A method and system for predicting the life-cycle operating cost of power equipment

Abstract

This invention discloses a method and system for predicting the life-cycle operating cost of substation equipment. It acquires visualized power data; utilizes temporal and spatial correlations to construct a strong-time-series dataset of substation equipment life-cycle operating costs based on the visualized power data; transforms non-stationary time-series data into stationary time-series data; and obtains final stationary time-series data based on the transformed stationary time-series data and the original stationary time-series data in the strong-time-series dataset of substation equipment life-cycle operating costs. The final stationary time-series data is then input into a pre-trained CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to obtain the cost prediction result. Advantages: This invention facilitates the analysis of large amounts of data, deeply explores the interrelationships and intrinsic connections within massive datasets, and improves power grid management.

Description

Technical Field

[0001] This invention relates to a method and system for predicting the operating cost of power equipment throughout its entire life cycle, belonging to the field of power analysis technology. Background Technology

[0002] Currently, with the rapid development of the national power system and the gradual advancement of digital grid construction, controlling the operating costs of substation equipment has become particularly important. Visual analysis technology applied to the power system displays power data in vivid and intuitive images or graphs, helping to understand and optimize the behavior and performance of the power system. This allows relevant departments of the power grid to promptly understand the relevant costs of substation equipment in a specific region, enabling better decision-making. It also provides a better understanding of the distribution and development trends of electricity customers, as well as user electricity consumption behavior. Machine learning-based prediction of the entire life-cycle operating costs of substation equipment is of great significance as an application of power big data. From a macro perspective, accurate data analysis can provide important information for national power development planning or major policy decision-makers; from a micro perspective, accurate data analysis is a crucial guarantee for power plants or power grid companies to rationally plan power generation or transmission and avoid power outages. Because of these advantages of big data, cost prediction based on big data is often faster and more accurate than traditional methods. Using power big data in conjunction with reasonable, economical, and efficient means for power data mining and prediction has strong practical significance. However, the management of this cost data is still relatively chaotic, the displayed data is quite complex, and the level of monitoring and management of the power grid is relatively low. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a method and system for predicting the operating cost of power equipment throughout its entire life cycle.

[0004] To address the aforementioned technical problems, this invention provides a method for predicting the full life-cycle operating cost of power equipment, comprising:

[0005] Obtain visualized power data;

[0006] A strong time-series dataset of the full life-cycle operating cost of substation equipment is constructed based on visualized power data, utilizing temporal and spatial correlations. This dataset includes both non-stationary and stationary time-series data.

[0007] Non-stationary time series data is transformed into stationary time series data. The final stationary time series data is obtained by combining the transformed stationary time series data with the original stationary time series data in the strong time series data set of the full life cycle operation cost of power equipment.

[0008] The final stationary time series data is input into a pre-trained hybrid model based on a fusion attention mechanism, CNN-MogrifierLSTM network, to obtain the cost prediction results.

[0009] Furthermore, obtaining visualized power data includes:

[0010] Collect operating cost data for the entire life cycle of power equipment to form a data warehouse; compress the data by clustering and deleting redundant data in the data warehouse to form power cost data;

[0011] Visualized electricity data can be obtained by using chart components based on visualization technology to create a visual interaction with electricity cost data.

[0012] Furthermore, the chart components based on visualization technology include pie charts, scatter plots, funnel charts, bubble charts, and thermodynamic graphs.

[0013] Furthermore, the method of constructing a strong time-series dataset of the entire life-cycle operating cost of substation equipment based on visualized power data, utilizing temporal and spatial correlations, includes:

[0014] Visualized power data is represented as F1 = {[g 11 ,…,g 1K ],…,[g N1 ,…,g NK ]},g nk This represents the k-th cost data in the n-th feature dimension, where n = 1, ..., N, k = 1, ..., K, and N and K represent the number of feature dimensions and the number of cost data, respectively.

[0015] The visualized power data is normalized and encoded to obtain encoded data F2, which is represented as:

[0016] F2={f 11 ,…,f 1K ],…,[f N1 ,…,f NK ]},f nk This represents the normalized and encoded data of the k-th cost data in the n-th feature dimension;

[0017] Historical environmental cost data is collected, and this data is normalized and encoded to obtain environmental data Y, represented as: Y = {[Y 11 ,…,Y 1K ],…,[Y N1 ,…,Y NK ]}, Y nk This represents the normalized and encoded cost data of the kth environmentally affected component in the nth feature dimension.

[0018] The correlation coefficient r between the encoded data F2 and the environmental data Y is calculated using the Pearson correlation coefficient formula and compared with a preset correlation coefficient threshold r0. If the correlation coefficient r is less than the correlation coefficient threshold r0, the corresponding f is removed from the encoded data F2. nk This allows us to obtain cost data related to environmental impacts.

[0019]

[0020] This represents the mean of all cost data after normalization coding. s represents the mean of environmental data. f With s Y These are the standard deviations of cost data and environmental data, respectively.

[0021] The cost dataset affected by the environment is represented as X1={x1,…,x i ,…,x I}, x i This represents cost data affected by the environment, i = 1, ..., I;

[0022] Cost data x affected by the environment i Perform data normalization to obtain normalized data. Represented as:

[0023]

[0024] μ β This is the average of all environmentally affected cost data. Let ∈ be the variance of all cost data affected by the environment, where ∈ is a positive number;

[0025] Based on the normalized data The dataset X2 representing the total lifecycle operating cost of substation equipment with strong time series characteristics is as follows:

[0026] Furthermore, the normalization encoding is performed based on whether the visualized power data is numerical or discrete. Numerical data uses Min-Max normalization encoding, while discrete data uses Catboost encoding.

[0027] Furthermore, the process of transforming non-stationary time series data into stationary time series data includes:

[0028] The difference method is used to transform non-stationary time series data into stationary time series data.

[0029] Furthermore, the final stationary time series data is input into a pre-trained CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to obtain cost prediction results, including:

[0030] The final stationary time series data is convolved in one dimension to obtain feature one;

[0031] Feature 1 is input into a CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to extract global high-level features;

[0032] The first feature and the global high-level features are merged, stretched, and then fed into the fully connected layer to output the cost prediction result.

[0033] Furthermore, the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism includes:

[0034] At least two CNN networks and a MogrifierLSTM network with fused attention mechanisms are connected sequentially;

[0035] The CNN network with the fusion attention mechanism includes a 1*1 convolutional layer, a 3*3 convolutional layer, a 3*3 convolutional layer, a 1*1 convolutional layer, and a fusion attention layer connected in sequence.

[0036] A machine learning-based system for predicting the life-cycle operating costs of substation equipment.

[0037] The acquisition module is used to acquire visualized power data;

[0038] A construction module is used to build a strong time-series dataset of the full life cycle operating cost of substation equipment based on visualized power data, utilizing temporal and spatial correlations; the strong time-series dataset of the full life cycle operating cost of substation equipment includes non-stationary time series data and stationary time series data;

[0039] The transformation module is used to transform non-stationary time series data into stationary time series data. Based on the transformed stationary time series data and the original stationary time series data in the strong time series power equipment life cycle operation cost dataset, the final stationary time series data is obtained.

[0040] The prediction module is used to input the final stationary time series data into a pre-trained CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to obtain cost prediction results.

[0041] Furthermore, the acquisition module,

[0042] This is used to collect operating cost data for the entire life cycle of power equipment, forming a data warehouse; by clustering and deleting redundant data in the data warehouse, the data is compressed to form power cost data; and based on visualization technology, chart components interact with the power cost data to obtain visualized power data.

[0043] Furthermore, the chart components based on visualization technology include pie charts, scatter plots, funnel charts, bubble charts, and thermodynamic graphs.

[0044] Furthermore, the constructed modules are used for

[0045] Visualize the power data as F1 = {[g 11 ,…,g 1K ],…,[g N1 ,…,g NK ]},g nk This represents the k-th cost data in the n-th feature dimension, where n = 1, ..., N, k = 1, ..., K, and N and K represent the number of feature dimensions and the number of cost data, respectively.

[0046] The visualized power data is normalized and encoded to obtain encoded data F2, which is represented as:

[0047] F2={f 11 ,…,f 1K ],…,[f N1 ,…,f NK ]},f nk This represents the normalized and encoded data of the k-th cost data in the n-th feature dimension;

[0048] Historical environmental cost data is collected, and this data is normalized and encoded to obtain environmental data Y, represented as: Y = {[Y 11 ,…,Y 1K ],…,[Y N1 ,…,Y NK ]}, Y nk This represents the normalized and encoded cost data of the kth environmentally affected component in the nth feature dimension.

[0049] The correlation coefficient r between the encoded data F2 and the environmental data Y is calculated using the Pearson correlation coefficient formula and compared with a preset correlation coefficient threshold r0. If the correlation coefficient r is less than the correlation coefficient threshold r0, the corresponding f is removed from the encoded data F2. nk This allows us to obtain cost data related to environmental impacts.

[0050]

[0051] This represents the mean of all cost data after normalization coding. s represents the mean of environmental data. f With s Y These are the standard deviations of cost data and environmental data, respectively.

[0052] The cost dataset affected by the environment is represented as X1={x1,…,x i ,…,x I}, x i This represents cost data affected by the environment, i = 1, ..., I;

[0053] Cost data x affected by the environment i Perform data normalization to obtain normalized data. Represented as:

[0054]

[0055] μ β This is the average of all environmentally affected cost data. Let ∈ be the variance of all cost data affected by the environment, where ∈ is a positive number;

[0056] Based on the normalized data The dataset X2 representing the total lifecycle operating cost of substation equipment with strong time series characteristics is as follows:

[0057] Furthermore, the normalization encoding is performed based on whether the visualized power data is numerical or discrete. Numerical data uses Min-Max normalization encoding, while discrete data uses Catboost encoding.

[0058] Furthermore, the conversion module,

[0059] It is used to transform non-stationary time series data into stationary time series data using the difference method.

[0060] Furthermore, the prediction module,

[0061] The final stationary time series data is convolved in one dimension to obtain feature 1; feature 1 is input into a CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to extract global high-level features; feature 1 and global high-level features are merged and stretched before being fed into a fully connected layer to output the cost prediction result.

[0062] Furthermore, the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism includes:

[0063] At least two CNN networks and a MogrifierLSTM network with fused attention mechanisms are connected sequentially;

[0064] The CNN network with the fusion attention mechanism includes a 1*1 convolutional layer, a 3*3 convolutional layer, a 3*3 convolutional layer, a 1*1 convolutional layer, and a fusion attention layer connected in sequence.

[0065] A computer-readable storage medium storing one or more programs, said one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods described.

[0066] A computing device, comprising,

[0067] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described.

[0068] The beneficial effects achieved by this invention are as follows:

[0069] This invention employs visualization technology to present cost data in chart form, transforming the non-stationary time series of visualized cost data into a stable time series, thus establishing the final stationary time series data. A hybrid model of CNN-MogrifierLSTM network based on a fusion attention mechanism is then used to process the final stationary time series data to obtain cost prediction results. This invention possesses independence, measurability, immediacy, flexibility, and practicality, helping to intuitively understand the operation of the power grid through visualized cost data, enabling real-time monitoring and management of the power grid, analyzing large amounts of data, deeply exploring the interrelationships and intrinsic connections within massive datasets, and improving the level of power grid management. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the process of this invention;

[0071] Figure 2 Flowchart of the CNN-MogrifierLSTM model based on the fusion attention mechanism of this invention;

[0072] Figure 3 Feature fusion model diagram. Detailed Implementation

[0073] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0074] like Figure 1As shown, a method for predicting the total life-cycle operating cost of power equipment includes:

[0075] S1. Use visualization technology to obtain visualized power data;

[0076] S2. Utilize temporal and spatial correlations to construct a strong time-series dataset of the entire life cycle operating cost of substation equipment based on visualized power data; the strong time-series dataset of the entire life cycle operating cost of substation equipment includes non-stationary time series data and stationary time series data; transform the non-stationary time series data into stationary time series data, and obtain the final stationary time series data based on the transformed stationary time series data and the original stationary time series data in the strong time-series dataset of the entire life cycle operating cost of substation equipment;

[0077] S3. Input the final stationary time series data into the pre-trained CNN-MogrifierLSTM hybrid model based on the fusion attention mechanism to obtain the cost prediction results.

[0078] The specific steps to obtain visualized power data are as follows:

[0079] S1.1 Collect relevant data on the operating costs of substation equipment throughout its entire life cycle, such as equipment costs, site costs, and maintenance losses, and merge multiple data sets into a data warehouse; compress the data by clustering and deleting redundant data in the data warehouse to form cost data; and use the Pearson correlation coefficient formula for scientific and reasonable analysis and judgment, realizing the scientific selection of influencing factors, completing the corresponding analysis and data collection, and incorporating them into the sample data.

[0080] S1.2. Visualize and interact with power data using chart components based on visualization technology to obtain visualized power data; the chart components based on visualization technology include pie charts, scatter plots, funnel charts, bubble charts, and thermodynamic diagrams.

[0081] The specific steps of S2 are as follows:

[0082] S2.1 The data in this project consists of a cost dataset with strong time properties, consisting of the total cost values ​​of equipment running continuously for one week. The features in the dataset are then divided into numerical and discrete types. Numerical features are encoded using Min-Max normalization, while discrete features are encoded using Catboost. The final format is as follows:

[0083]

[0084] S2.2 First, collect relevant environmental impact data, and calculate the correlation coefficient using the following Pearson correlation coefficient formula:

[0085]

[0086] Among them, s X With s Y Let X and Y be the standard deviations.

[0087] S2.3. Normalize the data, including environmental samples, to [0,1]. The calculation formula is as follows:

[0088] x i =(x i -x min ) / (x max -x min )

[0089] S2.4 To obtain more distinct data features, data normalization is also required as follows:

[0090]

[0091] Where μ β The mean, Let be its variance, and ∈ be a very small positive number.

[0092] The specific steps of S3 are as follows:

[0093] S31. First, perform a one-dimensional convolution on the stationary time series data to obtain feature one, as shown in the following formula.

[0094]

[0095] Where h(k) is the weight of the convolutional kernel, u(k) is the input to the convolutional layer, and y(k) is the network output.

[0096] S32. The features obtained from one-dimensional convolution are fed into a bidirectional LSTM (BiLSTM), and then into the self-attention module to further extract global high-level features.

[0097] The LSTM part mainly consists of a forget gate, an input gate, and an output gate.

[0098] The main calculations for the forgetting gate part are as follows:

[0099] f t =σ(U f h t-1 +W f x t )

[0100] k t =c t-1 ⊙f

[0101] Among them, U fWith W f Let x be the weight of the forgetting gate, f be the obtained forgetting factor, c be the memory factor, k be the main forgetting output, t and t-1 represent time, and σ represent the function form.

[0102] The main calculations for the input gate are as follows:

[0103] i t =σ(U i h t-1 +W i x t )

[0104] g t =tanh(U g h t-1 +W g x t )

[0105] j t =g t ⊙i t

[0106] c t =j t +k t

[0107] Among them, U i With W i Let x be the weight of the input gate, j be the size of the gating supply, c be the memory factor, i and g represent the intermediate states inside the input gate, and U be the weight of the input gate. g W g All of these are network weights.

[0108] The main calculations for the output gate are as follows:

[0109] o t =σ(U o h t-1 +W o x t )

[0110] h t =tanh(c t )⊙o t

[0111] Among them U O With W O Let be the weight of the output gate D, 'o' be the output size, and 'h' be the main output size.

[0112] S33. Based on the fusion attention mechanism, neurons with abundant neurons typically exhibit firing patterns different from surrounding neurons and inhibit surrounding neurons. Neurons with spatial inhibition effects should be assigned higher importance, as calculated below:

[0113]

[0114] in,

[0115] The current energy coefficient of the neuron, The calculation is as above, representing the sum of the L2 norms between neurons, where λ represents a very small number to ensure the denominator is not zero. This indicates the input to peripheral neurons.

[0116] S34. The features before and after self-attention are merged, stretched, and then fed into a fully connected layer for data regression, as shown below:

[0117]

[0118] Where Q, K, and V are all vectors, and d is the dimension of K.

[0119] The CNN-MogrifierLSTM hybrid model based on the fusion attention mechanism of this invention includes:

[0120] At least two CNN networks and a MogrifierLSTM network with fused attention mechanisms are connected sequentially;

[0121] The CNN network with the fusion attention mechanism includes a 1*1 convolutional layer, a 3*3 convolutional layer, a 3*3 convolutional layer, a 1*1 convolutional layer, and a fusion attention layer connected in sequence.

[0122] To improve the prediction accuracy of this invention, a hybrid CNN-MogrifierLSTM network model based on the fusion attention mechanism can be pre-trained. The training process is consistent with the basic CNN model training. A training set and a validation set are constructed using historical stationary time series data. Then, the hybrid CNN-MogrifierLSTM network model based on the fusion attention mechanism is trained based on the training set, the model parameters are updated, and the model is validated using the validation set to obtain a trained hybrid CNN-MogrifierLSTM network model based on the fusion attention mechanism.

[0123] Accordingly, this invention also provides a machine learning-based system for predicting the life-cycle operating cost of substation equipment.

[0124] The acquisition module is used to acquire visualized power data;

[0125] A construction module is used to build a strong time-series dataset of the full life cycle operating cost of substation equipment based on visualized power data, utilizing temporal and spatial correlations; the strong time-series dataset of the full life cycle operating cost of substation equipment includes non-stationary time series data and stationary time series data;

[0126] The transformation module is used to transform non-stationary time series data into stationary time series data. Based on the transformed stationary time series data and the original stationary time series data in the strong time series power equipment life cycle operation cost dataset, the final stationary time series data is obtained.

[0127] The prediction module is used to input the final stationary time series data into a pre-trained CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to obtain cost prediction results.

[0128] Furthermore, the acquisition module,

[0129] This is used to collect operating cost data for the entire life cycle of power equipment, forming a data warehouse; by clustering and deleting redundant data in the data warehouse, the data is compressed to form power cost data; and based on visualization technology, chart components interact with the power cost data to obtain visualized power data.

[0130] Furthermore, the chart components based on visualization technology include pie charts, scatter plots, funnel charts, bubble charts, and thermodynamic graphs.

[0131] Furthermore, the constructed modules are used for

[0132] Visualize the power data as F1 = {[g 11 ,…,g 1K ],…,[g N1 ,…,g NK ]},g nk This represents the k-th cost data in the n-th feature dimension, where n = 1, ..., N, k = 1, ..., K, and N and K represent the number of feature dimensions and the number of cost data, respectively.

[0133] The visualized power data is normalized and encoded to obtain encoded data F2, which is represented as:

[0134] F2={f 11 ,…,f 1K ],…,[f N1 ,…,f NK ]},f nk This represents the normalized and encoded data of the k-th cost data in the n-th feature dimension;

[0135] Historical environmental cost data is collected, and this data is normalized and encoded to obtain environmental data Y, represented as: Y = {[Y 11 ,…,Y 1K ],…,[Y N1 ,…,Y NK ]}, Y nk This represents the normalized and encoded cost data of the kth environmentally affected component in the nth feature dimension.

[0136] The correlation coefficient r between the encoded data F2 and the environmental data Y is calculated using the Pearson correlation coefficient formula and compared with a preset correlation coefficient threshold r0. If the correlation coefficient r is less than the correlation coefficient threshold r0, the corresponding f is removed from the encoded data F2. nk This allows us to obtain cost data related to environmental impacts.

[0137]

[0138] This represents the mean of all cost data after normalization coding. s represents the mean of environmental data. f With s Y These are the standard deviations of cost data and environmental data, respectively.

[0139] The cost dataset affected by the environment is represented as X1={x1,…,x i ,…,x I}, x i This represents cost data affected by the environment, i = 1, ..., I;

[0140] Cost data x affected by the environment i Perform data normalization to obtain normalized data. Represented as:

[0141]

[0142] μ β This is the average of all environmentally affected cost data. Let ∈ be the variance of all cost data affected by the environment, where ∈ is a positive number;

[0143] Based on the normalized data The dataset X2 representing the total lifecycle operating cost of substation equipment with strong time series characteristics is as follows:

[0144] Furthermore, the normalization encoding is performed based on whether the visualized power data is numerical or discrete. Numerical data uses Min-Max normalization encoding, while discrete data uses Catboost encoding.

[0145] Furthermore, the conversion module,

[0146] It is used to transform non-stationary time series data into stationary time series data using the difference method.

[0147] Furthermore, the prediction module,

[0148] The final stationary time series data is convolved in one dimension to obtain feature 1; feature 1 is input into a CNN-MogrifierLSTM hybrid model based on a fusion attention mechanism to extract global high-level features; feature 1 and global high-level features are merged and stretched before being fed into a fully connected layer to output the cost prediction result.

[0149] Furthermore, the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism includes:

[0150] At least two CNN networks and a MogrifierLSTM network with fused attention mechanisms are connected sequentially;

[0151] The CNN network with the fusion attention mechanism includes a 1*1 convolutional layer, a 3*3 convolutional layer, a 3*3 convolutional layer, a 1*1 convolutional layer, and a fusion attention layer connected in sequence.

[0152] Accordingly, the present invention also provides a computer-readable storage medium for storing one or more programs, said one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods described.

[0153] Accordingly, the present invention also provides a computing device, comprising,

[0154] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described.

[0155] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0157] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0158] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0159] 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 the operating cost of a power transformation device throughout its life cycle, characterized by, The application relates to a method for predicting the operation cost of a power transformation device, and belongs to the technical field of power transformation device operation cost prediction. The method comprises the following steps: acquiring visual power data; constructing a strong time sequence power transformation device whole life cycle operation cost dataset according to the visual power data by utilizing the correlation in time and space; the strong time sequence power transformation device whole life cycle operation cost dataset comprises non-stationary time sequence data and stationary time sequence data; transforming the non-stationary time sequence data into stationary time sequence data, and obtaining final stationary time sequence data according to the transformed stationary time sequence data and the original stationary time sequence data in the strong time sequence power transformation device whole life cycle operation cost dataset; inputting the final stationary time sequence data into a pre-trained CNN-MogrifierLSTM network hybrid model based on a fusion attention mechanism to obtain a cost prediction result; the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism comprises at least two CNN networks and MogrifierLSTM networks connected in sequence; the CNN network based on the fusion attention mechanism comprises a 1*1 convolution layer, a 3*3 convolution layer, a 3*3 convolution layer, a 1*1 convolution layer and a fusion attention layer connected in sequence; ； wherein, is the energy coefficient of the current neuron in the fusion attention mechanism, is the sum of the two-norms between neurons, the calculation formula of the energy coefficient of a neuron in the fusion attention mechanism of the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism is as follows: is a very small number, is the input of the surrounding neurons.

2. The machine learning based substation equipment life cycle operational cost prediction method of claim 1, wherein, the acquisition of the visual power data comprises the following steps: collecting power transformation device whole life cycle operation cost data to form a data warehouse; compressing data by clustering and deleting redundant data in the data warehouse to form power cost data; carrying out visual interaction between a chart component based on a visualization technology and the power cost data to obtain visual power data.

3. The machine learning based substation equipment life cycle operational cost prediction method of claim 2, wherein, The chart component based on the visualization technology comprises a pie chart, a scatter plot, a funnel chart, a bubble chart and a thermodynamic chart.

4. The machine learning based substation equipment life cycle operational cost prediction method of claim 1, wherein, The construction of the strong time sequence power transformation device whole life cycle operation cost dataset according to the visual power data by utilizing the correlation in time and space comprises the following steps: Visualized power data is represented as F1={[ g 11 , … , g 1K ],…,[ g N1 , … , g NK ]},g nk Indicates the first n The first feature dimension k Cost data, n =1,…, N , k =1,…, K , N and K These represent the number of feature dimensions and the number of cost data points, respectively. normalizing and encoding the visual power data to obtain encoding data F2, which is expressed as: F2={[ f 11 , … , f 1K ],…,[ f N1 , … , f NK ]}, f nk denotes the normalized data of the n th cost data in the k th feature dimension; Historical cost data related to environmental impacts are collected, and this historical cost data is normalized and encoded to obtain environmental data Y, represented as: Y = {[ Y 11 , … , Y 1K ],…,[ Y N1 , … , Y NK ]}, Y nk Indicates the first n The first feature dimension k The data after normalization and coding of cost data affected by the environment; The correlation coefficient of the coded data F2 and the environmental data Y is calculated using the Pearson correlation coefficient formula r , and a preset correlation coefficient threshold value r 0, the corresponding r is removed from the coded data F2 if the correlation coefficient r 0 is less than the correlation coefficient threshold value f nk , and the environmental-affected cost data is obtained ； a mean value representing the normalized encoded all cost data, a mean value representing the environmental data, s f with s Y the standard deviation of the cost data and the environmental data, respectively; The environment-influenced cost dataset, denoted as X1={ x 1,…, x i ,…, x I}, x i denotes the environment-influenced cost data, i =1,…, I ; Cost data influenced by the environment x i Data normalization is performed to obtain normalized data is expressed as: ； the normalization and encoding are performed according to whether the visual power data is numerical or discrete; the numerical data is normalized and encoded by using a Min-Max normalization encoding method, and the discrete data is normalized and encoded by using a Catboost encoding method. β the mean of all environment-affected cost data, the variance of all environment-affected cost data, is a positive number; According to the normalized data , the strong timing substation equipment life cycle operation cost data set X2 is determined, represented as: .

5. The machine learning based substation equipment life cycle operational cost prediction method of claim 4, wherein, The transformation of the non-stationary time sequence data into stationary time sequence data comprises the following steps:

6. The machine learning based substation equipment lifecycle operational cost prediction method of claim 1, wherein, the non-stationary time sequence data is transformed into stationary time sequence data by using a difference method. The input of the final stationary time sequence data into the pre-trained CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism to obtain the cost prediction result comprises the following steps:

7. The machine learning based substation equipment lifecycle operational cost prediction method of claim 1, wherein, one-dimensional convolution is performed on the final stationary time sequence data to obtain feature one; feature one is input into the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism to extract global high-level features; feature one and the global high-level features are combined and stretched and then input into a full connection layer to output the cost prediction result. ​ 8. A machine learning-based substation full life cycle operation cost prediction system, characterized in that, an acquisition module configured to acquire visualized power data; a construction module configured to construct a strong time series substation full life cycle operation cost dataset from the visualized power data according to the correlation in time and space; the strong time series substation full life cycle operation cost dataset comprises non-stationary time series data and stationary time series data; a conversion module configured to convert the non-stationary time series data into stationary time series data, and obtain final stationary time series data from the converted stationary time series data and the original stationary time series data in the strong time series substation full life cycle operation cost dataset; a prediction module configured to input the final stationary time series data into a pre-trained CNN-MogrifierLSTM network hybrid model based on a fusion attention mechanism to obtain a cost prediction result; the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism comprises at least two CNN networks and MogrifierLSTM networks connected in sequence, and the CNN network based on the fusion attention mechanism comprises 1*1 convolution layers, 3*3 convolution layers, 3*3 convolution layers, 1*1 convolution layers, and a fusion attention layer connected in sequence; the calculation formula of the energy coefficient of a neuron in the fusion attention mechanism of the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism is: ； wherein, is the energy coefficient of the current neuron in the fusion attention mechanism, is the sum of the two-norms between neurons, the acquisition module is configured to collect substation full life cycle operation cost data to form a data warehouse, compress the data by clustering and deleting redundant data in the data warehouse, and form power cost data; is a very small number, is the input of the surrounding neurons.

9. The machine learning based substation life cycle operation cost prediction system of claim 8, wherein, a chart component based on a visualization technology is configured to visually interact with the power cost data to obtain visualized power data. The chart component based on the visualization technology comprises a pie chart, a scatter plot, a funnel chart, a bubble chart, and a thermodynamic chart. The construction module is configured to 10. The machine learning based substation full life cycle operating cost prediction system of claim 9, wherein, normalize the visualized power data to obtain encoded data F2, which is represented as:

11. The machine learning based substation life cycle operation cost prediction system of claim 8, wherein, the normalization encoding is performed according to whether the visualized power data is numerical or discrete, the numerical data is normalized by Min-Max normalization encoding, and the discrete data is normalized by Catboost encoding. Visualize the power data as F1={[ g 11 , … , g 1K ],…,[ g N1 , … , g NK ]},g nk Indicates the first n The first feature dimension k Cost data, n =1,…, N , k =1,…, K , N and K These represent the number of feature dimensions and the number of cost data points, respectively. The conversion module is configured to convert the non-stationary time series data into stationary time series data by using a difference method. F2={[ f 11 , … , f 1K ],…,[ f N1 , … , f NK ]}, f nk Indicates the first n The first feature dimension k The cost data is normalized and coded; Historical cost data related to environmental impacts are collected, and this historical cost data is normalized and encoded to obtain environmental data Y, represented as: Y = {[ Y 11 , … , Y 1K ],…,[ Y N1 , … , Y NK ]}, Y nk Indicates the first n The first feature dimension k The data after normalization and coding of cost data affected by the environment; The correlation coefficient of the coded data F2 and the environmental data Y is calculated using the Pearson correlation coefficient formula r , and a preset correlation coefficient threshold value r 0, the corresponding r is removed from the coded data F2 if the correlation coefficient r is less than the correlation coefficient threshold value f nk , and the environmental-affected cost data is obtained ； a mean value representing the normalized encoded all cost data, a mean value representing the environmental data, s f with s Y the standard deviation of the cost data and the environmental data, respectively; The environment-influenced cost dataset, denoted as X1={ x 1,…, x i ,…, x I}, x i denotes the environment-influenced cost data, i =1,…, I ; Cost data influenced by the environment x i Data normalization is performed to obtain normalized data is expressed as: ； The prediction module is configured to perform one-dimensional convolution on the final stationary time series data to obtain feature one, input the feature one into the CNN-MogrifierLSTM network hybrid model based on the fusion attention mechanism to extract global high-level features, and combine and stretch the feature one and the global high-level features to send them to a fully connected layer to output a cost prediction result. β the mean of all environment-affected cost data, the variance of all environment-affected cost data, is a positive number; According to the normalized data , the strong timing substation equipment life cycle operation cost data set X2 is determined, represented as: .

12. The machine learning based substation life cycle operation cost prediction system of claim 11, wherein, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods of claims 1-7.

13. The machine learning based substation life cycle operation cost prediction system of claim 8, wherein, ​ ​ 14. The machine learning based substation life cycle operation cost prediction system of claim 8, wherein, ​ ​ ​ 15. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions for: ​ 16. A computing device, comprising: ​ One or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-7.

Images