CN-121998184-A - Intelligent prediction method and system for production stock of heat metering and regulating equipment

CN121998184ACN 121998184 ACN121998184 ACN 121998184ACN-121998184-A

Abstract

The application relates to the field of equipment production stock prediction, in particular to an intelligent prediction method and system for heat metering and regulating equipment production stock. The method comprises the steps of multi-source data acquisition and preprocessing, feature engineering construction, mixed model construction and training, prediction result output and feedback optimization, and a corresponding intelligent prediction system comprising data acquisition and preprocessing, feature construction, mixed prediction, output and feedback and other modules. The application can accurately predict the demand of each product category of the heat metering and regulating equipment, reasonably arrange production and stock according to the prediction result, and continuously improve the prediction accuracy through feedback optimization.

Inventors

LU TONGJIN
DONG YIJIE

Assignees

杭州中沛电子有限公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. The intelligent prediction method for the production stock of the heat metering and regulating equipment is characterized by comprising the following steps of: s1, multi-source data acquisition and pretreatment: Collecting historical order data, wherein the historical order data comprises product models, specifications, quantity, order placing time and a region where a customer is located; Acquiring historical and future forecast meteorological data, including heating degree days; Aligning and aggregating the historical order data with meteorological data in time and area; s2, feature engineering construction: Constructing characteristic variables based on the aligned multi-source data, wherein the characteristic variables comprise a historical sales hysteresis characteristic and a current heating degree date characteristic; S3, constructing and training a mixed model: predicting the total demand of a future period by adopting a gradient lifting tree model; Predicting the demand proportion distribution of each product class by adopting a sequence-to-sequence model; multiplying the total demand predicted by the gradient lifting tree model by the ratio of each product class predicted by the sequence-to-sequence model to obtain the predicted demand of each product class; s4, arranging production and stock according to the predicted demand of each product class.
2. The intelligent prediction method for production stock of heat metering and regulating equipment according to claim 1, which comprises the following steps: S5, prediction result output and feedback optimization: outputting a product demand forecast list of one or more forecast periods in the future; Collecting actual sales data and calculating a prediction error; And carrying out feedback optimization on the hybrid model based on the prediction error, and updating model parameters.
3. The intelligent prediction method for production stock of heat metering and regulating equipment according to claim 1, wherein the objective function of the gradient lifting tree model is to minimize mean square error, and regularization term is introduced to control model complexity.
4. The intelligent prediction method for production stock of heat metering and regulating equipment according to claim 1, wherein the sequence-to-sequence model comprises an encoder and a decoder, the encoder encodes a sales sequence of each historical product category into a context vector, and the decoder predicts the demand proportion of each product category in the future based on the context vector.
5. The intelligent prediction method for production stock of heat metering and regulating equipment according to claim 2, wherein the feedback optimization comprises: triggering a model optimization mechanism when the integrated error exceeds a first threshold; and when the error of a certain product class exceeds a second threshold value, performing special analysis and adjusting the model aiming at the product class.
6. An intelligent prediction system for production and stock of a heat metering and regulating device, which is applicable to the intelligent prediction method for production and stock of a heat metering and regulating device according to any one of claims 1 to 5, and is characterized by comprising: The data acquisition and preprocessing module is used for acquiring historical order data and meteorological data and aligning time with the area; the characteristic construction module is used for constructing characteristic variables based on the aligned data; a hybrid prediction module comprising: A gradient lifting tree prediction unit for predicting a total demand of a future period; the sequence-to-sequence prediction unit is used for predicting the demand proportion distribution of each product class; the fusion unit is used for multiplying the total demand by the proportion of each product category to obtain the predicted demand of each product category; and the output and feedback module is used for outputting a prediction result, collecting actual sales data, calculating errors and optimizing the hybrid prediction module.
7. The intelligent prediction system for production stock of heat metering and regulating equipment according to claim 6, wherein the gradient-lifting tree prediction unit adopts XGBoost or LightGBM model.
8. The intelligent prediction system for production stock of heat metering and regulating equipment according to claim 6, wherein the sequence-to-sequence prediction unit adopts an encoder-decoder structure, the encoder is a cyclic neural network or a transducer encoder, and the decoder is a cyclic neural network or a transducer decoder.
9. The intelligent prediction system for production and inventory of heat metering and regulating equipment according to claim 5, wherein the output and feedback module further comprises a dynamic weight adjustment unit for dynamically adjusting weights of absolute errors and relative errors in the integrated errors according to historical errors.
10. The intelligent prediction system for heat metering and regulating equipment production stock of claim 6, further comprising a model update management unit for performing a/B testing, performance evaluation, and new model deployment.

Description

Intelligent prediction method and system for production stock of heat metering and regulating equipment Technical Field The application relates to the field of equipment production stock prediction, in particular to an intelligent prediction method and system for heat metering and regulating equipment production stock. Background With the rapid development of smart cities and central heating systems, the market demand for heat metering devices is growing and is exhibiting significant seasonal, regional fluctuations. The method accurately predicts the production reserve quantity of the heat metering and regulating equipment, and is important for controlling the cost of production enterprises, shortening the delivery cycle and improving the customer satisfaction. As a production provider, scientific stock is a key to ensuring timely supply and reducing inventory costs. The heat metering equipment comprises a household heat meter, a building total heat meter, a temperature control valve, a temperature collector, a balance valve heat meter and the like, and the accurate prediction of market demands has important significance for the efficient operation of the whole heat supply industry chain. At present, in the aspect of predicting the production stock quantity of heat metering and regulating equipment, the closest prior art is a time sequence prediction method based on historical sales data. The data source of the method only depends on the annual sales data in the enterprise, and the prediction model adopts a traditional time sequence prediction model, such as a moving average method, an exponential smoothing method or an ARIMA (autoregressive integral moving average model). The method comprises the specific implementation steps of firstly collecting monthly or quarterly equipment sales data of the past years, then cleaning the data, removing abnormal values, finally inputting the processed data into a model, carrying out extrapolation prediction by identifying the trend and the seasonality of the data, and outputting the predicted demand of each quarter or month of the next year. However, this time series prediction method based on historical sales data has obvious drawbacks. The method has the advantages that firstly, the data dimension is single, only the annual sales data in enterprises are relied on, external driving factors are ignored, sudden changes of external environments cannot be responded, when weather forecast predicts that the winter is abnormal cold, acceleration completion of new construction projects can be promoted, and therefore equipment requirements are driven, but a pure time sequence model cannot capture the external factors. Secondly, prediction accuracy is low, adaptability is poor, for a heat metering market greatly influenced by climate, a prediction result of a traditional time sequence model is often delayed from actual market change, prediction deviation is large, the model needs long time to be adjusted, and adaptability is poor. Third, the subdivision prediction of the product types can not be carried out, the method can only predict the total demand, and the demand distribution of specific product types such as different models, different communication protocols and the like can not be accurately predicted, so that the blindness of production and stock is still caused. Disclosure of Invention The application aims to overcome the technical problems and provides an intelligent prediction method and system for production stock of heat metering and regulating equipment An intelligent prediction method for production stock of heat metering and regulating equipment comprises the following steps: s1, multi-source data acquisition and pretreatment: Collecting historical order data, wherein the historical order data comprises product models, specifications, quantity, order placing time and a region where a customer is located; Acquiring historical and future forecast meteorological data, including heating degree days; Aligning and aggregating the historical order data with meteorological data in time and area; s2, constructing characteristic engineering, namely constructing characteristic variables based on the aligned multi-source data, wherein the characteristic variables comprise historical sales hysteresis characteristics and current heating degree daily number characteristics; S3, building and training a mixed model: predicting the total demand of a future period by adopting a gradient lifting tree model; Predicting the demand proportion distribution of each product class by adopting a sequence-to-sequence model; Multiplying the total demand predicted by the gradient lifting tree model by the ratio of each product class predicted by the sequence-to-sequence model to obtain the predicted demand of each product class. By adopting the technical scheme, the multi-source data are collected and preprocessed, more comprehensive and accurate market information can be obtained, the charac