A transformer dissolved gas concentration prediction method based on D-S evidence theory optimization ensemble learning

Abstract

The present application relates to the technical field of transformer dissolved gas concentration prediction, in particular to a transformer dissolved gas concentration prediction method based on D-S evidence theory optimized ensemble learning, comprising the following steps: generating sub-datasets, training base learners on each sub-dataset to obtain multiple independent models, using the trained base learners to predict the to-be-predicted samples to obtain prediction results, calculating the basic probability assignment for the prediction results of each base learner, using D-S evidence theory to fuse the prediction results of all base learners and generating the final prediction value, and using the ensemble learning model optimized based on D-S evidence theory to predict the transformer dissolved gas concentration; the method is based on the Bagging method in ensemble learning and uses D-S evidence theory for the fusion of base learners, thereby improving the transformer dissolved gas concentration prediction accuracy of the ensemble learning model; the sequential least squares programming algorithm is adopted, which facilitates the processing of complex constraint conditions and improves the convergence speed and calculation efficiency.

Description

Technical Field

[0001] This invention relates to the technical field of transformer dissolved gas concentration prediction, and in particular to a method for predicting transformer dissolved gas concentration based on DS evidence theory-optimized ensemble learning. Background Technology

[0002] Transformers play a crucial role in power grid systems, primarily responsible for voltage transformation and power distribution. They convert high-voltage electricity generated by power plants into voltage levels suitable for transmission and distribution, ensuring efficient power transmission. Simultaneously, transformers provide electrical isolation, protecting equipment and personnel. However, transformers may face various faults during operation, making condition monitoring particularly important, with the prediction of dissolved gas concentration being critical. Changes in dissolved gas concentration in transformer oil can reflect internal fault conditions, such as overheating or breakdown. By monitoring and predicting gas concentrations, potential problems can be predicted in advance, allowing for the development of reasonable maintenance and repair plans and reducing unnecessary downtime.

[0003] Therefore, invention patents such as the one disclosed in patent CN118779810A, which describes a method and system for diagnosing transformer faults based on dissolved gas analysis in oil, and the one disclosed in patent CN117829822B, which describes a method and system for early warning of power transformer faults, have emerged to predict the dissolved gas concentration in transformers.

[0004] However, during use, it was found that since the prediction of dissolved gas concentration in transformers is a nonlinear problem, traditional statistical methods are generally ineffective in handling this problem. Deep learning methods have high data requirements and are computationally complex. Data in the power grid is difficult to meet their needs and can easily put too much computational pressure on the power grid system, resulting in poor practicality. Therefore, there is an urgent need for a transformer dissolved gas concentration prediction method based on DS evidence theory and optimized ensemble learning to improve the above problems. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention provides a method for predicting transformer dissolved gas concentration based on the Bagging method in ensemble learning and using DS evidence theory for the fusion of base learners. This improves the accuracy of the ensemble learning model in predicting transformer dissolved gas concentration and better handles uncertainties introduced by subset datasets, enhancing the model's robustness to noisy data. Furthermore, it employs a sequential least squares programming algorithm to iteratively update decision variables, ultimately finding the optimal solution that satisfies all constraints. This facilitates handling complex constraints and improves convergence speed and computational efficiency.

[0006] The present invention discloses a method for predicting dissolved gas concentration in transformers based on DS evidence theory optimization ensemble learning, comprising the following steps:

[0007] S1. Generate subsets, which means randomly selecting multiple independent subsets with replacement from the original dataset. In the Bagging method, random data samples in the training set are selected using the replacement method.

[0008] S2. Train a base learner for each subset of data to obtain multiple independent models, that is, train a decision tree model for each subset of data to obtain multiple independent decision tree models.

[0009] S3. Use the trained base learner to predict the sample to be predicted to obtain the prediction result. That is, use multiple decision tree models obtained in S2 to predict the sample to be predicted at the same time to obtain the prediction results of different decision tree models.

[0010] S4. Calculate the basic probability assignment for the prediction results of each base learner. The initial basic probability assignment value is the average value. Then, the basic probability assignment value is continuously updated according to the optimization algorithm with the goal of minimizing the mean square error of the prediction results.

[0011] S5. Use DS evidence theory to fuse the prediction results of all base learners and generate the final prediction value. After obtaining the fused prediction result, evaluate the result. When the mean square error of the prediction result tends to converge, proceed to S6; otherwise, return to S4 to recalculate the basic probability assignment.

[0012] S6. An ensemble learning model optimized based on DS evidence theory is used to predict the concentration of dissolved gas in transformers. The prediction results are compared with the actual results to analyze the accuracy of the prediction method.

[0013] S7. If the accuracy of the prediction meets the required standards after comparison, it can be directly applied to the prediction of dissolved gas concentration in actual transformers.

[0014] Preferably, the training process of the decision tree model in S2 includes:

[0015] Select a splitting feature, traverse all features and their possible splitting points, and select the optimal feature for splitting based on information gain;

[0016] Recursively constructing a tree involves repeating the feature selection and node splitting process for each subset until a stopping condition is met. The stopping conditions include the tree depth exceeding a preset maximum value, the number of samples in a node falling below a set threshold, and further splitting failing to significantly improve information gain.

[0017] Preferably, in the prediction process of S3, starting from the root node, the decision tree is traversed downwards based on the feature value. Each node makes a judgment based on the selected feature and threshold. If the feature value meets a certain condition, it moves downwards along one path; otherwise, it moves downwards along another path until a leaf node is reached, and the predicted value of that leaf node is returned.

[0018] Preferably, the specific method for calculating the basic probability assignment for the prediction results of each base learner in S4 includes:

[0019] During the optimization process, sequential least squares programming is used as the optimization algorithm, and its objective function satisfies the following relationship: Where Y i In this application, y represents the actual result value. i The aforementioned fusion prediction value.

[0020] Preferably, in the optimization process, the rule for updating the basic probability allocation value in each iteration satisfies the following relationship: m k+1 =m k +α k *d k , where α k Let d be the step size. k The core of the calculated direction is to construct the objective function of mean square error, set the constraint of the sum of basic probability assignment values, and find the optimal solution by iteratively updating the basic probability assignment values.

[0021] Preferably, in step S5, the specific method for fusing the prediction results of all base learners using DS evidence theory to generate the final prediction value includes:

[0022] In DS evidence theory, the basic probability assignment represents the confidence level of a hypothesis or event. For all possible events, the sum of the basic probability assignments is 1. When DS evidence theory is applied to the fusion of prediction results in the Bagging method, the fused prediction values ​​satisfy the following relationship: Where, m i f is the base probability assignment value for the i-th base learner. i For the i-th base learner, f i (x) represents the prediction result of the i-th base learner. The application of DS evidence theory in the code provides an effective result fusion mechanism for the Bagging model. By weighting the predictions of different models, information can be better integrated and the uncertainty of prediction can be reduced. By flexibly setting the trust assignment value, the prediction performance of the model can be improved. In particular, it effectively solves the problem of unstable fusion prediction results caused by the randomness of the subset dataset in ensemble learning.

[0023] Preferably, the specific method for predicting the dissolved gas concentration of the transformer using the ensemble learning model optimized based on DS evidence theory in step S6 includes:

[0024] The dataset to be predicted is divided into sub-datasets, and a base learner is used to predict the data for each sub-dataset. After obtaining the prediction results of each base learner, the prediction results of all base learners are fused using DS evidence theory based on the optimal basic probability assignment obtained through multiple iterations of S4 and S5. Finally, the fused prediction results are compared with the actual values.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1. Based on the Bagging method in ensemble learning and using DS evidence theory for the fusion of base learners, the accuracy of the ensemble learning model in predicting the dissolved gas concentration of transformers is improved, and it can better handle the uncertainty brought by the subset dataset, thus enhancing the robustness of the model to noisy data.

[0027] 2. The sequential least squares programming algorithm is adopted, which updates the decision variables step by step through iteration, and finally finds the optimal solution that satisfies all constraints. This makes it easier to handle complex constraints and improves the convergence speed and computational efficiency. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the transformer dissolved gas concentration prediction method based on DS evidence theory optimization ensemble learning according to the present invention.

[0029] Figure 2 This is a radar chart comparing the accuracy of the experimental prediction results of this invention. Detailed Implementation

[0030] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0031] Example

[0032] like Figure 1 As shown, a method for predicting dissolved gas concentration in transformers based on DS evidence theory-optimized ensemble learning includes the following steps:

[0033] S1. Generate subsets, which means randomly selecting multiple independent subsets with replacement from the original dataset. In the Bagging method, random data samples in the training set are selected using the replacement method.

[0034] S2. Train a base learner for each subset of data to obtain multiple independent models, that is, train a decision tree model for each subset of data to obtain multiple independent decision tree models. The training process of the decision tree model includes:

[0035] Select a splitting feature, traverse all features and their possible splitting points, and select the optimal feature for splitting based on information gain;

[0036] Recursively constructing a tree involves repeating the feature selection and node splitting process for each subset until a stopping condition is met. The stopping conditions include the tree depth exceeding a preset maximum value, the number of samples in a node falling below a set threshold, and further splitting failing to significantly improve information gain.

[0037] S3. Use the trained base learner to predict the sample to be predicted to obtain the prediction result. That is, use multiple decision tree models obtained in S2 to predict the sample to be predicted at the same time to obtain the prediction results of different decision tree models. When predicting, start from the root node and traverse down the decision tree according to the feature value. Each node will make a judgment based on the selected feature and threshold. If the feature value meets a certain condition, it will move down along one path. Otherwise, it will move down along another path until it reaches the leaf node and returns the prediction value of the leaf node.

[0038] S4. Calculate the basic probability assignment for the prediction results of each base learner. The initial basic probability assignment value is the average value. Then, based on the optimization algorithm, the basic probability assignment value is continuously updated with the goal of minimizing the mean square error of the prediction results. The specific methods include:

[0039] During the optimization process, sequential least squares programming is used as the optimization algorithm, and its objective function satisfies the following relationship: Where Y i In this application, y represents the actual result value. i The aforementioned fusion prediction value; the rule for updating the basic probability assignment value in each iteration satisfies the following relationship: m k+1 =m k +α k *d k , where α k Let d be the step size. k The core of the calculated direction is to construct the objective function of mean square error, set the constraint of the sum of basic probability assignment values, and find the optimal solution by iteratively updating the basic probability assignment values.

[0040] S5. Use DS evidence theory to fuse the prediction results of all base learners and generate the final prediction value. After obtaining the fused prediction result, evaluate the result. If the mean square error of the prediction result tends to converge, proceed to S6; otherwise, return to S4 to recalculate the basic probability assignment. The specific method includes:

[0041] In DS evidence theory, the basic probability assignment represents the confidence level of a hypothesis or event. For all possible events, the sum of the basic probability assignments is 1. When DS evidence theory is applied to the fusion of prediction results in the Bagging method, the fused prediction values ​​satisfy the following relationship: Where, m i f is the base probability assignment value for the i-th base learner. i For the i-th base learner, f i (x) represents the prediction result of the i-th base learner. The application of DS evidence theory in the code provides an effective result fusion mechanism for the Bagging model. By weighting the predictions of different models, information can be better integrated and the uncertainty of prediction can be reduced. By flexibly setting the trust assignment value, the prediction performance of the model can be improved. In particular, it effectively solves the problem of unstable fusion prediction results caused by the randomness of the subset dataset in ensemble learning.

[0042] S6. An ensemble learning model optimized based on DS evidence theory is used to predict the dissolved gas concentration in transformers. The prediction results are compared with actual results to analyze the accuracy of the prediction method. Specific methods include:

[0043] The dataset to be predicted is divided into sub-datasets, and the prediction of each sub-dataset is performed by a base learner. After obtaining the prediction results of each base learner, the prediction results of all base learners are fused using DS evidence theory based on the optimal basic probability allocation obtained by multiple iterations of S4 and S5. Finally, the fused prediction results are compared with the actual values.

[0044] S7. If the accuracy of the prediction meets the required standards after comparison, it can be directly applied to the prediction of dissolved gas concentration in actual transformers.

[0045] Experimental Example

[0046] An experiment was conducted based on 21 continuous time gas concentration change data from a substation, including four gases: H2, C2H6, C2H4, and CH4. The first 20 data points were used as a training set to train an ensemble learning model optimized based on the DS evidence theory. The trained model was then used to predict the concentration of each gas at the next moment.

[0047] Based on the above experimental setup, experiments were conducted, and the results are shown in Table 1. This application defines the following relationship for the prediction accuracy of each model for a certain gas:

[0048]

[0049] And the overall prediction accuracy of each model satisfies the following relationship:

[0050]

[0051] Where C represents the number of gas categories. Y represents the actual concentration of the i-th gas. i This is the value predicted by the model;

[0052] Table 1 Prediction Results

[0053]

[0054] Therefore, based on the above formulas, we calculated the prediction accuracies of the ensemble learning method and the ensemble learning method optimized based on DS evidence theory. In H2 gas prediction, the ensemble learning prediction accuracy was approximately 98.7%, while the ensemble learning method optimized based on DS evidence theory achieved approximately 99%. In C2H6 gas prediction, the ensemble learning prediction accuracy was approximately 98.1%, while the ensemble learning method optimized based on DS evidence theory achieved approximately 99.7%. In C2H4 gas prediction, the ensemble learning prediction accuracy was approximately 99.6%, while the ensemble learning method optimized based on DS evidence theory achieved approximately 99.9%. In CH4 gas prediction, the ensemble learning prediction accuracy was approximately 97.7%, while the ensemble learning method optimized based on DS evidence theory achieved approximately 96.8%. As shown in Table 2, although the ensemble learning method optimized based on DS evidence theory achieved slightly lower prediction accuracy than the ensemble learning method in CH4 gas prediction, it improved the overall accuracy.

[0055] Table 2 Prediction Accuracy Results

[0056]

[0057] like Figure 2 As shown, for easy and intuitive comparison, this application has drawn a radar chart based on the data in Table 2. From the magnitude of the values ​​of the two on a single axis and the area ratio of the two, it can be seen more intuitively that the prediction accuracy of the DS evidence theory optimized ensemble learning method is lower than that of the ensemble learning method only in the prediction of CH4 gas. The prediction accuracy of the other three gases and the overall prediction accuracy are significantly higher than that of the ensemble learning method. Therefore, it can be proved that the DS evidence theory optimized ensemble learning transformer dissolved gas concentration prediction method proposed in this application is effective and improves the accuracy and robustness of the original ensemble learning method.

[0058] Since the objective function of the sequential least squares programming algorithm included in the DS evidence theory-optimized ensemble learning transformer dissolved gas concentration prediction method proposed in this application is to minimize the overall mean square error, this application calculates the mean square error values ​​of the prediction results of the DS evidence theory-optimized ensemble learning and ensemble learning methods based on experimental results, thereby indirectly verifying the effectiveness of the optimization algorithm. The mean square error is calculated according to the following relationship: The results calculated according to the above formula are shown in Table 3. It can be seen that the overall mean square error of the prediction results based on DS evidence theory optimization ensemble learning is only about 22% of the mean square error of the prediction results based on ensemble learning. This shows that the DS evidence theory optimization ensemble learning method has greatly reduced the mean square error of the overall prediction results. At the same time, it can also show the stability of the DS evidence theory optimization ensemble learning transformer dissolved gas concentration prediction method proposed in this application, as well as the effectiveness of the sequence least squares programming optimization algorithm.

[0059] Table 3. Prediction Mean Square Error Results

[0060]

[0061] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting dissolved gas concentration in transformers based on DS evidence theory-optimized ensemble learning, characterized in that, Includes the following steps: S1. Generate subsets, which means randomly selecting multiple independent subsets with replacement from the original dataset. In the Bagging method, random data samples in the training set are selected using the replacement method. S2. Train a base learner for each subset of data to obtain multiple independent models, that is, train a decision tree model for each subset of data to obtain multiple independent decision tree models. S3. Use the trained base learner to predict the sample to be predicted to obtain the prediction result. That is, use multiple decision tree models obtained in S2 to predict the sample to be predicted at the same time to obtain the prediction results of different decision tree models. S4. Calculate the basic probability assignment for the prediction results of each base learner. The initial basic probability assignment value is the average value. Then, the basic probability assignment value is continuously updated according to the optimization algorithm with the goal of minimizing the mean square error of the prediction results. S5. Use DS evidence theory to fuse the prediction results of all base learners and generate the final prediction value. After obtaining the fused prediction result, evaluate the result. When the mean square error of the prediction result tends to converge, proceed to S6; otherwise, return to S4 to recalculate the basic probability assignment. S6. An ensemble learning model optimized based on DS evidence theory is used to predict the concentration of dissolved gas in transformers. The prediction results are compared with the actual results to analyze the accuracy of the prediction method. S7. If the accuracy of the prediction meets the required standards after comparison, it can be directly applied to the prediction of dissolved gas concentration in actual transformers. In S5, the specific method for fusing the prediction results of all base learners using DS evidence theory to generate the final prediction value of 0 includes: In DS evidence theory, the basic probability assignment represents the confidence level of a hypothesis or event. For all possible events, the sum of the basic probability assignments is 1. When DS evidence theory is applied to the fusion of prediction results in the Bagging method, the fused prediction values ​​satisfy the following relationship: ,in, Assign basic probability values ​​to the i-th base learner. For the i-th base learner, For the prediction result of the i-th base learner, the application of DS evidence theory in the code provides an effective result fusion mechanism for the Bagging model. By weighting the predictions of different models, information can be better integrated and the uncertainty of prediction can be reduced. By flexibly setting the trust assignment value, the prediction performance of the model can be improved. In particular, it effectively solves the problem of unstable fusion prediction results caused by the randomness of the subset dataset in ensemble learning. The specific method for calculating the basic probability assignment for the prediction results of each base learner in S4 includes: During the optimization process, sequential least squares programming is used as the optimization algorithm, and its objective function satisfies the following relationship: ,in The values ​​used in this application are actual results. The aforementioned fusion prediction value.

2. In the optimization process, the rules for updating the basic probability allocation values ​​in each iteration satisfy the following relationship: ,in Step size, The core of the calculated direction is to construct the objective function of mean square error, set the constraint of the sum of basic probability assignment values, and find the optimal solution by iteratively updating the basic probability assignment values.

3. The method for predicting dissolved gas concentration in transformers based on DS evidence theory optimization ensemble learning as described in claim 1, characterized in that, The training process of the decision tree model in S2 includes: Select a splitting feature, traverse all features and their possible splitting points, and select the optimal feature for splitting based on information gain; Recursively constructing a tree involves repeating the feature selection and node splitting process for each subset until a stopping condition is met. The stopping conditions include the tree depth exceeding a preset maximum value, the number of samples in a node falling below a set threshold, and further splitting failing to significantly improve information gain.

4. The method for predicting dissolved gas concentration in transformers based on DS evidence theory optimization ensemble learning as described in claim 1, characterized in that, In the prediction process described in S3, starting from the root node, the system traverses downwards along the decision tree based on the feature values. Each node makes a judgment based on the selected features and thresholds. If the feature value meets a certain condition, it moves downwards along one path; otherwise, it moves downwards along another path until a leaf node is reached, at which point the predicted value of that leaf node is returned.

5. The method for predicting dissolved gas concentration in transformers based on DS evidence theory optimization ensemble learning as described in claim 1, characterized in that, The specific method for predicting the dissolved gas concentration in transformers using the ensemble learning model optimized based on DS evidence theory in S6 includes: The dataset to be predicted is divided into sub-datasets, and a base learner is used to predict the data for each sub-dataset. After obtaining the prediction results of each base learner, the prediction results of all base learners are fused using DS evidence theory based on the optimal basic probability assignment obtained through multiple iterations of S4 and S5. Finally, the fused prediction results are compared with the actual values.

Images