Data-knowledge combined driven low-voltage active transformer area fault identification method and system

Abstract

The invention provides a data-knowledge combined driven low-voltage active transformer area fault identification method and system, and the method comprises the steps: predicting low-voltage distributed photovoltaic power, adding the consideration of weather factors in the prediction process, and building a low-voltage distribution network simulation model through combining a low-voltage active transformer area topological structure; performing fault simulation by using the low-voltage distribution network simulation model to obtain fault data, and extracting time domain and frequency domain characteristics in the fault data; establishing a low-voltage active transformer area fault identification model based on an extreme learning machine, and optimizing the initial weight and deviation of the low-voltage active transformer area fault identification model by using a particle swarm optimization algorithm in combination with experience knowledge of low-voltage power distribution transformer area fault identification to obtain a data-knowledge combined driving model; and processing the real-time fault data or the target fault data by using a data-knowledge combined driving model to realize fault identification of the low-voltage active transformer area. According to the method, the uncertainty of distributed photovoltaic output and the complexity of low-voltage transformer area topology are considered, and the accuracy of an identification result is improved.

Description

Technical Field

[0001] The present invention belongs to the field of low-voltage distribution transformer substations in distribution networks, and specifically relates to a data-knowledge jointly driven low-voltage active substation fault identification method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the gradual depletion of fossil fuel resources and the deterioration of the global environment, more and more countries are adopting solar energy development strategies. Photovoltaic power generation is known for its cleanliness, low carbon footprint and sustainability, and has become one of the most promising forms of renewable energy generation.

[0004] However, due to the inherent uncertainty of weather conditions, photovoltaic power generation has the characteristics of randomness and intermittency. In the low-voltage distribution area, the integration of low-voltage distributed photovoltaic into the low-voltage distribution network has brought a series of new problems. The low-voltage distribution area has gradually evolved from the traditional form of one-way energy transmission, single user characteristics, and clear "source-grid-load" interface to a new form with two-way flow, complex roles of producers and consumers, and coexistence of large and small power points, which poses a severe challenge to the safe use of low-voltage electricity.

[0005] In terms of distributed photovoltaic power generation prediction, current research is limited to the prediction of power generation of centralized photovoltaic power stations, while there is little research on the prediction of low-voltage distributed photovoltaics, and the applicability of the prediction model after access to the grid is rarely considered. Therefore, it is necessary to first consider the prediction of distributed photovoltaic power generation and access the grid for the next step of fault identification. In terms of the identification of low-voltage area faults after the intervention of distributed photovoltaics, some research has been conducted on the fault identification of low-voltage distribution networks based on smart meters and learning algorithms, but none of them considers the impact of distributed photovoltaic access on low-voltage distribution network faults, nor the new risk status brought about. Therefore, in the case of large-scale access of sources and loads, there is still a large technical gap that needs to be studied urgently. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a data-knowledge jointly driven low-voltage active area fault identification method and system. The present invention takes into account the uncertainty of distributed photovoltaic output and the complexity of low-voltage area topology, thereby improving the accuracy of the identification results.

[0007] According to some embodiments, the present invention adopts the following technical solutions:

[0008] A data-knowledge-jointly driven low-voltage active area fault identification method comprises the following steps:

[0009] Obtain historical data of low-voltage distributed photovoltaic power generation, use the grey prediction model to predict it, add weather factors into the prediction process, and obtain the prediction results of distributed photovoltaic power generation;

[0010] Based on the prediction results of distributed photovoltaic power generation and combined with the topological structure of low-voltage active substations, a low-voltage distribution network simulation model is established;

[0011] Use the low-voltage distribution network simulation model to perform fault simulation, obtain fault data, and extract time domain and frequency domain features;

[0012] A low-voltage active substation fault identification model based on extreme learning machine is established. Combined with the empirical knowledge of low-voltage distribution substation fault identification, the particle swarm optimization algorithm is used to optimize its initial weights and deviations to obtain a data-knowledge joint driving model.

[0013] The data-knowledge joint driving model is used to process real-time fault data or target fault data to realize low-voltage active area fault identification.

[0014] As an optional implementation method, a grey prediction model is used to predict it, and weather factors are taken into consideration during the prediction process. The process of obtaining the distributed photovoltaic power generation power prediction result includes: accumulating the historical data of low-voltage distributed photovoltaic power generation to form a data sequence, constructing a differential equation based on the data sequence, calculating the parameters of the grey prediction model, obtaining the low-voltage distributed photovoltaic output expression, discretizing the low-voltage distributed photovoltaic output expression, and dividing it into four states according to weather conditions, constructing a state transfer probability matrix according to the transfer probability of different conditions, and calculating the result of the target step state transfer according to the initial state.

[0015] As an optional implementation method, the process of establishing a low-voltage distribution network simulation model includes establishing a topological model based on the typical topological structure of the low-voltage substation, and considering the randomness and uncertainty of distributed photovoltaic output, connecting the predicted value of distributed photovoltaic output to the substation to form a low-voltage distribution network simulation model.

[0016] As an optional implementation, the process of extracting its frequency domain characteristics includes performing spectrum analysis on the current signal of the fault phase to obtain the frequency components of the signal, and then extracting the frequency domain characteristics of the signal.

[0017] As an optional implementation method, the process of establishing a low voltage active area fault identification model based on an extreme learning machine includes:

[0018] Build an extreme learning machine model and define model parameters;

[0019] The extracted waveform features are used as feature vector data, and the vector label of each ground fault cause is defined to divide the training set and the test set;

[0020] The model constructed by training the training set is used, and the initial weights and deviations are optimized using the particle swarm optimization algorithm to complete the training and optimization of the model.

[0021] As an optional implementation method, the process of combining the experience knowledge of low-voltage distribution area fault identification includes: integrating the experience knowledge of low-voltage active area fault identification and actual needs, and refining the following knowledge:

[0022] (1) If the zero-sequence current does not change suddenly before and after the fault occurs, it is definitely not a single-phase grounding fault;

[0023] (2) If the sudden change of multi-phase current when a fault occurs is less than the empirical threshold a, it is definitely not a phase-to-phase short circuit fault.

[0024] As an optional implementation, the knowledge function of the data-knowledge joint driven model includes two parts: a loss function that characterizes the error between the output of the pure data-driven model and the actual labels of the training set samples, and a conflict function that characterizes the conflict between the output of the pure data-driven model and domain experience knowledge.

[0025] As a further implementation, the loss function uses a softmax function to map the elements of the model output vector to the interval (0, 1), and uses cross entropy to calculate its loss function.

[0026] As a further implementation, the conflict function determines the mutation degree of the zero-sequence current and the fault phase current of each fault sample based on the empirical knowledge in the field of low-voltage active area fault identification, and determines the corresponding conflict function calculation rules:

[0027] (1) If A=1&&B=1, the value of the conflict function is +1;

[0028] (2) If A=1&&C=1, the value of the conflict function is +1;

[0029] Among them, A, B and C are data sets. For the output vector of the extreme learning machine model, if the maximum number of columns is the set value, A is set to 1, otherwise it is 0; if there is a fault zero-sequence current, B is set to 1, otherwise it is set to 0; if the multi-phase fault current mutation degree of the fault sample is less than a, C is set to 1, otherwise it is set to 0; A = 1 means that for a set of fault input samples, the fault type is most likely to be a phase-to-phase short circuit fault; B = 1 and C = 1 mean that the fault input sample cannot be a phase-to-phase short circuit fault determined by domain empirical knowledge.

[0030] As a further implementation, the knowledge function is constructed using the following method:

[0031]

[0032] Among them, h(θ) is the identification result based on the machine learning model, t j is the true cause mark of the jth fault sample, H() is used to measure the error between the predicted result and the true value, using the cross entropy function; P() is used to quantitatively measure the error between the machine learning model result and the identification rule, θ is the parameter set of the machine learning model, R(x j ) is based on the identification rule for the sample x j The recognition result of

[0033]

[0034] Set the concept of random factors and introduce trust β i The specific value of this parameter is:

[0035]

[0036] The random factor ε is introduced to reflect the degree of participation of empirical knowledge in the fusion model training process, so as to adjust the model's application of existing identification rules and data mining.

[0037] A data-knowledge-jointly driven low-voltage active substation fault identification system, comprising:

[0038] The prediction module is configured to obtain historical data of low-voltage distributed photovoltaic power generation, predict it using a grey prediction model, add weather factors into the prediction process, and obtain a prediction result of distributed photovoltaic power generation;

[0039] The simulation module is configured to establish a low-voltage distribution network simulation model based on the distributed photovoltaic power generation power prediction results and in combination with the low-voltage active area topology structure;

[0040] The feature extraction module is configured to perform fault simulation using a low-voltage distribution network simulation model, obtain fault data, and extract time domain and frequency domain features therein;

[0041] The data-knowledge joint driving module is configured to establish a low-voltage active substation fault identification model based on an extreme learning machine, and optimize its initial weights and deviations using a particle swarm optimization algorithm in combination with the empirical knowledge of low-voltage distribution substation fault identification, thereby obtaining a data-knowledge joint driving model;

[0042] The fault identification module is configured to utilize the data-knowledge joint driving model to process real-time fault data or target fault data to achieve low-voltage active area fault identification.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention establishes a data-knowledge joint driven low voltage active area fault identification model.

[0045] Considering the uncertainty and randomness of distributed photovoltaic output, a low-voltage distributed photovoltaic prediction model based on grey Markov chain is established to ensure the accuracy of distributed photovoltaic output prediction.

[0046] The present invention establishes a refined low-voltage area simulation model, connects the established photovoltaic prediction model to the established low-voltage area, considers the uncertainty and randomness of distributed photovoltaics, simulates the abnormal risk state and common faults of the low-voltage area, establishes a fault simulation set, and ensures the accuracy, applicability and typicality of the data;

[0047] The present invention extracts multi-dimensional transient and steady-state features based on simulation data and establishes a data-driven fault identification model based on an extreme learning machine. The empirical knowledge for fault identification of active low-voltage distribution stations is extracted and combined with the data-driven identification model, thereby improving the model identification accuracy and having strong generalization.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0050] Figure 1 is an overall architecture of an embodiment;

[0051] Figure 2 A photovoltaic power generation access system solution of an embodiment;

[0052] Figure 3 is a schematic diagram of a typical topological structure of a low-voltage area in an embodiment;

[0053] Figure 4 is a schematic diagram of a low-voltage area radial power grid model of an embodiment;

[0054] Figure 5 is an example of data transmission of a low voltage area fault detection system of an embodiment;

[0055] Figure 6 is a schematic diagram of an extreme learning machine network architecture of an embodiment;

[0056] Figure 7 The present invention is a schematic diagram of a low voltage active area fault identification model training process based on ELM data drive according to an embodiment;

[0057] Figure 8 A schematic diagram of a data-knowledge-jointly driven low-voltage active substation fault identification model training process according to an embodiment;

[0058] Fig. 9 is a schematic diagram of a fitness curve of a particle swarm algorithm of an embodiment;

[0059] Fig.10 1 is a schematic diagram of confusion matrices of a training set and a test set of a low-voltage active substation fault identification model driven by ELM data according to an embodiment;

[0060] Fig.11 The diagram is a confusion matrix diagram of a training set and a test set of a low-voltage active substation fault identification model driven by data-knowledge combination according to an embodiment. DETAILED DESCRIPTION

[0061] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0062] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0063] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0064] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0065] Embodiment 1

[0066] A data-knowledge-jointly driven low-voltage active area fault identification method comprises the following steps:

[0067] Obtain historical data of low-voltage distributed photovoltaic power generation, use the grey prediction model to predict it, add weather factors into the prediction process, and obtain the prediction results of distributed photovoltaic power generation;

[0068] Based on the prediction results of distributed photovoltaic power generation and combined with the topological structure of low-voltage active substations, a low-voltage distribution network simulation model is established;

[0069] Use the low-voltage distribution network simulation model to perform fault simulation, obtain fault data, and extract time domain and frequency domain features;

[0070] A low-voltage active substation fault identification model based on extreme learning machine is established. Combined with the empirical knowledge of low-voltage distribution substation fault identification, the particle swarm optimization algorithm is used to optimize its initial weights and deviations to obtain a data-knowledge joint driving model.

[0071] The data-knowledge joint driving model is used to process real-time fault data or target fault data to realize low-voltage active area fault identification.

[0072] The details are described below.

[0073] First, the historical data of low-voltage distributed photovoltaic power generation is obtained, and the grey prediction model is used to predict it. Weather factors are taken into consideration during the prediction process to obtain the prediction results of distributed photovoltaic power generation.

[0074] In this embodiment, the grey prediction model is a model for predicting a system with uncertain factors, and only a small amount of data is needed to achieve high-precision prediction. The modeling steps of the grey GM (1,1) photovoltaic prediction model are:

[0075] (1) Assume that the historical output of low-voltage distributed photovoltaic power generation is the original data

[0076] P (0) =(P (0) (1),P (0) (2),…,P (0) (n)) (1)

[0077] (2) Obtain a new sequence based on the accumulation of original data

[0078]

[0079] (3) Constructing differential equations

[0080]

[0081] (4) Parameter calculation: The least squares estimate of the GM(1,1) parameter sequence L = [ab]T is:

[0082] [ab] T =(B T B) -1 B T Y (4)

[0083]

[0084] (5) Obtaining a low-voltage distributed photovoltaic output model

[0085]

[0086] For low-voltage distributed photovoltaics, the daily output is uploaded by the power consumption information collection system every 15 minutes. The daily low-voltage distributed photovoltaic output can be composed of 96 discrete points. Since weather factors are an important factor affecting photovoltaic power, this paper divides weather conditions into four states, as shown in Table 1, namely sunny, cloudy, overcast, and rainy.

[0087] Table 1

[0088]

[0089]

[0090] The state transition probability matrix P is constructed by the transition probabilities of different conditions:

[0091] P ij =H ij / H i ,i=1,2,3,…,n (7)

[0092] Where P ij represents the probability of transitioning from state i to state j; H ij represents the number of original data transferred from state i to state j; H i Indicates the number of original data in state i.

[0093]

[0094] Determine the initial state

[0095]

[0096] Through the state transfer formula U (t) =U (0) ·P (t) , find the result of the state transfer in the kth step and take the state with a larger probability.

[0097] At this point, the photovoltaic random output model considering weather and historical factors has been constructed.

[0098] Next, build a topology simulation model.

[0099] This embodiment adopts the principle of 380V voltage level access to the user's internal power grid, combined with the photovoltaic project plan provided by the user and the main wiring situation of the 10kV distribution room in the photovoltaic park, and refers to the multi-point access typical design plan XGF380-Z-Z1 in the "Typical Design of Distributed Photovoltaic Power Generation Access System of the State Grid", and plans to adopt the following Figure 2The access system scheme shown in the figure. This scheme is mainly applicable to photovoltaic power stations that generate electricity for their own use / surplus power is connected to the grid (connected to the user's grid). The reference installed capacity of a single grid connection point is not more than 300kW, and three-phase access is adopted; for installed capacity of 8kW and below, single-phase access can be adopted. The access of distributed photovoltaic in this paper takes into account the randomness and uncertainty of distributed photovoltaic output, and connects its predicted value to the substation area. According to Figure 3 The typical topological structure of low-voltage area is established through field visits and research on low-voltage area wiring methods. Figure 4 The low-voltage area radial power grid model shown has a line voltage level of 380V and considers the three-phase five-wire low-voltage area topology.

[0100] In this embodiment, the line model is LGJ-25mm2, with a total of three outgoing lines. The active power consumed by each user ranges from 8kW to 12kW. The overall power factor of the substation is set to 0.92. The distance between each two households ranges from 0.02 to 0.04km, and the line unit length impedance is 1.131+j0.393Ω / km.

[0101] Based on the built model, the model is trained and tested. This embodiment uses the interaction between Matlab and Simulink to realize batch simulation of faults. A Matlab custom script can be written to call the Simulink automation library, abstract the model components and assign values ​​to their parameters, and traverse the parameter combination through a loop structure, and finally generate and export simulation data.

[0102] With the development of smart meters, the real-time data collection of low-voltage distribution networks has been improved. By introducing artificial edge machine algorithm calculations, processing distribution network data nearby and diagnosing faults, the efficiency of fault processing can be effectively improved. Figure 5 shown.

[0103] Then, a data-driven low-voltage active area fault identification model is first constructed. Before introducing the machine learning algorithm, the fault data collected by the smart meter in the distribution network is firstly extracted. This paper extracts 9 time domain statistical features, including 5 dimensional features and 4 dimensionless features. The current signal of the fault phase is subjected to spectrum analysis to obtain the frequency component of the signal, and then the frequency domain features of the signal are extracted. Four frequency domain features are extracted, and a total of 13-dimensional time-frequency domain features are used for identification. The calculation formula for feature extraction is shown in Table 2.

[0104] Table 2

[0105]

[0106]

[0107] Extreme Learning Machine (ELM) is a machine learning algorithm used to train a single hidden layer feedforward neural network (SLFN). Its basic principles are as follows:

[0108] Given an input sample {(x i ,t i ),i=1,2,…,N}, N represents the number of samples, x i ={x i1 ,x i2 ,…,x im} T ,t i ={t i1 ,t i2 ,…,t im} T , respectively represent the input vector and corresponding sample label of the i-th sample. Its network structure is as follows Figure 6 shown.

[0109] This embodiment uses a particle swarm optimization algorithm to optimize its initial weights and deviations. The steps for constructing the low-voltage active area fault identification model based on ELM are:

[0110] (1) Define the ELM model parameters. Specifically, this paper takes the selected waveform features as input, and the number of neurons in the input layer is set to 13. Combined with the empirical formula, the number of hidden layer neurons is set to 11. The five identified fault types are taken as output, as shown in Table 3, and the number of neurons in the output layer is set to 5. The activation function of the hidden layer neurons is selected as the Sigmoid function, and the output vector is processed by the Softmax function, converting the value of each element in the output vector to between 0 and 1.

[0111] Table 3

[0112]

[0113] (2) Divide the training set and the test set, then input the feature vector data composed of waveform features and define the vector label of each ground fault cause. Use the particle swarm optimization algorithm (PSO) to optimize the initial weights and deviations to complete the training and optimization of the model;

[0114] In this embodiment, by changing the fault transmission location, the location where the distributed power source is connected, the random distributed photovoltaic output, and different fault transition resistances, each group of faults is finally simulated with 120 groups of samples, totaling 600 groups of samples as the fault sample set. Among them, 83% of each group of samples is randomly selected as the training set, and the remaining 17% is used as the test set, and the fault classification model is used for training and classification.

[0115] The PSO algorithm is used to optimize the initial value of ELM. The learning factors c1 and c2 are both set to 1.5, and the number of iterations is set to 1000. The optimization results are shown in Fig. 9 shown.

[0116] (3) Use the test set data to evaluate the model performance.

[0117] The overall training process of the data-driven low-voltage active area fault identification model is as follows: Figure 7 shown.

[0118] Then, a low-voltage active substation fault identification model based on data-knowledge joint drive is established, which specifically includes:

[0119] Taking into account the experience and knowledge in this field and the actual needs of low-voltage active area fault identification, the following two pieces of knowledge are initially extracted:

[0120] (1) If the zero-sequence current does not change suddenly before and after the fault occurs, it is definitely not a single-phase grounding fault;

[0121] (2) If the sudden change of multi-phase current when a fault occurs is less than the empirical threshold a, it is definitely not a phase-to-phase short circuit fault.

[0122] The knowledge function of the fusion model is composed of two parts: one is the loss function that represents the error between the output of the pure data-driven model and the actual label of the training set samples, and the other is the conflict function that represents the contradiction and conflict between the output of the pure data-driven model and the domain experience knowledge. The knowledge function of the data-knowledge joint driven model can be obtained by adding the two together.

[0123] The loss function can be constructed as follows:

[0124] Cross entropy is usually used to characterize the distance between the actual output and the true label in machine learning algorithms. The smaller the cross entropy, the closer the probability distribution of the two is. Its calculation formula is shown in formula (15).

[0125]

[0126] Where: i is the dimension of the output vector, p(i) and q(i) are the values ​​of the i-th element in the actual label vector and the predicted output vector, respectively.

[0127] The data-driven model developed in this paper uses the softmax function to map the elements of the model output vector to the (0,1) interval, and then uses cross entropy to calculate its loss function.

[0128] The conflict function can be constructed in the following way:

[0129] According to the empirical knowledge in the field of low-voltage active substation fault identification, the mutation degree of zero-sequence current and fault phase current of each fault sample is determined, and the corresponding conflict function calculation rules are determined:

[0130] Rule 1: If A=1&&B=1, the value of the conflict function is +1;

[0131] Rule 2: If A=1&&C=1, the value of the conflict function is +1;

[0132] A, B, and C are data sets. For the ELM model output vector, if the maximum number of columns is 5, A is set to 1, otherwise it is 0. If there is a fault zero-sequence current, B is set to 1, otherwise it is set to 0; if the multi-phase fault current mutation degree of the fault sample is less than a, C is set to 1, otherwise it is set to 0. A = 1 means that for a set of fault input samples, the pure data-driven model believes that the fault type is most likely to be a phase-to-phase short circuit fault. B = 1 and C = 1 mean that the fault input sample cannot be a phase-to-phase short circuit fault determined by domain empirical knowledge.

[0133] The knowledge function of the joint data-knowledge driven model can be obtained by adding the two. The knowledge function can reflect both the error between the data-driven algorithm and the actual label, and the conflict between the data-driven algorithm and domain expertise.

[0134] The knowledge function can be constructed in the following way:

[0135]

[0136] Among them, h(θ) is the identification result based on the machine learning model, t j is the true cause mark of the jth fault sample, H() is used to measure the error between the predicted result and the true value, and the cross entropy function can be used; P() is used to quantitatively measure the error between the machine learning model result and the identification rule, θ is the parameter set of the machine learning model, R(x j ) is based on the identification rule for the sample x j The recognition result.

[0137]

[0138] Set the concept of random factors and introduce trust β i The specific value of this parameter is:

[0139]

[0140] The random factor ε is introduced here to reflect the degree of participation of empirical knowledge in the fusion model training process, so as to adjust the model's application of existing identification rules and data mining.

[0141] The random initial parameters of the machine learning model are optimized using optimization algorithms such as particle swarm. For the data-mechanism joint drive model, particle swarm and other optimization algorithms are also used for model optimization. The overall training process is as follows: Figure 8 shown.

[0142] Fig.10 The confusion matrix of the fault classification results of the ELM model on the training set and the test set has a classification accuracy of 93.6% and 89% respectively, which proves that the data-driven active low-voltage area fault extreme learning machine identification model proposed in this embodiment has a good classification effect.

[0143] Embodiment 2

[0144] A data-knowledge-jointly driven method for fault identification in low-voltage active substations.

[0145] The following steps are involved:

[0146] Step 1: Use the low-voltage active area radial model constructed in Example 1, extract features consistent with it, and the training set and test set samples are also consistent with Example 1.

[0147] Step 2: The structure and parameters of the ELM model and the PSO algorithm are the same as those in the first embodiment. In the data-knowledge joint driven model, the fitness function of the PSO optimization algorithm is the knowledge function shown in formula (16). For the parameters in the knowledge function, the value of the random factor ε is set to 0.1. Considering that the two pieces of empirical knowledge introduced have high credibility, the trust β i The specific value c is taken as 0.75, which is a fixed value and does not change with the number of iterations.

[0148] Step 3: Data-knowledge joint driven model, except that its fitness function is the cross entropy loss function, its structure and parameter settings are the same as those of the pure data driven model. Figure 8 Train it using the model building process shown.

[0149] Step 4: The data-knowledge joint-driven model still uses the 600 simulated data sets to identify fault types. The identification confusion matrix of the low-voltage active substation fault identification model driven by data-knowledge joint is as follows: Fig.11 shown.

[0150] Step 5: In the training process of the data-knowledge joint-driven model, in addition to paying attention to the empirical risk, the value of the conflict function is also paid attention to. In addition, the two empirical knowledge propositions proposed in this disclosure have strong universality, which makes up for the shortcomings of the data-driven algorithm in mapping the relationship between undetected fault types and fault characteristics. Therefore, for the test set composed of samples, the performance of the joint-driven model exceeds that of the pure data-driven model. Tables 4 and 5 show the accuracy of the two models in identifying 5 types of faults on the training set and test set, respectively.

[0151] Table 4

[0152]

[0153]

[0154] Table 5

[0155]

[0156] Step 6: After adding the above two pieces of empirical knowledge, the ELM algorithm conflicts with the empirical knowledge diagnosis result after the diagnosis error, and guides the data-knowledge joint driving model to correct the error, especially in terms of the fault type of phase-to-phase short circuit fault, to make a more accurate classification. The accuracy of the training set and test set recognition based on data-driven is 93.6% and 89% respectively, while the accuracy of the data-knowledge joint driving model is as high as 95.2% and 97% respectively, which fully proves the effectiveness and feasibility of the model.

[0157] Embodiment 3

[0158] A data-knowledge-jointly driven low-voltage active substation fault identification system, comprising:

[0159] The prediction module is configured to obtain historical data of low-voltage distributed photovoltaic power generation, predict it using a grey prediction model, add weather factors into the prediction process, and obtain a prediction result of distributed photovoltaic power generation;

[0160] The simulation module is configured to establish a low-voltage distribution network simulation model based on the distributed photovoltaic power generation power prediction results and in combination with the low-voltage active area topology structure;

[0161] The feature extraction module is configured to perform fault simulation using a low-voltage distribution network simulation model, obtain fault data, and extract time domain and frequency domain features therein;

[0162] The data-knowledge joint driving module is configured to establish a low-voltage active substation fault identification model based on an extreme learning machine, and optimize its initial weights and deviations using a particle swarm optimization algorithm in combination with the empirical knowledge of low-voltage distribution substation fault identification, thereby obtaining a data-knowledge joint driving model;

[0163] The fault identification module is configured to utilize the data-knowledge joint driving model to process real-time fault data or target fault data to achieve low-voltage active area fault identification.

[0164] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0166] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0167] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A data-knowledge-jointly driven low-voltage active area fault identification method, characterized in that: The following steps are involved: Obtain historical data of low-voltage distributed photovoltaic power generation, use the grey prediction model to predict it, add weather factors into the prediction process, and obtain the prediction results of distributed photovoltaic power generation; Based on the prediction results of distributed photovoltaic power generation and combined with the topological structure of low-voltage active substations, a low-voltage distribution network simulation model is established; Use the low-voltage distribution network simulation model to perform fault simulation, obtain fault data, and extract time domain and frequency domain features; A low-voltage active substation fault identification model based on extreme learning machine is established. Combined with the empirical knowledge of low-voltage distribution substation fault identification, the particle swarm optimization algorithm is used to optimize its initial weights and deviations to obtain a data-knowledge joint driving model. The data-knowledge joint driving model is used to process real-time fault data or target fault data to realize low-voltage active area fault identification.

2. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 1, characterized in that: The grey prediction model is used to predict it, and weather factors are taken into consideration during the prediction process. The process of obtaining the distributed photovoltaic power generation power prediction result includes: accumulating the historical data of low-voltage distributed photovoltaic power generation to form a data sequence, constructing a differential equation based on the data sequence, calculating the parameters of the grey prediction model, obtaining the low-voltage distributed photovoltaic output expression, discretizing the low-voltage distributed photovoltaic output expression, and dividing it into four states according to weather conditions, constructing a state transfer probability matrix according to the transfer probability of different conditions, and calculating the result of the target step state transfer according to the initial state.

3. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 1, characterized in that: The process of establishing a low-voltage distribution network simulation model includes establishing a topological model based on the typical topological structure of the low-voltage substation, considering the randomness and uncertainty of distributed photovoltaic output, connecting the predicted value of distributed photovoltaic output to the substation, and forming a low-voltage distribution network simulation model.

4. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 1, characterized in that: The process of extracting its frequency domain characteristics includes performing spectrum analysis on the current signal of the fault phase to obtain the frequency components of the signal and then extracting the frequency domain characteristics of the signal.

5. The data-knowledge-jointly driven low-voltage active area fault identification method according to claim 1, characterized in that: The process of establishing a low-voltage active area fault identification model based on extreme learning machine includes: Build an extreme learning machine model and define model parameters; The extracted waveform features are used as feature vector data, and the vector label of each ground fault cause is defined to divide the training set and the test set; The model constructed by training the training set is used, and the initial weights and deviations are optimized using the particle swarm optimization algorithm to complete the training and optimization of the model.

6. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 1, characterized in that: The process of combining the experience knowledge of low-voltage distribution area fault identification includes: integrating the experience knowledge of low-voltage active area fault identification and actual needs, and refining the following knowledge: (1) If the zero-sequence current does not change suddenly before and after the fault occurs, it is definitely not a single-phase grounding fault; (2) If the sudden change of multi-phase current when a fault occurs is less than the empirical threshold a, it is definitely not a phase-to-phase short circuit fault.

7. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 1, characterized in that: The knowledge function of the data-knowledge joint driven model includes two parts: a loss function that characterizes the error between the output of the pure data driven model and the actual labels of the training set samples, and a conflict function that characterizes the contradiction and conflict between the output of the pure data driven model and domain experience knowledge.

8. A data-knowledge-jointly driven low-voltage active area fault identification method as claimed in claim 7, characterized in that: The loss function uses the softmax function to map the elements of the model output vector to the interval (0,1), and uses cross entropy to calculate its loss function; The conflict function determines the mutation degree of the zero-sequence current and the fault phase current of each fault sample based on the empirical knowledge in the field of low-voltage active area fault identification, and determines the corresponding conflict function calculation rules: (1) If A=1&&B=1, the value of the conflict function is +1; (2) If A=1&&C=1, the value of the conflict function is +1; Among them, A, B and C are data sets. For the output vector of the extreme learning machine model, if the maximum number of columns is the set value, A is set to 1, otherwise it is 0; if there is a fault zero-sequence current, B is set to 1, otherwise it is set to 0; if the multi-phase fault current mutation degree of the fault sample is less than a, C is set to 1, otherwise it is set to 0; A = 1 means that for a set of fault input samples, the fault type is most likely to be a phase-to-phase short circuit fault; B = 1 and C = 1 mean that the fault input sample cannot be a phase-to-phase short circuit fault determined by domain empirical knowledge.

9. A data-knowledge-jointly driven low voltage active area fault identification method as claimed in claim 7, characterized in that: The knowledge function is constructed using the following method: Among them, h(θ) is the identification result based on the machine learning model, t j is the true cause mark of the jth fault sample, H() is used to measure the error between the predicted result and the true value, using the cross entropy function; P() is used to quantitatively measure the error between the machine learning model result and the identification rule, θ is the parameter set of the machine learning model, R(x j ) is based on the identification rule for the sample x j The recognition result of Set the concept of random factors and introduce trust β i The specific value of this parameter is: The random factor ε is introduced to reflect the degree of participation of empirical knowledge in the fusion model training process, so as to adjust the model's application of existing identification rules and data mining.

10. A data-knowledge-jointly driven low-voltage active area fault identification system, characterized in that it includes: The prediction module is configured to obtain historical data of low-voltage distributed photovoltaic power generation, predict it using a grey prediction model, add weather factors into the prediction process, and obtain a distributed photovoltaic power generation power prediction result; The simulation module is configured to establish a low-voltage distribution network simulation model based on the distributed photovoltaic power generation power prediction results and in combination with the low-voltage active area topology structure; The feature extraction module is configured to perform fault simulation using a low-voltage distribution network simulation model, obtain fault data, and extract time domain and frequency domain features therein; The data-knowledge joint driving module is configured to establish a low-voltage active substation fault identification model based on an extreme learning machine, and optimize its initial weights and deviations using a particle swarm optimization algorithm in combination with the empirical knowledge of low-voltage distribution substation fault identification, thereby obtaining a data-knowledge joint driving model; The fault identification module is configured to utilize the data-knowledge joint driving model to process real-time fault data or target fault data to achieve low-voltage active area fault identification.

Images