CN-121998267-A - Intelligent demand prediction method and system for natural gas clients

CN121998267ACN 121998267 ACN121998267 ACN 121998267ACN-121998267-A

Abstract

The invention discloses a natural gas customer intelligent demand prediction method and a system, which belong to the technical field of energy demand prediction, wherein the prediction system comprises a data acquisition and processing module for processing acquired natural gas demand influence factors and natural gas demand data; the system comprises a correlation analysis module, a model selection module, a model training module, a parameter adjustment module, a model evaluation module, a prediction module and a customer natural gas demand prediction module, wherein the correlation analysis module analyzes the relation between all influence factors and natural gas demands and screens out input influence factors, the model selection module is used for selecting an algorithm model of the input influence factors, the model training module trains the algorithm model according to the input influence factors, the parameter adjustment module adjusts parameters of the algorithm model in training to determine the parameters of the algorithm model, the model evaluation module evaluates the performance of the algorithm model after the parameters are adjusted, screens out the algorithm model with the highest score as a prediction model, and the prediction module predicts the customer natural gas demands according to the prediction model. The prediction system provided by the invention can obviously improve the demand prediction precision and the data processing and analyzing capacity.

Inventors

ZHANG HAN
ZHANG XI
PAN KAI
XIE XIANG
HAN KEJIANG
ZHANG YUANTAO
LI WEI

Assignees

中国石油天然气股份有限公司

Dates

Publication Date: 20260508
Application Date: 20241104

Claims (11)

1. An intelligent demand prediction system for natural gas customers, comprising: the data acquisition and processing module is used for processing the acquired natural gas demand influence factors and the natural gas demand data; The correlation analysis module is used for analyzing the relation between all influence factors and natural gas requirements and screening out input influence factors; The model selection module is used for selecting an algorithm model for inputting influence factors; the model training module is used for training the algorithm model according to the input influence factors; the parameter adjustment module is used for adjusting parameters of the algorithm model in training and determining the parameters of the algorithm model; the model evaluation module is used for performing performance evaluation on the algorithm model with the adjusted parameters, and screening out the algorithm model with the highest score as a prediction model; and the prediction module is used for predicting the natural gas demand of the customer according to the prediction model.
2. The intelligent demand prediction system of a natural gas customer according to claim 1, wherein the data acquisition and processing module is specifically configured to: collecting natural gas demand influencing factors and natural gas demand data; carrying out data cleaning and data complementation on the natural gas demand influencing factors and the natural gas demand data; preprocessing and converting the natural gas demand influence factors and the natural gas demand data after data complementation; and carrying out data alignment and customer alignment on the natural gas demand influence factors and the natural gas demand data after the data conversion, so that all the natural gas demand influence factors and the natural gas demand data are mapped to corresponding customers.
3. The natural gas customer intelligent demand prediction system according to claim 1 or 2, wherein the correlation analysis module comprises a pearson correlation coefficient analysis unit, a SHAP value analysis unit, and a regression analysis unit; The pearson correlation coefficient analysis unit is used for calculating pearson correlation coefficients based on each piece of selected influence factor data and natural gas demand data, sequencing all absolute values of the pearson correlation coefficients obtained through calculation from large to small, and selecting influence factors 30% -70% in front of the ranking; The SHAP value analysis unit is used for determining the contribution value of each influence factor to the natural gas demand prediction based on a SHAP value algorithm, sequencing all the contribution values and selecting the influence factors with the top ranking of 30% -70%; The regression analysis unit is used for establishing a plurality of regression models based on each influence factor and the natural gas demand data, calculating the decision coefficient of each regression model, screening the maximum decision coefficient from the plurality of decision coefficients obtained by calculating different regression models of each influence factor, sequencing the maximum decision coefficient of each influence factor, and selecting the influence factors with the ranking of 30% -70%; The correlation analysis module is further used for taking the union of the influence factors respectively screened by the pearson correlation coefficient analysis unit, the SHAP value analysis unit and the regression analysis unit as input influence factors.
4. The natural gas customer intelligent demand prediction system of claim 1, wherein the algorithm model comprises a random forest algorithm, a support vector regression model, a BP neural network algorithm, a long-term memory network algorithm, a Lasso regression algorithm, a ridge regression algorithm, an extreme gradient enhancement algorithm, a propset time series prediction algorithm, a lightweight gradient lifting machine learning algorithm, a gradient lifting decision tree model, a linear regression algorithm, a gradient lifting algorithm, a bayesian ridge regression algorithm, an active correlation decision theory algorithm, and an adaptive lifting algorithm.
5. The intelligent demand prediction system of claim 1, wherein the model training module is configured to split the input influencing factors and the natural gas demand data into a training set, a validation set, and a test set, and to use the training set data to train the selected algorithm model to fit the algorithm model parameters.
6. The intelligent demand prediction system of claim 1, wherein the parameter adjustment module is configured to perform an exhaustive search in a predefined hyper-parameter space to find a combination of parameters for each algorithm model as it is trained.
7. The intelligent demand prediction system of claim 5, wherein the model evaluation module evaluates the performance of the algorithm model after the adjustment parameters using the validation set to calculate an average absolute error, an average absolute percentage error, a root mean square error, and a decision coefficient; Normalizing the average absolute error, the average absolute percentage error and the root mean square error; And comprehensively scoring the algorithm model after the parameters are adjusted based on the determined coefficient, the average absolute error, the average absolute percentage error and the root mean square error after normalization processing, and screening out the algorithm model with the highest score as a prediction model.
8. The natural gas customer intelligent demand prediction system according to claim 7, wherein the composite score is determined by the following formula: composite score = R 2 -RMSE norm -MAPE norm -MAE norm Wherein R 2 is a determination coefficient, RMSE norm is a normalized root mean square error, MAPE norm is a normalized mean absolute percentage error, and MAE norm is a normalized mean absolute error.
9. The intelligent demand prediction system of claim 1, wherein the prediction module is configured to select a prediction period, periodically update data, and predict customer natural gas demand.
10. The intelligent demand prediction system of claim 1, further comprising a presentation and application module for visually presenting the prediction model and the prediction results for each customer.
11. The intelligent demand prediction method for the natural gas client is characterized by comprising the following steps of: The acquired natural gas demand influence factors and the natural gas demand data are processed through a data acquisition and processing module; Analyzing the relation between all influence factors and natural gas requirements through a correlation analysis module, and screening out input influence factors; selecting an algorithm model for inputting influence factors through a model selection module; training an algorithm model according to the input influence factors by a model training module; adjusting parameters of the algorithm model in training through a parameter adjusting module, and determining parameters of the algorithm model; Performing performance evaluation on the algorithm model with the parameters adjusted through a model evaluation module, and screening out the algorithm model with the highest score as a prediction model; predicting the natural gas demand of the customer according to a prediction model through a prediction module; and displaying the prediction model, the prediction result of the natural gas demand and the model parameters through the display and application module.

Description

Intelligent demand prediction method and system for natural gas clients Technical Field The invention belongs to the technical field of energy demand prediction, and particularly relates to an intelligent demand prediction method and system for natural gas clients. Background With the continuous deep reform of domestic natural gas market, market main body develops towards diversification, market competition is increasingly vigorous, and an aggravated situation is presented. In the situation of diversified resources, how to accurately identify and change, rapidly strain and actively ask for changes in strong market competition of natural gas suppliers, and accurate cognition of customers is urgently needed. Under such a background, the conventional natural gas demand prediction method gradually exposes its limitations, and is difficult to adapt to a rapidly changing market environment. In order to maintain advantages in competition, there is a need for a demand forecast analysis method and system for improving forecast accuracy in a dynamic response to market changes in natural gas suppliers and related businesses. Natural gas suppliers and sales enterprises face challenges in how to improve customer demand prediction accuracy. The characteristics of the air consumption of customers have complex characteristics of volatility, structure, industry and economy, and the characteristics are comprehensive manifestations of the air consumption fluctuation of different end users. Meanwhile, the external factors influencing the user's aerodynamic fluctuation have a multi-element coupling cross relation, so that the traditional prediction method is difficult to accurately capture the complex interrelationships, and a significant deviation exists between the prediction result and the actual demand. In the current market, the natural gas demand prediction method mainly takes a national or provincial area as a prediction main body, and lacks accurate demand prediction on individual customer levels. Currently, the current state of the art for natural gas demand prediction is focused primarily on the national or provincial regional level. Predictive models based on macro-economic data, which typically use macro-economic data, demographic data, historical natural gas consumption data, and the like, to predict future natural gas demand. The method is suitable for large-scale demand prediction and is suitable for the regional level of nationwide or provincial areas. However, these models cannot capture the demand differences of individual customers, and are difficult to apply to more refined customer-level demand predictions. For example, chinese patent (publication No. CN105894113 a) discloses a method for predicting short-term demand of natural gas, on the basis of collecting meteorological parameters, historic loads and gas utilization structures in different regions, performing correlation analysis to determine load influencing factors, respectively performing prediction by using an artificial neural network, a support vector machine, a principal component analysis prediction, a hybrid regression analysis, a node multiple ratio method, an error correction model and an autoregressive distribution hysteresis model, and forming a set of complete short-term demand prediction method by using a prediction decision theory and an optimization technology based on a traditional classical statistical theory and an artificial intelligent algorithm, so as to predict daily natural gas demand in the future month in different regions. Time series based prediction methods such as ARIMA model (sum autoregressive moving average model) or exponential smoothing rely primarily on time series analysis of historical data for prediction. Such methods perform well in processing historical data, but suffer from a lack of responsiveness to future market changes and industry dynamics, and particularly in the face of complex and diverse market environments, the accuracy of predictions tends to be unsatisfactory. In addition, in recent years, machine learning techniques have been widely used in natural gas demand prediction, such as linear regression, support Vector Machines (SVM), random forests, and the like. These methods can process more complex data and can improve prediction accuracy to some extent. However, these models have limited generalization ability and cannot meet the complex and changeable market demands. Especially when facing multidimensional data and nonlinear relations, the prediction result of a single model is often not ideal enough, and personalized and dynamic prediction is difficult to realize. The current natural gas market has a large number of clients, and the problems of high research and development cost, complex data management, large model maintenance and updating difficulty and the like are faced to research and development of a demand prediction model for each client. Thus, a new method and system for predicting natural