Method for predicting whole life cycle cost of main equipment of transformer substation

By normalizing, convolution processing and multi-dimensional transformation of the LCC data of the main equipment of the substation, combining the prediction of the fusion model and the GRU model, the problems of data anomalies and insufficient timing length are solved, and the accuracy of the full life cycle cost is improved. It provides more reliable economic comparison support for power grid project decisions.

CN119027150APending Publication Date: 2024-11-26ECONOMIC TECH RES INST OF STATE GRID HENAN ELECTRIC POWER +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410992423.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the prior art, the accuracy of the calculation results of the full life cycle cost of the main equipment of the substation is insufficient, mainly due to data abnormalities and insufficient data timing length, resulting in insufficient support for power grid project decision-making.

Method used

A method of predicting the life cycle of the substation main equipment is adopted, including obtaining LCC data, performing normalization processing, convolution processing, adding timing position information, generating composite data, similarity analysis, multi-dimensional to single-dimensional conversion, sliding window strategy , fusion model prediction and GRU model secondary prediction to ensure the accuracy of the prediction results.

Benefits of technology

By improving data quality and model prediction accuracy, enhancing the impact of relevant parameters on the prediction results, and improving the accuracy of the life cycle cost of the substation main equipment, providing more reliable economic comparison support for power grid project decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027150A_ABST
    Figure CN119027150A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of transformer main equipment cost prediction, and discloses a transformer substation main equipment whole life cycle cost prediction method. During use, main equipment LCC data of a transformer substation are selected and subjected to normalization processing; the problem that the model prediction precision is low due to the fact that the number of parameter data of the LCC data of the transformer substation is different can be solved; in addition, the preprocessed data is subjected to convolution processing, the similarity between the data is solved, the data is converted from multiple dimensions to one dimension, the influence of relevant parameters on the final prediction result can be enhanced, and therefore the quality of the LCC data is improved; and finally, learning prediction is carried out through the fusion model, prediction is carried out after fusion model learning is completed to obtain a fourth data set, and a difference value between the fourth data set and the third data set is sent to the GRU model for secondary prediction, so that the accuracy of the whole life cycle cost of the main equipment of the transformer substation can be improved. And thus, the method has important significance for successfully realizing economical efficiency comparison and selection of the main equipment of the transformer substation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of transformer main equipment cost prediction, and in particular to a method for predicting the full life cycle cost of transformer main equipment. Background Art

[0002] In the power system, substation is an indispensable part, which plays the role of transforming voltage, realizing electric energy conversion, regulating power quality, distributing electric energy, protecting power system safety, power control and electric energy metering.

[0003] At present, as the economic selection of substation main equipment becomes increasingly important for power grid project decision-making, the calculation results of the life cycle cost (LCC) of the main equipment, which is the determining factor of economic selection, are also very important.

[0004] However, the current life cycle cost calculation results of main equipment are mainly determined by the quality of LCC raw data. However, due to the short operation time of some substation main equipment or equipment replacement, the LCC raw data of the main equipment has data anomalies and insufficient data time series length. The emergence of these problems leads to the inaccuracy of the life cycle cost calculation results of the main equipment, making it difficult to compare the economic efficiency of substation main equipment, resulting in insufficient support for power grid project decision-making. Summary of the invention

[0005] In view of the shortcomings of the background technology, the present invention provides a method for predicting the full life cycle cost of the main equipment of a substation, which can achieve accurate calculation of the full life cycle cost calculation results of the main equipment of the substation.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for predicting the full life cycle cost of substation main equipment, comprising the following steps:

[0007] S1: Acquire LCC data of the substation, where the LCC data is data related to the full life cycle cost of the main equipment, including multiple parameter data;

[0008] S2: normalizing the LCC data to generate preprocessed data;

[0009] S3: performing convolution processing on the preprocessed data to generate convolution data;

[0010] Adding time sequence position information to the convolution data based on the time sequence relationship of the LCC data to generate first composite data;

[0011] reconstructing the LCC data by generating a function, and setting a ratio of the reconstructed LCC data in the overall data, and superimposing the ratio with the first composite data to generate second composite data;

[0012] Acquire similarities between different types of data in the LCC data, and convert the second composite data from multi-dimensional to single-dimensional based on the similarities to generate dimension-converted data;

[0013] S4: reconstructing the dimension of the dimension conversion data by using a sliding window strategy to generate a first data set, and dividing the first data set into a second data set and a third data set according to a division ratio;

[0014] S5: performing time series prediction on the second data set by adding the fusion model with the improved strategy to obtain a time prediction result set;

[0015] S6: comparing the time prediction result set with the second data set, and judging whether the fusion model has fully learned the features of the second data set based on the difference between the time prediction result set and the second data set, if the fusion model has fully learned the features of the second data set, executing step S7, otherwise, increasing the number of second data sets, and then re-executing step S5;

[0016] S7: predicting the third data set by using the learned fusion model, and waiting for the fourth data set;

[0017] S8: Send the difference between the fourth data set and the third data set to the GRU model for prediction to obtain a prediction difference, and combine the prediction difference with the fourth data set as a final prediction data set.

[0018] In some embodiments, the LCC data includes the voltage level of the substation, the geographical location of the substation, the pollution level of the substation, the equipment commissioning time, the original value of the main equipment assets, the daily material costs of the substation, labor costs, maintenance and inspection costs, personnel inspection costs, emergency repair costs, insurance premiums, vehicle usage fees and other costs.

[0019] In some implementations, the normalization process in step S2 is as follows:

[0020] The parameter data of LCC data is recorded as parameter sequence U i =[u1,u2,...,u n ],u n is the nth parameter, and the parameter sequence U is calculated by the following formula i Perform normalization processing to obtain preprocessed data U′;

[0021]

[0022] In a certain embodiment, in step S3, before performing convolution processing on the preprocessed data U′, some year data are first extracted from the preprocessed data U′ by random sampling, and the remaining data is recorded as U″;

[0023] When performing convolution processing, the following formula is used for convolution processing:

[0024] X = conv1d(U″)

[0025] Wherein, X is the first composite data; convld is the convolution function;

[0026] In step S3, the expression of the added timing position relationship is as follows:

[0027]

[0028] Among them, P t is the time series position information function of LCC data; i∈(1,N), j∈(1,D), N and D are the data dimensions of the convolution data, N represents the horizontal dimension of the time length of the convolution data, and D represents the vertical dimension of the convolution data; k is a positive integer;

[0029] The generation formula of the first composite data is as follows:

[0030] X′=X+P t

[0031] Where X′ is the first composite data;

[0032] The way to reconstruct LCC data is as follows:

[0033] q = sort(randperm(L))

[0034] Where q is the randomly sorted generated data set from the 1st to the Lth year; sort is the time series sorting function; randperm is the random integer generation function; L represents the horizontal dimension of the time length of the LCC data;

[0035] The method for setting the proportion of reconstructed LCC data in the overall data is as follows:

[0036] r = round(rate(q))

[0037] Wherein r represents the proportion; round is the rounding function; rate is the set ratio value;

[0038] Add the first composite data X′ and the proportion to obtain the second composite data X″;

[0039] The formula for obtaining the similarity between different similar data in LCC data is as follows:

[0040]

[0041] where a i is the correlation degree, i∈(1,n); η is the correlation degree threshold;

[0042] The process of converting the second composite data from multi-dimensional to single-dimensional by using the similarity is as follows:

[0043] First based on a i To generate the correlation matrix A, the expression of the correlation matrix A is as follows:

[0044] A=[a1,a2,…,a n ] T

[0045] Then generate the dimension transformation data Y based on the following formula:

[0046] Y=X″A.

[0047] In some implementations, the loss function of the fusion model in step S5 is a dual-structure association loss function.

[0048] In some embodiments, the dual structure association loss function LOSS is expressed as follows:

[0049]

[0050] The expression formula of XS1 is as follows:

[0051]

[0052] The expression formula of XS1 is as follows:

[0053]

[0054] The expression formula of G is as follows:

[0055]

[0056] Among them, XS1 represents the individual similarity of the real data before the fusion model prediction, XS2 represents the individual similarity of the predicted data after the fusion model prediction, G represents the step size in the data collection process, λ and K represent the influencing parameters, and N represents the prediction step size of this prediction task.

[0057] In some implementations, the formula for calculating the prediction difference in step S8 is as follows:

[0058]

[0059] Where n1 represents the predicted value, n2 represents the actual value, P represents the absolute value function, and Err(n) represents the amplitude error of the final prediction. If the amplitude error is less than the error threshold, the prediction is successful, and the predicted difference is combined with the fourth data set as the final predicted data set. Otherwise, a new prediction is performed.

[0060] In some implementations, in step S6, the difference between the time prediction result set and the second data set is judged by using goodness of fit, root mean square error or direct error.

[0061] In a certain implementation manner, in step S5, the TCN model suitable for time series prediction is selected in the fusion model and the Transformer model is fused according to the weight entropy assignment strategy, and the weight entropy assignment process is as follows:

[0062] First, the preliminary prediction results and their errors of different meta-learner models of the fusion model are calculated. The calculation formula is as follows:

[0063] P1={Y1,…,Y N}

[0064] P2={Y1,…,Y N}

[0065] Q={T1,...,T N}

[0066] Where P1 and P2 represent the preliminary prediction results of different meta-learners, N is the prediction step length, and set Q represents the true value;

[0067] Then, the initial prediction results of different meta-learners are subtracted from the true values ​​to obtain the initial prediction error set E. The calculation formula is as follows:

[0068]

[0069] Then calculate the sum of all errors in the prediction step size according to the following calculation formula:

[0070]

[0071] Finally, the allocation weights of different meta-learners are calculated according to the following formula.

[0072] C M =ln(1+e M i )-ln(R+e M i )

[0073] Among them C M Represents the predicted weight values ​​of different meta-learners.

[0074] Compared with the prior art, the present invention has the following beneficial effects: firstly, by selecting the LCC data of the main equipment of the substation and normalizing the LCC data, the present invention can eliminate the problem of low model prediction accuracy caused by different quantitative dimensions of parameter data of the substation LCC data;

[0075] By performing convolution processing on the preprocessed data and obtaining the similarity between the data to achieve the transformation of the data from multi-dimensional to single-dimensional, the influence of the relevant parameters on the final prediction results can be enhanced, thereby improving the quality of the LCC data;

[0076] Finally, learning and prediction are carried out through the fusion model, and the fourth data set is predicted after the fusion model learning is completed. The difference between the fourth data set and the third data set is sent to the GRU model for secondary prediction. This can improve the accuracy of the full life cycle cost of the main equipment of the substation, which is of great significance for the successful realization of the economic selection of the main equipment of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 is a flow chart of the present invention in an embodiment;

[0078] Figure 2 This is a schematic diagram of the prediction results of the present invention in the examples. DETAILED DESCRIPTION

[0079] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0080] like Figure 1 As shown, a method for predicting the full life cycle cost of main equipment of a substation includes the following steps:

[0081] S1: Obtain the LCC data of the substation. The LCC data is data related to the life cycle cost of the main equipment, including multiple parameter data.

[0082] Specifically, in this embodiment, the LCC data includes data such as the voltage level of the substation, the geographical location of the substation, the pollution level of the substation, the time of equipment commissioning, the original value of the main equipment assets, the daily material cost of the substation, labor cost, maintenance and repair cost, personnel inspection cost, emergency repair cost, insurance premium, vehicle usage fee and other costs;

[0083] Among them, data such as substation voltage level, substation geographical location, substation location pollution level, equipment commissioning time, main equipment asset original value, substation daily material costs, labor costs, maintenance and repair costs, personnel inspection costs, emergency repair costs, insurance premiums, vehicle usage fees and other costs are all equivalent to one parameter data.

[0084] S2: Normalize the LCC data to generate preprocessed data.

[0085] Specifically, in this embodiment, the normalization process is as follows:

[0086] The parameter data of LCC data is recorded as parameter sequence U i =[u1,u2,...,u n ],u n is the nth parameter, and the parameter sequence U is calculated by the following formula i Perform normalization processing to obtain preprocessed data U′;

[0087]

[0088] In actual use, by normalizing the LCC data, the problem of low model prediction accuracy caused by different quantitative dimensions of parameter data of substation LCC data can be eliminated.

[0089] S3: performing convolution processing on the preprocessed data to generate convolution data;

[0090] Adding time sequence position information to the convolution data based on the time sequence relationship of the LCC data to generate first composite data;

[0091] Reconstructing the LCC data by generating a function, and setting the ratio of the reconstructed LCC data in the overall data, and superimposing the ratio with the first composite data to generate second composite data;

[0092] The similarities between different types of data in the LCC data are obtained, and the second composite data is converted from multi-dimensional to single-dimensional based on the similarities to generate dimension-converted data.

[0093] Specifically, in this embodiment, in step S3, before the pre-processed data U′ is subjected to convolution processing, some year data are first extracted from the pre-processed data U′ by random sampling, and the remaining data is recorded as U″;

[0094] When performing convolution processing, the following formula is used for convolution processing:

[0095] X = conv1d(U″)

[0096] Wherein, X is the first composite data; convld is the convolution function;

[0097] In step S3, the expression of the added timing position relationship is as follows:

[0098]

[0099] Among them, Pt is the time series position information function of LCC data; i∈(1,N), j∈(1,D), N and D are the data dimensions of the convolution data, N represents the horizontal dimension of the time length of the convolution data, and D represents the vertical dimension of the convolution data; k is a positive integer;

[0100] The generation formula of the first composite data is as follows:

[0101] X′=X+P t

[0102] Where X′ is the first composite data;

[0103] The way to reconstruct LCC data is as follows:

[0104] q = sort(randperm(L))

[0105] Where q is the randomly sorted generated data set from the 1st to the Lth year; sort is the time series sorting function; randperm is the random integer generation function; L represents the horizontal dimension of the time length of the LCC data;

[0106] The method for setting the proportion of reconstructed LCC data in the overall data is as follows:

[0107] r = round(rate(q))

[0108] Where r represents the percentage; round is the rounding function; rate is the set ratio value;

[0109] Add the first composite data X′ and the proportion to obtain the second composite data X″;

[0110] The formula for obtaining the similarity between different similar data in LCC data is as follows:

[0111]

[0112] where a i is the correlation degree, i∈(1,n); η is the correlation degree threshold;

[0113] The process of converting the second composite data from multidimensional to single-dimensional by similarity is as follows:

[0114] First based on a i To generate the correlation matrix A, the expression of the correlation matrix A is as follows:

[0115] A=[a1,a2,...,a n ] T

[0116] Then generate the dimension transformation data Y based on the following formula:

[0117] Y=X″A.

[0118] In actual use, step S3 can highlight the characteristics of LCC data to facilitate subsequent data prediction; in addition, by converting data from multi-dimensional to single-dimensional, the impact of relevant parameters on the results of cost prediction can be enhanced.

[0119] S4: reconstructing the dimension conversion data by a sliding window strategy to generate a first data set, and dividing the first data set into a second data set and a third data set according to a division ratio.

[0120] Specifically, in this embodiment, the division ratio of the first data set can be set according to actual needs. Exemplarily, the second data set and the third data set can be divided in a ratio of seven to three.

[0121] S5: By adding the fusion model of the improved strategy, time series learning and prediction are performed on the second data set to obtain a time prediction result set.

[0122] Specifically, in this embodiment, the loss function of the fusion model in step S5 is a dual-structure association loss function.

[0123] Specifically, in this embodiment, the expression formula of the dual structure association loss function LOSS is as follows:

[0124]

[0125] The expression formula of XS1 is as follows:

[0126]

[0127] The expression formula of XS1 is as follows:

[0128]

[0129] The expression formula of G is as follows:

[0130]

[0131] Among them, XS1 represents the individual similarity of the real data before the fusion model prediction, XS2 represents the individual similarity of the predicted data after the fusion model prediction, G represents the step size in the data collection process, λ and K represent the influencing parameters, which are usually calculated based on the engineering needs, with a value range of 0 to 1, and N represents the prediction step size of this prediction task; the real data set at the current time is y t-N+i ; The prediction data set at the current time is y t-N+i+A .

[0132] Specifically, in this embodiment, in step S5, the TCN model suitable for time series prediction is selected in the fusion model and the Transformer model is fused according to the strategy of weight entropy assignment. The process of weight entropy assignment is as follows:

[0133] First, calculate the preliminary prediction results and errors of different meta-learner models of the fusion model. The calculation formula is as follows:

[0134] P1={Y1,...,Y N}

[0135] P2={Y1,...,Y N}

[0136] Q={T1,...,T N}

[0137] Where P1 and P2 represent the preliminary prediction results of different meta-learners, N is the prediction step length, and set Q represents the true value;

[0138] Then, the initial prediction results of different meta-learners are subtracted from the true values ​​to obtain the initial prediction error set E. The calculation formula is as follows:

[0139]

[0140] Then calculate the sum of all errors in the prediction step size according to the following calculation formula:

[0141]

[0142] Finally, the allocation weights of different meta-learners are calculated according to the following formula.

[0143] C M =ln(1+e M i )-ln(R+e M i )

[0144] Among them C M Represents the predicted weight values ​​of different meta-learners.

[0145] S6: Compare the time prediction result set with the second data set, and determine whether the fusion model has fully learned the characteristics of the second data set based on the difference between the time prediction result set and the second data set. If the fusion model has fully learned the characteristics of the second data set, execute step S7; otherwise, increase the number of second data sets and then re-execute step S5.

[0146] Specifically, in this embodiment, in step S6, the difference between the time prediction result set and the second data set is processed by using goodness of fit, root mean square error or direct error to determine whether the fusion model has fully learned the characteristics of the second data set.

[0147] The prediction results P1 and P2 are integrated to obtain data set S4, which is expressed as follows:

[0148] S4={S 41 ,...,S 4i}

[0149] In the formula, S 41 represents the first data in data set S4, S 4i Represents the i-th data in data set S4.

[0150] The goodness of fit is calculated as follows:

[0151]

[0152]

[0153] The calculation formula of the root mean square error is as follows:

[0154]

[0155] The calculation formula of direct error is as follows:

[0156]

[0157] S7: predicting the third data set by using the learned fusion model, and waiting for the fourth data set;

[0158] S8: Send the difference between the fourth data set and the third data set to the GRU model for prediction to obtain a prediction difference, and combine the prediction difference with the fourth data set as a final prediction data set.

[0159] Specifically, in this embodiment, the formula for calculating the prediction difference in step S8 is as follows:

[0160]

[0161] Where n1 represents the predicted value, n2 represents the actual value, P represents the absolute value function, and Err(n) represents the amplitude error of the final prediction. If the amplitude error is less than the error threshold, the prediction is successful, and the predicted difference is combined with the fourth data set as the final predicted data set. Otherwise, a new prediction is performed.

[0162] Specifically, in this embodiment, the size of the error threshold can be set according to actual needs. Exemplarily, the error threshold can be 5%.

[0163] In actual use, the present invention can eliminate the problem of low model prediction accuracy caused by different quantitative dimensions of parameter data of substation LCC data by selecting LCC data of main equipment of substation and normalizing the LCC data;

[0164] Secondly, by performing convolution processing on the preprocessed data and obtaining the similarity between the data to achieve the transformation of the data from multi-dimensional to single-dimensional, the influence of the relevant parameters on the final prediction results can be enhanced, thereby improving the quality of the LCC data;

[0165] In addition, the prediction result schematic diagram of the present invention is as follows Figure 2 As shown, from Figure 2 It can be obtained that learning and prediction are carried out through the fusion model, and the fourth data set is predicted after the fusion model learning is completed. The difference between the fourth data set and the third data set is sent to the GRU model for secondary prediction. This can improve the accuracy of the full life cycle cost of the main equipment of the substation, which is of great significance for the successful realization of the economic comparison and selection of the main equipment of the substation.

[0166] The above is based on the present invention as an inspiration. Through the above description, relevant staff can make various changes and modifications without departing from the technical idea of ​​this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for predicting the full life cycle cost of substation main equipment, characterized in that: The steps include: S1: Acquire LCC data of the substation, where the LCC data is data related to the full life cycle cost of the main equipment, including multiple parameter data; S2: normalizing the LCC data to generate preprocessed data; S3: performing convolution processing on the preprocessed data to generate convolution data; Adding time sequence position information to the convolution data based on the time sequence relationship of the LCC data to generate first composite data; reconstructing the LCC data by generating a function, and setting a ratio of the reconstructed LCC data in the overall data, and superimposing the ratio with the first composite data to generate second composite data; Acquire similarities between different types of data in the LCC data, and convert the second composite data from multi-dimensional to single-dimensional based on the similarities to generate dimension-converted data; S4: reconstructing the dimension of the dimension conversion data by using a sliding window strategy to generate a first data set, and dividing the first data set into a second data set and a third data set according to a division ratio; S5: performing time series prediction on the second data set by adding the fusion model with the improved strategy to obtain a time prediction result set; S6: comparing the time prediction result set with the second data set, and judging whether the fusion model has fully learned the features of the second data set based on the difference between the time prediction result set and the second data set, if the fusion model has fully learned the features of the second data set, executing step S7, otherwise, increasing the number of second data sets, and then re-executing step S5; S7: predicting the third data set by using the learned fusion model, and waiting for the fourth data set; S8: Send the difference between the fourth data set and the third data set to the GRU model for prediction to obtain a prediction difference, and combine the prediction difference with the fourth data set as a final prediction data set.

2. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: The LCC data includes substation voltage level, substation geographical location, substation site pollution level, equipment commissioning time, main equipment asset original value, substation daily material costs, labor costs, maintenance and repair costs, personnel inspection costs, emergency repair costs, insurance premiums, vehicle usage fees and other costs.

3. A method for predicting the full life cycle cost of substation main equipment according to claim 1 or 2, characterized in that: The normalization process in step S2 is as follows: The parameter data of LCC data is recorded as parameter sequence U i =[u1,u2,...,u n ],u n is the nth parameter, and the parameter sequence U is calculated by the following formula i Perform normalization processing to obtain preprocessed data U′; 4. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: In step S3, before performing convolution processing on the preprocessed data U′, some year data are extracted from the preprocessed data U′ by random sampling, and the remaining data is recorded as U″; When performing convolution processing, the following formula is used for convolution processing: X = conv1d(U″) Wherein, X is the first composite data; convld is the convolution function; In step S3, the expression of the added timing position relationship is as follows: Among them, P t is the time series position information function of LCC data; i∈(1,N), j∈(1,D), N and D are the data dimensions of the convolution data, N represents the horizontal dimension of the time length of the convolution data, and D represents the vertical dimension of the convolution data; k is a positive integer; The generation formula of the first composite data is as follows: X’=X+P t Where X′ is the first composite data; The way to reconstruct LCC data is as follows: q = sort(randperm(L)) Where q is the randomly sorted generated data set from the 1st to the Lth year; sort is the time series sorting function; randperm is the random integer generation function; L represents the horizontal dimension of the time length of the LCC data; The method for setting the proportion of reconstructed LCC data in the overall data is as follows: r = round(rate(q)) Wherein r represents the proportion; round is the rounding function; rate is the set ratio value; Add the first composite data X′ and the proportion to obtain the second composite data X″; The formula for obtaining the similarity between different types of data in LCC data is as follows: where a i is the correlation degree, i∈(1,n); η is the correlation degree threshold; The process of converting the second composite data from multi-dimensional to single-dimensional using the similarity is as follows: First based on a i To generate the correlation matrix A, the expression of the correlation matrix A is as follows: <h2 style=";text-align:left;direction:ltr">A=[a1,a2,…,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ]<h2 style=";text-align:left;direction:ltr"> T Then generate the dimension transformation data Y based on the following formula: Y=X″A.

5. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: The loss function of the fusion model in step S5 is a dual-structure association loss function.

6. A method for predicting the full life cycle cost of substation main equipment according to claim 5, characterized in that: The expression formula of the dual structure association loss function LOSS is as follows: The expression formula of XS1 is as follows: The expression formula of XS1 is as follows: The expression formula of G is as follows: Among them, XS1 represents the individual similarity of the real data before the fusion model prediction, XS2 represents the individual similarity of the predicted data after the fusion model prediction, G represents the step size in the data collection process, λ and K represent the influencing parameters, and N represents the prediction step size of this prediction task.

7. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: The formula for calculating the prediction difference in step S8 is as follows: Where n1 represents the predicted value, n2 represents the actual value, P represents the absolute value function, and Err(n) represents the amplitude error of the final prediction. If the amplitude error is less than the error threshold, the prediction is successful, and the predicted difference is combined with the fourth data set as the final predicted data set. Otherwise, a new prediction is performed.

8. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: In step S6, the difference between the time prediction result set and the second data set is processed by using goodness of fit, root mean square error or direct error, so as to determine whether the fusion model has fully learned the characteristics of the second data set.

9. A method for predicting the full life cycle cost of substation main equipment according to claim 1, characterized in that: In step S5, the TCN model suitable for time series prediction is selected in the fusion model and the Transformer model is fused according to the weight entropy assignment strategy. The weight entropy assignment process is as follows: First, the preliminary prediction results and their errors of different meta-learner models of the fusion model are calculated. The calculation formula is as follows: P1={Y1,…,Y N } P2={Y1,...,Y N } Q={T1,…,T N } Where P1 and P2 represent the preliminary prediction results of different meta-learners, N is the prediction step length, and set Q represents the true value; Then, the initial prediction results of different meta-learners are subtracted from the true values ​​to obtain the initial prediction error set E. The calculation formula is as follows: Then calculate the sum of all errors in the prediction step size according to the following calculation formula: Finally, the allocation weights of different meta-learners are calculated according to the following calculation formula: C M ln(1+e M i )-ln(R+e M i ) Among them C M Represents the predicted weight values ​​of different meta-learners.