A method, device and system for evaluating transformer status
Through the joint training and data fusion of multiple probabilistic neural network models, the accuracy and reliability problems of a single neural network model in transformer state recognition are solved, and more accurate failure probability evaluation and state evaluation are achieved, improving the transformer operating state evaluation accuracy and calculation accuracy.
Patent Information
- Application Number
- CN202211344415.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-31
- Publication Date
- 2024-12-24
- Estimated Expiration
- 2042-10-31
AI Technical Summary
The existing single neural network model has low accuracy and reliability in transformer state recognition and fault assessment, and it is difficult to effectively reflect the complex operating state and fault characteristics of the transformer.
Multiple probabilistic neural network models are used for training and data preprocessing, combined with data fusion method, normalized processing is performed through the starting layer of the probabilistic neural network model, conditional probability is calculated and probability summed, and finally, failure occurs through data fusion Probability.
It improves the accuracy and reliability of transformer status evaluation, simplifies the data processing process, reduces the complexity of transformer operation data, and enhances the computing accuracy and reliability of fault diagnosis.
Smart Images
Figure CN115713027B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power equipment status diagnosis, and in particular to a transformer status evaluation method, device and system. Background Art
[0002] In recent years, the number of transformers in operation has continued to grow. As key equipment in power plants and substations, the operating status of transformers may affect the safety and stability of the entire power system at any time. At present, many transformers in my country are close to their service life or even far exceed their design life, resulting in an increase in the failure rate of some transformers year by year. In order to ensure the safe and reliable operation of transformers and save maintenance costs, power grid companies have gradually changed the maintenance method of transformers from accident maintenance and regular maintenance to condition maintenance. Condition maintenance is based on the results of condition assessment, so it is urgent to carry out transformer condition assessment.
[0003] For example, patent document CN114118288A discloses a transformer state assessment method based on deep learning, including the following steps: 1) data acquisition; 2) data preprocessing; 3) establishing a CNN deep convolutional neural network model; 4) model training; 5) self-learning; This method uses the learning algorithm of deep learning theory to complete the extraction of fault characteristic gas, and then uses the trained deep convolutional neural network model to identify and evaluate the state of the transformer. However, the transformer state quantities are numerous and the mechanisms are complex. The operating state assessment is a comprehensive evaluation process and has certain randomness, fuzziness and uncertainty. Therefore, the accuracy and reliability of transformer state recognition relying on a single neural network model are not high. Summary of the invention
[0004] The present invention provides a transformer state assessment method, device and system, which can perform more effective state identification on the transformer and obtain more accurate fault probability and state assessment results.
[0005] A transformer condition assessment method, comprising:
[0006] Collect historical transformer status data and label the status to form a sample training set;
[0007] Establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training respectively until the probabilistic neural network models converge;
[0008] Collect multiple sets of transformer status data corresponding to multiple moments within a preset time period;
[0009] Inputting multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation, respectively, to obtain multiple probability state sets;
[0010] Data fusion is performed on a plurality of the probability state sets to obtain the probability of a fault occurring.
[0011] Furthermore, the transformer status data includes operating data of the transformer and / or index data obtained by calculating the transformer status.
[0012] Furthermore, the probabilistic neural network model includes a starting layer, and the starting layer is used to preprocess the sample training set and the transformer state data.
[0013] Furthermore, the preprocessing of the sample training set and the transformer state data includes normalization processing, which is expressed by the following formula:
[0014]
[0015] Among them, x is the transformer state data set or sample training set, x i is the i-th transformer status data or the i-th historical transformer status data in the sample training set, To calculate the normalized value obtained, N is the total number of data in the transformer state data set or sample training set, max(x) represents the maximum data in the transformer state data set or sample training set, and min(x) represents the minimum data in the transformer state data set or sample training set.
[0016] Furthermore, the probabilistic neural network model also includes a sample layer and a summation layer;
[0017] Multiple sets of transformer state data are input into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets, including:
[0018] Inputting the plurality of sets of transformer state data into the starting layers of the trained plurality of probabilistic neural network models respectively for preprocessing;
[0019] The probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state;
[0020] The conditional probabilities of each state corresponding to each sample point are input into the summation layer and summed up respectively to obtain the probability values corresponding to each state;
[0021] According to the probability values corresponding to each state, a probability state set is obtained;
[0022] Multiple probabilistic neural network models output probability state sets to obtain multiple probability state sets.
[0023] Furthermore, the conditional probability is calculated by the following formula:
[0024]
[0025] Among them, σ is the smoothing factor, N is the total number of sample points, and x is the normal value of a certain state when the transformer is running. is the normalized value of the i-th transformer state data or the i-th historical transformer state data in the sample training set, is the probability value of the i-th transformer state data or the i-th historical transformer state data in the sample training set being a certain state, and π is a constant.
[0026] Furthermore, data fusion is performed on the plurality of probability state sets to obtain the probability of a fault occurrence, including:
[0027] Selecting two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set;
[0028] A state probability state set is selected from the remaining probability state sets to be merged with the fused probability state set to obtain a new fused probability state set, and this step is repeated until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the fault occurrence probability.
[0029] Further, two probability state sets are selected from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated by the following formula:
[0030] z r =p r q r / (1-∑p r q r );
[0031] r=1,2,…,m;
[0032] Among them, z r is the probability of the rth state in the fusion probability state set, p r and q r are the probabilities of the rth state in the two probability state sets respectively.
[0033] A transformer state assessment device applied to the above method comprises:
[0034] The sample module is used to collect historical transformer status data and label the status to form a sample training set;
[0035] The model training module is used to establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training, until the probabilistic neural network models converge;
[0036] A collection module, used to collect multiple sets of transformer status data corresponding to multiple moments in a preset time period;
[0037] A state calculation module is used to input multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets;
[0038] The fusion module is used to perform data fusion on the multiple probability state sets to obtain the probability of fault occurrence.
[0039] Furthermore, the transformer status data includes operating data of the transformer and / or index data obtained by calculating the transformer status.
[0040] Furthermore, the probabilistic neural network model includes a starting layer, and the starting layer is used to preprocess the sample training set and the transformer state data.
[0041] Furthermore, the preprocessing of the sample training set and the transformer state data by the model training module and the state calculation module includes normalization processing, which is expressed by the following formula:
[0042]
[0043] Among them, x is the transformer state data set or sample training set, x i is the i-th transformer status data or the i-th historical transformer status data in the sample training set, To calculate the normalized value obtained, N is the total number of data in the transformer state data set or sample training set, max(x) represents the maximum data in the transformer state data set or sample training set, and min(x) represents the minimum data in the transformer state data set or sample training set.
[0044] Furthermore, the probabilistic neural network model also includes a sample layer and a summation layer;
[0045] The state calculation module is also used to input multiple groups of transformer state data into the starting layers of multiple trained probabilistic neural network models for preprocessing respectively; the probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state; the conditional probability of each state corresponding to each sample point is input into the summation layer for summation respectively, and the probability value corresponding to each state is obtained; according to the probability value corresponding to each state, a probability state set is obtained; and multiple probabilistic neural network models output the probability state set to obtain multiple probability state sets.
[0046] Furthermore, the conditional probability is calculated by the following formula:
[0047]
[0048] Among them, σ is the smoothing factor, N is the total number of sample points, and x is the normal value of a certain state when the transformer is running. is the normalized value of the i-th transformer state data or the i-th historical transformer state data in the sample training set, is the probability value of the i-th transformer state data or the i-th historical transformer state data in the sample training set being a certain state, and π is a constant.
[0049] Furthermore, the fusion module is also used to select two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set; select a state probability state set from the remaining probability state sets to fuse with the fused probability state set to obtain a new fused probability state set, and repeat this step until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the probability of fault occurrence.
[0050] Further, two probability state sets are selected from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated by the following formula:
[0051] z r =p r q r / (1-∑p r q r );
[0052] r=1,2,…,m;
[0053] Among them, z r is the probability of the rth state in the fusion probability state set, p r and q r are the probabilities of the rth state in the two probability state sets respectively.
[0054] A transformer state assessment system comprises a processor and a storage device, wherein the storage device stores a plurality of instructions, and the processor is used to read the plurality of instructions and execute the above method.
[0055] The transformer status assessment method, device and system provided by the present invention at least include the following
[0056] Beneficial effects:
[0057] (1) The transformer operating status fault probability model constructed using probabilistic neural networks can not only reflect the transformer operating status, but also achieve a quantitative description of the distribution range of transformer historical data. The data fusion method used can effectively analyze and process multi-source data and fuse them into a set of results. It can make full use of multiple information to obtain status results and fault probabilities, and can improve the accuracy of transformer operating status assessment.
[0058] (2) The probabilistic neural network is improved. The operating data of the transformer is normalized at the starting layer of the probabilistic neural network model to simplify the data processing process and reduce the complexity of the transformer operating data. When the indicator of the transformer operating data is a one-dimensional vector, the conditional probability calculation expression is simplified to make the network structure simple.
[0059] (3) The improved probabilistic neural network model and data fusion are applied in the fault diagnosis of transformers, realizing the reasonable and efficient combination of probabilistic neural network and data fusion methods. The transformer state assessment probability model finally generated effectively improves the calculation accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 The present invention provides a flow chart of an embodiment of a transformer status assessment method.
[0061] Figure 2 A schematic structural diagram of an embodiment of a probabilistic neural network model in the transformer state assessment method provided by the present invention.
[0062] Figure 3 The present invention provides a flowchart of an embodiment of probabilistic neural network model data processing in the transformer state assessment method provided by the present invention.
[0063] Figure 4 The present invention provides a flowchart of an embodiment of data fusion in the transformer state assessment method provided by the present invention.
[0064] Figure 5 A schematic structural diagram of an embodiment of a transformer state assessment device provided by the present invention.
[0065] Figure 6 A schematic structural diagram of an embodiment of a transformer status assessment system provided by the present invention. DETAILED DESCRIPTION
[0066] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0067] refer to Figure 1 In some embodiments, a transformer state assessment method is provided, comprising:
[0068] S1, collect historical transformer status data and mark the status to form a sample training set;
[0069] S2, establishing multiple probabilistic neural network models, and inputting different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training respectively, until the probabilistic neural network models converge;
[0070] S3, collecting multiple sets of transformer status data corresponding to multiple moments in a preset time period;
[0071] S4, inputting the multiple sets of transformer state data into the trained multiple probabilistic neural network models for preprocessing, conditional probability calculation and probability summation, to obtain multiple probability state sets;
[0072] S5. Perform data fusion on the multiple probability state sets to obtain the probability of fault occurrence.
[0073] Specifically, in step S1, historical transformer state data can be obtained through a variety of monitoring means, including transformer operation data and / or indicator data obtained by calculating the transformer state. For example, historical transformer state data may include but is not limited to current, resistance, heat generation, power consumption, vibration intensity, etc. when operating under various states. Such transformer state data is easy to obtain and can quickly respond to changes in the transformer state. The indicator data obtained by calculating the transformer state may include, for example, the peak value, waveform index, margin index, kurtosis index, etc. of a certain operating data. Such indicator data can well characterize the operating state of the transformer, and the corresponding historical transformer state data under different states are analyzed and labeled to form a sample training set. The formed sample training set includes rich data, which can improve the accuracy of subsequent model training.
[0074] Furthermore, in step S2, multiple probabilistic neural network models are established, and different multiple groups of sample training sets are respectively input into the multiple probabilistic neural network models for preprocessing and training, wherein the values of the operating data and the values of the index data in each of the multiple groups of sample training sets are different, for example, historical transformer state data at multiple moments can be collected to form multiple groups of sample training sets. Multiple groups of different sample training sets can make full use of multiple information to obtain state results and fault probabilities, and the prediction accuracy and precision of the obtained multiple neural network probability models are higher.
[0075] Among them, reference Figure 2 The probabilistic neural network model includes a starting layer 1, which is used to preprocess the sample training set and the transformer state data.
[0076] Specifically, the preprocessing of the sample training set and the transformer state data includes normalization processing, which is expressed by the following formula:
[0077]
[0078] Among them, x is the transformer state data set or sample training set, x i is the i-th transformer status data or the i-th historical transformer status data in the sample training set, To calculate the normalized value obtained, N is the total number of data in the transformer state data set or sample training set, max(x) represents the maximum data in the transformer state data set or sample training set, and min(x) represents the minimum data in the transformer state data set or sample training set.
[0079] The normalized sample training set and the transformer state data are easier to form sample points of the sample layer in the probabilistic neural network.
[0080] Further, refer to Figure 2 , the probabilistic neural network model also includes a sample layer 2 and a summation layer 3;
[0081] refer to Figure 3 In step S3, multiple sets of transformer state data are respectively input into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets, including:
[0082] S31, respectively inputting the multiple groups of transformer state data into the starting layers of the trained multiple probabilistic neural network models for preprocessing;
[0083] S32, the probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state;
[0084] S33, inputting the conditional probabilities of each state corresponding to each sample point into the summation layer for summing up respectively, to obtain the probability value corresponding to each state;
[0085] S34, obtaining a probability state set according to the probability values corresponding to each state;
[0086] S35. Multiple probabilistic neural network models output probability state sets to obtain multiple probability state sets.
[0087] Specifically, in step S31, multiple sets of transformer state data are respectively input into the starting layers of multiple trained probabilistic neural network models for normalization processing according to formula (1). The transformer operation data is normalized at the starting layer of the probabilistic neural network model, and the multi-dimensional operation data is simplified into one-dimensional data, which simplifies the data processing process and reduces the complexity of the transformer operation data.
[0088] Furthermore, in step S32, the conditional probability is calculated by the following formula:
[0089]
[0090] Among them, σ is the smoothing factor, N is the total number of sample points, and x is the normal value of a certain state when the transformer is running. is the normalized value of the i-th transformer state data or the i-th historical transformer state data in the sample training set, is the probability value of the i-th transformer state data or the i-th historical transformer state data in the sample training set being a certain state, and π is a constant.
[0091] Specifically, π refers to the magnitude of pi. The smoothing factor is obtained by multiplying the average minimum distance between sample points by a coefficient using an empirical formula, and the coefficient ranges from 1.1 to 1.4.
[0092] The average minimum distance between sample points is obtained by the following formula:
[0093]
[0094] d ij =|x i -x j |,i,j=1,2,…,N i≠j
[0095] Then the smoothing factor is obtained as:
[0096] g=1.1~1.4
[0097] Specifically, after calculating the conditional probability, the probability that a certain sample point, that is, a certain transformer state data belongs to a certain state can be obtained.
[0098] Furthermore, in step S33, the conditional probabilities of each state corresponding to each sample point are input into the summation layer for summation respectively to obtain the probability value corresponding to each state, that is, all the probabilities belonging to the state obtained by formula (2) in a certain state are added together to obtain the probability value corresponding to the state.
[0099] Furthermore, in step S34, the probability values corresponding to the multiple states are a probability state set.
[0100] Further, in step S35, each probabilistic neural network model outputs a probability state set, and multiple probabilistic neural network models output multiple probability state sets.
[0101] Further, refer to Figure 4 In step S5, data fusion is performed on the plurality of probability state sets to obtain the probability of a fault occurrence, including:
[0102] S51, selecting two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set;
[0103] S52, select a state probability state set from the remaining probability state sets to fuse with the fused probability state set to obtain a new fused probability state set, repeat this step until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the fault occurrence probability.
[0104] Among them, two probability state sets are selected from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated by the following formula:
[0105] z r =p r q r / (1-∑p r q r ); (3)
[0106] r=1,2,…,m;
[0107] Among them, z r is the probability of the rth state in the fusion probability state set, p r and q r are the probabilities of the rth state in the two probability state sets respectively.
[0108] Specifically, for example, suppose there are n probability state sets, each of which contains the possibility of m states occurring. The first probability state set is p = {p 1 ,p 2 ,…,p r}(r=1,2,…,m), the second probability state set is q={q 1 ,q 2 ,…,q r}(r=1,2,…,t), and after calculation by formula (3), we get the fusion probability z={z 1 ,z 2 ,…,z r}, the obtained fusion probability is fused with the third probability state set, and so on, until all probability state sets are fused, and the final fused probability state set is the probability of fault occurrence.
[0109] The transformer status assessment method provided in the above embodiment has at least the following beneficial effects:
[0110] (1) The transformer operating status fault probability model constructed using probabilistic neural networks can not only reflect the transformer operating status, but also achieve a quantitative description of the distribution range of transformer historical data; the data fusion method used can improve the accuracy of transformer operating status assessment;
[0111] (2) The probabilistic neural network is improved. When the indicator of the transformer operation data is a one-dimensional vector, the conditional probability calculation expression is simplified to make the network structure simpler.
[0112] (3) The improved probabilistic neural network model and data fusion are applied in the fault diagnosis of transformers, realizing the reasonable and efficient combination of probabilistic neural network and data fusion methods. The transformer state assessment probability model finally generated effectively improves the calculation accuracy and reliability.
[0113] refer to Figure 5 In some embodiments, a transformer state assessment device applied to the above method is also provided, comprising:
[0114] Sample module 201, used to collect historical transformer status data and perform status annotation to form a sample training set;
[0115] The model training module 202 is used to establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training respectively, until the probabilistic neural network models converge;
[0116] The acquisition module 203 is used to acquire multiple sets of transformer status data corresponding to multiple moments in a preset time period;
[0117] A state calculation module 204 is used to input multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets;
[0118] The fusion module 205 is used to perform data fusion on the plurality of probability state sets to obtain the probability of a fault occurring.
[0119] Furthermore, the transformer status data includes operating data of the transformer and / or index data obtained by calculating the transformer status.
[0120] Furthermore, the probabilistic neural network model includes a starting layer, and the starting layer is used to preprocess the sample training set and the transformer state data.
[0121] Furthermore, the model training module 202 and the state calculation module 204 preprocess the sample training set and the transformer state data including normalization, which is calculated by formula (1), which will not be repeated here.
[0122] Furthermore, the probabilistic neural network model also includes a sample layer and a summation layer;
[0123] The state calculation module 204 is also used to input multiple groups of transformer state data into the starting layers of multiple trained probabilistic neural network models for preprocessing respectively; the probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state; the conditional probability of each state corresponding to each sample point is input into the summation layer for summation respectively, and the probability value corresponding to each state is obtained; according to the probability value corresponding to each state, a probability state set is obtained; multiple probabilistic neural network models output the probability state set to obtain multiple probability state sets.
[0124] Furthermore, the conditional probability is calculated by formula (2), which will not be described in detail here.
[0125] Furthermore, the fusion module 205 is also used to select two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set; select a state probability state set from the remaining probability state sets to fuse with the fused probability state set to obtain a new fused probability state set, and repeat this step until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the probability of fault occurrence.
[0126] Furthermore, two probability state sets are selected from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated using formula (3) and formula (4), which will not be repeated here.
[0127] The transformer state assessment device provided in the above embodiment has at least the following beneficial effects:
[0128] (1) The transformer operating status fault probability model constructed using probabilistic neural network can not only reflect the transformer operating status, but also quantitatively describe the distribution range of transformer historical data; the data fusion method adopted can improve the accuracy of transformer operating status assessment;
[0129] (2) The probabilistic neural network is improved. When the indicator of the transformer operation data is a one-dimensional vector, the conditional probability calculation expression is simplified to make the network structure simpler.
[0130] (3) The improved probabilistic neural network model and data fusion are applied in the fault diagnosis of transformers, realizing the reasonable and efficient combination of probabilistic neural network and data fusion methods. The transformer state assessment probability model finally generated effectively improves the calculation accuracy and reliability.
[0131] refer to Figure 6 In some embodiments, a transformer state assessment system is further provided, comprising a processor 301 and a storage device 302, wherein the storage device stores a plurality of instructions, and the processor is used to read the plurality of instructions and perform the above method, for example:
[0132] Collect historical transformer status data and label the status to form a sample training set;
[0133] Establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training respectively until the probabilistic neural network models converge;
[0134] Collect multiple sets of transformer status data corresponding to multiple moments within a preset time period;
[0135] Inputting multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation, respectively, to obtain multiple probability state sets;
[0136] Data fusion is performed on a plurality of the probability state sets to obtain the probability of a fault occurring.
[0137] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. A transformer status assessment method, characterized in that: include: Collect historical transformer status data and label the status to form a sample training set; Establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training respectively until the probabilistic neural network models converge; Collect multiple sets of transformer status data corresponding to multiple moments within a preset time period; Inputting multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation, respectively, to obtain multiple probability state sets; Performing data fusion on a plurality of the probability state sets to obtain a probability of a fault occurring; The probabilistic neural network model includes a starting layer, and the starting layer is used to preprocess the sample training set and the transformer state data; The probabilistic neural network model also includes a sample layer and a summation layer; Multiple sets of transformer state data are input into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets, including: Inputting the plurality of sets of transformer state data into the starting layers of the trained plurality of probabilistic neural network models respectively for preprocessing; The probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state; The conditional probabilities of each state corresponding to each sample point are input into the summation layer and summed up respectively to obtain the probability values corresponding to each state; According to the probability values corresponding to each state, a probability state set is obtained; Multiple probabilistic neural network models output probability state sets to obtain multiple probability state sets; Performing data fusion on a plurality of the probability state sets to obtain the probability of a fault occurrence includes: Selecting two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set; Select a state probability state set from the remaining probability state sets to fuse with the fused probability state set to obtain a new fused probability state set, repeat this step until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the probability of fault occurrence; Select two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated by the following formula: z r =p r q r / (1-∑p r q r ); r＝1,2,…,m； Among them, z r is the probability of the rth state in the fusion probability state set, p r and q r are the probabilities of the rth state in the two probability state sets respectively; Each probability state set contains the possibility of m states occurring. The first probability state set is p = {p1, p2, ..., p r }(r=1,2,…,m), the second probability state set is q={q1,q2,…,q r }(r=1,2,…,t), through the formula z r =p q q r / (1-∑p r q r ) is calculated to obtain the fusion probability, and the obtained fusion probability is fused with the third probability state set, and so on, until all probability state sets are fused, and the final fusion probability state set is the probability of fault occurrence.
2. The method according to claim 1, characterized in that The transformer status data includes transformer operation data and / or index data obtained by calculating the transformer status.
3. The method according to claim 1, characterized in that The preprocessing of the sample training set and the transformer state data includes normalization processing, which is expressed by the following formula: Among them, x is the transformer state data set or sample training set, x i is the i-th transformer status data or the i-th historical transformer status data in the sample training set, is the normalized value obtained by calculation, N is the total number of data in the transformer state data set or sample training set, max(x) represents the maximum data in the transformer state data set or sample training set, and min(x) represents the minimum data in the transformer state data set or sample training set.
4. The method according to claim 1, characterized in that The conditional probability is calculated by the following formula: Among them, σ is the smoothing factor, N is the total number of sample points, and x is the normal value of a certain state when the transformer is running. is the normalized value of the i-th transformer state data or the i-th historical transformer state data in the sample training set, is the probability value of the i-th transformer state data or the i-th historical transformer state data in the sample training set being a certain state, and π is a constant.
5. A transformer state assessment device applied to the method according to any one of claims 1 to 4, characterized in that: include: The sample module is used to collect historical transformer status data and label the status to form a sample training set; The model training module is used to establish multiple probabilistic neural network models, and input different groups of sample training sets into the multiple probabilistic neural network models for preprocessing and training, until the probabilistic neural network models converge; A collection module, used to collect multiple sets of transformer status data corresponding to multiple moments in a preset time period; A state calculation module is used to input multiple sets of transformer state data into multiple trained probabilistic neural network models for preprocessing, conditional probability calculation and probability summation to obtain multiple probability state sets; The fusion module is used to perform data fusion on the multiple probability state sets to obtain the probability of fault occurrence.
6. The device according to claim 5, characterized in that The preprocessing of the sample training set and the transformer state data by the model training module and the state calculation module includes normalization processing, which is expressed by the following formula: Among them, x is the transformer state data set or sample training set, x i is the i-th transformer status data or the i-th historical transformer status data in the sample training set, is the normalized value obtained by calculation, N is the total number of data in the transformer state data set or sample training set, max(x) represents the maximum data in the transformer state data set or sample training set, and min(x) represents the minimum data in the transformer state data set or sample training set.
7. The device according to claim 5, characterized in that The probabilistic neural network model also includes a sample layer and a summation layer; The state calculation module is also used to input multiple groups of transformer state data into the starting layers of multiple trained probabilistic neural network models for preprocessing respectively; the probabilistic neural network model inputs the transformer state data preprocessed by the starting layer into the sample layer as sample points, and calculates the conditional probability of each sample point corresponding to each state; the conditional probability of each state corresponding to each sample point is input into the summation layer for summation, and the probability value corresponding to each state is obtained; and the probability state set is obtained according to the probability value corresponding to each state; Multiple probabilistic neural network models output probability state sets to obtain multiple probability state sets.
8. The device according to claim 5, characterized in that The fusion module is also used to select two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set; select a state probability state set from the remaining probability state sets to fuse with the fused probability state set to obtain a new fused probability state set, and repeat this step until the number of remaining probability state sets is 0, and the final fused probability state set obtained is the probability of fault occurrence.
9. The device according to claim 5, characterized in that Select two probability state sets from the multiple probability state sets for fusion to obtain a fused probability state set, which is calculated by the following formula: z r =p r q r / (1-∑p r q r ); r＝1,2,…,m； Among them, z r is the probability of the rth state in the fusion probability state set, p r and q r are the probabilities of the rth state in the two probability state sets respectively.
10. A transformer status assessment system, characterized in that: It comprises a processor and a storage device, wherein the storage device stores a plurality of instructions, and the processor is used to read the plurality of instructions and execute the method according to any one of claims 1 to 4.