CN-122018361-A - Multivariable time sequence prediction and dynamic optimization method and device

CN122018361ACN 122018361 ACN122018361 ACN 122018361ACN-122018361-A

Abstract

The invention relates to a multivariable time sequence prediction and dynamic optimization method and device, which are applied to real-time data prediction and control optimization in an industrial process. The time sequence data of the sensor modules are collected, the deep learning models such as a long-short-term memory network model and the like are adopted for real-time prediction, and the parameters of the control system are dynamically adjusted, so that the production efficiency is improved, the resource use is optimized, and the energy consumption is reduced. The method is suitable for a wide range of industrial processes and has high application value.

Inventors

LI SHEN
XIONG RUI
CUI HAILONG
LIU JIANPENG
LIU XIANG
LIU YAN
MA XUELIN
WU CHEN
HAO JUNHUA
DONG BAOCHENG

Assignees

唐钢国际工程技术有限公司
唐山钢铁集团有限责任公司
河钢乐亭钢铁有限公司

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (9)

1. A multivariable time sequence prediction and dynamic optimization method is characterized by comprising the following steps: S1, configuring a data acquisition and sensor module, simultaneously acquiring time sequence data of multiple dimensions in real time through multiple sensor modules, and collecting all acquired time sequence data into a central data processing unit so as to carry out subsequent processing; S2, data preprocessing, namely performing time alignment and normalization on the data of different sensors, wherein the time alignment ensures that the data of the different sensors can be synchronously compared on the same time dimension; s3, model training and optimization are carried out, the preprocessed data are input into a prediction model, the prediction model for deep learning is trained, the model parameters are optimized and adjusted by using an optimization algorithm in the training process, and the prediction model after training is evaluated; S4, the model is lightened and deployed, the trained prediction model is subjected to light weight treatment, and the calculation burden and storage pressure of the prediction model are reduced; s5, dynamic optimization control, wherein the control system dynamically adjusts the operation parameters of industrial equipment according to the key parameters predicted in real time based on the real-time data and the prediction result of the prediction model, so that the optimization of the industrial production process is realized; s6, system integration and upper layer data interaction ensure that real-time prediction and optimization results can be shared with the upper layer system, and production scheduling optimization or quality control is achieved.
2. The method of multivariate timing prediction and dynamic optimization of claim 1, wherein S1, the collected timing data includes, but is not limited to, one or more of temperature, pressure, flow, concentration, and composition.
3. The method of multivariate timing prediction and dynamic optimization of claim 2, wherein in S2, the normalization process is to convert all data to a unified standard range of 0 to 1, and the normalization process includes but is not limited to Min-Max normalization.
4. The method for predicting and dynamically optimizing multi-variable time sequence according to claim 2, wherein in S3, the prediction model uses a long-short-time memory network model as the prediction model, so that long-term dependency in time sequence data is effectively captured.
5. The method for predicting and dynamically optimizing multi-variable time sequence according to claim 4, wherein in S3, a back propagation algorithm is used for optimization in the training process, model parameters are adjusted through error back propagation, differences between predicted values and actual values are measured through a loss function until the predicted model converges, the loss function comprises but is not limited to a mean square error function, and the trained predicted model is evaluated through a cross verification method to ensure the predicted result of the predicted model and generalization capability under different scenes.
6. The method for predicting and dynamically optimizing multi-variable time sequence according to claim 5, wherein in S4, the step of performing light weight processing on the prediction model after training means that the long-short-time memory network model after training is converted into ONNX format, and the edge equipment comprises one or more of an industrial personal computer and an embedded system.
7. The method for predicting and dynamically optimizing multi-variable time sequence according to claim 1, wherein in S5, algorithms selected for dynamic optimization control include, but are not limited to, PID control and fuzzy control, and the dynamic optimization control specifically means that when the impending temperature rise or pressure change is predicted, a control system can automatically adjust key process parameters of the rotating speed, input quantity and flow of industrial equipment so as to ensure stable operation of an industrial production process.
8. The method of claim 1, wherein S6, the industrial communication protocol includes but is not limited to OPC, modbus, TCP/IP, MQTT, the same or multiple of the same.
9. A multivariable timing prediction and dynamic optimization apparatus utilizing the method of any one of claims 1-8, comprising: The sensor modules are multiple in number and are used for collecting time sequence data of multiple dimensions in the industrial process; the data processing module is used for preprocessing data; The prediction module is internally provided with a prediction model, and the pre-processed data is predicted in real time through the prediction model; The light-weight and deployment module is used for carrying out light-weight processing and deployment on the prediction model after training; the control system dynamically adjusts the operation parameters of the industrial equipment based on the prediction result; and the integration and interaction module is used for sharing the real-time prediction and optimization result with the upper layer system to realize production scheduling optimization or quality control.

Description

Multivariable time sequence prediction and dynamic optimization method and device Technical Field The patent application belongs to the technical field of industrial process control, and particularly relates to a prediction and optimization method and device based on time sequence data, which are applied to dynamic regulation and optimization control in an industrial process. The method and the device can be used for processing multi-source heterogeneous time sequence data, and the efficiency and the quality of the production process are improved by predicting and dynamically adjusting system parameters in real time. Background With the development of industrial automation, more and more industrial processes involve the collection and processing of various real-time monitoring data. These data are typically represented as time series data, such as temperature, pressure, flow, composition, and the like, in multiple dimensions. The existing prediction and control methods often depend on static rules or simple regression models, and lack comprehensive analysis and real-time processing capacity on multi-variable time sequence data. How to effectively combine time sequence data to predict and dynamically optimize control parameters in the production process becomes a technical problem to be solved in the current industrial field. Disclosure of Invention The technical problem to be solved by the invention is to provide the multivariable time sequence prediction and dynamic optimization method and device, which can predict multidimensional time sequence data in the industrial process in real time and dynamically adjust parameters of a control system according to a prediction result, thereby optimizing production efficiency, improving product quality and reducing energy consumption and material consumption. Specifically, the technical scheme of the invention realizes accurate prediction and dynamic adjustment of key process parameters in industrial process control, so that the production process can adapt to different working condition changes, and the purposes of high efficiency, energy saving and resource use optimization are achieved. In order to solve the problems, the invention adopts the following technical scheme: A multivariable time sequence prediction and dynamic optimization method comprises the following steps: step one, data acquisition and sensor module configuration In industrial processes, there are many different sensors, such as temperature sensors, pressure sensors, flow sensors, etc., which can collect data of various physical or chemical quantities in real time and transmit the data to a data processing system. In order to ensure accuracy and timeliness of data, the first step of the invention is to collect time sequence data of multiple dimensions in real time through multiple sensor modules. The collected data may include, but is not limited to, temperature, pressure, flow, concentration, composition, and the like. All collected sensor data will be pooled into a central data processing unit for further preprocessing and analysis. Step two, data preprocessing The data originally collected are often data of different sensors at different time points, and the problems of data missing, data noise or inconsistent data time sequence and the like can occur, so that data preprocessing is needed. In the data preprocessing process, time alignment is carried out on the data of different sensors, so that the data of different sensors can be synchronously compared in the same time dimension. The normalization technology is adopted to process the data, so that the data with different dimensions and different ranges can be input and compared in the same model, and the data can be ensured to adapt to the input requirement of a subsequent deep learning model (prediction model). Step three, model training and optimization The preprocessed data is used to train a deep learning predictive model. And a long-time memory network model is used as a prediction model, so that long-term dependency relationship in time series data is effectively captured. The input of the prediction model is past time sequence data, and the output is a predicted key parameter. During model training, the input data is typically historical data over the past 60 minutes, which is used as an input feature to help the predictive model to predict key parameters such as temperature, pressure, flow, etc. for future time periods. The long-short-term memory network model learns to identify potential time sequence characteristics based on the change rule of the historical data, and predicts the future production process by using the characteristics. In the training process, the model continuously adjusts the internal parameters so as to reduce the prediction error. Step four, model light weight and deployment The well-trained long and short term memory network model contains a large number of parameters and computational requirements. In order to