Power grid material type selection method and system, terminal equipment and storage medium
By constructing a multi-dimensional correlation heterogeneous network and graph attention neural network model of engineering materials, the problem of weak information conversion into procurement plans in the early stage of the project is solved, the accuracy and precision of material selection and procurement plans are improved, and the supply of power grid materials is optimized Quality and efficiency.
Patent Information
- Application Number
- CN202410925446.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-08
AI Technical Summary
The existing technology is difficult to convert weak information in the early stage of the project into strong information guiding material plans and procurement lists, resulting in poor quality and efficiency of material supply, especially when the demand forecast of new power equipment and the procurement time window is unclear.
Build a multi-dimensional correlation heterogeneous network of engineering materials, use the graph attention neural network model to predict material selection, and generate a set of material procurement models through the intrinsic correlation between engineering preferences and material characteristics, and optimize the procurement plan.
It improves the accuracy of material selection and the precision of procurement plans, overcomes the limitations of high-dimensional sparseness and traditional data dimensionality reduction methods, and achieves more accurate demand forecasting and determination of procurement time windows.
Smart Images

Figure CN118917577A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of engineering material demand matching, and in particular to a power grid material selection method, system, terminal equipment and storage medium. Background Art
[0002] The power grid is an important infrastructure related to the national economy and people's livelihood. The current period is a critical period for the construction of a new power system. The unified management of procurement demand across the entire network is an important part of the management of modern intelligent supply chain of materials. Demand planning is an important prerequisite for material procurement and a key factor in optimizing procurement costs.
[0003] At present, a series of empirical work has been carried out in demand forecasting, but there are also some pain points that need to be addressed: due to the lack of historical procurement information, the demand forecast for new power equipment is relatively biased; only macro-project information can be obtained, and there is a lack of more detailed material demand information, especially the lack of demand information under the material coding dimension, which makes it difficult to proactively carry out material reserves; the unclear procurement time window means that we can only respond passively by purchasing as early as possible.
[0004] Therefore, in order to transform passive response into active "planning-scheduling-procurement", further research is needed on how to convert the weak information transmitted in the early stage of the project into strong information that can guide material planning and procurement lists. Summary of the invention
[0005] The technical problem to be solved by the present invention is how to convert the weak information transmitted in the early stage of the project into strong information to guide the material plan and procurement list, and generate a material procurement model set, so as to guide the subsequent "planning-planning-procurement" mapping "demand window" construction work and ensure the quality and efficiency of material supply. In order to solve the above technical problems, the present invention provides a power grid material selection method, system, terminal device and storage medium.
[0006] In a first aspect, an embodiment of the present invention provides a method for selecting power grid materials, including:
[0007] Based on the intrinsic association between project preferences and material characteristics, a multi-dimensional heterogeneous network of project materials is constructed. The intrinsic association includes the demand interaction relationship between projects and materials, the similarity relationship between projects, and the complementary substitution relationship between materials. The multi-dimensional heterogeneous network of project materials includes a project material selection interaction matrix, a project similarity matrix, a material complementary matrix, and a material substitution matrix.
[0008] According to the grid engineering construction standards, the weak information of the engineering project and the procurement plan information are strengthened to obtain the grid engineering construction standard information, wherein the grid engineering construction standards include general design, general equipment, general cost and standard process, and the weak information of the engineering project includes the basic information of the engineering project, investment scale and construction content;
[0009] Using the standard information of the power grid engineering construction, graph sampling is performed on the multi-dimensional associated heterogeneous network of engineering materials, and the generated subgraphs are represented as graph structure training data;
[0010] Building an initial material selection prediction model based on a graph attention neural network, and training the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model;
[0011] The engineering characteristics and material characteristics are input into the target material selection prediction model to obtain the selection probability of each material characteristic, and the material characteristics whose selection probability is higher than a preset threshold are selected to form a material model set of the engineering project corresponding to the engineering characteristics.
[0012] Preferably, the construction of a multi-dimensional heterogeneous network of engineering materials based on the intrinsic association between engineering preferences and material characteristics includes:
[0013] Construct a project material selection interaction matrix based on the interactive relationship between engineering and material requirements;
[0014] Construct a project similarity matrix based on the similarity relationship between projects;
[0015] Based on the complementary and substitutive relationship between materials, the material complementary matrix and material substitution matrix are constructed respectively;
[0016] The project material selection interaction matrix, project similarity matrix, material complementarity matrix and material substitution matrix are integrated to obtain a multi-dimensional associated heterogeneous network of engineering materials.
[0017] Preferably, the following formula is used to obtain the project material selection interaction matrix:
[0018]
[0019] Among them, R represents the project material selection interaction matrix, r MN Indicates project u M For supplies N The selection probability, M represents the number of items, and N represents the number of materials;
[0020] The project similarity matrix is obtained using the following formula:
[0021]
[0022] in, Represents the item similarity matrix, SimU M1 Indicates project u M Cosine similarity with item u1;
[0023] The material complementarity matrix is obtained using the following formula:
[0024]
[0025] in, represents a material complementarity matrix, each element of which represents the frequency of materials of different categories appearing in the same project at the same time;
[0026] The material substitution matrix is obtained using the following formula:
[0027]
[0028] in, represents the material substitution matrix, Indicates suppliesv N The number of features that are the same as item v1, and D represents the feature dimension of the item.
[0029] Preferably, the strengthening of weak information of engineering projects and procurement plan information according to the grid engineering construction standards includes:
[0030] Using power grid engineering construction standards to strengthen weak information of engineering projects, the strengthening at least includes data cleaning and data verification;
[0031] Enhance procurement planning information based on natural language processing and hierarchical analysis.
[0032] Preferably, the strengthening of procurement plan information based on natural language processing and hierarchical analysis includes:
[0033] Materials are grouped based on natural language processing to group materials of the same type into the same material category;
[0034] Use hierarchical analysis to determine whether materials of the same material category should be included in the procurement plan.
[0035] Preferably, the step of performing graph sampling on the multi-dimensional associated heterogeneous network of engineering materials by using the standard information of the power grid engineering construction includes:
[0036] PinSAGE was used to perform graph sampling on the multi-dimensional associated heterogeneous network of engineering materials.
[0037] Preferably, the initial material selection prediction model is constructed based on the graph attention neural network, and the initial material selection prediction model is trained according to the graph structure training data to obtain the target material selection prediction model, including:
[0038] The objective function of the target material selection prediction model is obtained using the following formula:
[0039]
[0040] Among them, L(y pred ,y true ) represents the objective function, y pred Represents the predicted material model set, y true represents the actual material model set, N represents the material quantity, y true,i Represents the i-th element of the actual material model set, y pred,i Represents the i-th element of the predicted material model set.
[0041] In a second aspect, an embodiment of the present invention provides a power grid material selection system, which is applicable to the power grid material selection method as described above, and includes:
[0042] A multi-dimensional associated heterogeneous network construction module is used to construct a multi-dimensional associated heterogeneous network of engineering materials based on the intrinsic association between engineering preferences and material characteristics. The intrinsic association includes the demand interaction relationship between engineering and materials, the similarity relationship between engineering projects, and the complementary substitution relationship between materials. The engineering material multi-dimensional associated heterogeneous network includes a project material selection interaction matrix, a project similarity matrix, a material complementary matrix, and a material substitution matrix.
[0043] A weak information strengthening module is used to strengthen the weak information of the project and the procurement plan information according to the grid project construction standards to obtain the grid project construction standard information. The grid project construction standards include general design, general equipment, general cost and standard process. The weak information of the project includes the basic information of the project, investment scale and construction content.
[0044] A graph sampling module, for performing graph sampling on the multi-dimensional associated heterogeneous network of engineering materials using the standard information of the power grid engineering construction, and representing the generated subgraphs as graph structure training data;
[0045] A material selection prediction model construction module is used to construct an initial material selection prediction model based on a graph attention neural network, and train the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model;
[0046] The material model set acquisition module is used to input the engineering characteristics and material characteristics into the target material selection prediction model, obtain the selection probability of each material characteristic, and select the material characteristics whose selection probability is higher than a preset threshold to form the material model set of the engineering project corresponding to the engineering characteristics.
[0047] In a third aspect, an embodiment of the present invention provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the power grid material selection method as described above when executing the computer program.
[0048] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the power grid material selection method as described above.
[0049] Compared with the prior art, the power grid material selection method, system, terminal device and storage medium of the embodiment of the present invention have the following beneficial effects: according to the intrinsic relationship between engineering preferences and material characteristics, a multidimensional heterogeneous association network between engineering projects and material models is constructed, which effectively overcomes the high-dimensional sparsity of engineering material procurement data and the limitations of traditional data dimensionality reduction methods; a graph attention neural network model is used to predict material selection, which can finely model and learn the importance of nodes in the graph. By introducing the attention mechanism, the model can better capture the relationship between nodes and feature information, thereby improving the accuracy in material selection prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 It is a flow chart of a method for selecting power grid materials according to an embodiment of the present invention;
[0051] Figure 2 It is a schematic diagram of a process of constructing a multi-dimensional associated heterogeneous network of engineering materials according to an embodiment of the present invention;
[0052] Figure 3 It is a schematic diagram of a multi-dimensional associated heterogeneous network of engineering materials according to an embodiment of the present invention;
[0053] Figure 4 It is a schematic diagram of a process of strengthening weak information according to an embodiment of the present invention;
[0054] Figure 5 It is a schematic diagram of a process for strengthening procurement plan information according to an embodiment of the present invention;
[0055] Figure 6 It is a structural schematic diagram of a power grid material selection system according to an embodiment of the present invention;
[0056] Figure 7 It is a schematic diagram of the structure of a terminal device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0057] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0058] like Figure 1 As shown, an embodiment of the present invention provides a method for selecting power grid materials, comprising the steps of:
[0059] S1. Based on the intrinsic relationship between engineering preferences and material characteristics, a multi-dimensional heterogeneous network of engineering materials is constructed;
[0060] It should be noted that the material selection mechanism based on multi-dimensional heterogeneous association includes: 1) Since similar projects often have similar selection preferences, historical project groups similar to the target project can be mined based on weak information, and the material selection of similar project groups can be referred to to guide the selection of new projects; 2) Materials with complementary relationships will have similar selection probabilities, so the material models with complementary relationships can be packaged for the target project; 3) The more similar the material characteristic values are, the closer the selection probabilities between the materials are; 4) The proportion of selection preferences for standardized engineering materials with common design, common equipment, common cost and standard process is increasing, and the prediction model should reflect this management orientation.
[0061] Specifically, the internal correlation includes the demand interaction relationship between projects and materials, the similarity relationship between projects and the complementary substitution relationship between materials. The multidimensional correlation heterogeneous network of project materials includes the project material selection interaction matrix, project similarity matrix, material complementary matrix and material substitution matrix.
[0062] Furthermore, if Figure 2 As shown, step S1 includes:
[0063] S101. Construct a project material selection interaction matrix based on the interactive relationship between engineering and material requirements;
[0064] Specifically, the following formula is used to obtain the project material selection interaction matrix:
[0065]
[0066] Among them, R represents the project material selection interaction matrix, r MN Indicates project u M For supplies N The selection probability, M represents the number of items, and N represents the number of materials.
[0067] Furthermore, using q ij Indicates project ui For supplies j The purchase quantity of item u i For supplies j The selection probability r ij It is expressed as follows:
[0068]
[0069] It can be understood that, for each element in the project material selection interaction matrix R, its value is 0 or 1.
[0070] S102, constructing a project similarity matrix based on similarity relationships between projects;
[0071] Let project u i The feature representation of The characteristic data comes from the comprehensive plan information table, including the basic information of the project, investment scale and construction content. The specific fields involved are shown in Table 1:
[0072] Table 1 Comprehensive planning information table of engineering projects
[0073]
[0074] Use one-hot encoding to convert a part of the character data fields in the basic information of the project into numeric variables. One-hot encoding converts discrete features into binary vectors, uses an N-bit state register to encode N states, and converts features with n different values into a binary vector of length n, in which only one element is 1 and the rest are 0. Thus, each value is represented as an independent dimension, avoiding the influence of the size relationship between different values on the model.
[0075] Furthermore, PCA dimensionality reduction method is used to compress the data dimension through principal component analysis. Specifically, the n-dimensional features are mapped to k-dimensional features reconstructed on the basis of the original n-dimensional features (new orthogonal features are also called principal components), and the principal components are reduced to the dimensional features that only retain most of the variance, ignoring the feature dimensions with almost zero variance, so as to achieve logarithmic
[0076] Dimensionality reduction of data features.
[0077] Calculate the cosine similarity between items based on the reduced-dimensional item feature vectors:
[0078]
[0079] The diagonal elements of the project similarity matrix are set to 0 to obtain the project similarity matrix. Specifically, the project similarity matrix is obtained using the following formula:
[0080]
[0081] in, Represents the item similarity matrix, SimU M1 Indicates project u M Cosine similarity with item u1.
[0082] S103, constructing a material complementarity matrix and a material substitution matrix based on the complementary and substitution relationship between materials;
[0083] The complementary relationship between materials refers to the frequency of materials of different categories appearing in the same project at the same time. The higher the frequency of materials of different categories appearing in the same project at the same time, the stronger the complementary relationship between the projects. Specifically, the following formula is used to obtain the complementary matrix of materials:
[0084]
[0085] in, Represents the material complementarity matrix. Each element of the material complementarity matrix represents the frequency of different categories of materials appearing in the same project at the same time. Its diagonal elements are set to 1, indicating a strong substitution relationship between the same materials.
[0086] The substitution relationship between materials is obtained by calculating the similarity of material characteristics. Represents the D-dimensional features of materials. First, the vector feature values are labeled and encoded, and then the substitution relationship between materials is obtained by calculating the Manhattan distance between each pair of materials:
[0087]
[0088] in, Indicates suppliesv j With suppliesv k The Manhattan distance between Indicates suppliesv j The d-th dimension feature of Indicates suppliesv k The d-th dimension feature of .
[0089] Furthermore, the material substitution matrix is obtained using the following formula:
[0090]
[0091] in, represents the material substitution matrix, Indicates suppliesv N The number of features that are the same as those of material v1, and D represents the feature dimension of the material. It can be understood that the material substitution matrix The elements in represent the number of identical features between the two materials, which can be obtained by calculating the Manhattan distance between the two materials. The larger the value, the more identical features the two materials have, and the stronger the substitution relationship between the two materials.
[0092] S104. Integrate the project material selection interaction matrix, project similarity matrix, material complementarity matrix and material substitution matrix to obtain a multi-dimensional associated heterogeneous network of engineering materials.
[0093] Specifically, the multi-dimensional heterogeneous network of engineering materials can be found in Figure 3 .
[0094] Since the historical engineering-material procurement data set presents high-dimensional and sparse characteristics, by constructing a multi-dimensional heterogeneous association network between engineering projects and material models, new projects with weak information are effectively associated with massive material models, thereby increasing data density.
[0095] S2. Strengthen the weak information of the project and the procurement plan information according to the grid engineering construction standards to obtain grid engineering construction standard information;
[0096] Specifically, the construction standards of power grid projects include general design, general equipment, general cost and standard process, and the weak information of engineering projects includes basic information of engineering projects, investment scale and construction content. The specific weak information of engineering projects can be found in Table 1.
[0097] Furthermore, if Figure 4 As shown, step S2 includes:
[0098] S201. Use power grid engineering construction standards to strengthen weak information of engineering projects;
[0099] Among them, strengthening includes at least data cleaning and data verification. Specifically, the main steps include:
[0100] 1) Clean the missing values and determine the range of missing values;
[0101] The missing value ratio of each characteristic field of weak information of engineering projects is calculated, and then strategies are formulated according to the missing ratio and the importance of characteristic fields.
[0102] 2) Delete unnecessary feature fields;
[0103] When cleaning data, each step needs to be backed up, or the full data can be processed only after successful experiments on small-scale data.
[0104] 3) Fill in missing content;
[0105] The missing content is filled in using the calculation results of the mean, median, mode, etc. of the same feature field, combined with the power grid engineering construction standards (universal design, universal equipment, universal cost, standard process).
[0106] 4) Re-obtain data;
[0107] Identify certain feature fields that are of high importance and high missing rates, communicate with data collectors or business personnel, and obtain relevant data from other channels.
[0108] 5) Data relevance verification;
[0109] If the data comes from multiple sources, correlation verification is required; if the data is not related, data cleaning is required.
[0110] 6) Data matching verification.
[0111] Specifically, data matching verification includes the following steps:
[0112] a. According to the data in the input table and the output table, import a number of differentiated samples and sample generation model parameters, and configure the model parameters to the model table;
[0113] The input table is used to standardize the parameters available for model training, including project characteristics, material characteristics, procurement quantity, procurement frequency, procurement price, procurement entity, procurement area, and supply method. The output table is used to standardize the prediction target of the model and train relevant results based on the input data. The output data includes procurement type, procurement scale, and procurement time range.
[0114] b. Make a legal judgment on the primary key of the query data;
[0115] c. Determine the data type of non-null values and query to generate records;
[0116] d. Based on the deep self-learning algorithm, identify the data in all records and generate a universal matching model;
[0117] The preferred deep self-learning algorithm in this embodiment is a graph neural network. The general matching model is a matching relationship automatically constructed by the model itself during the training process. Based on the characteristics of the neural network, this relationship cannot be controlled.
[0118] e. Use ETL cleaning to read the model table to clean the data.
[0119] S202. Strengthen procurement plan information based on natural language processing and hierarchical analysis.
[0120] Specifically, Figure 5 As shown, step S202 includes:
[0121] S202-A, grouping materials based on natural language processing so that materials of the same type are grouped into the same material category;
[0122] Since the ERP system covers almost all material information, and based on the stability and periodicity of materials in the power industry, the types of materials will not change significantly in the short term. This embodiment uses a rule-based and dictionary-based method to identify named entities and group materials of the same type into the same material category.
[0123] S202-B. Use hierarchical analysis to determine whether to include materials of the same material category into the procurement plan.
[0124] After the materials are grouped, the analytic hierarchy process is used to conduct a comprehensive evaluation of the materials' procurement characteristics, supplier characteristics, and other aspects from multiple angles and levels.
[0125] First, a hierarchical analysis structure model is constructed to conduct qualitative and quantitative analysis. The hierarchical relationship is that the lower layer affects the upper layer, and the factors within the layer basically do not affect each other. The hierarchical analysis structure model includes the target layer, the criterion layer, and the solution layer from bottom to top. Among them, the target layer is used to score the procurement of materials, the criterion layer is used to obtain evaluation indicators, and the solution layer is used to select materials.
[0126] Secondly, construct the judgment matrix between the upper and lower layers to determine the affiliation relationship between the upper and lower layers. The judgment matrix has the following properties: each element is a positive number, b ij and b ji The elements are inversely related, and the diagonal elements are all 1. Among them, b ij Indicates that relative to the upper layer element, element b i Element b j The present embodiment adopts the 1-9 scale method, for example, b ij =1 means that factor i and factor j are equally important, b ij The larger the value, the more important factor i is than factor j.
[0127] Furthermore, the eigenvalue method is used to calculate the relative weights of the elements under a single criterion.
[0128] Finally, there are many factors that influence the procurement demand of power grid materials, and the elements of the above judgment matrix are obtained subjectively through the pairwise comparison of two factors. In order to test the conclusions that violate the convention, the consistency index and consistency ratio are calculated respectively:
[0129] C I =(λ max -n) / (n-1)
[0130] C R =C I / R I
[0131] Among them, C I represents the consistency index, C R represents the consistency ratio, λ max represents the maximum eigenvalue of the judgment matrix, n represents the number of factors in the target layer or criterion layer, R I represents the average random consistency index. R I The results can be queried based on the existing table, which is the R of different orders in the hierarchical analysis method. I The query table of values. R When <0.1, the judgment matrix is considered to have good consistency, otherwise the values of the judgment matrix elements should be adjusted.
[0132] S3, using the standard information of power grid engineering construction to sample the multi-dimensional associated heterogeneous network of engineering materials, and representing the generated subgraphs as graph structure training data;
[0133] Specifically, PinSAGE is used to perform graph sampling on the multi-dimensional associated heterogeneous network of engineering materials.
[0134] The sampling process includes:
[0135] 1) Randomly select one or several nodes as node 0;
[0136] 2) Taking node 0 as the starting node, start random walk generation sequence through DeepWalk method;
[0137] 3) During the random walk, if the standard information of power grid project construction is encountered, the probability p is sampled again;
[0138] 4) Count the nodes that appear frequently in the random walk as neighbors of node 0 to generate a new subgraph;
[0139] 5) Count the high-frequency nodes in the new round of random walks and connect them to the original subgraph as new nodes;
[0140] 6) Repeat the above process k times, where k is a hyperparameter. The generated new subgraph is input into the graph neural network as a small batch sample for the next task.
[0141] PinSAGE was used to sample the multi-dimensional correlated heterogeneous network of engineering materials, and the weight of engineering projects selected using power grid engineering construction standards (universal design, universal equipment, universal cost, and standard process) was strengthened, coupling management trends in the form of a model.
[0142] S4. Construct an initial material selection prediction model based on the graph attention neural network, and train the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model;
[0143] The graph attention neural network introduces the attention mechanism into the graph convolutional network to extract key information of the graph structure. The information of the neighborhood feature layer is integrated by stacking nodes, and the weight coefficients of different nodes in the neighborhood are calculated. The specific algorithm flow is as follows:
[0144] 1) Train the shared weight matrix W;
[0145] 2) Calculate the weight of each neighbor node. The weight matrix will map the relationship between input features and output features.
[0146] 3) During the calculation process, the shared weight matrix W is used to map the representations of node i and node j respectively, and the resulting vectors are concatenated;
[0147] 4) Through feedforward neural network Map the vector to real numbers and activate it through the LeakyReLU function;
[0148] 5) After normalization, the attention coefficient is obtained, and the calculation formula is as follows:
[0149]
[0150] Here, the symbol || represents the concatenation of two vectors.
[0151] Specifically, the objective function of the target material selection prediction model is obtained using the following formula:
[0152]
[0153] Among them, L(y pred ,y true ) represents the objective function, y pred Represents the predicted material model set, y true represents the actual material model set, N represents the material quantity, y true,i Represents the i-th element of the actual material model set, y pred,i Represents the i-th element of the predicted material model set.
[0154] Furthermore, the evaluation index of material selection in this embodiment uses precision, recall and F1 value as the measure of accuracy. Among them, F1 value is a comprehensive evaluation of the precision and recall of the selection result, which can more comprehensively express the accuracy of the recommendation list. The higher the precision, recall and F1 value, the better the accuracy of the selection result. Specifically, the calculation formulas of precision, recall and F1 value are as follows:
[0155]
[0156] Among them, precision represents the precision rate, recall represents the recall rate, and V true Represents the actual material model set, V pred Represents the forecast material model set.
[0157] S5. Input the engineering characteristics and material characteristics into the target material selection prediction model to obtain the selection probability of each material characteristic, and select material characteristics whose selection probability is higher than a preset threshold to form a material model set of the engineering project corresponding to the engineering characteristics.
[0158] It is understandable that the problem of the material selection task for an engineering project is to find the material set with the greatest demand intensity for any given project. This embodiment inputs engineering features and material features, constructs a mapping relationship and outputs a selection probability, thereby obtaining a material model set for the target project.
[0159] In order to further verify the effectiveness of the above-mentioned power grid material selection method, in a specific embodiment, 6275 engineering projects with a total of 105,310 bidding records were extracted as experimental data. Among the projects covered by the experimental data, there are 1972 power transmission and transformation projects, 1223 AC lines, 805 substation projects, and 2275 other types (project feature fields are missing). The experimental data is classified according to project voltage, with 110kV projects being the largest, reaching 2507; followed by 35kV and 220kV, with 1248 and 1090 respectively.
[0160] This embodiment involves 9 categories of materials, including poles and towers, cables, ground wires, AC transformers, complete sets of substation equipment, combined electrical appliances, switch cabinets (boxes), cable accessories, and insulators, involving a total of 1,638 material codes. In this embodiment, the positive samples are 105,310 selection records, and the negative samples are obtained by randomly sampling unselected materials under different subcategories to obtain negative samples of the same proportion. Therefore, the ratio of positive and negative samples in this embodiment is 1:1, with a total of 210,620 experimental samples.
[0161] Since the purpose of material selection prediction is to select a batch of material models that best match the target project from the candidate material set, the focus is to let the model learn the matching pattern between project and materials. Different from the prediction of procurement quantity, the matching probabilities of different materials for the same project are independent of each other. Specifically, the procurement quantities of two types of materials that replace each other for the same project are increasing and decreasing, and the procurement quantities of the two types of materials are interdependent, while the matching probabilities of the two do not increase and decrease. When the characteristics of the two types of materials are known, the matching degree of the two for the same project is independent of each other. Therefore, in the material selection prediction task, the division of the training set and the test set is not divided according to the project, but the randomly mixed procurement record samples are divided according to 4:1.
[0162] This embodiment is experimentally verified and analyzed on the preprocessed data set. When the selection probability threshold is selected as 0.2, the material models with a selection probability higher than 0.2 are selected by the target project, and the material models with a selection probability lower than 0.2 are not selected by the target project. At this time, the average precision, recall and F1 values of the material selection model are 75%, 83% and 79% respectively. Among them, the precision rate measures how many material models predicted by the model to be selected are actually selected. On average, nearly 75% of the material models predicted by the model to be selected are eventually selected in real procurement, reflecting the accuracy of the prediction model; the recall rate measures the ability of the selection model to identify positive samples. In the real selection, 83% of the material models can be predicted by the model, reflecting the completeness of the prediction model. The F1 value, as the harmonic mean of the precision rate and the recall rate, comprehensively considers the accuracy and completeness of the model.
[0163] Since the precision rate, recall rate and F1 value are affected by the selection probability threshold, in order to comprehensively measure the performance of the selection model under different thresholds, the AUC value of the material selection model is calculated. The value range of AUC is between 0 and 1. The closer to 1, the better the model performance. AUC of 0.5 means that the model performance is equivalent to random guessing. The AUC value of the material selection model in this embodiment is 0.82. Considering that the number of samples available for training is limited and the candidate material models are large, the AUC value of 0.82 indicates that the performance of the model in material selection is relatively reliable.
[0164] The selection prediction capability of the model under different types of material dimensions is shown in Table 2.
[0165] Table 2 Prediction performance of material selection prediction model under the category dimension
[0166] Class Name Precision Recall F1 value Sample size Material code number Tower 88.25% 83.58% 85.85% 17914 67 Substation complete equipment 77.31% 94.23% 84.94% 4346 34 Cable accessories 78.65% 79.16% 78.91% 33243 185 Conductor, ground wire 71.19% 87.21% 78.39% 12432 144 Switch cabinet (box) 71.35% 86.28% 78.11% 46153 188 AC Transformer 74.22% 81.00% 77.46% 6492 186 cable 74.94% 77.51% 76.20% 48007 290 insulator 73.13% 75.76% 74.42% 22637 235 Combined electrical appliances 70.23% 78.53% 74.15% 19397 309
[0167] Among the nine types of core materials, the prediction ability for poles and towers is the strongest, with an F1 value of 85.75%. The prediction ability for combined electrical appliances is relatively weak, with an F1 value of 74.42%. From the experimental results, the model selection ability is affected by two factors: the number of samples and the number of material codes. The more samples there are, the more materials of this type are used for training, thereby improving the model's learning and generalization capabilities for the selection mode of this type of material. The number of material codes reflects the difficulty of material selection. The larger the number of material codes, the more candidate material models there are, and the more difficult the selection task is. At present, the overall accuracy of the material selection model ranges from 70% to 90%. For material categories with poor prediction ability, the model's selection ability can be further improved by accumulating material selection data samples of this category.
[0168] The prediction ability of the model under different engineering category dimensions is shown in Table 3.
[0169] Table 3 Prediction performance of material selection prediction model under engineering category dimension
[0170] Project Category Precision Recall F1 value Sample size Power transmission and transformation engineering 78.24% 86.46% 82.14% 102629 Line Engineering 71.20% 87.95% 78.69% 39481 Substation Engineering 78.36% 82.57% 80.41% 19213
[0171] Among the three types of projects, the selection model has the strongest prediction ability for power transmission and transformation projects, with an F1 value of 82%; the prediction ability for line projects is the weakest, with an F1 value of 79%. Overall, the F1 value of the material selection prediction ability for different types of projects is basically greater than 80%, which has a good prediction ability.
[0172] The prediction ability of the model under different voltage level dimensions is shown in Table 4.
[0173] Table 4 Prediction performance of material selection prediction model under engineering voltage level dimension
[0174] Engineering voltage level Precision Recall F1 value Sample size 750kV 77.12% 89.07% 82.67% 890 500kV 76.42% 89.08% 82.27% 5921 330kV 70.27% 86.19% 77.42% 2443 220kV 73.89% 87.98% 80.32% 43392 110kV 75.21% 86.95% 80.66% 79063 66kV 72.11% 85.30% 78.15% 4993 35kV 66.35% 79.96% 72.52% 24818
[0175] Among the three types of projects, the selection model has the strongest prediction ability for projects with a voltage level of 750kV, with an F1 value of 83%, and the weakest prediction ability for projects with a voltage level of 35kV, with an F1 value of 73%. Overall, there is a large gap in the prediction ability of material selection for projects with different voltage levels.
[0176] The embodiment of the present invention provides a method for selecting materials for a power grid. According to the intrinsic correlation between engineering preferences and material characteristics, a multi-dimensional heterogeneous correlation network between engineering projects and material models is constructed, which effectively overcomes the high-dimensional sparsity of engineering material procurement data and the limitations of traditional data dimensionality reduction methods. A graph attention neural network model is used to predict material selection, which can finely model and learn the importance of nodes in the graph. By introducing the attention mechanism, the model can better capture the relationship and feature information between nodes, thereby improving the accuracy in material selection prediction.
[0177] like Figure 6 As shown, an embodiment of the present invention provides a power grid material selection system, which is applicable to the power grid material selection method as described above, and includes:
[0178] The multi-dimensional correlation heterogeneous network construction module 1 is used to construct a multi-dimensional correlation heterogeneous network of engineering materials based on the intrinsic correlation between engineering preferences and material characteristics. The intrinsic correlation includes the demand interaction relationship between engineering and materials, the similarity relationship between engineering projects and the complementary substitution relationship between materials. The multi-dimensional correlation heterogeneous network of engineering materials includes a project material selection interaction matrix, a project similarity matrix, a material complementary matrix and a material substitution matrix.
[0179] Weak information strengthening module 2 is used to strengthen the weak information of engineering projects and procurement plan information according to the power grid engineering construction standards to obtain the power grid engineering construction standard information. The power grid engineering construction standards include general design, general equipment, general cost and standard process. The weak information of engineering projects includes the basic information of engineering projects, investment scale and construction content.
[0180] Graph sampling module 3, used to perform graph sampling on the multi-dimensional associated heterogeneous network of engineering materials using the standard information of power grid engineering construction, and to represent the generated subgraphs as graph structure training data;
[0181] The material selection prediction model construction module 4 is used to construct an initial material selection prediction model based on the graph attention neural network, and train the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model;
[0182] The material model set acquisition module 5 is used to input the engineering characteristics and material characteristics into the target material selection prediction model, obtain the selection probability of each material characteristic, and select the material characteristics with a selection probability higher than a preset threshold to form the material model set of the engineering project corresponding to the engineering characteristics.
[0183] It should be noted that each module in the above-mentioned power grid material selection system can be fully or partially implemented through software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a power grid material selection system, please refer to the definition of a power grid material selection method above. The two have the same functions and effects, which will not be repeated here.
[0184] An embodiment of the present invention further provides a terminal device, the terminal device comprising:
[0185] processor, memory, and bus;
[0186] The bus is used to connect the processor and the memory;
[0187] The memory is used to store operation instructions;
[0188] The processor is used to call the operation instruction, and the executable instruction enables the processor to perform the operation corresponding to the above-mentioned power grid material selection method of the present invention.
[0189] In an optional embodiment, a terminal device is provided, such as Figure 7 As shown, Figure 7The terminal device 5000 shown includes: a processor 5001 and a memory 5003. The processor 5001 and the memory 5003 are connected, such as through a bus 5002. Optionally, the terminal device 5000 may also include a transceiver 5004. It should be noted that in actual applications, the transceiver 5004 is not limited to one, and the structure of the terminal device 5000 does not constitute a limitation on the embodiments of the present invention.
[0190] Processor 5001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 5001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.
[0191] The bus 5002 may include a path to transmit information between the above components. The bus 5002 may be a PCI bus or an EISA bus, etc. The bus 5002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0192] The memory 5003 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.
[0193] The memory 5003 is used to store application code for executing the solution of the present invention, and the execution is controlled by the processor 5001. The processor 5001 is used to execute the application code stored in the memory 5003 to implement the content shown in any of the above method embodiments.
[0194] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method for selecting power grid materials described above is implemented.
[0195] Yet another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed on a computer, enables the computer to execute the corresponding contents of the aforementioned method embodiments.
[0196] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.
[0197] In summary, the embodiments of the present invention provide a method, system, terminal device and storage medium for selecting materials for a power grid. According to the intrinsic correlation between engineering preferences and material characteristics, a multidimensional heterogeneous association network between engineering projects and material models is constructed, which effectively overcomes the high-dimensional sparsity of engineering material procurement data and the limitations of traditional data dimensionality reduction methods. A graph attention neural network model is used to predict material selection, which can finely model and learn the importance of nodes in the graph. By introducing the attention mechanism, the model can better capture the relationship between nodes and feature information, thereby improving the accuracy in material selection prediction.
[0198] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0199] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.
Claims
1. A method for selecting power grid materials. It is characterized in that include: Based on the intrinsic association between project preferences and material characteristics, a multi-dimensional heterogeneous network of project materials is constructed. The intrinsic association includes the demand interaction relationship between projects and materials, the similarity relationship between projects, and the complementary substitution relationship between materials. The multi-dimensional heterogeneous network of project materials includes a project material selection interaction matrix, a project similarity matrix, a material complementary matrix, and a material substitution matrix. According to the grid engineering construction standards, the weak information of the engineering project and the procurement plan information are strengthened to obtain the grid engineering construction standard information, wherein the grid engineering construction standards include general design, general equipment, general cost and standard process, and the weak information of the engineering project includes the basic information of the engineering project, investment scale and construction content; Using the standard information of the power grid engineering construction, graph sampling is performed on the multi-dimensional associated heterogeneous network of engineering materials, and the generated subgraphs are represented as graph structure training data; Building an initial material selection prediction model based on a graph attention neural network, and training the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model; The engineering characteristics and material characteristics are input into the target material selection prediction model to obtain the selection probability of each material characteristic, and the material characteristics whose selection probability is higher than a preset threshold are selected to form a material model set of the engineering project corresponding to the engineering characteristics.
2. The method for selecting power grid materials according to claim 1, It is characterized in that The multi-dimensional heterogeneous network of engineering materials is constructed based on the intrinsic relationship between engineering preferences and material characteristics, including: Construct a project material selection interaction matrix based on the interactive relationship between engineering and material requirements; Construct a project similarity matrix based on the similarity relationship between projects; Based on the complementary and substitutive relationship between materials, the material complementary matrix and material substitution matrix are constructed respectively; The project material selection interaction matrix, project similarity matrix, material complementarity matrix and material substitution matrix are integrated to obtain a multi-dimensional associated heterogeneous network of engineering materials.
3. The method for selecting power grid materials according to claim 2, It is characterized in that The following formula is used to obtain the material selection interaction matrix of the project: Among them, R represents the project material selection interaction matrix, r MN Indicates project u M For supplies N The selection probability, M represents the number of items, and N represents the number of materials; The project similarity matrix is obtained using the following formula: in, Represents the item similarity matrix, SimU M1 Indicates project u M With Project U 1 The cosine similarity of The material complementarity matrix is obtained using the following formula: in, represents a material complementarity matrix, each element of which represents the frequency of materials of different categories appearing in the same project at the same time; The material substitution matrix is obtained using the following formula: in, represents the material substitution matrix, Indicates suppliesv N With suppliesv 1 The number of identical features between them, D represents the characteristic dimension of the material.
4. The method for selecting power grid materials according to claim 1, It is characterized in that The aforementioned strengthening of weak project information and procurement plan information according to the grid engineering construction standards includes: Using power grid engineering construction standards to strengthen weak information of engineering projects, the strengthening at least includes data cleaning and data verification; Enhance procurement planning information based on natural language processing and hierarchical analysis.
5. The method for selecting power grid materials according to claim 4, It is characterized in that The procurement plan information is enhanced based on natural language processing and hierarchical analysis, including: Materials are grouped based on natural language processing to group materials of the same type into the same material category; Use hierarchical analysis to determine whether materials of the same material category should be included in the procurement plan.
6. The method for selecting power grid materials according to claim 1, It is characterized in that The method of using the power grid engineering construction standard information to perform graph sampling on the engineering material multi-dimensional associated heterogeneous network includes: PinSAGE was used to perform graph sampling on the multi-dimensional associated heterogeneous network of engineering materials.
7. The method for selecting power grid materials according to claim 1, It is characterized in that The initial material selection prediction model is constructed based on the graph attention neural network, and the initial material selection prediction model is trained according to the graph structure training data to obtain the target material selection prediction model, including: The objective function of the target material selection prediction model is obtained using the following formula: Among them, L(y pred ,y true ) represents the objective function, y pred Represents the predicted material model set, y true represents the actual material model set, N represents the material quantity, y true,i Represents the i-th element of the actual material model set, y pred,i Represents the i-th element of the predicted material model set.
8. A power grid material selection system, applicable to the power grid material selection method according to any one of claims 1 to 7, It is characterized in that include: A multi-dimensional associated heterogeneous network construction module is used to construct a multi-dimensional associated heterogeneous network of engineering materials based on the intrinsic association between engineering preferences and material characteristics. The intrinsic association includes the demand interaction relationship between engineering and materials, the similarity relationship between engineering projects, and the complementary substitution relationship between materials. The engineering material multi-dimensional associated heterogeneous network includes a project material selection interaction matrix, a project similarity matrix, a material complementary matrix, and a material substitution matrix. A weak information strengthening module is used to strengthen the weak information of the project and the procurement plan information according to the grid project construction standards to obtain the grid project construction standard information. The grid project construction standards include general design, general equipment, general cost and standard process. The weak information of the project includes the basic information of the project, investment scale and construction content. A graph sampling module, for performing graph sampling on the multi-dimensional associated heterogeneous network of engineering materials using the standard information of the power grid engineering construction, and representing the generated subgraphs as graph structure training data; A material selection prediction model construction module is used to construct an initial material selection prediction model based on a graph attention neural network, and train the initial material selection prediction model according to the graph structure training data to obtain a target material selection prediction model; The material model set acquisition module is used to input the engineering characteristics and material characteristics into the target material selection prediction model, obtain the selection probability of each material characteristic, and select the material characteristics whose selection probability is higher than a preset threshold to form the material model set of the engineering project corresponding to the engineering characteristics.
9. A terminal device, It is characterized in that The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for selecting power grid materials as claimed in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the power grid material selection method according to any one of claims 1 to 7.
Citation Information
Cited By
Electric power material relation graph drawing and demand prediction method and system based on entropy value
CN119990572A
