Material feeding control method, system and equipment for cadmium oxide production line and medium

Abstract

The invention belongs to the technical field of industrial control, and discloses a material feeding control method, system and equipment of a cadmium oxide production line and a medium. The sensor data is first sensor data collected at a plurality of time scales or second sensor data collected at a single time scale; fusing the first sensor data to obtain a multi-scale fusion vector of each index collected at a plurality of time scales; inputting all the multi-scale fusion vectors and the second sensor data into an LSTM network to obtain time sequence features output by the LSTM network; inputting the time sequence features into a BP neural network to obtain optimized PID parameters output by the BP neural network; and inputting the optimized PID parameters into a PID controller so as to optimize and adjust the feeding proportion of the material. The feeding control precision and stability of the cadmium oxide production line can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial control, and in particular relates to a material feeding control method, system, equipment and medium for a cadmium oxide production line. Background Art

[0002] Cadmium oxide (CdO), as an important inorganic compound, is widely used in semiconductor materials, optoelectronic devices, pigments, transparent conductive films and other fields. Its production process usually involves the oxidation reaction of cadmium metal or cadmium compounds in oxygen or air. The production process of cadmium oxide is relatively complex, especially in industrial-scale production, involving precise ratio control of multiple raw materials, including cadmium metal, oxygen, catalysts and other additives.

[0003] In modern industrial production, PID automatic control based on BP neural network has become a widely used intelligent control method, especially in complex process industries such as chemical industry and materials. BP neural network uses its powerful nonlinear mapping ability to adjust PID controller parameters in real time to help the system maintain stability in an uncertain environment.

[0004] However, with the complexity and refinement of production processes, especially in application scenarios with high precision requirements such as feed control, BP neural networks have exposed some shortcomings in processing real-time data and maintaining control accuracy: the dynamic relationship between various data in real-time data is constantly changing in the production environment, and BP neural networks are difficult to adapt in real time, often resulting in delayed control response; lack of sensitivity to time series and multi-scale data, unable to accurately capture and predict mutation signals or trend changes, affecting control accuracy.

[0005] Therefore, the accuracy and stability of current feed control need to be improved. Summary of the invention

[0006] The object of the present invention is to provide a material feed control method, system, equipment and storage medium for a cadmium oxide production line, which can improve the accuracy and stability of feed control.

[0007] The first aspect of the present invention discloses a material feeding control method for a cadmium oxide production line, comprising: Collecting sensor data corresponding to various indicators in the production process, wherein the sensor data is first sensor data collected at multiple time scales or second sensor data collected at a single time scale; fusing the first sensor data to obtain a multi-scale fusion vector of each indicator collected at multiple time scales; Inputting all the multi-scale fusion vectors and the second sensor data into the LSTM network to obtain the time series features of the LSTM network output; Input the time series characteristics into the BP neural network to obtain the optimized PID parameters output by the BP neural network; The optimized PID parameters are input into a PID controller to optimize and adjust the feed ratio of the material.

[0008] In some embodiments, after collecting sensor data corresponding to various indicators in the production process, the method further includes: Calculating the correlation between the sensor data and the control target, and adjusting the sensor acquisition frequency and / or the weight corresponding to the sensor data according to the correlation; and / or, The covariance between the sensor data is calculated, and data dimension reduction is performed according to the covariance and / or the weight corresponding to the sensor data is adjusted.

[0009] In some embodiments, after collecting sensor data corresponding to various indicators in the production process, the method further includes: Eliminating noise and outliers in the sensor data to obtain processed sensor data; The processed sensor data is input into a PID controller to make preliminary adjustments to the feed ratio of the material.

[0010] In some embodiments, after collecting sensor data corresponding to various indicators in the production process, the sensor data is sampled multiple times using different time scales, and the first sensor data is generated based on the sensor data.

[0011] In some embodiments, the time series features include features extracted from sensor data corresponding to the feed rates of cadmium metal, oxygen, and catalyst, and features extracted from sensor data corresponding to reactor temperature, reactor pressure, and oxygen concentration; the optimized PID parameters include proportional parameters, integral parameters, and differential parameters corresponding to cadmium metal, oxygen, and catalyst, respectively.

[0012] In some embodiments, all of the multi-scale fusion vectors and the second sensor data are input into an LSTM network to obtain a time series feature output by the LSTM network, including: In the LSTM network, a forget gate is used to forget the interference of sudden fluctuations on control parameters in the stable production stage, an input gate is used to filter the data of the memory unit added to the LSTM network in the multi-scale fusion vector and the second sensor data, and an output gate is used to output key features of multiple time steps according to the memory unit of the LSTM network to form the time series features.

[0013] The second aspect of the present invention discloses a material feeding control system for a cadmium oxide production line, comprising: A perception layer, used to collect sensor data corresponding to multiple indicators in the production process, wherein the sensor data is first sensor data collected at multiple time scales or second sensor data collected at a single time scale; A processing layer, used for fusing the first sensor data to obtain a multi-scale fusion vector of each indicator collected at multiple time scales; The intelligent optimization layer inputs all the multi-scale fusion vectors and the second sensor data into the LSTM network to obtain the time series features output by the LSTM network; inputs the time series features into the BP neural network to obtain the optimized PID parameters output by the BP neural network, and inputs the optimized PID parameters into the control layer; The control layer is used to optimize and adjust the feed ratio of the material according to the optimized PID parameters; The execution layer is used to control the valves, pumps and material conveying devices to control the material feed ratio according to the adjustment instructions of the control layer.

[0014] In some embodiments, the processing layer is further used to calculate the correlation between the sensor data and the control target, and adjust the sensor acquisition frequency and / or the weight corresponding to the sensor data according to the correlation; and / or, The covariance between the sensor data is calculated, and data dimension reduction is performed according to the covariance and / or the weight corresponding to the sensor data is adjusted.

[0015] The third aspect of the present invention discloses an electronic device, comprising a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the material feeding control method for the cadmium oxide production line disclosed in the first aspect.

[0016] The fourth aspect of the present invention discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the material feed control method for a cadmium oxide production line disclosed in the first aspect.

[0017] The beneficial effect of the present invention is that by adopting multiple time scales to collect sensor data, it is possible to extract richer information on short-term and long-term changes, and by capturing time series characteristics through an LSTM network, it is possible to accurately capture and predict mutation signals or trend changes, and then by converting the time series characteristics into optimized PID parameters through a BP neural network, it is possible to cope with the continuous changes in the dynamic relationship between various data in the production environment, and improve the accuracy and stability of the feed control of the cadmium oxide production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings herein show specific examples of the technical solutions described in the present invention, and together with the specific implementation methods constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.

[0019] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.

[0020] Figure 1 It is an architecture diagram of a material feeding control system of a cadmium oxide production line disclosed in an embodiment of the present invention; Figure 2 It is a flow chart of a material feeding control method for a cadmium oxide production line disclosed in an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. In combination with the technical solution of the present invention in a realistic scenario, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. The "first, second..." used herein is only used to distinguish the names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0022] It should be noted that when a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be a central component; when an component is considered to be "connected to" another component, it can be directly connected to the other component or there can be a central component at the same time; when an component is considered to be "installed on" another component, it can be directly installed on the other component or there can be a central component at the same time. When an component is considered to be "set on" another component, it can be directly set on the other component or there can be a central component at the same time.

[0023] Unless otherwise specified or defined, the "said" and "the" used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms "including" and "having" and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0024] At present, the stability and accuracy of production line feed control are still not ideal, which limits the application effect of traditional BP neural networks in high-precision, multi-variable industrial control.

[0025] In view of the above problems, the present invention uses dynamic correlation monitoring, covariance analysis, multi-scale data fusion and incremental data cleaning methods to process data, improve data correlation, data quality and time dependence of the model, and then uses a hybrid control model formed by combining LSTM (Long Short-Term Memory Network) network and BP (Back propagation Neural Network) neural network to control feeding, forming an intelligent, high-precision and robust material feeding control system. It can respond to the dynamic relationship changes between data in a timely manner, accurately capture and predict mutation signals or trend changes based on time series and multi-scale data, and improve control accuracy and stability.

[0026] For situations where the precise ratio of raw materials in the cadmium oxide production line is more difficult to control, such as Figure 1 As shown, this embodiment provides a material feeding control system, and the overall architecture of the control system can be divided into a perception layer, a processing layer, an intelligent optimization layer, a control layer and an execution layer.

[0027] The perception layer collects sensor data corresponding to various indicators in the production process (i.e., key parameters in the production process). Some indicators are collected at multiple time scales (such as 1 second and 5 seconds) to obtain first sensor data, and some indicators are collected at a single time scale to obtain second sensor data. These real-time collected sensor data provide basic data input for control and optimization, and are also used for PID control and learning and optimization of hybrid control models. The perception layer of this embodiment includes the following hardware: flow sensor, temperature sensor, pressure sensor, gas sensor, material level sensor, edge processing gateway and PLC (programmable logic controller), among which, the flow sensor is used to monitor the flow and feed rate of materials (cadmium metal, oxygen, catalyst, etc.) to ensure that the input of each material is accurately controllable; the temperature sensor is used to monitor the temperature in the reactor to ensure that the reaction environment meets the requirements; the pressure sensor is used to monitor the pressure in the reactor to prevent the process conditions from exceeding the safety range; the gas sensor is used to detect the concentration of oxygen or other gases during the reaction to ensure that the gas supply ratio meets the reaction requirements; the material level sensor is used to monitor the inventory of materials to avoid insufficient or excessive materials affecting the reaction; the edge processing gateway is used to collect and transmit data fed back by the sensor to ensure that the real-time data of the sensor is transmitted to the PLC; the PLC is used to preliminarily process the sensor data and feed back the processing results to the PID controller and the processing layer, and can also control the basic production process.

[0028] The processing layer preprocesses and optimizes the collected sensor data, performs multi-level and multi-scale processing, and provides high-quality, highly correlated and noise-filtered data input for the hybrid control model; the first sensor data is also fused to obtain a multi-scale fusion vector for each indicator collected at multiple time scales. In this embodiment, the processing layer calculates the correlation between the sensor data and the control target, dynamically adjusts the sensor acquisition frequency and / or adjusts the weight corresponding to the sensor data according to the correlation; and / or calculates the covariance between the sensor data, performs data dimension reduction according to the covariance and / or adjusts the weight corresponding to the sensor data; the processing layer can also eliminate noise and outliers in the sensor data to obtain processed sensor data, and simultaneously input the processed sensor data into the control layer.

[0029] The intelligent optimization layer inputs all multi-scale fusion vectors and second sensor data into the LSTM network to obtain the time series features of the LSTM network output; the time series features are input into the BP neural network to obtain the optimized PID parameters output by the BP neural network, and the optimized PID parameters are input into the control layer to dynamically adjust the parameters of the PID controller through the BP neural network. The intelligent optimization layer is deployed on high-performance industrial servers or industrial computers to ensure the computing power of the LSTM network and BP neural network. The BP neural network will continuously learn the nonlinear behavior of the system based on historical data and real-time production data, and optimize the PID parameters to cope with changes in process conditions and nonlinear responses.

[0030] The control layer includes: PID controller and valve controller. Among them, the PID controller is the main controller responsible for adjusting the material input in the entire production process. It optimizes and adjusts the material feed ratio according to the optimized PID parameters. It is the core control unit of the entire system and is responsible for adjusting the material feed ratio in real time. The valve controller is connected to the PID controller to adjust the valve opening in the material conveying system and control the flow of materials and gases.

[0031] The PID controller can also use processed sensor data to preliminarily adjust the proportions of multiple materials. That is, according to the processed sensor data, the difference between the set target value (desired feed ratio, temperature or pressure) and the processed sensor data is used to make proportional, integral and differential adjustments, and output control instructions to the valve controller.

[0032] The execution layer is the actual implementation unit of the control instructions. According to the control instructions of the control layer, it controls valves, pumps and material conveying devices to adjust the material feed ratio. The execution layer includes: pneumatic valves, conveying pumps and dosing devices, among which pneumatic valves are used to control the flow of fluids or gases, such as the feed control of cadmium metal, oxygen, additives and other materials; conveying pumps are used to accurately control the feed amount of solid or liquid materials; dosing devices are used to accurately feed powder or granular materials. The execution layer equipment will act according to the instructions issued by the PID controller, and transmit the execution results back to the perception layer through the feedback loop to form a closed-loop control.

[0033] When the material feed control system is running, a material feed control method for a cadmium oxide production line according to an embodiment of the present invention is executed to control the material feed of the cadmium oxide production line. In order to facilitate understanding of the present invention, a specific embodiment of the present invention will be described in more detail below with reference to the accompanying drawings of the specification.

[0034] like Figure 2 As shown, the method comprises the following steps: Step S100: collecting sensor data corresponding to various indicators in the production process, wherein the sensor data is first sensor data collected at multiple time scales or second sensor data collected at a single time scale; Among them, the indicators are key process parameters in production, such as: cadmium metal, oxygen, catalyst feed rate, reactor temperature, reactor pressure and oxygen concentration, etc. Sensor data corresponding to each indicator is collected through various sensors (such as flow, temperature, and pressure sensors).

[0035] When collecting sensor data in this embodiment, for indicators with significant dynamic changes in the production process, such as fluctuations in feed rate or changes in oxygen concentration, multiple time scales are used for collection, that is, the same indicator is collected at two time scales, short time and long time, to obtain the first sensor data. For indicators that change slowly in the production process, a single time scale is used to collect and obtain the second sensor data. The short time scale can capture the rapid fluctuations of the indicator; the long time scale can capture the long-term trend of the indicator. For example: for the feed rate, the short time scale is collected once per second, which is used to detect sudden changes (such as sudden adjustments to the valve opening), and the long time scale is to take an average value every 5 seconds, which is used to analyze long-term change trends (such as the gradual increase or decrease of the feed rate over a period of time). Through the combination of long-term and short-term collection, the multi-level change characteristics of the indicator can be captured, especially in dynamic control scenarios, which can not only respond to short-term fluctuations in a timely manner, but also adjust the control strategy according to the long-term trend. Moreover, the first sensor data contains rich information, which is conducive to the LSTM network learning the dependencies of sensor data in different time series.

[0036] In some embodiments, for indicators with significant dynamic changes in the production process, after collecting sensor data, the sensor data is sampled multiple times using different time scales, for example, re-sampling the sensor data at a time scale of 1 second or 5 seconds, thereby generating first sensor data based on the sensor data.

[0037] Step S200: fusing the first sensor data to obtain a multi-scale fusion vector of each indicator collected at multiple time scales; This step is called multi-scale data fusion, which fuses the short-term and long-term data in the first sensor data into a multi-scale fusion vector for use in the subsequent LSTM network. Assume that the short-term data vector is , the long-term data vector is , then the multi-scale fusion vector is: .in, Represents short-term data vectors, such as rapidly changing values ​​of feed rate; Represents a long-term data vector, such as the long-term trend of feed rate; Represents a multi-scale fusion vector, which provides information on the changes of each indicator at different time scales.

[0038] In cadmium oxide production, the first sensor data containing data of different time scales is fused and input into the LSTM network. This can not only quickly respond to feed fluctuations, but also track the long-term trend of feed, thus enhancing the accuracy of the control system.

[0039] Step S300: inputting all multi-scale fusion vectors and second sensor data into the LSTM network to obtain the time series features output by the LSTM network; Industrial feed control usually involves data changes at different time scales. For example, temperature changes slowly, but oxygen concentration and feed rate change quickly. BP neural network is weak in processing data at different time scales and lacks sensitivity to time series and multi-scale data features, which results in the control system being unable to accurately capture and predict mutation signals or trend changes, affecting control accuracy.

[0040] Therefore, in this embodiment, all multi-scale fusion vectors and second sensor data are input into the LSTM network, and the LSTM network can learn the temporal characteristics of the input data. The LSTM network contains input gates, forget gates, and output gates, which enable it to selectively retain or forget historical information, solve the gradient vanishing problem of traditional neural networks, and is suitable for processing sequence data with long-term dependencies. The time series features output by the LSTM network are then used as the input of the BP neural network.

[0041] In the LSTM network, the input of each time step will be processed by three key gates, namely the forget gate, input gate and output gate. The forget gate is used to forget the interference of sudden fluctuations on the control parameters in the stable production stage. The input gate is used to filter the data added to the memory unit of the LSTM network in the multi-scale fusion vector and the second sensor data. The output gate is used to output the key features of multiple time steps according to the memory unit of the LSTM network to form time series features.

[0042] Specifically, the forget gate determines how much past information the LSTM network needs to discard in the current time step. In feed control, the forget gate determines whether to forget certain previous historical features, such as forgetting the interference of sudden fluctuations on control parameters in the stable production stage. The calculation formula of the forget gate is: ,in, It is the output of the forget gate, with a value between 0 and 1, indicating the proportion of forgetting, 0 means complete forgetting, and 1 means complete retention; is the weight matrix of the forget gate; is the hidden state at the previous time step. is the input of the current time step (i.e., time series data such as temperature and feed rate); is the bias of the forget gate; It is the Sigmoid activation function, which maps the value to between 0 and 1.

[0043] The input gate determines how much new information (i.e., current sensor data) needs to be added to the memory cell at the current time step. This step is particularly critical for feed control with high real-time requirements, allowing the LSTM network to respond quickly when the control system changes (such as a sudden increase in feed rate). The calculation process of the input gate is: 1. Control the "intensity of adding" new information through the input gate: ,in, is the output of the input gate, is the weight matrix of the input gate, is the hidden state at the previous time step, is the input of the current time step (i.e., time series data such as temperature and feed rate), is the bias of the input gate. 2. Generate a candidate memory content , which is the information that can be added to the memory at the current time step: , is the weight matrix of the memory gate, is the bias of the memory gate; 3. Obtain the calculation result of the input gate: . is the memory cell state at the current time step; is the memory state of the previous time step; It is the historical information part controlled by the forget gate, indicating how much memory of the previous moment is retained; It is the current information part controlled by the input gate, indicating how much information is introduced into the current input.

[0044] The output gate determines which information will be extracted from the current memory cell and output as the input for the next time step. This output will serve as the hidden state of the current time step. , and passed to the next time step. For feed control, the output gate can help the LSTM network output the key features of each time step based on a series of time series data to generate PID control parameters. The output gate calculation formula is: , then, based on the activation value of the memory cell state and the weight of the output gate, the final hidden state It can be calculated as: .in, is the weight matrix of the output gate, is the bias of the output gate; It is the hidden state of the current time step, that is, the feature extraction result of the current time step, which contains the dynamic timing information of the input data; It is the activation value of the memory unit, compressing the state data to between -1 and 1; is the output control ratio of the output gate.

[0045] At each time step , the hidden state of the LSTM network will contain the key features of the previous time steps. These hidden states form a time series feature vector, that is, the time series feature , used for subsequent BP neural network input.

[0046] In this embodiment, the hidden state of the LSTM network output The input of the BP neural network includes six key features closely related to the feeding control of three materials, features extracted from sensor data corresponding to the feeding rate of cadmium metal, oxygen, and catalyst, and features extracted from sensor data corresponding to reactor temperature, reactor pressure, and oxygen concentration.

[0047] Step S400: inputting the time series features into the BP neural network to obtain the optimized PID parameters output by the BP neural network; BP neural network is mainly used to dynamically adjust the PID parameters of PID controller (including: P, proportional parameter, I, integral parameter, D, differential parameter), that is, the timing characteristics learned from the time series characteristics are further converted into PID parameters to achieve adaptive optimization control of the feed ratio of multiple materials in cadmium oxide production.

[0048] The BP neural network consists of an input layer, a hidden layer, and an output layer. It includes 6 key features, so the input layer has 6 neurons, corresponding to 6 input variables: , is the feed rate of cadmium metal, is the oxygen delivery rate, is the catalyst feed rate, is the reactor temperature, is the reactor pressure, is the oxygen concentration. The hidden layer is responsible for extracting the features of the input information and learning the complex relationship between input and output through nonlinear mapping. In this embodiment, the hidden layer has 10 neurons, and the output of the hidden layer is calculated as: , represents the weight from the ith neuron in the input layer to the jth neuron in the hidden layer. The weight reflects the influence of the ith input variable in the input layer on the output of the jth neuron in the hidden layer. represents the ith input value of the input layer. The input variables include the feed rates of cadmium metal, oxygen and catalyst, reactor temperature, pressure and oxygen concentration, so may correspond to one of these variables respectively; represents the bias term of the jth neuron in the hidden layer, represents the weighted sum of all input variables in the input layer, and is the weighted input value of the jth neuron in the hidden layer. The output vector of the hidden layer is: The output layer is responsible for generating the PID parameters (P, I, D) of the PID controller for each material, which is used to adjust the feed rate of each material. Since three materials (cadmium metal, oxygen, catalyst) are controlled, the output layer contains 9 neurons. The optimized PID parameters are .in, is the PID control parameter of cadmium metal, is the PID control parameter of oxygen, is the PID control parameter of the catalyst. The calculation formula of the output layer is ,in, is the weight from the hidden layer to the output layer, is the bias term of the output layer.

[0049] Step S500: inputting the optimized PID parameters into the PID controller to optimize and adjust the feed ratio of the material.

[0050] After receiving the optimized PID parameters output by the BP neural network, the PID controller outputs control signals to pneumatic valves, delivery pumps, dosing devices and other equipment through proportional (P), integral (I) and differential (D) algorithms, according to the deviation between the optimized PID parameter value and the actual PID parameter value, to adjust the opening of the material feed valve or the flow of the pump, optimize the material feed ratio, and achieve precise control of the material ratio. The action results of pneumatic valves, delivery pumps, dosing devices and other equipment (such as changes in material flow, temperature or pressure changes in the reactor) are collected and fed back through sensors to form a closed-loop control.

[0051] In this embodiment, it is also adjusted continuously according to real-time data and feedback to ensure that the production process is always in the optimal state. The BP neural network continues to learn online, that is, during the generation process, after each set of new sensor data is obtained, forward propagation and back propagation will be performed to update the weights and biases. This gradual adjustment method enables the network to adapt to the ever-changing data pattern, thereby coping with complex production conditions. As production continues, the control effect becomes more and more precise and intelligent.

[0052] In summary, by collecting sensor data at multiple time scales, we can extract richer information on short-term and long-term changes, capture time series features through the LSTM network, and convert the time series features into optimized PID parameters through the BP neural network, which can improve the accuracy and stability of the feed control of the cadmium oxide production line.

[0053] Traditional BP neural networks mainly rely on historical data for training. Although they have certain learning capabilities, they are unable to quickly and adaptively capture the dynamic correlation between key variables when faced with real-time data changes in complex processes. For example, in the production process of cadmium oxide, the feed rate, temperature, pressure and oxygen concentration have a large mutual influence, and this dynamic relationship is constantly changing in the production environment. Traditional BP neural networks are difficult to adapt in real time, often resulting in delayed control response. Moreover, in multivariable control systems, each data source has a different degree of influence on the control target. Traditional BP neural networks treat input data equally and cannot adjust the priority of data sources according to changes in actual working conditions. This may cause key data to be ignored and secondary data to be overly concerned. This causes the BP neural network to fail to control key parameters such as feed rate, which in turn affects production efficiency and product quality.

[0054] Therefore, after collecting sensor data corresponding to various indicators in the production process, this embodiment also performs dynamic correlation detection and / or covariance analysis. The dynamic correlation detection is used to calculate the correlation between the sensor data and the control target, and dynamically adjust the sensor collection frequency and / or the weight corresponding to the sensor data according to the correlation; the covariance analysis is used to calculate the covariance between the sensor data, and perform data dimensionality reduction and / or adjust the weight corresponding to the sensor data according to the covariance.

[0055] Specifically, dynamic correlation detection identifies which indicators have a more significant impact on control targets (such as feed ratio and reaction rate) under current production conditions. The frequency and priority of data collection are dynamically adjusted according to the monitored correlation to ensure that key data sources are given priority.

[0056] First, the correlation calculation is performed, and the Pearson Correlation Coefficient is used to calculate the linear relationship between the indicator and the control target. The calculation formula of the correlation coefficient is: , in, and They represent the sensor data of the indicator and the observed value of the control target, respectively. and Represent the sensor data of the indicator and the mean of the control target, The value range is between [-1, 1]. The closer the value is to 1 or -1, the stronger the correlation between the two variables, and the closer it is to 0, the weaker the correlation.

[0057] Then make a priority adjustment. Every other sampling period, calculate the relevance of each indicator to the control target. If the relevance of some indicators exceeds the preset threshold (such as 0.7), increase the data collection frequency of the indicator. For example, if the correlation between oxygen concentration and feed rate reaches 0.8 at a certain moment, increase the frequency of oxygen concentration collection to obtain more critical information.

[0058] Covariance analysis is used to identify covariance relationships between indicators in a feed control system, i.e., two or more indicators have similar trends in numerical changes. For example, temperature and oxygen concentration may show strong synergistic variation relationships under certain operating conditions. When certain indicators show strong covariance relationships, they can be subjected to dimensionality reduction (such as principal component analysis, PCA) or smoothing filtering to reduce noise interference in the data input. And / or, the weights corresponding to the sensor data can be adjusted by adjusting the weight matrix in the LSTM network.

[0059] Real-time data collected in industrial production often contains various noises and sudden interferences. Traditional BP neural networks do not have special processing for industrial noise, which leads to noise and interference in input data affecting the control effect. Especially in the case of large data volatility, the network output is prone to instability, making the feed ratio control inaccurate and even causing system instability.

[0060] Therefore, this embodiment also performs incremental data cleaning to eliminate noise and outliers in the sensor data. For example, after detecting the rationality of the sensor data, the data is corrected or smoothed according to the preset rules to obtain the processed sensor data. The processed sensor data can be directly input into the PID controller to make a preliminary adjustment to the feed ratio of the material; and the processed sensor data is fused and input into the LSTM network to ensure that the data entering the LSTM network maintains high quality.

[0061] For example, the process of outlier detection is: setting a reasonable threshold for each indicator based on historical data. Assuming that the reasonable range of temperature is , for temperature data ,like The process of data correction or smoothing is to replace the outlier with the mean of the previous moment and the next moment. ; or, use linear interpolation to fill in outliers.

[0062] In this embodiment, the feed control system in cadmium oxide production collects the feed rate once per second. Assume that during a production conversion, the system records an abnormal increase in the fluctuation of the feed rate. Incremental cleaning will detect the new data of each second one by one according to the current fluctuation situation, and apply interpolation or mean smoothing to the data that exceeds the reasonable range, so as to ensure that the sensor data is always stable and the production efficiency will not be affected by abnormal fluctuations.

[0063] Through the above-mentioned methods such as dynamic correlation detection, covariance analysis and incremental data cleaning, the quality of sensor data can be improved, the sensor data can be ensured to be accurate and effective, and the accuracy and stability of feed control of the cadmium oxide production line can be improved.

[0064] In this embodiment, a hybrid control model is composed of an LSTM network and a BP neural network. It is easy to understand that a data processing module can also be added to the hybrid control model to perform operations such as dynamic correlation detection, covariance analysis, incremental data cleaning, and multi-scale data fusion in the data processing module.

[0065] Before using the hybrid control model, the LSTM network and BP neural network need to be jointly trained. The specific training process is: the sensor data such as feed rate, temperature, humidity and oxygen concentration collected in real time are pre-processed by dynamic correlation detection, covariance analysis, incremental data cleaning and multi-scale data fusion, and then input into the LSTM network. The LSTM network processes the input data, extracts time series features, and generates feature vectors. , input the feature vector output by the LSTM network into the BP network, the BP network generates PID parameters (P, I, D), compares the generated PID parameters with the target value, and calculates the MSE loss. Back propagates from the BP network output layer to the LSTM layer layer by layer, calculates the gradient of the loss to the weight of each layer, updates the weights of the LSTM and BP networks according to the gradient, completes a training iteration, and repeats the above process until the loss function converges and the control effect of the hybrid control model meets the expected requirements.

[0066] Among them, the loss function reflects the deviation between the PID parameters generated by the hybrid control model and the desired control parameters, so that the hybrid control model can be adjusted towards the optimal control direction. Assuming that the target PID parameters are , the model output PID parameters are , then the MSE loss function can be expressed as: , Where N is the number of PID parameters (e.g., P, I, and D for 3 materials, a total of 9). and They represent the kth PID parameter and target value output by the hybrid control model respectively.

[0067] In order to ensure that the hybrid control model maintains strong robustness to important parameters such as feed rate and temperature, a regularization term is added to prevent the hybrid control model from overfitting the noise data. The loss function after regularization is: , Where M is the total number of all weight parameters in the hybrid control model, Represents the i-th weight parameter.

[0068] During the training process, the addition of a regularized loss function affects the model's gradient calculations, thereby automatically adjusting weights during back propagation. Regularization reduces the value of high-weight parameters by imposing smoothing constraints, making the model more stable and insensitive to noise. For multivariable PID control problems, the regularized loss function can improve the generalization and robustness of the model, making it more reliable in dynamic feed control and environmental changes.

[0069] In joint training, the gradient between the LSTM and BP neural networks is updated through the back propagation algorithm. The gradient of the loss function is passed layer by layer, from the BP layer back to the LSTM layer. The specific gradient calculation steps are as follows: 1. Calculate the gradient of the output layer of the BP neural network. In the output layer of the BP network, calculate the partial derivative of the loss function with respect to the PID parameters. Assuming that the output layer is a fully connected layer, the gradient calculation formula is: , in, It represents the gradient of the loss function to the kth output neuron of the BP neural network. It is the partial derivative of the error between the current network output and the target value, and represents the key value used to update the network parameters during the back propagation process. By calculating this gradient, the model can adjust the weights and biases of the output layer to gradually reduce the loss function, thereby optimizing the output. Denotes the loss function for the k-th output The partial derivative of indicates the contribution of the k-th output to the overall loss. It reflects the gap between the current output and the expected value and is used to guide the gradient update in back propagation. This item is used to adjust the output Related parameters make the model output gradually approach the target value .

[0070] 2. Pass the output layer gradient to the hidden layer of the BP network and calculate the gradient of each neuron in the hidden layer.

[0071] 3. Pass the gradient to the LSTM layer. The output feature vector of the LSTM layer is is the input of the BP network, so the loss function is The partial derivative of is: , in, Represents the feature vector output from LSTM To the weight of the BP network output layer.

[0072] After obtaining the partial derivative of the loss with respect to the LSTM output, continue to backpropagate the gradient through the various gates of the LSTM structure (input gate, forget gate, output gate).

[0073] The gradient calculation process of LSTM is as follows: 1) Calculate the gradient of the forget gate. The calculation formula is: ,in, Represents the state of the memory unit at the previous time step. The task of the forget gate is to determine the memory state of the previous time step. How much information is passed to the current time step, so It will directly affect the gradient of the forget gate; Represents the output value of the forget gate. This is the retention ratio (value between 0 and 1) calculated by the forget gate based on the input data in the current time step, indicating the memory of the previous time step How much to keep; The derivative of the Sigmoid activation function output by the forget gate. The Sigmoid function outputs a value between 0 and 1. The derivative part determines the rate of change of the forget gate output, which is very important for the smoothness and stability of the gradient transfer. is the partial derivative of the loss function with respect to the LSTM memory unit.

[0074] 2) Calculate the input gate gradient, the calculation formula is: ,in, Represents the candidate memory cell state, which is determined by the input data of the current time step and the hidden state of the previous time step. The input gate controls how much new information is Added to the memory unit, so the candidate memory state It will affect the gradient of the input gate; Represents the output value of the input gate. This is the ratio calculated by the input gate, with a value between 0 and 1, indicating how much of the current input is added to the memory unit; It is the derivative of the Sigmoid activation function output by the input gate. This derivative represents the sensitivity of the input gate to changes in input data and is used to smooth the increment of input data. is the partial derivative of the loss function with respect to the LSTM memory unit.

[0075] 3) Calculate the output gate gradient, the calculation formula is: ,in, Indicates the state of the memory unit go through The output of the activation function. The output gate determines how much information of the memory unit is passed to the hidden state, so the current memory state The activation value of will affect the gradient of the output gate; Represents the output value of the output gate, which is between 0 and 1. The output gate controls how much information flows from the memory cell to the hidden state, so Will directly affect the gradient; : This is the derivative of the Sigmoid activation function of the output gate, which is used to smooth the increment of the output gate output and ensure the stability of the model when the parameters are updated. is the partial derivative of the loss function with respect to the hidden state.

[0076] After the hybrid control model is trained, the BP neural network learns the nonlinear relationship between the material feed ratio and process parameters under different working conditions. After training, the BP neural network will be able to continuously adjust its weights and biases according to real-time data to optimize the PID controller parameters. Specifically, after the hybrid control model is deployed on the cadmium oxide production line, when the cadmium oxide production line starts running, the operator first determines the target value according to production requirements, such as the material feed ratio, reaction temperature, pressure, etc., and sets the initial PID parameters according to standard process conditions. During operation, the LSTM network extracts the time series characteristics of the sensor data, and the BP neural network continuously receives the time series characteristics output by the LSTM network. The BP neural network conducts online learning through these data, identifies the nonlinear relationship of the system under different working conditions, and outputs the optimized PID parameters to achieve adaptive dynamic control.

[0077] like Figure 3 As shown, an embodiment of the present invention discloses an electronic device, including a memory 401 storing executable program codes and a processor 402 coupled to the memory 401; The processor 402 calls the executable program code stored in the memory 401 to execute the material feeding control method for the cadmium oxide production line described in the above embodiments.

[0078] An embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the material feeding control method for the cadmium oxide production line described in the above embodiments.

[0079] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to make the public understand the disclosed content of the present invention more thoroughly and comprehensively, and it does not limit the scope of protection of the present invention.

[0080] The above embodiments are not exhaustive enumerations of the present invention, and there may be multiple other implementations not listed. Any replacement and improvement made without violating the concept of the present invention shall fall within the protection scope of the present invention.

Claims

1. A material feeding control method for a cadmium oxide production line, characterized in that: include: Collecting sensor data corresponding to various indicators in the production process, wherein the sensor data is first sensor data collected at multiple time scales or second sensor data collected at a single time scale; fusing the first sensor data to obtain a multi-scale fusion vector of each indicator collected at multiple time scales; Inputting all the multi-scale fusion vectors and the second sensor data into the LSTM network to obtain the time series features of the LSTM network output; Input the time series characteristics into the BP neural network to obtain the optimized PID parameters output by the BP neural network; The optimized PID parameters are input into a PID controller to optimize and adjust the feed ratio of the material.

2. The material feed control method for the cadmium oxide production line according to claim 1, characterized in that: After collecting sensor data corresponding to various indicators in the production process, it also includes: Calculating the correlation between the sensor data and the control target, and adjusting the sensor acquisition frequency and / or the weight corresponding to the sensor data according to the correlation; and / or, The covariance between the sensor data is calculated, and data dimension reduction is performed according to the covariance and / or the weight corresponding to the sensor data is adjusted.

3. The material feed control method for the cadmium oxide production line according to claim 1, characterized in that: After collecting sensor data corresponding to various indicators in the production process, it also includes: Eliminating noise and outliers in the sensor data to obtain processed sensor data; The processed sensor data is input into a PID controller to make preliminary adjustments to the feed ratio of the material.

4. The material feed control method for the cadmium oxide production line according to claim 1, characterized in that: After collecting sensor data corresponding to various indicators in the production process, the sensor data is sampled multiple times using different time scales, and the first sensor data is generated according to the sensor data.

5. The material feed control method for the cadmium oxide production line according to claim 1, characterized in that: The time series features include features extracted from sensor data corresponding to the feed rate of cadmium metal, oxygen, and catalyst, and features extracted from sensor data corresponding to reactor temperature, reactor pressure, and oxygen concentration; the optimized PID parameters include proportional parameters, integral parameters, and differential parameters corresponding to cadmium metal, oxygen, and catalyst, respectively.

6. The material feed control method for a cadmium oxide production line according to claim 1, characterized in that: Inputting all the multi-scale fusion vectors and the second sensor data into the LSTM network to obtain the time series features output by the LSTM network, including: In the LSTM network, a forget gate is used to forget the interference of sudden fluctuations on control parameters in the stable production stage, an input gate is used to filter the data of the memory unit added to the LSTM network in the multi-scale fusion vector and the second sensor data, and an output gate is used to output key features of multiple time steps according to the memory unit of the LSTM network to form the time series features.

7. A material feeding control system for a cadmium oxide production line, characterized in that: include: A perception layer, used to collect sensor data corresponding to multiple indicators in the production process, wherein the sensor data is first sensor data collected at multiple time scales or second sensor data collected at a single time scale; A processing layer, used for fusing the first sensor data to obtain a multi-scale fusion vector of each indicator collected at multiple time scales; The intelligent optimization layer inputs all the multi-scale fusion vectors and the second sensor data into the LSTM network to obtain the time series features of the LSTM network output; Input the time series characteristics into the BP neural network, obtain the optimized PID parameters output by the BP neural network, and input the optimized PID parameters into the control layer; The control layer is used to optimize and adjust the feed ratio of the material according to the optimized PID parameters; The execution layer is used to control valves, pumps and material conveying devices to adjust the material feed ratio according to the control instructions of the control layer.

8. The material feed control system of the cadmium oxide production line according to claim 7, characterized in that: The processing layer is further used to calculate the correlation between the sensor data and the control target, and adjust the sensor acquisition frequency and / or the weight corresponding to the sensor data according to the correlation; and / or, The covariance between the sensor data is calculated, and data dimension reduction is performed according to the covariance and / or the weight corresponding to the sensor data is adjusted.

9. An electronic device, characterized in that: It comprises a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the material feed control method for the cadmium oxide production line according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the material feed control method for a cadmium oxide production line according to any one of claims 1 to 6.

Images