Electric power payment prediction method and equipment based on multi-dimensional classification
By using multidimensional classification and combined prediction models, the shortcomings of traditional electricity payment prediction methods in terms of scenario adaptability and heterogeneous data processing are solved, achieving higher accuracy in payment prediction and adapting to complex environments and business dynamics.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-04-14
AI Technical Summary
Traditional electricity payment prediction methods are ill-suited to complex external environments and internal business dynamics, have poor scenario adaptability, are inefficient in processing heterogeneous data, and single algorithms struggle to effectively integrate features from multiple heterogeneous data sources, resulting in insufficient prediction accuracy.
A multidimensional classification method is adopted to divide business payment data into project dimensions and non-project dimensions, establish a business classification system, extract features through machine learning and combined prediction models, integrate features of multi-source heterogeneous data, and match the corresponding prediction models for accurate prediction.
It improves the accuracy and reliability of electricity payment forecasting, enabling more accurate prediction of payment data for various businesses and adapting to complex external environments and internal business dynamics.
Smart Images
Figure CN121860108A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power technology, and in particular to a method and device for predicting electricity payments based on multidimensional classification. Background Technology
[0002] As an asset-intensive enterprise, State Grid Corporation of China has a massive amount of interest-bearing debt. Reasonable forecasting of cash inflows and outflows is crucial for ensuring the company's financial health. Traditional payment forecasting methods rely on manual forecasting and static financial models, which are ill-suited to handling complex external environments and dynamic internal business operations. (1) Poor scenario adaptability: Enterprises usually rely on historical data and experience to make budgets. They lack the ability to track market dynamics, policy changes and project execution progress in real time. They cannot effectively quantify the synergistic effects of business rule complexity, historical behavior similarity and independent cash flow, resulting in low prediction accuracy and weak generalization ability, leading to a large deviation rate between budget and actual execution.
[0003] (2) Inefficient processing of heterogeneous data: Payment data usually covers multimodal information such as structured (contract amount), unstructured (policy text), and time series (interest rate fluctuations). Traditional methods rely on manual feature engineering, which makes it difficult to automatically integrate high-value information.
[0004] (3) Traditional single algorithms are difficult to effectively integrate the features of multi-source heterogeneous data, resulting in insufficient prediction accuracy: In the scenario of fund prediction, the payment behavior of project dimension is affected by multiple factors such as time-series fluctuations, contract rule constraints, and external policy disturbances. Due to inherent limitations, traditional single algorithm models are difficult to fully capture the time-series and periodic patterns of project payments, and the built-in rule transformation of unstructured data such as contract terms is not accurate enough. This limitation is mainly due to the fundamental conflict between the single modeling dimension of the single algorithm and the multimodal complexity of the business scenario, resulting in limited prediction accuracy. Summary of the Invention
[0005] In view of this, embodiments of the present invention provide a method and device for predicting electricity payments based on multidimensional classification, in order to solve the problem that traditional electricity payment prediction methods are difficult to cope with complex external environments and internal business dynamics.
[0006] In a first aspect, embodiments of the present invention provide a method for predicting electricity payments based on multidimensional classification, comprising: Obtain business payment data and its related data; The business payment data is divided into project dimensions and non-project dimensions through multidimensional classification, and a business classification system for the business payment data is established based on the associated data. Based on the business classification system, the first payment features of each business payment data in the project dimension are extracted, and the predicted payment data of each business in the project dimension is determined based on the first payment features of each business payment data and the first type of preset prediction model corresponding to each business. Based on the business classification system, the second payment features of each business payment data in the non-project dimension are extracted, and the predicted payment data of each business in the non-project dimension is determined based on the second payment features of each business payment data and the second type of preset prediction model corresponding to each business.
[0007] In one possible implementation, the first type of preset prediction model is a machine learning model; the step of determining the predicted payment data for each business in the project dimension based on the first payment characteristics of each business payment data and the first type of preset prediction model corresponding to each business includes: Obtain the policy factors and environmental factors corresponding to each business segment of the project; Based on the first payment characteristics, policy factors, and environmental factors corresponding to each business in the project dimension, a prediction reference dataset is established for each business in the project dimension. The prediction reference datasets corresponding to each business in the project dimension are input into the machine learning models corresponding to each business to obtain the predicted payment data for each business in the project dimension.
[0008] In one possible implementation, the second type of preset prediction model is: a combined prediction model with at least two sub-prediction models connected in parallel; The step of determining the predicted payment data for each business in the non-project dimension based on the second payment characteristics of each business payment data and the second type of preset prediction model corresponding to each business includes: Based on the month to be predicted, determine the sub-prediction models included in the combined prediction model, and obtain the target combined prediction model corresponding to the month to be predicted; Obtain the policy factors and environmental factors corresponding to each business in the non-project dimension; Based on the second payment characteristics, policy factors, and environmental factors corresponding to each business in the non-project dimension, a prediction reference dataset is established for each business in the non-project dimension. The prediction reference datasets corresponding to each business in the non-project dimension are input into the target combination prediction model corresponding to each business to obtain the prediction payment data of each business in the non-project dimension.
[0009] In one possible implementation, the step of inputting the prediction reference dataset corresponding to each business in the non-project dimension into the target combination prediction model corresponding to each business to obtain the prediction payment data for each business in the non-project dimension includes: For any business in the non-project dimension, the prediction reference dataset corresponding to the business is input into each sub-prediction model in the target combination prediction model corresponding to the business, and the prediction value output by each sub-prediction model is obtained. The predicted values of each sub-prediction model are weighted and summed to obtain the predicted payment data for this business.
[0010] In one possible implementation, the sub-prediction model includes: a linear regression model, a log-linear regression model, a seasonal autoregressive integral moving average model, and a long short-term memory network prediction model.
[0011] In one possible implementation, before establishing the business classification system for the business payment data, the method further includes: performing data preprocessing on the business payment data and its associated data; The data preprocessing includes: If the feature dimension of the business payment data and its associated data is greater than the preset dimension threshold, the K-nearest neighbor interpolation algorithm is used to select the nearest neighbor samples of similar businesses to fill in the missing values. If the feature dimension of the business payment data and its associated data is less than or equal to the dimension threshold, then a multiple linear regression imputation algorithm is used to construct a prediction model to fill in the missing values.
[0012] In one possible implementation, the data preprocessing further includes: Calculate the standard deviation of the business payment data, and identify outliers based on the standard deviation; The impact of the outlier is evaluated. If the impact evaluation result of the outlier meets the preset conditions, the outlier is retained; otherwise, the outlier is deleted. The outliers are reviewed based on business rules to determine whether they should be deleted.
[0013] In one possible implementation, the preset conditions include: The proportion of outliers in the business payment data exceeds a first preset threshold; or, after deleting the outliers, the change in variance of the business payment data is greater than a second preset threshold.
[0014] In one possible implementation, the first payment feature and the second payment feature include at least one of the following: key features, core time-series features, and derived trend features.
[0015] In a second aspect, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.
[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: In this embodiment of the invention, a business classification system for business payment data is established by using business payment data and its associated data, and features of multi-source heterogeneous data are integrated to improve prediction accuracy. Then, through multi-dimensional classification, payment business data is explicitly classified according to project type and non-project type, and features are extracted by dimension to match the corresponding preset prediction model, thereby enabling more accurate prediction of payment data for each business. Attached Figure Description
[0017] Figure 1 This is a schematic diagram illustrating the implementation process of a multidimensional classification-based electricity payment prediction method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the project-dimensional business classification system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the salary prediction model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a data processing flow provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] See Figure 1 The flowchart illustrating the implementation of the electricity payment prediction method based on multidimensional classification provided in this embodiment of the invention is described in detail below: Step S101: Obtain business payment data and its related data.
[0020] In this embodiment, data can be collected through systems such as financial control platforms, enterprise resource planning (ERP) platforms, and smart shared finance platforms. To ensure the accuracy of the data, data collection specifications are formulated: the principle of comprehensive data sources, the principle of full process coverage, and the principle of data standardization and uniformity.
[0021] Business payment data includes both project-based and non-project-based categories.
[0022] For example, the core data collection for project-level payments includes the following: Basic fields include project number, project type, project start and end dates, service confirmation / delivery confirmation time, contract number, order number, and order type.
[0023] Transaction attributes include: payment amount, payment frequency, payment method (cash, bank transfer, bank draft, etc.), payee bank information (bank name, interbank code, payee account number, payee unit), and structured identifiers of the payment cycle (payment amount within 1 to n months for the first payment, and the initial payment falling within the total project amount for the current month).
[0024] Time dimension: payment initiation time (accurate to the second), actual payment time, etc.
[0025] For example, the core data collection for non-project-based payment details includes the following information: Basic fields include: applicant unit code (used to identify power plant type) / name (applicant unit identifier), payer unit code / name, business type classification (such as power plant power purchase / inter-grid power purchase and sale / financing business / service procurement / payroll business / material procurement / renewable energy subsidies / tax business / other payables / bill payment / transmission fees / employee reimbursement / electricity fee refund, etc.), and order type (such as service procurement / material procurement, etc.).
[0026] Transaction attributes: payment amount, payment frequency (monthly, quarterly, annually), payment method (cash, bank transfer, banker's acceptance bill, etc.), payee bank information (bank name, interbank code, payee account number, payee unit), and structured tags for payment summary.
[0027] Time dimension: payment initiation time (accurate to the second), actual payment time.
[0028] For example, the associated data mainly involves organizational structure data and payment financial account data. Organizational structure data reveals the chain of responsible parties and decision-making paths for payment behavior, while payment financial account data transforms the business substance into a quantifiable and modelable feature vector.
[0029] Organizational structure data can include: hierarchical relationships (group-provincial company-municipal company), organization type (branch, subsidiary, provincial-level industrial unit, etc.), effective timestamp, etc.
[0030] Payment financial account data can include: accounting accounts (such as accounting accounts: electricity purchase fees payable, transmission and distribution costs, taxes payable - renewable energy surcharge, engineering fees - ultra-high voltage infrastructure; cash flow accounts: cash flow from electricity sales, cash flow from purchasing goods, cash flow from purchasing and constructing fixed assets, cash flow from debt repayment, etc.), auxiliary accounting items, and account version number.
[0031] Step S102: Divide business payment data into project dimensions and non-project dimensions through multi-dimensional classification, and establish a business classification system for business payment data based on related data.
[0032] In this embodiment, a mapping between business types and accounting subjects can be pre-established. Business types cover project-level materials and service payments, as well as non-project-level costs such as salary costs, electricity purchase costs, and fund taxes. In constructing the correspondence between business types and accounting subjects, business types must simultaneously carry dual-dimensional attributes: Basic business dimensions (such as 12 categories including payroll and material procurement): These can be directly matched with accounting subjects (such as payroll - employee compensation payable). This dimension is used for the classification of raw payments during the data collection phase.
[0033] Prediction logic dimension (project-based / non-project-based): This is dynamically determined using a multidimensional classification method for routing selection in the prediction model. The multidimensional classification method refers to a classification approach that dynamically determines the data's classification (project-based / non-project-based) based on multiple core feature dimensions of business payment data, establishing judgment rules (non-fixed mappings). These multiple dimensions are key evaluation indicators (such as complexity, independence, and correlation) designed in advance to assess the relationship between payment data and projects.
[0034] The mapping from business type to accounting subject is driven by static rules (e.g., the business type "material procurement" is fixedly mapped to the accounting subject "raw materials"). The division from business type to prediction dimension is driven by dynamic algorithms, allowing the same business type to be adapted to different prediction models.
[0035] In this embodiment, the classification system includes: (1) A multidimensional classification architecture is implemented using an L1-L2 cascade classifier. First, spatial-physical dual-axis positioning: taking renewable energy subsidies as an example, the L1 level is divided into multiple prefecture-level cities in the province according to the geographical location of the payment unit, and the L2 level is subdivided into corporate entities according to the nature of the account entity.
[0036] Second, text semantic-statistical feature fusion: Taking payroll business as an example, the L1 level matches projects such as enterprise annuity and housing provident fund through payment summary keywords, and the L2 level uses natural language processing technology to extract high-frequency words and generate sub-labels based on the amount distribution clustering.
[0037] 2) Special Classification of Electricity Purchase Business Establish a fingerprint database of transaction entity characteristics; for power transmission and purchase, lock the State Grid account entity by account name; divide provincial power plants and central enterprise power plants at the L1 level, and add installed capacity tags to power plants at the L2 level; and directly link inter-grid power purchase to the State Grid account entity.
[0038] 3) Evolutionary Mechanism of Unclassified Data Businesses without pre-defined categories initially enter the dynamic buffer pool, triggering the BERT model to analyze payment summaries. When the cumulative number exceeds 100, a three-level tag generation process is initiated (extracting core entities, matching similar business category features, generating and validating L1 / L2 level tags, with an error rate threshold of <3%).
[0039] For example, see the project-based classification system. Figure 2 As shown.
[0040] Step S103: Based on the business classification system, extract the first payment features of each business payment data in the project dimension, and determine the predicted payment data of each business in the project dimension based on the first payment features of each business payment data and the first type of preset prediction model corresponding to each business.
[0041] The detailed process is similar to step S104; please refer to step S104 below.
[0042] Step S104: Based on the business classification system, extract the second payment features of each business payment data in the non-project dimension, and determine the predicted payment data of each business in the non-project dimension based on the second payment features of each business payment data and the second type of preset prediction model corresponding to each business.
[0043] In this embodiment, the standardized payment data, after data cleaning, anomaly handling, and classification coding, can have its second payment feature extracted in the following way: 1) Retain key features such as salaries, taxes, and electricity purchase costs; 2) The historical payment series is decomposed using a seasonal and trend decomposition algorithm based on local weighted regression, retaining core time-series features such as monthly volatility and seasonality index; 3) Determine derived trend characteristics such as moving average slope and year-on-year abrupt change detection; 4) Remove features with residual abnormal volatility >50%, use cumulative sum control charts to identify structural changes, retain the direction and magnitude of mutations, and remove isolated noise points.
[0044] In this embodiment, key factors are extracted from policy documents through policy factor quantification engineering, and environmental factors are obtained by aligning meteorological-payment data using a city-level grid matching algorithm. Based on the second payment features, policy factors, and environmental factors corresponding to each business in the non-project dimension, a prediction reference dataset corresponding to each business in the non-project dimension is established and input into the second type of preset prediction model corresponding to each business to obtain the predicted payment data for each business in the non-project dimension.
[0045] In this embodiment of the invention, a business classification system for business payment data is established by using business payment data and its associated data, and features of multi-source heterogeneous data are integrated to improve prediction accuracy. Then, through multi-dimensional classification, payment business data is explicitly classified according to project type and non-project type, and features are extracted by dimension to match the corresponding preset prediction model, thereby enabling more accurate prediction of payment data for each business.
[0046] In some embodiments, before establishing a business classification system for business payment data, the method further includes: data preprocessing of the business payment data and its associated data.
[0047] Here, data preprocessing includes: 1) Null value handling mechanism: If the feature dimensions of the business payment data and its related data exceed a preset dimension threshold (e.g., 10), the K-nearest neighbor interpolation algorithm is used to select the nearest neighbor samples from similar businesses to fill in the missing values. For example, in the renewable energy subsidy business, missing geographical location values are interpolated using data from similar accounts in neighboring cities.
[0048] If the feature dimensions of the business payment data and its related data are less than or equal to the dimension threshold, a multiple linear regression imputation algorithm is used to construct a predictive model to fill in the missing values. For example, a predictive model can be constructed using payment month and amount as independent variables to predict the number of payments.
[0049] 2) Outlier handling process: Calculate the standard deviation of business payment data σ And outliers are determined based on the standard deviation.
[0050] Outliers are assessed for their impact. If the impact assessment results meet preset conditions, the outliers are retained; otherwise, they are deleted. For example, if the proportion of outliers is greater than 5% of the total sample, they are retained. Then, a model sensitivity test is conducted. If the change in prediction variance after removing outliers is greater than 15%, the outliers are retained.
[0051] Outliers are reviewed based on business rules to determine whether they should be deleted. For example, exemption amounts are considered for electricity bill refund data, and the bonus field in payroll allows ±5%. σ float.
[0052] In some embodiments, for payment prediction at the project level, this embodiment uses a machine learning model as the prediction model.
[0053] Through validation analysis of different models, it was found that under the current limitation of small sample data, the overall prediction performance of traditional machine learning models is significantly better than that of deep learning models. The reason is that deep learning models typically require massive amounts of data to fully learn and utilize their powerful representational capabilities, while small sample data easily leads to overfitting or underfitting. In contrast, machine learning models have relatively lower data requirements, stronger generalization ability in small sample scenarios, and their structure is easier to interpret and adjust. This embodiment comprehensively considers nonlinear modeling capability, resource consumption, and actual prediction accuracy, and selects the XGBoost algorithm.
[0054] Regarding the time granularity of the forecast: the model performs excellently in predicting six-month cycles, maintaining the error within a reasonable range, indicating a good grasp of long-term total monetary trends. However, its accuracy is insufficient when predicting monthly fluctuations, with slight fluctuations in monthly granularity prediction errors. For medium- to long-term (e.g., six-month) total monetary forecasts, the weekly granularity model is currently the optimal choice. However, for forecasting tasks requiring precision down to the details of monthly fluctuations, the model's capabilities still need improvement, possibly due to more complex, non-periodic fluctuation patterns or data noise within the month.
[0055] In some embodiments, for non-project-based payment forecasting, and considering situations with numerous business types and significant differences in their time-series patterns, a forecasting model architecture employing parallel algorithms such as linear regression, log-linear regression, Seasonal Autoregressive Integrated Moving Average (SARIMA), and Long Short-Term Memory Network Prediction (LSTM) is used. The specific combined forecasting model is determined based on the month to be predicted. Regression models are better at capturing the long-term linear trends of historical payment data. SARIMA models are better at capturing the cyclical trends of historical payment data, such as the annual cycle of salaries and the quarterly cycle of some tax declarations. LSTM models can better balance the long-term and short-term dependencies of historical data and can better capture components with strong randomness or non-linear fluctuations in historical data.
[0056] The prediction model accurately reflects historical data from different payment item models across various time periods and specific characteristics. It innovatively proposes a dynamic weighting approach to model construction. Higher weights are assigned to models with pronounced linear and periodic characteristics, or when the errors between the regression and SARIMA models are small. Conversely, the weight of the LSTM model is increased when nonlinear characteristics are prominent. This dynamic weighting allows the combined model to better adapt to different prediction scenarios and reduces overall prediction bias.
[0057] For example, such as Figure 3 As shown, for the salary forecasting model: based on the stable growth trend of annual salary payments and the regular fluctuation pattern of fixed months, different algorithm combinations are established for salary forecasting according to the forecast month. March payments increase year by year, but the overall growth rate gradually slows down; therefore, a combination model of linear regression and log-linear regression is considered. Non-March payments show stable growth; for all months except March, a combination model of linear regression and SARIMA time series regression is used. Weights are allocated to different algorithms based on the training set testing results. For example, regarding the tax prediction model: Tax payment involves multiple tax types and stages, including 10 major categories such as fund payments, value-added tax (VAT), corporate income tax, individual income tax, property tax, land tax, and deed tax. Among them, fund payments and VAT payments account for a relatively high proportion, with stable monthly payment amounts and slight fluctuations in individual months, mainly affected by the fund payment period and payment rules. Therefore, linear regression, SARIMA time series model, and LSTM prediction are used. VAT is directly related to annual electricity sales, and the payment amount increases in tandem with the growth of electricity sales. In recent years, due to policy impacts, there have been suspensions of payment in the first half of the year, lasting for 3-4 months, while payments are stable in other months. The remaining tax types account for a smaller proportion, but the monthly payment dates are relatively fixed, and the overall amount does not fluctuate much. Therefore, a combination of SARIMA and LSTM algorithms is used.
[0058] Figure 4 This is another embodiment of the present invention providing the implementation process of a multi-dimensional classification-based electricity payment prediction method: First, the complexity of business rules is quantified, and power grid payment businesses are explicitly classified according to business type and project type, completing a detailed mapping of payment relationships. Addressing the difficulty in efficiently integrating structured, unstructured, and time-series multimodal information in payment data, the method first performs automatic extraction of multimodal data, achieving feature extraction and dimension summarization; second, various heterogeneous data are aligned, and data of different dimensions are split and summarized according to business type, enabling automatic updates and real-time calculation of payment data, and dynamic updates to the feature library. Based on a multi-algorithm fusion model for different business types, multi-dimensional prediction is carried out. Through multi-dimensional prediction, the advantages of each algorithm in different business scenarios are fully utilized, thereby accurately grasping the complex characteristics of project payments. This overcomes the inherent limitations of traditional single-algorithm models in failing to fully capture the temporal and periodic patterns of project payments, improving the accuracy and reliability of predictions.
[0059] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0060] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 5 As shown, the electronic device 5 of this embodiment includes a processor 50 and a memory 51. The memory 51 stores a computer program 52. When the processor 50 executes the computer program 52, it implements the steps in the various method embodiments described above. Alternatively, when the processor 50 executes the computer program 52, it implements the functions of each module in the various device embodiments described above.
[0061] For example, computer program 52 may be divided into one or more modules / units, which are stored in memory 51 and executed by processor 50 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 52 in electronic device 5.
[0062] Electronic device 5 may include, but is not limited to, processor 50 and memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 5 and does not constitute a limitation on electronic device 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 5 may also include input / output devices, network access devices, buses, etc.
[0063] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.
[0064] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.
[0065] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method for predicting electricity payments based on multidimensional classification, characterized in that, include: Obtain business payment data and its related data; The business payment data is divided into project dimensions and non-project dimensions through multidimensional classification, and a business classification system for the business payment data is established based on the associated data. Based on the business classification system, the first payment features of each business payment data in the project dimension are extracted, and the predicted payment data of each business in the project dimension is determined based on the first payment features of each business payment data and the first type of preset prediction model corresponding to each business. Based on the business classification system, the second payment features of each business payment data in the non-project dimension are extracted, and the predicted payment data of each business in the non-project dimension is determined based on the second payment features of each business payment data and the second type of preset prediction model corresponding to each business.
2. The electricity payment prediction method based on multidimensional classification according to claim 1, characterized in that, The first type of preset prediction model is a machine learning model; The step of determining the predicted payment data for each business in the project dimension based on the first payment characteristics of each business payment data and the first type of preset prediction model corresponding to each business includes: Obtain the policy factors and environmental factors corresponding to each business segment of the project; Based on the first payment characteristics, policy factors, and environmental factors corresponding to each business in the project dimension, a prediction reference dataset is established for each business in the project dimension. The prediction reference datasets corresponding to each business in the project dimension are input into the machine learning models corresponding to each business to obtain the predicted payment data for each business in the project dimension.
3. The electricity payment prediction method based on multidimensional classification according to claim 1, characterized in that, The second type of preset prediction model is a combined prediction model consisting of at least two sub-prediction models connected in parallel. The step of determining the predicted payment data for each business in the non-project dimension based on the second payment characteristics of each business payment data and the second type of preset prediction model corresponding to each business includes: Based on the month to be predicted, determine the sub-prediction models included in the combined prediction model, and obtain the target combined prediction model corresponding to the month to be predicted; Obtain the policy factors and environmental factors corresponding to each business in the non-project dimension; Based on the second payment characteristics, policy factors, and environmental factors corresponding to each business in the non-project dimension, a prediction reference dataset is established for each business in the non-project dimension. The prediction reference datasets corresponding to each business in the non-project dimension are input into the target combination prediction model corresponding to each business to obtain the prediction payment data of each business in the non-project dimension.
4. The electricity payment prediction method based on multidimensional classification according to claim 3, characterized in that, The step of inputting the prediction reference dataset corresponding to each business in the non-project dimension into the target combination prediction model corresponding to each business to obtain the prediction payment data for each business in the non-project dimension includes: For any business in the non-project dimension, the prediction reference dataset corresponding to the business is input into each sub-prediction model in the target combination prediction model corresponding to the business, and the prediction value output by each sub-prediction model is obtained. The predicted values of each sub-prediction model are weighted and summed to obtain the predicted payment data for this business.
5. The electricity payment prediction method based on multidimensional classification according to claim 3, characterized in that, The sub-prediction models include: linear regression model, seasonal autoregressive integral moving average model, and long short-term memory network prediction model.
6. The electricity payment prediction method based on multidimensional classification according to any one of claims 1-5, characterized in that, Before establishing the business classification system for the business payment data, the method further includes: performing data preprocessing on the business payment data and its associated data; The data preprocessing includes: If the feature dimension of the business payment data and its associated data is greater than the preset dimension threshold, the K-nearest neighbor interpolation algorithm is used to select the nearest neighbor samples of similar businesses to fill in the missing values. If the feature dimension of the business payment data and its associated data is less than or equal to the dimension threshold, then a multiple linear regression imputation algorithm is used to construct a prediction model to fill in the missing values.
7. The electricity payment prediction method based on multidimensional classification according to claim 6, characterized in that, The data preprocessing also includes: Calculate the standard deviation of the business payment data, and identify outliers based on the standard deviation; The impact of the outlier is evaluated. If the impact evaluation result of the outlier meets the preset conditions, the outlier is retained; otherwise, the outlier is deleted. The outliers are reviewed based on business rules to determine whether they should be deleted.
8. The electricity payment prediction method based on multidimensional classification according to claim 7, characterized in that, The preset conditions include: The proportion of outliers in the business payment data exceeds a first preset threshold; or, after deleting the outliers, the change in variance of the business payment data is greater than a second preset threshold.
9. The electricity payment prediction method based on multidimensional classification according to any one of claims 1-5, characterized in that, The first payment feature and the second payment feature include at least one of the following: Key features, core time-series features, and derived trend features.
10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 9.