A power outage fault warning method and system for a substation area

Abstract

The present invention discloses a method and system for early warning of power outage faults in a power distribution area. The method includes: obtaining various power distribution area data and normalizing them to obtain various data to be warned; respectively performing equipment fault prediction on various data to be warned through a target DBN model to generate multiple equipment fault probability vectors; performing data fusion using all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector; calculating the belief function values corresponding to each equipment comprehensive fault probability in the equipment comprehensive fault probability vector, and selecting the equipment label to which the maximum belief function value belongs as the power distribution area fault equipment label; calculating the fault early warning time corresponding to the power distribution area fault equipment label based on the comprehensive fault probability, power outage coefficient, and service life corresponding to the power distribution area fault equipment label; and generating and outputting early warning information using the power distribution area fault equipment label and the fault early warning time. This method can effectively utilize multi-source information in the power distribution area and improve the accuracy of power outage fault early warning in the power distribution area.

Description

Technical Field

[0001] The present invention relates to the technical field of power outage fault warning in low - voltage distribution areas, and particularly relates to a power outage fault warning and system for a distribution area. Background Art

[0002] As a key link in the power distribution system, the safety of the low - voltage distribution area is related to the safe and stable operation of the entire power distribution system and the vital interests of users. Due to its complex structure, the main equipment in the distribution area has a complex working environment and numerous interference factors, resulting in frequent failures. Moreover, most distribution areas lack effective power outage warning and positioning functions, so they are extremely vulnerable to power outages due to various factors. According to relevant statistics, more than 80% of user power outage events are caused by failures in the distribution network system. The failures in the distribution network system will not only cause power outage losses to users, but also cause immeasurable economic losses and safety hazards to the upstream power generation and transmission systems. Therefore, strengthening the research on power outage warning and positioning of distribution area faults has great practical significance.

[0003] Currently, for the power outage warning technology of the distribution area, it mainly focuses on the overall monitoring of the large power grid. There is relatively more research on the fault warning of medium - and high - voltage power grids, and less on the power outage fault warning of low - voltage distribution areas. The existing technology for fault warning diagnosis of the distribution network mainly judges and analyzes the action information of switch protection to find the faulty components.

[0004] However, this type of method does not consider the impact on fault diagnosis when the protection or switch refuses to operate, malfunctions, or there is information distortion during a fault, and cannot comprehensively and effectively utilize the multi - source information of the distribution area, resulting in a low accuracy of power outage fault warning in the distribution area. Summary of the Invention

[0005] The present invention provides a power outage fault warning method and system for a distribution area, which solves the technical problems in the prior art that the fault warning diagnosis of the distribution network does not consider the impact on fault diagnosis when the protection or switch refuses to operate, malfunctions, or there is information distortion during a fault, and cannot comprehensively and effectively utilize the multi - source information of the distribution area, resulting in a low accuracy of power outage fault warning in the distribution area.

[0006] A power outage fault warning method provided by the present invention includes:

[0007] Obtain various distribution area data and perform normalization to obtain various data to be warned;

[0008] Respectively perform equipment fault prediction on various data to be warned through a target DBN model to generate multiple equipment fault probability vectors;

[0009] Use all the equipment fault probability vectors for data fusion to generate an equipment comprehensive fault probability vector;

[0010] Calculate the belief function values corresponding to the comprehensive failure probabilities of each device in the comprehensive failure probability vector of the device, and select the device label corresponding to the maximum belief function value as the substation area failure device label;

[0011] Based on the comprehensive failure probability, power outage coefficient, and service life corresponding to the substation area failure device label, calculate the failure warning time corresponding to the substation area failure device label;

[0012] Generate and output a warning message using the substation area failure device label and the failure warning time.

[0013] Optionally, the method further includes:

[0014] Construct an initial DBN model using a multi-layer RBM network structure and a single-layer BP network structure;

[0015] Obtain the historical data of the substation area and normalize it to obtain historical training data;

[0016] Train the initial DBN model with the historical training data to generate a target DBN model.

[0017] Optionally, the step of training the initial DBN model with the historical training data to generate a DBN model includes:

[0018] Input the historical training data into the first layer of the RBM network structure;

[0019] Generate target neuron activation state data through the multi-layer RBM network structure and transmit it to the BP network structure, and perform reverse fine-tuning on each layer of the RBM network structure through the BP network structure;

[0020] If the number of times of reverse fine-tuning has not reached the preset iteration threshold, jump to execute the step of inputting the historical training data into the first layer of the RBM network structure;

[0021] If the number of times of reverse fine-tuning reaches the preset iteration threshold, use the multi-layer RBM network structure and the BP network structure at the current moment to generate a target DBN model.

[0022] Optionally, the RBM network structure includes a visible layer and a hidden layer; the step of generating target neuron activation state data through the multi-layer RBM network structure and transmitting it to the BP network structure, and performing reverse fine-tuning on each layer of the RBM network structure through the BP network structure includes:

[0023] Select the first visible layer from the multi-layer RBM network structure as the head visible layer at the current moment;

[0024] Input the historical training data into the head visible layer;

[0025] Calculate the intermediate neuron activation state data of the associated hidden layer through the head visible layer;

[0026] Select an activation probability from the intermediate neuron activation state data according to a preset rule to reconstruct the head visible layer, generating an updated visible layer;

[0027] Calculate the neuron activation probability of the hidden layer associated with the updated visible layer and reconstruct it to generate an updated hidden layer;

[0028] Use the intermediate neuron activation state data as new historical training data, and select the first visible layer from the unselected RBM network structure as the new head visible layer at the current moment;

[0029] Jump to execute the step of inputting the historical training data into the head visible layer until all the RBM network structures are selected.

[0030] Optionally, the step of using all the device failure probability vectors for data fusion to generate a device comprehensive failure probability vector includes:

[0031] When the device failure probability vector is not an empty set, input the respective device failure probabilities into the fusion function to generate device comprehensive failure probabilities respectively;

[0032] Use all the device comprehensive failure probabilities to generate a device comprehensive failure probability vector;

[0033] Wherein, the fusion function is:

[0034]

[0035] In the formula: is the device failure state label corresponding to the k-th device after fusion at time T l , and are any subsets of is the probability that the k-th device in the s system fails at time T l , is the device failure state label corresponding to the k-th device at time T l , taking the value of 1 indicates the failure state, A value of 0 indicates a non-fault state. s = 1, 2, 3 represent the distribution automation system, the user electricity consumption information acquisition system, and the meteorological information acquisition system respectively. k is the ID of the device, k ∈ K = {1, 21, …, 2i, 31, …, 3j, 4}. k = 1 represents the transformer device label, k = 2i represents the line i device label, k = 3j represents the user j device label, and k = 4 represents the device in the normal state label.

[0036] Optionally, the step of calculating the belief function values corresponding to the comprehensive fault probabilities of each device in the device comprehensive fault probability vector and selecting the device label corresponding to the maximum belief function value as the substation area fault device label includes:

[0037] Calculate the belief function values corresponding to the comprehensive fault probabilities of each device in the device comprehensive fault probability vector;

[0038] Select the device label corresponding to the maximum belief function value as the substation area fault device label;

[0039] The belief function value is:

[0040]

[0041] In the formula: represents the lower limit estimate of the fault belief of device k after fusion at time T l , represents the comprehensive fault probability of device k after fusion at time T l .

[0042] Optionally, the step of calculating the fault warning time corresponding to the substation area fault device label based on the comprehensive fault probability, power outage coefficient, and service life corresponding to the substation area fault device label includes:

[0043] Construct a fitting function between the historical statistical power outage fault warning time, power outage coefficient, and equipment service life corresponding to each device comprehensive fault probability;

[0044] Use the least squares method to calculate the parameters to be solved in the fitting function and generate a target fitting function;

[0045] Input the comprehensive fault probability, power outage coefficient, and service life corresponding to the substation area fault device label into the target fitting function to generate the fault warning time corresponding to the substation area fault device label;

[0046] The fitting function is:

[0047]

[0048] Where: t is the power outage fault warning time of historical statistics, σ is the power outage coefficient, p is the comprehensive equipment failure probability, n is the equipment service life, and α and β are parameters to be determined.

[0049] The present invention also provides a power outage fault warning system for a distribution area, including: a data acquisition module, configured to acquire various distribution area data and perform normalization to obtain various data to be warned;

[0050] A fault prediction module, configured to respectively perform equipment fault prediction on various data to be warned through a target DBN model, and generate multiple equipment fault probability vectors;

[0051] A data fusion module, configured to perform data fusion by using all the equipment fault probability vectors to generate a comprehensive equipment fault probability vector;

[0052] A fault location module, configured to calculate the belief function values corresponding to the comprehensive equipment fault probabilities of each equipment in the comprehensive equipment fault probability vector, and select the equipment label corresponding to the maximum belief function value as the distribution area fault equipment label;

[0053] A fault warning time module, configured to calculate the fault warning time corresponding to the distribution area fault equipment label based on the comprehensive fault probability, power outage coefficient and service life corresponding to the distribution area fault equipment label;

[0054] A fault information warning output module, configured to generate and output a warning information by using the distribution area fault equipment label and the fault warning time.

[0055] Optionally, the system further includes:

[0056] An initial model construction module, configured to construct an initial DBN model by using a multi-layer RBM network structure and a single-layer BP network structure;

[0057] A training data acquisition module, configured to acquire distribution area historical data and perform normalization to obtain historical training data;

[0058] A model training module, configured to train the initial DBN model by using the historical training data to generate a target DBN model.

[0059] Optionally, the model training module includes:

[0060] A training data input sub-module, configured to input the historical training data into the first-layer RBM network structure;

[0061] A data adjustment sub-module, configured to generate target neuron activation state data through the multi-layer RBM network structure and transmit it to the BP network structure, and perform reverse fine-tuning on each layer of the RBM network structure through the BP network structure;

[0062] A return sub-module, configured to, if the number of times of the reverse fine-tuning does not reach a preset iteration threshold, jump to execute the step of inputting the historical training data into the first-layer RBM network structure;

[0063] A target model generation sub-module, configured to, if the number of times of the reverse fine-tuning reaches the preset iteration threshold, generate a target DBN model by using the multi-layer RBM network structure and the BP network structure at the current moment.

[0064] As can be seen from the above technical solutions, the present invention has the following advantages:

[0065] The present application provides a method and a system for warning of power outage faults in a substation area. The method includes: obtaining various substation area data and normalizing them to obtain various data to be warned; respectively performing equipment fault prediction on various data to be warned through a target DBN model, generating multiple equipment fault probability vectors and performing data fusion, so as to generate an equipment comprehensive fault probability vector; calculating the belief function values corresponding to the respective equipment comprehensive fault probabilities in the equipment comprehensive fault probability vector, and selecting the equipment label to which the maximum belief function value belongs as the substation area fault equipment label; calculating the fault warning time corresponding to the substation area fault equipment label based on the comprehensive fault probability, power outage coefficient and service life corresponding to the substation area fault equipment label; and generating and outputting a warning message by using the substation area fault equipment label and the fault warning time, thereby solving the technical problems in the prior art that the influence of protection or switch refusal to operate, misoperation and information distortion on fault diagnosis in the case of faults is not considered in the fault warning diagnosis of the distribution network, and the multi-source information of the substation area cannot be fully and effectively utilized, resulting in low accuracy of the substation area fault warning. Description of the Drawings

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0067] Figure 1 It is a flowchart of the steps of a method for warning of power outage faults in a substation area provided by an embodiment of the present invention;

[0068] Figure 2 It is a schematic structural diagram of a target DBN model provided by an embodiment of the present invention;

[0069] Figure 3 It is a flowchart of the steps of a method for warning of power outage faults in a substation area provided by an optional embodiment of the present invention;

[0070] Figure 4 A typical low-voltage topology diagram of a substation area provided by an alternative embodiment of the present invention;

[0071] Figure 5 A schematic structural diagram of a power outage fault warning system for a substation area provided by an embodiment of the present invention. Detailed implementation manners

[0072] The embodiments of the present invention provide a power outage fault warning method and system for a substation area, which are used to solve the technical problems in the prior art that the fault warning diagnosis of the distribution network does not consider the influence on the fault diagnosis caused by the refusal or misoperation of protection or switches and information distortion under fault conditions, and cannot comprehensively and effectively utilize the multi-source information of the substation area, resulting in low accuracy of power outage fault warning in the substation area.

[0073] In order to make the invention objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0074] Please refer to Figure 1 , Figure 1 A step flow chart of a power outage fault warning method for a substation area provided by an embodiment of the present invention.

[0075] A power outage fault warning method provided by the present invention includes:

[0076] Step 101: Obtain various substation area data and perform normalization to obtain various data to be warned.

[0077] In the embodiments of the present application, to obtain various substation area data, it can be collected through the substation area distribution automation system, the user electricity consumption information collection system, and the meteorological collection system. It is worth mentioning that before inputting various warning data into the corresponding target DBN model respectively, the equipment power outage time and power outage times in the data to be warned corresponding to the user electricity consumption information collection system need to be filled into the data to be warned corresponding to the substation area distribution automation system and the meteorological collection system.

[0078] It should be noted that the data to be pre-warned corresponding to the substation area power distribution automation system may include: three-phase voltage, current active power, reactive power, and power factor of the transformer; the data to be pre-warned corresponding to the user power consumption information acquisition system may include: three-phase voltage, current, active power, reactive power, power factor, electricity consumption (the difference in electricity consumption compared to the previous sampling point), historical power outage time, and number of power outages of the user; the data to be pre-warned corresponding to the meteorological acquisition system may include: temperature, humidity, pressure, light intensity, and average light hours at the location of the substation area.

[0079] Step 102: Perform equipment fault prediction on various data to be pre-warned respectively through the target DBN model, and generate multiple equipment fault probability vectors.

[0080] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of a target DBN model provided by an embodiment of the present invention; it should be noted that the target DBN model is mainly composed of multiple layers of restricted Boltzmann machines (RBM) and one layer of backpropagation neural network (BPNN).

[0081] After obtaining various data to be pre-warned, input the various data to be pre-warned into the corresponding target DBN model respectively for equipment fault prediction, so as to generate multiple equipment fault probability vectors.

[0082] Step 103: Perform data fusion using all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector.

[0083] In the embodiment of the present application, performing data fusion on all the equipment fault probability vectors is based on the DS evidence theory, so as to generate an equipment comprehensive fault probability, and then use all the comprehensive fault probabilities to generate an equipment comprehensive fault probability vector.

[0084] It should be noted that the DS evidence theory fusion is to fuse the representation information of multiple systems for the same event into one piece of information representing the event according to its fusion rule, so as to obtain a consistent description of the event by multiple systems.

[0085] Step 104: Calculate the belief function values corresponding to the respective equipment comprehensive fault probabilities in the equipment comprehensive fault probability vector, and select the equipment label to which the maximum belief function value belongs as the substation area fault equipment label.

[0086] In the embodiments of the present application, the belief function refers to the sum of the probabilities of all subsets of a device failure in a power distribution area, representing the probability that the device failure is true. Since there is only one failure probability corresponding to a device in the power distribution area after fusion, that is, the comprehensive failure probability of a device in the power distribution area, the belief function value of the failure probability of a device in the power distribution area is equal to its comprehensive failure probability, that is

[0087]

[0088] In the formula: represents the lower limit estimate of the failure belief of the fused device k at time T l ; represents the comprehensive failure probability of the fused device k at time T l .

[0089] It should be noted that the device labels include: transformer failure label, line i device failure label, user j device failure label, and device normal label.

[0090] Step 105: Calculate the failure warning time corresponding to the power distribution area failure device label based on the comprehensive failure probability, power outage coefficient, and service life corresponding to the power distribution area failure device label.

[0091] It should be noted that the power outage coefficient is statistically obtained from the ratio of the historical power outage times corresponding to the power distribution area devices to the total power outage times of the power distribution area. The larger the power outage coefficient, the greater the probability that the failure causes a power outage in the power distribution area.

[0092] In the embodiments of the present application, a fitting function between the historical statistical power outage failure warning time, power outage coefficient, and device service life corresponding to the comprehensive failure probability of each device is constructed, that is

[0093]

[0094] In the formula, t is the historical statistical power outage failure warning time, σ is the power outage coefficient, ρ is the comprehensive failure probability of the device, n is the service life of the device, and α and β are parameters to be determined.

[0095] The parameters to be determined in the fitting function are calculated by the least squares method to generate a target fitting function; the comprehensive failure probability, power outage coefficient, and service life corresponding to the power distribution area failure device label are input into the target fitting function to generate the failure warning time corresponding to the power distribution area failure device label.

[0096] Step 106: Generate and output a warning message using the power distribution area failure device label and the failure warning time.

[0097] In the embodiments of the present application, a warning message is generated and output using the power distribution area failure device label and the failure warning time, thereby realizing the positioning of the power distribution area failure device and the time warning of future failures of the failure device.

[0098] In the embodiment of the present application, a variety of substation area data are acquired and normalized to obtain a variety of data to be warned; the target DBN model is used to respectively perform equipment fault prediction on various data to be warned, generate multiple equipment fault probability vectors and perform data fusion, so as to generate an equipment comprehensive fault probability vector; calculate the belief function values corresponding to the respective equipment comprehensive fault probabilities in the equipment comprehensive fault probability vector, and select the equipment label to which the maximum belief function value belongs as the substation area fault equipment label; based on the comprehensive fault probability, power outage coefficient and service life corresponding to the substation area fault equipment label, calculate the fault warning time corresponding to the substation area fault equipment label; use the substation area fault equipment label and the fault warning time to generate and output a warning message, thereby solving the technical problem in the prior art that for the fault warning diagnosis of the distribution network, the influence on the fault diagnosis caused by the refusal or malfunction of protection or switches and information distortion in the case of faults is not considered, and the multi-source information of the substation area cannot be fully and effectively utilized, resulting in low accuracy of the substation area fault warning.

[0099] Please refer to Figure 3 , Figure 3 which is the step flow chart of a substation area power outage fault warning method provided by an optional embodiment of the present invention.

[0100] The present invention provides a substation area power outage fault warning method, and the method includes the following steps:

[0101] Step 201, acquire a variety of substation area data and normalize them to obtain a variety of data to be warned.

[0102] In an optional embodiment of the present application, the specific implementation process of step 201 is similar to the above step 101, and will not be elaborated here.

[0103] Step 202, construct an initial DBN model by using a multi-layer RBM network structure and a single-layer BP network structure.

[0104] It should be noted that the initial DBN model is mainly composed of a multi-layer Restricted Boltzmann Machine (RBM) and a single-layer Back Propagation Neural Network (BPNN). During the training process, first use the unsupervised learning method of RBM for pre-training, and then use the supervised BP network for fine-tuning, fully combining the advantages of unsupervised learning and supervised learning, not only solving the problem that the multi-layer training of traditional neural networks is not applicable, but also solving the problem of unsupervised RBM training.

[0105] Step 203, acquire the substation area historical data and normalize them to obtain historical training data.

[0106] In an alternative embodiment of the present application, to obtain various historical data of the substation area, it can be collected through the substation area distribution automation system, the user electricity consumption information collection system, and the meteorological collection system. It is worth mentioning that before inputting the historical training data into the initial DBN model, the equipment power outage time and the number of power outages in the historical training data corresponding to the user electricity consumption information collection system need to be filled into the historical training data corresponding to the substation area distribution automation system and the meteorological collection system.

[0107] Step 204: Train the initial DBN model with the historical training data to generate a target DBN model.

[0108] Step 204 includes the following sub-steps S1 - S4:

[0109] S1: Input the historical training data into the first-layer RBM network structure;

[0110] S2: Generate target neuron activation state data through the multi-layer RBM network structure and transmit it to the BP network structure, and perform reverse fine-tuning on each layer of the RBM network structure through the BP network structure;

[0111] It should be noted that the RBM network structure is composed of a visible layer v (Visual Layer) and a hidden layer h (HiddenLayer). Neurons between the visible layer and the hidden layer are fully connected bidirectionally, and neurons within the layer are not connected.

[0112] In an alternative embodiment of the present application, select the first visible layer from the multi-layer RBM network structure as the leading visible layer at the current moment, and input the historical training data into the leading visible layer; calculate the intermediate neuron activation state data of the associated hidden layer through the leading visible layer; select an activation probability from the intermediate neuron activation state data to reconstruct the leading visible layer according to a preset rule to generate an updated visible layer; calculate the neuron activation probability of the associated hidden layer of the updated visible layer and reconstruct it to generate an updated hidden layer; use the intermediate neuron activation state data as the new historical training data, and select the first visible layer from the unselected RBM network structure as the new leading visible layer at the current moment; jump to execute the step of inputting the historical training data into the leading visible layer until all RBM network structures are selected.

[0113] In a specific implementation, it is set that the RBM network structure has 57 visible neurons and 28 hidden neurons. In the initial state (1,1), its energy joint distribution is:

[0114]

[0115] In the formula, v i 、h j are respectively the i-th visible neuron of the visible layer and the j-th hidden neuron of the hidden layer, ai and b j are the bias values of the i-th visible neuron and the j-th hidden neuron respectively, and W ij is the connection weight between the i-th visible neuron and the j-th hidden neuron.

[0116] Using the defined energy function, the joint probability distribution of state (1, 1) can be given, that is

[0117]

[0118]

[0119] where Z θ is the normalization factor, also known as the partition function.

[0120] When the j-th hidden neuron h j is activated, its activation probability is:

[0121]

[0122] where Sigmoid is the activation function, that is

[0123]

[0124] Since there is a bidirectional connection between the visible layer and the hidden layer, when the hidden layer neurons are activated, the corresponding neurons in the visible layer are also activated, and the activation probability is:

[0125]

[0126] The RBM uses the Contrastive Divergence (CD) algorithm for training. By calculating the activation probability of the hidden neuron h j and setting μ = 0.8 as the activation threshold, when the activation probability of h j is not less than μ, then h j is activated, denoted as h j = 1; otherwise it is not activated, denoted as h j = 0, that is

[0127] P(h j |x) ≥ 0.8, h j = 1

[0128] P(h j |x) < 0.8, h j = 0

[0129] Using Gibbs sampling, a sample is drawn from the calculated probability distribution of the hidden neuron activations to reconstruct the visible layer, thereby calculating the probability that the visible layer neurons are activated; then the probability that the hidden layer neurons are activated is recalculated using the same sampling method.

[0130] S3. If the number of times of reverse fine-tuning has not reached the preset iteration threshold, jump to execute the step of inputting the historical training data into the first-layer RBM network structure.

[0131] S4. If the number of times of reverse fine-tuning reaches the preset iteration threshold, use the current multi-layer RBM network structure and BP network structure to generate the target DBN model.

[0132] Step 205: Perform equipment fault prediction on various data to be warned by using the target DBN model, and generate multiple equipment fault probability vectors.

[0133] In an optional embodiment of the present application, the specific implementation process of step 205 is similar to the above step 102 and will not be elaborated here.

[0134] Step 206: Perform data fusion on all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector.

[0135] Optionally, step 206 includes the following sub-steps:

[0136] When the equipment fault probability vector is not an empty set, input each equipment fault probability into the fusion function to generate the equipment comprehensive fault probability respectively.

[0137] In an optional embodiment of the present application, a can be set as the fusion framework of three systems. represents a set of possible values of the fault probability of each system. is the equipment fault status label corresponding to device k at time T. l The value of 1 indicates the fault status, and the value of 0 indicates the non-fault status. s = 1, 2, 3 respectively represent the distribution automation system, the user electricity consumption information acquisition system, and the meteorological information acquisition system. k is the ID of the device. k = 1 represents the transformer device label, k = 2i represents the line i device label, k = 3j represents the user j device label, and k = 4 represents the device in the normal state label. The function m: 2

[0138] →[0, 1] (2 U →[0, 1] (2 U is the set composed of all subsets of U) needs to satisfy the following conditions.

[0139]

[0140] Wherein, m is the probability distribution function of the substation area fault, is the probability of the kth device failing at time T in the s system, that is, l At the moment, is the device fault probability within the device fault probability vector generated by the target DBN model, is an empty set.

[0141] When is At this time, the comprehensive device fault probability after fusing the device fault probabilities of each type in the fusion framework U is 0; when is not At this time, the comprehensive device fault probability after fusing the device fault probabilities of each type in the fusion framework U is:

[0142]

[0143] Wherein, and are Any subset of, is for T l The device fault status label corresponding to the kth device after fusion at the moment, is the probability of the kth device failing in the s system at time T l At the moment.

[0144] Generate a device comprehensive fault probability vector using the comprehensive device fault probabilities of all devices.

[0145] Step 207, calculate the belief function values corresponding to the comprehensive device fault probabilities of each device in the device comprehensive fault probability vector, and select the device label to which the maximum belief function value belongs as the substation area fault device label.

[0146] Optionally, step 207 includes the following sub-steps:

[0147] Calculate the belief function values corresponding to the comprehensive device fault probabilities of each device in the device comprehensive fault probability vector;

[0148] Select the device label to which the maximum belief function value belongs as the substation area fault device label.

[0149] Step 208, based on the comprehensive fault probability, power outage coefficient, and service life corresponding to the substation area fault device label, calculate the fault warning time corresponding to the substation area fault device label.

[0150] Optionally, step 208 includes the following sub-steps:

[0151] Construct a fitting function between the historical statistical power outage fault warning time, power outage coefficient, and equipment service life corresponding to the comprehensive failure probability of each device;

[0152] Use the least squares method to calculate the parameters to be determined in the fitting function and generate the target fitting function;

[0153] Input the comprehensive failure probability, power outage coefficient, and service life corresponding to the fault device label in the substation area into the target fitting function to generate the fault warning time corresponding to the fault device label in the substation area.

[0154] In an alternative embodiment of the present application, referring to Figure 4 , taking the low-voltage topology of a typical substation area as an example, the substation area line is divided into several segments, and the substation area includes several users. The power outage coefficients corresponding to different fault types in this substation area are shown in Table 1.

[0155] Table 1 Power outage coefficients for different fault types

[0156]

[0157] Taking the typical substation area topology as an example, a fitting sample set at a moment consists of A total of 22 sample points are constructed. Taking the historical statistical sample set at T moments as the fitting sample set, there are a total of 22T fitting sample points, and construct The fitting function with t g,k , that is

[0158]

[0159] In the formula, t is the historical statistical power outage fault warning time, σ is the power outage coefficient, p is the comprehensive failure probability of the device, n is the service life of the device, and α and β are the parameters to be determined.

[0160] Use the least squares method to obtain the parameters to be determined α and β, and the sum of the squared residuals of the fitting function is:

[0161]

[0162] In the formula, g = 1,..., T, k ∈ K = {1, 21,..., 2i, 31,..., 3j, 4}, t g,k Is the historical statistical power outage fault warning time of device k at the g-th moment, α and β are the parameters to be determined, n k Is the service life of device k, σ g,k Is the power outage coefficient of device k at the g-th moment, p g,k Is the comprehensive failure probability of device k at the g-th moment.

[0163] To minimize the residual as much as possible, the right - hand side function of the sum of squared residuals of the fitting function is partially differentiated with respect to α and β respectively and set equal to 0, obtaining the fitting equation system, that is

[0164]

[0165] In the formula, g = 1, …, T, k ∈ K = {1, 21, …, 2i, 31, …, 3j, 4}, 22T is the number of sampling points of the fitting samples, n k is the service life of equipment k, σ g,k is the power outage coefficient of equipment k at the g - th moment, p g,k is the comprehensive failure probability of equipment k at the g - th moment, t g,k is the historical statistical power outage fault warning time of equipment k at the g - th moment.

[0166] Substitute the 22T fitting sample points into the fitting equation system to obtain the parameters α and β to be solved; substitute the calculated parameters α and β to be solved into the fitting function to obtain the target fitting function of the typical distribution area, that is

[0167]

[0168] In the formula, t is the equipment fault warning time, σ is the power outage coefficient, n is the service life of the equipment, and p is the comprehensive failure probability of the equipment.

[0169] After the target fitting function is determined, during the operation stage, input the comprehensive failure probability, power outage coefficient, and service life corresponding to the distribution area fault equipment label into the target fitting function to generate the fault warning time corresponding to the distribution area fault equipment label.

[0170] Step 209: Generate and output warning information by using the distribution area fault equipment label and the fault warning time.

[0171] In the embodiment of the present application, the specific implementation process of step 209 is similar to the above - mentioned step 106 and will not be elaborated here.

[0172] In the embodiments of the present application, a variety of substation area data are acquired and normalized to obtain a variety of data to be warned; an initial DBN model is constructed by using a multi-layer RBM network structure and a single-layer BP network structure, and the initial DBN model is trained by using historical training data to generate a target DBN model; the target DBN model is used to respectively perform equipment fault prediction on various data to be warned, generate multiple equipment fault probability vectors and perform data fusion, so as to generate an equipment comprehensive fault probability vector; the belief function values corresponding to the respective equipment comprehensive fault probabilities in the equipment comprehensive fault probability vector are calculated, and the equipment label to which the maximum belief function value belongs is selected as the substation area fault equipment label; based on the comprehensive fault probability, power outage coefficient and service life corresponding to the substation area fault equipment label, the fault warning time corresponding to the substation area fault equipment label is calculated; the substation area fault equipment label and the fault warning time are used to generate and output a warning message, thereby solving the technical problems in the prior art that the influence of protection or switch refusal to operate, misoperation and information distortion on fault diagnosis is not considered in the fault warning diagnosis of the distribution network, and the multi-source information of the substation area cannot be fully and effectively utilized, resulting in low accuracy of substation area fault warning.

[0173] Please refer to Figure 5 , Figure 5 A substation area power outage fault warning system provided by an embodiment of the present invention, including:

[0174] A data acquisition module 501, configured to acquire a variety of substation area data and normalize them to obtain a variety of data to be warned;

[0175] A fault prediction module 502, configured to respectively perform equipment fault prediction on various data to be warned through a target DBN model to generate multiple equipment fault probability vectors;

[0176] A data fusion module 503, configured to perform data fusion by using all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector;

[0177] A fault location module 504, configured to calculate the belief function values corresponding to the respective equipment comprehensive fault probabilities in the equipment comprehensive fault probability vector, and select the equipment label to which the maximum belief function value belongs as the substation area fault equipment label;

[0178] A fault warning time module 505, configured to calculate the fault warning time corresponding to the substation area fault equipment label based on the comprehensive fault probability, power outage coefficient and service life corresponding to the substation area fault equipment label;

[0179] A fault information warning output module 506, configured to generate and output a warning message by using the substation area fault equipment label and the fault warning time;

[0180] Optionally, the system further includes:

[0181] An initial model construction module for constructing an initial DBN model using a multi-layer RBM network structure and a single-layer BP network structure;

[0182] A training data acquisition module for obtaining historical data of the power distribution area and normalizing it to obtain historical training data;

[0183] A model training module for training the initial DBN model with the historical training data to generate a target DBN model.

[0184] Optionally, the model training module includes:

[0185] A training data input sub-module for inputting the historical training data into the first-layer RBM network structure;

[0186] A data adjustment sub-module for generating target neuron activation state data through the multi-layer RBM network structure and delivering it to the BP network structure, and performing reverse fine-tuning on each layer of the RBM network structure through the BP network structure;

[0187] A return sub-module for, if the number of times of reverse fine-tuning has not reached the preset iteration threshold, jumping to execute the step of inputting the historical training data into the first-layer RBM network structure;

[0188] A target model generation sub-module for, if the number of times of reverse fine-tuning reaches the preset iteration threshold, using the multi-layer RBM network structure and the BP network structure at the current moment to generate a target DBN model.

[0189] Optionally, the data adjustment sub-module includes:

[0190] An input unit for inputting the historical training data into the front visible layer;

[0191] An activation unit for calculating the intermediate neuron activation state data of the associated hidden layer through the front visible layer;

[0192] A visible layer update unit for selecting an activation probability from the intermediate neuron activation state data according to a preset rule to reconstruct the front visible layer and generate an updated visible layer;

[0193] A hidden layer update unit for calculating and reconstructing the neuron activation probability of the hidden layer associated with the updated visible layer to generate an updated hidden layer;

[0194] A selection unit for using the intermediate neuron activation state data as new historical training data and selecting the first visible layer from the unselected RBM network structures as the new front visible layer at the current moment;

[0195] A jump unit for jumping to execute the step of inputting historical training data into the front-view layer until all RBM network structures are selected.

[0196] Optionally, the data fusion module includes:

[0197] A device comprehensive failure probability generation sub-module for inputting each device failure probability into a fusion function to generate a device comprehensive failure probability respectively when the device failure probability vector is not an empty set;

[0198] A device comprehensive failure probability vector generation sub-module for generating a device comprehensive failure probability vector using all the device comprehensive failure probabilities.

[0199] Optionally, the fault location module includes:

[0200] A belief function value calculation module for calculating the belief function values corresponding to each device comprehensive failure probability in the device comprehensive failure probability vector;

[0201] A substation area fault device label selection module for selecting the device label to which the maximum belief function value belongs as the substation area fault device label.

[0202] Optionally, the fault warning time module includes:

[0203] A fitting function construction module for constructing a fitting function between the historical statistical power outage fault warning time, power outage coefficient, and device service life corresponding to each device comprehensive failure probability;

[0204] A target fitting function generation module for calculating the parameters to be determined in the fitting function using the least squares method to generate a target fitting function;

[0205] A fault warning time calculation module for inputting the comprehensive failure probability, power outage coefficient, and service life corresponding to the substation area fault device label into the target fitting function to generate the fault warning time corresponding to the substation area fault device label.

[0206] The power outage fault warning system provided by the embodiment of the present application obtains various data to be warned according to the data acquisition module, inputs the various data to be warned into the fault prediction module for fault prediction, and generates multiple device fault probability vectors; inputs the multiple device fault probability vectors into the data fusion module to generate a comprehensive device fault probability vector and inputs it into the fault location module to obtain the power distribution area fault device label; generates a fault warning time through the fault warning time module based on the power distribution area fault device label; based on the power distribution area fault device label and the fault warning time, inputs them into the fault information warning output module to generate and output a warning message, thereby solving the technical problem in the prior art that the fault warning diagnosis for the distribution network does not consider the impact on fault diagnosis when protection or switches refuse to operate, malfunction, or information distortion occurs in case of a fault, and cannot comprehensively and effectively utilize the multi-source information of the power distribution area, resulting in a low accuracy of power distribution area fault warning.

[0207] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, module, sub-module, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0208] In several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be in an electrical, mechanical, or other form.

[0209] As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or equivalently replace some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for warning of power outage faults in a transformer substation area, characterized in that, Including: Obtain various substation area data and perform normalization to obtain various data to be warned; Respectively perform equipment fault prediction on various data to be warned through a target DBN model to generate multiple equipment fault probability vectors; Perform data fusion using all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector; Calculate the belief function values corresponding to the respective equipment comprehensive fault probabilities within the equipment comprehensive fault probability vector, and select the equipment label to which the maximum belief function value belongs as the substation area fault equipment label; Based on the comprehensive fault probability, power outage coefficient, and service life corresponding to the substation area fault equipment label, calculate the fault warning time corresponding to the substation area fault equipment label; Generate a warning message and output it using the substation area fault equipment label and the fault warning time.

2. The power outage fault warning method for the substation area according to claim 1, wherein, The method further includes: Construct an initial DBN model using a multi-layer RBM network structure and a single-layer BP network structure; Obtain substation area historical data and perform normalization to obtain historical training data; Train the initial DBN model through the historical training data to generate a target DBN model.

3. The warning method for the power outage fault in the transformer area according to claim 2, characterized in that, The step of training the initial DBN model through the historical training data to generate a DBN model includes: Input the historical training data into the first-layer RBM network structure; Generate target neuron activation state data through the multi-layer RBM network structure and transmit it to the BP network structure, and perform reverse fine-tuning on each layer of the RBM network structure through the BP network structure; If the number of times of reverse fine-tuning has not reached the preset iteration threshold, jump to execute the step of inputting the historical training data into the first-layer RBM network structure; If the number of times of reverse fine-tuning reaches the preset iteration threshold, use the multi-layer RBM network structure and the BP network structure at the current moment to generate a target DBN model.

4. The method for warning about power outage faults in a substation area according to claim 3, wherein, The RBM network structure includes a visible layer and a hidden layer; the step of generating target neuron activation state data through the multi-layer RBM network structure and transmitting it to the BP network structure, and performing reverse fine-tuning on each layer of the RBM network structure through the BP network structure includes: Select the first visible layer from the multi-layer RBM network structure as the head visible layer at the current moment; Input the historical training data into the head visible layer; Calculate the intermediate neuron activation state data of the associated hidden layer through the head visible layer; Select an activation probability from the intermediate neuron activation state data according to a preset rule to reconstruct the head visible layer to generate an updated visible layer; Calculate the neuron activation probability of the hidden layer associated with the updated visible layer and reconstruct it to generate an updated hidden layer; Use the intermediate neuron activation state data as new historical training data, and select the first visible layer from the unselected RBM network structure as the new head visible layer at the current moment; Jump to execute the step of inputting the historical training data into the head visible layer until all the RBM network structures are selected.

5. The warning method for power outage faults in a substation area according to claim 1, characterized in that, The step of performing data fusion using all the equipment fault probability vectors to generate an equipment comprehensive fault probability vector includes: When the device failure probability vector is not an empty set, input the device failure probabilities into the fusion function to generate the comprehensive device failure probabilities respectively; Generate a comprehensive device failure probability vector using all the comprehensive device failure probabilities; Among them, the fusion function is: Wherein: is T l the device fault status label corresponding to the k-th device after fusion at time T, and is any subset of, the probability that the k-th device in the s system fails at time T l , is T l the device fault status label corresponding to the k-th device at time T, taking the value of 1 indicates the fault status, taking the value of 0 indicates the non-fault status. s = 1, 2, 3 represent the distribution automation system, the user electricity consumption information acquisition system, and the meteorological information acquisition system respectively. k is the ID of the device, k ∈ K = {1, 21,..., 2i, 31,..., 3j, 4}. k = 1 represents the transformer device label, k = 2i represents the line i device label, k = 3j represents the user j device label, and k = 4 represents the device in the normal state label.

6. The method for predicting the power outage fault in the substation area according to claim 1, wherein, The step of calculating the belief function values corresponding to the comprehensive device failure probabilities in the comprehensive device failure probability vector and selecting the device label to which the maximum belief function value belongs as the substation area failure device label includes: Calculate the belief function values corresponding to the comprehensive device failure probabilities in the comprehensive device failure probability vector respectively; Select the device label to which the maximum belief function value belongs as the substation area failure device label; The belief function value is: Wherein: represents the lower limit estimate of the fault trust degree of device k after fusion at time T l ; represents the comprehensive fault probability of device k after fusion at time T l .

7. The method for warning of power outage faults in a transformer substation area according to claim 1, characterized in that, The step of calculating the failure warning time corresponding to the substation area failure device label based on the comprehensive failure probability, power outage coefficient and service life corresponding to the substation area failure device label includes: Construct a fitting function between the historical statistical power outage failure warning time, power outage coefficient and equipment service life corresponding to each of the comprehensive device failure probabilities; Use the least squares method to calculate the parameters to be solved in the fitting function to generate a target fitting function; Input the comprehensive failure probability, power outage coefficient and service life corresponding to the substation area failure device label into the target fitting function to generate the failure warning time corresponding to the substation area failure device label; The fitting function is: In the formula: t is the historical statistical power outage failure warning time, σ is the power outage coefficient, p is the comprehensive device failure probability, n is the equipment service life, and α and β are the parameters to be solved.

8. A power outage fault warning system for a power distribution area, characterized in that, Include: A data acquisition module for acquiring various substation area data and normalizing them to obtain various data to be warned; A fault prediction module for respectively performing device fault prediction on various data to be warned through a target DBN model to generate multiple device failure probability vectors; A data fusion module for fusing data using all the device failure probability vectors to generate a comprehensive device failure probability vector; A fault location module for calculating the belief function values corresponding to the comprehensive device failure probabilities in the comprehensive device failure probability vector respectively and selecting the device label to which the maximum belief function value belongs as the substation area failure device label; A fault warning time module for calculating the failure warning time corresponding to the substation area failure device label based on the comprehensive failure probability, power outage coefficient and service life corresponding to the substation area failure device label; A fault information warning output module for generating and outputting a warning information using the substation area failure device label and the failure warning time.

9. The warning system for power outage faults in the substation area according to claim 8, characterized in that The system further includes: An initial model construction module for constructing an initial DBN model using a multi-layer RBM network structure and a single-layer BP network structure; A training data acquisition module for acquiring substation area historical data and normalizing them to obtain historical training data; A model training module for training the initial DBN model through the historical training data to generate a target DBN model.

10. The power outage fault warning system for the substation area according to claim 9, wherein, The model training module includes: A training data input sub-module for inputting the historical training data into the first-layer RBM network structure; A data adjustment sub-module, which is used to generate target neuron activation state data through multiple layers of the RBM network structure and transmit it to the BP network structure, and perform reverse fine-tuning on each layer of the RBM network structure through the BP network structure; A return sub-module, which is used to jump to execute the step of inputting the historical training data into the first layer of the RBM network structure if the number of times of the reverse fine-tuning does not reach the preset iteration threshold; A target model generation sub-module, which is used to generate a target DBN model by using the RBM network structure and the BP network structure of multiple layers at the current moment if the number of times of the reverse fine-tuning reaches the preset iteration threshold.

Images