Distribution network material consumption prediction method and equipment based on multi-dimensional data mining and closed-loop optimization

Abstract

The invention discloses a distribution network material consumption prediction method and device based on multi-dimensional data mining and closed-loop optimization, and relates to the field of distribution network material supply chain management.The method comprises the steps that firstly, a distribution network material basic information base is constructed, and data multi-dimensional structuring is achieved by establishing a triple composed of time, space and engineering type labels; typical consumption modes of different areas are extracted through area clustering analysis, then a material matching matrix is generated by using a mining algorithm, and a material consumption prediction model is constructed and trained according to the material matching matrix. And finally, based on a prediction result, generating a supply chain intelligent decision-making scheme in combination with the material matching matrix. According to the method and the system, the spatial-temporal characteristics of the material consumption are mined by fusing the multi-source heterogeneous data, and the distribution network material prediction model with the dynamic adaptive capability is constructed, so that the efficiency, the precision and the intelligent level of the distribution network material supply chain management are improved, and the high-strength and multi-change engineering requirements in the construction of a novel electric power system are met.

Description

Technical Field

[0001] This application relates to the field of distribution network material supply chain management, and in particular to a method and equipment for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization. Background Technology

[0002] As the terminal network of the power system, the distribution network is a crucial infrastructure for ensuring power transmission and supply quality. With the accelerated construction of new power systems, the scale and investment intensity of distribution network projects are constantly increasing, resulting in a greater variety of materials, higher procurement frequency, and wider geographical distribution, significantly increasing the complexity of material supply chain management.

[0003] In some situations, demand forecasting and procurement planning for power distribution network materials rely heavily on manual experience and static statistical methods, such as extrapolating from historical average consumption, centrally summarizing lists submitted by construction units, or passively replenishing inventory based on safety thresholds. While these methods are simple and easy to implement, they often suffer from insufficient forecast accuracy, low data utilization, and slow response times. Especially in scenarios involving concurrent construction of multiple projects, seasonal load variations, unstable supplier performance, or sudden emergency repairs, traditional methods struggle to adjust procurement plans in a timely manner, leading to either excessive stockpiling or frequent shortages. Furthermore, existing methods lack the integrated utilization of multi-source data on procurement, inventory, consumption, and supplier performance, neglecting the differences in material consumption across time, space, and project type, and lacking a closed-loop mechanism for forecast error monitoring and adaptive optimization of model parameters. This not only causes forecast accuracy to decline over time but also limits the realization of intelligent decision-making, failing to automatically identify the combination relationships of supporting materials or provide optimization basis for fund allocation and inventory transfer.

[0004] In summary, the current management process of power distribution network materials suffers from problems such as the inability to achieve refined and intelligent management, and low efficiency in supply chain management. Summary of the Invention

[0005] The purpose of this application is to provide a method and equipment for predicting the consumption of distribution network materials based on multidimensional data mining and closed-loop optimization, which can improve the efficiency, accuracy and intelligence level of distribution network material supply chain management, and meet the high-intensity and highly variable engineering needs in the construction of new power systems.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for predicting the consumption of distribution network materials based on multidimensional data mining and closed-loop optimization. The method includes: constructing a basic information database for distribution network materials; constructing structured label triples for distribution network materials based on historical distribution network material data; the structured label triples are used for multidimensional data indexing; the structured label triples specifically include: time labels, spatial labels, and project type labels; using the structured label triples to perform multidimensional structured processing on material consumption records in historical distribution network material data to obtain a multidimensional label dataset; and performing regional clustering analysis on the multidimensional label dataset to extract typical material consumption intensity classes for different regions. The method involves: using a data mining algorithm to identify pre-defined material combination relationships in a basic information database to obtain a material allocation matrix; constructing a material consumption prediction model based on a sequence prediction model according to the typical material consumption intensity types in different regions; using historical distribution network material data, optimizing the material consumption prediction model in a closed-loop manner with the goal of minimizing the loss function, and obtaining a trained material consumption prediction model; using the trained material consumption prediction model to predict the material consumption of the target distribution network materials, and obtaining the material consumption prediction results; and using the material allocation matrix to make intelligent supply chain decisions for the target distribution network materials based on the material consumption prediction results.

[0007] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for predicting the consumption of distribution network materials based on multidimensional data mining and closed-loop optimization.

[0008] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application constructs a basic information database for distribution network materials integrating multi-source heterogeneous data and establishes a triplet index system composed of time, space, and project type labels, achieving deep integration and efficient utilization of multi-dimensional data and improving the accuracy and representativeness of model input data. Cluster analysis is used to extract typical material consumption intensity types in different regions, improving the accuracy and regional adaptability of demand feature identification. Simultaneously, a material allocation matrix is ​​generated through data mining algorithms, enhancing the representational ability of the combination relationships of supporting materials and providing a key basis for coordinated procurement. A material consumption prediction model is constructed based on a sequence prediction model, and a closed-loop optimization mechanism with the goal of minimizing the loss function is introduced, enabling the model to achieve error self-awareness and parameter self-adjustment, improving the model's prediction accuracy and long-term stability. Finally, based on the prediction results and combined with the material allocation matrix, a supply chain intelligent decision-making scheme is generated, effectively reducing the risks of inventory backlog and stockouts. This application improves the efficiency, accuracy, and intelligence level of distribution network material supply chain management through full-chain optimization of data fusion, feature mining, adaptive prediction, and intelligent decision-making, meeting the high-intensity and highly variable engineering needs in the construction of new power systems. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a distribution network material consumption prediction method based on multidimensional data mining and closed-loop optimization, provided as an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] Example 1, such as Figure 1As shown in the figure, this embodiment provides a method for predicting the consumption of distribution network materials based on multidimensional data mining and closed-loop optimization. The method includes the following steps.

[0015] S1. Construct a basic information database for distribution network materials.

[0016] Furthermore, the basic information database includes the following fields: material number, material name, specifications, supplier, historical procurement data, historical consumption data, contract number, project location, and implementing unit information.

[0017] In practical applications, the construction process of the basic information database for distribution network materials is as follows.

[0018] 1) Construct a unique material identification number M: used to uniquely identify each material without duplication.

[0019] 2) Construct the material name Nm: used to represent the Chinese name of the material.

[0020] 3) Construct Model Specification Sm: Represents the specific model and specification parameters of the material.

[0021] 4) Construct Supplier ID Vm: Used to identify the supplier providing the material.

[0022] 5) Construct a historical procurement time series {tp(i)}: where tp(i) is the date and timestamp of the i-th procurement, s.

[0023] 6) Construct a historical procurement quantity sequence {qp(i)}: where qp(i) represents the quantity of the i-th procurement, in units.

[0024] 7) Construct a historical consumption time series {tc(j)}: where tc(j) is the timestamp of the j-th time the material is consumed at the engineering site.

[0025] 8) Construct a historical consumption quantity sequence {qc(j)}: where qc(j) represents the quantity consumed in the j-th time, in units.

[0026] 9) Construction Project Number Em: Indicates the specific power distribution network project to which the material belongs in the history of consumption.

[0027] Perform field validation and deduplication on the data to ensure that: ∀i,qp(i))≥0, ∀j,qc(j)≥0; and unify the timestamp format to "YYYY-MM-DD HH:MM:SS".

[0028] Finally, the cleaned data is stored in a relational database, and an index is created with M as the primary key, while allowing for fast querying and aggregation analysis based on multi-dimensional conditions such as time t, region L, and project type K.

[0029] S2. Construct structured tag triples for distribution network materials based on historical distribution network material data; the structured tag triples are used for multidimensional data indexing; the structured tag triples specifically include: time tag, spatial tag and project type tag.

[0030] Furthermore, step S2 specifically includes: S21. Time tag construction: Extract the Gregorian calendar year and month from the timestamps of historical distribution network material data, and calculate the quarter number through the month to construct the time tag.

[0031] S22. Spatial Label Construction: Based on the construction site, construct a hierarchical spatial location identifier composed of four levels of administrative and organizational codes: province, city, district / county, and power supply unit, as a spatial label.

[0032] S23. Construction of Project Type Labels: Read the project type field from the basic information database of distribution network materials, map the project type field to a standardized category set, perform standardization processing on the categories, and construct project type labels.

[0033] Optionally, project type tags include transformer installation, low-voltage renovation, and line relocation.

[0034] In practical applications, the process of generating structured tag triples (multidimensional tagging systems) includes: 1) Time tag generation.

[0035] Extract time features from the timestamp t of each material record: The year label Ty∈N takes values ​​from the Gregorian calendar year.

[0036] Monthly label Tm∈{1,2,…,12}.

[0037] The quarterly tag Tq=⌈Tm / 3⌉ indicates the quarter number to which the record belongs.

[0038] 2) Spatial Tag Generation A four-level code is generated based on the construction location of the project to which the material belongs: L=(Lp,Lc,Ld,Lu).

[0039] Where: Lp is the provincial administrative code; Lc is the municipal administrative code; Ld is the district / county administrative code; and Lu is the power supply unit code.

[0040] 3) Generation of project type labels.

[0041] Read the project type field from the project management system and map the original type to a standardized set: K∈{new transformer installation, low-voltage upgrade, meter installation, cable laying, distribution network relocation}.

[0042] 4) Tag binding.

[0043] The time tag T, spatial tag L, and project type tag K are bound to each historical record to form a tag triple: ⟨T, L, K>, and an index table is generated for subsequent use in cluster analysis and predictive modeling.

[0044] Tag generation has the following advantages.

[0045] 1. By uniformly encoding and labeling time, space, and project type, the originally scattered and unstructured material consumption data is transformed into structured and aggregable multidimensional data, which facilitates efficient processing by subsequent machine learning algorithms and improves data utilization and analysis efficiency.

[0046] 2. Time tags (year, month, quarter) support seasonal and periodic analysis; spatial tags (province, city, district, power supply unit) support regional difference analysis; project type tags support project category difference analysis. This makes the identification of material consumption patterns more accurate, enabling the capture of consumption patterns under different dimensions and improving the input quality of the prediction model.

[0047] 3. By establishing index tables and binding multi-dimensional tags, the system supports rapid querying, filtering, and aggregation analysis based on conditions such as time, region, and project type. This improves system response speed and supports real-time or near real-time decision support, such as quickly generating regional procurement suggestions or inventory allocation plans.

[0048] S3. Use structured label triples to perform multidimensional structured processing on the material consumption records in the historical distribution network material data to obtain a multidimensional label dataset.

[0049] Furthermore, step S3 specifically includes: S31. Use structured tag triples to tag historical distribution network material data to obtain tagged historical distribution network material data.

[0050] S32. Convert the tagged historical distribution network material data into a multidimensional tensor.

[0051] S33. Use a sparse storage structure to store multidimensional tensors, retain the index keys, and obtain a multidimensional label dataset.

[0052] In practical applications, multidimensional structured processing includes: 1) Convert the tagged material records (historical distribution network material data) into a multidimensional tensor: .

[0053] Where m is the material number index; t is the time tag index; l is the spatial tag index; k is the project type tag index; and xp is the p-th feature value.

[0054] The characteristic variables include: x1 is the average consumption per unit project (pieces / project); x2 is the consumption cycle volatility; x3 is the consumption growth rate of the previous cycle; x4 is the planned procurement quantity; x5 is the real-time inventory (pieces); x6 is the delivery lead time (days); x7 is the project volume (ten thousand yuan); x8 is the construction density (projects / square kilometer).

[0055] 2) Use a sparse storage structure to reduce storage space and retain the index key: Key=(m, t, l, k), which facilitates fast batch reading by machine learning algorithms.

[0056] S4. Perform regional clustering analysis on the multidimensional labeled dataset to extract typical material consumption intensity types in different regions.

[0057] Furthermore, the clustering algorithm includes at least one of the following: K-means, DBSCAN, and Gaussian mixture model.

[0058] In practical applications, the cluster analysis process is as follows.

[0059] 1) Extracting eigenvectors from tensors Xm,t,l,k: .

[0060] 2) Use the K-means clustering algorithm for classification, with the objective function being: .

[0061] Where: C is the number of clusters; Nc is the number of samples in the c-th cluster; , which is the feature vector of the i-th sample belonging to class c; μc: the centroid vector of class c.

[0062] 3) Store the consumption pattern label Zl of each region l into the result table for use as the regional feature input of the subsequent prediction model.

[0063] S5. Use mining algorithms to identify the preset material combination relationships in the basic information database to obtain the material allocation matrix.

[0064] Furthermore, the mining algorithms include: frequent itemset mining algorithms and association rule mining algorithms; the association rule mining algorithms include at least: either the Apriori algorithm or the FP-Growth algorithm.

[0065] Furthermore, step S5 specifically includes the following steps.

[0066] S51. Extract the set of material numbers corresponding to each project from the basic information database.

[0067] S52. Using mining algorithms, calculate the support and confidence of preset material combination relationships in the material number set, and filter strong rules that meet the threshold.

[0068] S53. The strong rules discovered are quantified into a co-occurrence probability matrix between materials to obtain the material allocation matrix.

[0069] In practical applications, the process of obtaining the material allocation matrix is ​​as follows.

[0070] 1) Objective: To identify the material combinations that frequently appear together in the same power distribution network project in history, and to provide a basis for the coordinated procurement of supporting materials in subsequent procurement plans.

[0071] 2) Data Input: Extract the material set under each project number E from the distribution network material basic information database: .

[0072] in, Number the supplies. This refers to the number of types of materials used in the project.

[0073] 3) Handling method: Define a global resource set: .

[0074] For any two subsets of resources A⊂M, B⊂M, calculate the support: .

[0075] Confidence level: .

[0076] in, , , the number of times itemset X appears in all projects; T, the total number of historical projects.

[0077] 4) Filtering criteria: Retain those that meet the following conditions: .

[0078] Where θs is the minimum support threshold and θc is the minimum confidence threshold, with typical values ​​of θs=0.1 and θc=0.6.

[0079] 5) Output: Generate material allocation matrix: .

[0080] Where Rm,n∈[0,1] represents the probability that material m and material n co-occur in the same project. This matrix serves as a constraint condition for generating the procurement plan.

[0081] Among them, after analyzing the engineering data, steps 3)-5) calculate the support (the number of times they appear together is high, for example, if the utility pole and the crossarm appear together 8 times in 10 projects, the support is high) and the confidence (if the utility pole is used, the crossarm is very likely to be used), and generate the matching matrix (derive the best combination based on historical data).

[0082] S6. Based on the typical material consumption intensity types in different regions, construct a material consumption prediction model using a sequence prediction model.

[0083] Furthermore, the sequence prediction model includes at least one of the following: long short-term memory network, extreme gradient boosting regression model, and seasonal time series model.

[0084] Optionally, the inputs to the material consumption prediction model include: historical power distribution network material data (historical consumption sequence), project planning intensity, construction period, climate factors, regional engineering construction index and other characteristic variables.

[0085] In practical applications, the process of constructing the prediction model is as follows: 1) Objective: To predict the demand for materials in the future based on historical power distribution network material data (historical consumption data), engineering characteristics and seasonal factors.

[0086] 2) Data Input: For each type of material m, the historical consumption time series under region l and project type k: .

[0087] Where dti is the actual consumption (units) at time ti.

[0088] Additional input variables: pt represents the planned project intensity at time t, number of projects per month; et represents the project type code (integer mapping) as st, seasonality factor (value range [-1,1]).

[0089] 3) Handling method: A prediction model is constructed using a Long Short-Term Memory (LSTM) network, with the following input sequence: .

[0090] Output predicted values: .

[0091] Where H represents the forecast period (months).

[0092] The training loss function is the weighted mean squared error: .

[0093] in, The weighting coefficient for period h (0 < ≤1).

[0094] 4) Output: The predicted consumption sequence of each material in the next H periods, used for generating procurement plans.

[0095] S7. Using historical distribution network material data, with the goal of minimizing the loss function, an adaptive learning rate adjustment strategy is adopted to perform closed-loop optimization training on the material consumption prediction model, resulting in a well-trained material consumption prediction model.

[0096] In practical applications, step S7 compares the prediction results with the actual material consumption data, calculates the prediction error, and adjusts the model parameters based on the error feedback mechanism. The feedback mechanism adopts an adaptive learning rate adjustment strategy and supports the cumulative optimization of historical error sequences.

[0097] The error feedback mechanism includes: 1) Objective: To dynamically monitor prediction accuracy and trigger model optimization when the error exceeds the standard.

[0098] 2) Data input: Predicted value sequence With actual value sequence , where h = 1, 2, ..., H.

[0099] 3) Handling method: Calculate the mean absolute error (MAE): .

[0100] Calculate the mean absolute percentage error (MAPE): .

[0101] When MAPE > ε for W consecutive periods, model optimization is triggered, including: Reload the latest dataset and train the model.

[0102] Adjust the number of hidden layer units and the time window length of the LSTM network.

[0103] Automatically select the best-performing candidate model from the set of candidate models (LSTM, XGBoost, seasonal time series models) to replace the current model.

[0104] 4) Output results: Optimized prediction model weights and structure, enabling continuous adaptive prediction.

[0105] S8. Use the trained material consumption prediction model to predict the material consumption of the target distribution network and obtain the material consumption prediction results.

[0106] S9. Based on the material consumption forecast results, use the material allocation matrix to make intelligent supply chain decisions for the target distribution network materials.

[0107] Furthermore, intelligent supply chain decision-making includes: intelligent procurement planning, intelligent inventory allocation, and intelligent fund allocation; intelligent decisions include: recommended procurement quantity, suggested replenishment cycle, fund priority ranking, and inventory threshold early warning.

[0108] In practical applications, step S9 generates procurement plan suggestions (intelligent procurement plan decision-making), inventory transfer suggestions (intelligent inventory transfer decision-making), and fund allocation suggestions (intelligent fund allocation decision-making) by region and project type.

[0109] The procurement plan generation includes: 1) Objective: To generate a multi-dimensional procurement plan based on predicted consumption, combined with inventory, delivery time and safety factor.

[0110] 2) Data Input: Predicted consumption Current inventory level Im,l; unit price Pm; safety reserve coefficient β∈[0.05,0.2].

[0111] 3) Handling method: Calculate the recommended purchase quantity: .

[0112] Calculate the recommended procurement funds: .

[0113] The system automatically replenishes the suggested quantities of materials for coordinated procurement based on the material allocation matrix Rm,n.

[0114] 4) Output results: Generate procurement plan tables by region and by material, with fields including: material number, suggested purchase quantity, budget funds, suggested supplier, and delivery cycle requirements. The plan can be directly imported into the ERP system to generate draft orders.

[0115] The technical effects of this application are as follows: This application constructs a basic information database for distribution network materials, incorporating heterogeneous data from multiple sources, including procurement, inventory, consumption, and supplier fulfillment. It also establishes a unified time, space, and project type tagging system, enabling the integrated utilization of multi-dimensional information. Compared to existing forecasting methods that rely solely on historical consumption data, this application comprehensively characterizes the material demand features of different regions and project types, significantly improving forecast accuracy and reducing forecast bias caused by limited data.

[0116] This application utilizes clustering algorithms to identify regional material consumption patterns and analyzes the combination relationships between typical materials through association rule mining, thereby extracting regional, seasonal, and complementary features of material demand. This feature recognition mechanism enables the prediction model to have stronger generalization ability, adapting to cross-regional and multi-project parallel construction scenarios, thus maintaining stable prediction performance under complex conditions.

[0117] This application introduces an error monitoring and feedback mechanism into the prediction model. When the prediction error in consecutive periods exceeds a set threshold, it automatically triggers model retraining and structural optimization, forming a closed-loop process of prediction-validation-optimization. This dynamic optimization capability enables the model to quickly adapt to uncertainties such as adjustments to project schedules, supply chain fluctuations, and sudden changes in demand, maintaining a high level of long-term prediction accuracy.

[0118] In terms of decision execution, this application combines forecast results with inventory status, safety factors, and material allocation matrices to generate regional and project-type procurement plans that can be directly imported into ERP systems, achieving automation and intelligence in procurement execution. This method ensures material supply security while reducing redundant inventory and capital occupation, improving capital utilization efficiency and supply chain operational efficiency. Furthermore, through a microservice architecture, it supports real-time data updates and rapid response, significantly shortening the reaction time for procurement adjustments.

[0119] In summary, this application outperforms existing technologies in terms of data utilization depth, prediction accuracy, model adaptability, decision intelligence, and response speed. It constructs an efficient, accurate, and flexible distribution network material supply chain management system that can meet the material management needs of high-intensity and highly variable projects in the construction of new power systems.

[0120] Example 2: This application also provides a computer device, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 2 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores and processes data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the methods described above.

[0121] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0123] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are 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.

[0124] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the consumption of distribution network materials based on multidimensional data mining and closed-loop optimization, characterized in that, The method includes: Establish a basic information database for power distribution network materials; Structured tag triples for distribution network materials are constructed based on historical distribution network material data; the structured tag triples are used for multidimensional data indexing; the structured tag triples specifically include: time tag, spatial tag, and project type tag; The material consumption records in historical distribution network material data are processed in a multidimensional structure using structured label triplets to obtain a multidimensional label dataset. Perform regional clustering analysis on the multidimensional labeled dataset to extract typical material consumption intensity types in different regions; A mining algorithm is used to identify the preset material combination relationships in the basic information database to obtain the material allocation matrix; Based on the typical material consumption intensity types in different regions, a material consumption prediction model is constructed using a sequence prediction model. Using historical distribution network material data, with the goal of minimizing the loss function, an adaptive learning rate adjustment strategy is adopted to perform closed-loop optimization training on the material consumption prediction model, resulting in a well-trained material consumption prediction model. The material consumption prediction model is used to predict the material consumption of the target distribution network and the prediction results are obtained. Based on the predicted material consumption, the material allocation matrix is ​​used to make intelligent supply chain decisions for the target distribution network materials.

2. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The basic information database includes the following fields: material number, material name, specifications, supplier, historical procurement data, historical consumption data, contract number, project location, and implementing unit information.

3. The distribution network material consumption prediction method based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, Based on historical distribution network material data, structured tag triples for distribution network materials are constructed, specifically including: Time tag construction: Extract the Gregorian calendar year and month from the timestamps of historical distribution network material data, and calculate the quarter number through the month to construct the time tag; Spatial tag construction: Based on the construction site, a hierarchical spatial location identifier composed of four levels of administrative and organizational codes (province, city, district / county, and power supply unit) is constructed as a spatial tag; Project type label construction: Read the project type field from the basic information database of distribution network materials, map the project type field to a standardized category set, perform standardization processing on the categories, and construct the project type label.

4. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The material consumption records in historical distribution network material data are processed using structured label triples to obtain a multidimensional label dataset, which specifically includes: Historical distribution network material data is tagged using structured tag triples to obtain tagged historical distribution network material data; Transform the tagged historical distribution network material data into a multidimensional tensor; A sparse storage structure is used to store multidimensional tensors while retaining the index keys, resulting in a multidimensional label dataset.

5. The distribution network material consumption prediction method based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The clustering algorithm includes at least one of K-means, DBSCAN, and Gaussian mixture model.

6. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The mining algorithm includes: frequent itemset mining algorithm and association rule mining algorithm; the association rule mining algorithm includes at least: Apriori algorithm and FP-Growth algorithm.

7. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The data mining algorithm is used to identify the preset material combination relationships in the basic information database to obtain the material allocation matrix, which specifically includes: Extract the set of material numbers corresponding to each project from the basic information database; Using mining algorithms, the support and confidence of preset material combination relationships in the material number set are calculated, and strong rules that meet the threshold are selected. The strong rules discovered are quantified into a co-occurrence probability matrix between materials, thus obtaining the material allocation matrix.

8. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The sequence prediction model includes at least one of the following: long short-term memory network, extreme gradient boosting regression model, and seasonal time series model.

9. The method for predicting distribution network material consumption based on multidimensional data mining and closed-loop optimization according to claim 1, characterized in that, The intelligent supply chain decision-making includes: intelligent procurement planning, intelligent inventory allocation, and intelligent fund allocation; the intelligent decision-making includes: recommended procurement quantity, suggested replenishment cycle, fund priority ranking, and inventory threshold early warning.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the distribution network material consumption prediction method based on multidimensional data mining and closed-loop optimization as described in any one of claims 1-9.

Images