CN-121314502-B - Cooperative control method and system for chemical reaction kettle clusters and computer equipment

CN121314502BCN 121314502 BCN121314502 BCN 121314502BCN-121314502-B

Abstract

The application discloses a cooperative control method, a cooperative control system and computer equipment for a chemical reaction kettle cluster, and relates to the field of high-end chemical flexible manufacturing. The collaborative control method comprises the steps of obtaining first time sequence data of each reaction kettle based on edge controllers of each reaction kettle in a reaction kettle cluster, synthesizing a first state feature vector with a preset dimension based on the first time sequence data, uploading the first state feature vector to a cloud server by the edge controllers, performing decentralization processing on the first state feature vector based on a mathematical model stored in the cloud server to obtain global state features of the reaction kettle cluster, and sending the global state features to each edge controller by the cloud server to enable the edge controllers to determine target actions executed by the corresponding reaction kettles according to the global state features. By the method, the communication bandwidth occupied by the transmission data is reduced, and the decision efficiency of the cloud end, the anti-interference capability in a dynamic environment and the rapid adaptation capability are improved.

Inventors

ZHANG JIAN
REN BIN
LU QI

Assignees

浙江大学
中国空分工程有限公司

Dates

Publication Date: 20260512
Application Date: 20251212

Claims (9)

1. The cooperative control method of the chemical reaction kettle cluster is characterized by comprising the following steps of: acquiring first time sequence data of each reaction kettle based on an edge controller of each reaction kettle in a reaction kettle cluster, wherein the first time sequence data comprises temperature, pressure, stirring rotating speed and/or feeding flow in the reaction kettle; Determining a feature set in a sliding time window based on the first time sequence data, and combining the feature set into a first state feature vector with a preset dimension, wherein the preset dimension is smaller than the dimension of the first time sequence data, and the feature set comprises a statistical feature, a time sequence feature and a frequency domain feature; the edge controller uploads a first state feature vector to a cloud server, and performs decentralization processing on the first state feature vector based on a mathematical model stored in the cloud server so as to obtain global state features of the reaction kettle cluster; The cloud server sends the global state characteristics to each edge controller, so that the edge controllers determine corresponding target actions executed by the reaction kettles according to the global state characteristics; The method comprises the steps of determining a feature set in a sliding time window based on the first time sequence data, combining the feature set into a first state feature vector with a preset dimension, and comprises the following steps: The method comprises the steps of respectively preprocessing first time sequence data based on at least two residual blocks stacked in a time sequence neural network, taking the first time sequence data as input data of the residual blocks, enabling the input data to pass through two layers of expansion causal convolution, adding output data of the two layers of expansion causal convolution with the input data to obtain an advanced feature sequence, and carrying out global average pooling on the at least two advanced feature sequences to convert the first time sequence data with variable length into a first state feature vector with preset dimension.
2. The cooperative control method according to claim 1, characterized in that, Each of the residual blocks contains at least two layers of the causal convolution of expansion inside.
3. The cooperative control method according to claim 1, characterized in that, Before the cloud server sends the global state features to each edge controller, the cooperative control method includes: Determining a current production target of the reaction kettle cluster, and using the first state feature vector and the current production target as input data to enable a graph neural network model stored in the cloud server to perform similarity comparison on the input data and a historical process map; Wherein the graphic neural network is obtained by aggregation of process nodes and process edges in the historical process map, and/or the process nodes and the process nodes adjacent to the process nodes, the process nodes are used for indicating the operation type, the temperature range, the pressure range and/or the duration of the reaction kettle, and the process edges are used for indicating the transfer relation and the conditions of the reaction kettle when the reaction kettle is operated; And determining the global state features based on the historical process map with the maximum similarity.
4. The cooperative control method according to claim 3, characterized in that, The cooperative control method further comprises the following steps: Acquiring second time sequence data of the reaction kettle after the target action is executed based on the edge controller, and determining a second state feature vector according to the second time sequence data; updating the graph neural network based on the second state feature vector and the current production target.
5. The cooperative control method according to claim 1, characterized in that, The performing the decentration processing on the first state feature vector based on the mathematical model stored in the cloud server includes: establishing an experience playback pool based on a historical process map stored in the cloud server, and randomly sampling training data from the experience playback pool; calculating global Q value of the joint action cost function according to the training data based on a value decomposition network, minimizing the loss function of the global Q value through a time difference error by a gradient descent algorithm, and updating all agents Network parameters of (a); the intelligent agent is used for indicating a digital twin model which is stored in the cloud server and corresponds to the reaction kettle, and the network parameters are used for indicating the intelligent agent Weights in the joint action cost function.
6. The cooperative control method according to claim 5, characterized in that, The joint action cost function The following relationship is satisfied: ; In the formula, The first state feature vectors received for all of the agents; The target actions executed for all the agents are used for updating the first state feature vector, wherein N is the number of the agents; the target action performed for one of the agents; Is the ith intelligent agent Is included in the joint action cost function.
7. The cooperative control method according to claim 1, characterized in that, The cooperative control method further comprises the following steps: Determining network states of the edge controller and the cloud server, and adjusting rule parameters of the edge controller according to the network states, wherein the rule parameters are used for indicating PID parameters of a PID controller built in the edge controller; When the edge controller and the cloud server are in a communication state, the edge controller adjusts the PID parameters based on the global state characteristics sent by the cloud server; When the edge controller is in a disconnected state with the cloud server, the edge controller adjusts the PID parameters based on a pre-established priority rule, wherein the priority rule comprises a first rule and a second rule, and when the first time sequence exceeds a critical state, the first rule is triggered preferentially; The first rule refers to that the edge controller skips the PID controller to control the reaction kettle to trigger a shutdown mechanism, and the second rule refers to that the edge controller determines errors and error change rates according to process variables after the reaction kettle executes the target action and dynamically adjusts the PID parameters through fuzzy reasoning according to the errors and the error change rates.
8. The utility model provides a cooperative control system of chemical industry reation kettle cluster which characterized in that includes: the reaction kettle cluster at least comprises a plurality of reaction kettles; The edge controllers are respectively connected with the corresponding reaction kettles and are used for acquiring first time sequence data of each reaction kettle, wherein the first time sequence data comprises temperature, pressure, stirring rotation speed and/or feeding flow rate in the reaction kettles; determining a feature set in a sliding time window based on the first time sequence data, combining the feature set into a first state feature vector with a preset dimension, wherein the preset dimension is smaller than the dimension of the first time sequence data, the feature set comprises a statistical feature, a time sequence feature and a frequency domain feature, wherein the feature set in the sliding time window is determined based on the first time sequence data, and the feature set is combined into a first state feature vector with the preset dimension, and the method comprises the steps of preprocessing the first time sequence data based on at least two residual blocks stacked in a time sequence neural network, taking the first time sequence data as input data of the residual blocks, enabling the input data to undergo two-layer expansion causal convolution, adding output data with the two-layer expansion causal convolution with the input data to obtain a high-level feature sequence, and carrying out global average pooling on the high-level feature sequence to convert the first time sequence data with variable length into the first state feature vector with the preset dimension; the cloud server is used for receiving the first state feature vector sent by the edge controller, and performing decentralization processing on the first state feature vector based on a mathematical model stored in the cloud to obtain global state features of the reaction kettle cluster; The edge controller is further used for determining a corresponding target action executed by the reaction kettle according to the global state characteristics and controlling an execution terminal in the reaction kettle to execute the target action.
9. A computer device, comprising: A memory and a processor, the memory having a computer program stored therein, which when executed by the processor performs the method of cooperative control of a chemical reaction kettle cluster of any one of claims 1-7.

Description

Cooperative control method and system for chemical reaction kettle clusters and computer equipment Technical Field The application relates to the field of high-end chemical flexible manufacturing, in particular to a cooperative control method, a cooperative control system and computer equipment for a chemical reaction kettle cluster. Background In the field of high-end chemical flexible manufacturing, the adoption of miniaturized and modularized reaction kettles (such as 10L standard reaction kettles) to form a production cluster has become an important development direction. The existing system generally directly uploads the original data (temperature, pressure and the like) collected by the edge end (reaction kettle) to the cloud end. The data carrier mode not only occupies a large amount of communication bandwidth, but also is more critical in that the cloud receives bottom data which is not deeply processed, and the real-time process state of the reaction kettle (such as whether the reaction is in a severe heat release period, whether the heat transfer efficiency is reduced or not) cannot be directly perceived, so that the cloud decision is lagged, and the rapid-change process is difficult to deal with. The reconfiguration logic of the system (e.g., which autoclaves are grouped together to perform a particular process) is dependent on a predetermined rule or fixed model. The system lacks the ability to learn online and dynamically adjust the reconstruction strategy to a limited extent when it is directed to unexpected operating conditions (e.g., batch differences, slight equipment performance decay) or when it is desired to achieve multi-objective optimization (e.g., minimum energy consumption, maximum yield, minimum production cycle, etc.). Disclosure of Invention In order to solve the defects of the prior art, the application aims to provide a cooperative control method, a cooperative control system and computer equipment for a chemical reaction kettle cluster, which can reduce the communication bandwidth occupied by transmission data, and improve the decision-making efficiency of a cloud end, the anti-interference capability and the rapid adaptation capability in a dynamic environment. In order to achieve the above purpose, the application adopts the following technical scheme: In a first aspect, the application provides a cooperative control method for a chemical reaction kettle cluster, which comprises the following steps: Acquiring first time sequence data of each reaction kettle based on an edge controller of each reaction kettle in the reaction kettle cluster, wherein the first time sequence data comprises temperature, pressure, stirring rotation speed and/or feeding flow in the reaction kettle; determining a feature set in a sliding time window based on first time sequence data, and combining the feature set into a first state feature vector with a preset dimension, wherein the preset dimension is smaller than the dimension of the first time sequence data, and the feature set comprises statistical features, time sequence features and frequency domain features; the edge controller uploads the first state feature vector to the cloud server, and performs decentralization processing on the first state feature vector based on a mathematical model stored in the cloud server so as to obtain global state features of the reaction kettle cluster; and the cloud server sends the global state characteristics to each edge controller, so that the edge controllers determine the target actions executed by the corresponding reaction kettles according to the global state characteristics. In some implementations, determining a feature set within a sliding time window based on first time sequence data and combining the feature set into a first state feature vector of one preset dimension includes: Preprocessing first time sequence data based on at least two residual blocks stacked in a time sequence neural network respectively to obtain an advanced feature sequence, wherein each residual block internally comprises at least two layers of expansion causal convolution; global averaging is performed on at least two advanced feature sequences to convert first time-ordered data of variable length into a first state feature vector. In some implementations, preprocessing the first time series data based on at least two residual blocks stacked in the time series neural network, respectively, to obtain an advanced feature sequence includes: Inputting first time sequence data as input data into a residual block, and enabling the input data to undergo two-layer causal convolution; the output data after two layers of causal convolution of expansion is added to the input data to obtain an advanced feature sequence. In some implementations, before the cloud server sends the global status feature to each edge controller, the cooperative control method includes: determining a current production target of the reaction kettle cluster