with the actual label set of the test set. If the accuracy of the test set is low, the parameters in the deep reinforcement learning DQN model are adjusted, and then the features under the new working condition are processed in the same way and input into the model for fault diagnosis to obtain the accuracy under the new working condition. If the accuracy is low, the parameters in the deep reinforcement learning DQN model are adjusted, and the old working condition during training is retested using the updated model until the fault detection accuracy under both the new and old working conditions meets the requirements.

6. According to claim 1, a method for configuring an arc prevention device under multi-source load system coupling interference conditions, characterized in that: Determine the structure and key monitoring areas of the power system, deploy arc prevention devices on node i, 1 means deployment, 0 means no deployment, there are n arc prevention devices on the power system nodes, the detection range is r, and its maximum coverage range is defined as the set of nodes within the circle with radius r and centered on node i, including i, and define the objective function as the coverage rate of the area covered by the formula: Where: m is the number of nodes that the arc protection device can cover, i is the value of each node when deployed, The constraints include: node quantity constraint: the number of arc prevention devices cannot exceed the number of nodes in the power system; detection range constraint: the arc prevention device must cover key nodes within its detection range; resource limitation: consider quantity limitation.

7. An arc prevention device under multi-source load system coupling interference conditions, characterized in that: Obtained based on the configuration method described in any one of claims 1-7.

8. A simulation configuration system for arc prevention device under multi-source load system coupling interference conditions, characterized in that: It includes a data acquisition module, a feature extraction module, a feature enhancement module, a fault detection module, a simulation configuration module and an optimization module; The data acquisition module is used to acquire multi-source arc data; according to the frequency f s The current signal output by the same system is sampled to obtain the fault current waveforms under different working conditions, and the difference in the collected power is also measured; The feature extraction module takes equidistant sampling of the original current based on TT transformation, obtains the TT sequence three-dimensional graph by sampling before and after the fault, detects the fault by judging the current signal amplitude near the main diagonal, extracts the fault arc characteristics, and transfers to the feature enhancement module for feature enhancement after the fault is detected according to the fault arc characteristics, otherwise it transfers to the multi-source arc data of the data acquisition module; The feature enhancement module is used to optimize the maximum correlation kurtosis deconvolution algorithm using the improved sparrow algorithm to enhance the feature when the fault arc feature is weakened and transfer it to the fault detection module, otherwise it is directly transferred to the fault detection module for detection; The fault detection module uses the DQN model based on deep reinforcement learning to perform adaptive detection of DC fault arcs. By testing different working conditions, the fault detection accuracy meets the requirements and goes to step e. If the fault detection accuracy does not meet the requirements, the model parameters are updated and retested until the fault detection accuracy meets the requirements. The simulation configuration module takes the target distribution network that needs to be configured with arc protection devices as the object, establishes the optimization objective function and constraint conditions based on the coverage effect and deployment quantity of arc protection equipment in the power system, and uses the optimization algorithm to solve the optimization objective function to obtain the configuration scheme of arc protection devices under the conditions of multi-source load system coupling interference; The optimization module is used to deploy arc prevention devices and analyze the coverage effect. If the coverage effect is achieved, the DQN model based on deep reinforcement learning is written into the arc prevention device system to configure the arc prevention device. If the coverage effect is not achieved, it is determined whether the installation is restricted. If the installation is restricted, the number of nodes is appropriately increased, and finally the optimal configuration scheme among the non-unique configuration schemes is selected. Otherwise, the simulation configuration module is turned to optimize the algorithm or adjust the expected optimization target.

9. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the method for configuring an arc prevention device under the coupling interference conditions of a multi-source-load system as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that: A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, the method for configuring an arc prevention device under the conditions of coupled interference of a multi-source-load system as described in any one of claims 1 to 6 can be implemented.