Washing composition production regulation and control method and system based on multi-parameter intelligent analysis

Abstract

The invention provides a washing composition production regulation and control method and system based on multi-parameter intelligent analysis, and relates to the technical field of production control, and the method comprises the steps: collecting process parameters to form a data flow, inputting the data into a cascade self-coding deep learning network to construct a dynamic association map, grouping parameters based on the map, and calculating collaborative change characteristics, a layered progressive intelligent controller is constructed to generate an optimization regulation and control strategy, adaptive parameter adjustment and closed-loop optimization control are realized, the problem that a traditional regulation and control method is difficult to cope with multi-parameter collaborative change is solved, and the production process stability and the product quality consistency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production control, and particularly to a production regulation method and system for washing compositions based on multi-parameter intelligent analysis. Background Art

[0002] Washing compositions are products widely used in modern industry and daily life. Their production processes involve complex technological links such as the proportioning, mixing, and reaction of various raw materials. With the improvement of consumers' requirements for product quality and the increasingly strict environmental protection standards, the production regulation of washing compositions faces higher technical challenges. Traditional production of washing compositions mainly relies on fixed process parameters and empirical regulation. With the development of automation technology and intelligent analysis methods, data-driven intelligent regulation methods have gradually been applied to the production process of washing compositions. Research in the field of production regulation of washing compositions mainly focuses on parameter monitoring, data analysis, and control algorithms. Existing data analysis methods can already perform real-time monitoring and simple adjustment of single process parameters. However, the existing technology still has deficiencies in the analysis of the multi-parameter correlation in the production process of washing compositions, being unable to effectively capture the dynamic correlation between parameters in complex processes, lacking an effective response mechanism to external factors such as environmental changes and raw material differences, and having insufficient prediction accuracy for the change trend of process parameters, thus being unable to provide accurate decision-making basis for production regulation, etc. Therefore, there is an urgent need for a solution to solve the problems existing in the prior art. Summary of the Invention

[0003] Embodiments of the present invention provide a production regulation method and system for washing compositions based on multi-parameter intelligent analysis, which can at least solve some of the problems existing in the prior art.

[0004] In the first aspect of the embodiments of the present invention, a production regulation method for washing compositions based on multi-parameter intelligent analysis is provided, including: Collect process parameters in the production process of washing compositions, generate a time-series data stream of process parameters and perform data preprocessing to obtain standard process parameter data; Input the standard process parameter data into a cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fuse feature information at different scales in combination with a residual connection structure, and perform priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; Group the process parameters based on the dynamic association map, generate process parameter groups and obtain the parameter change trend through online monitoring, and calculate the co-variation characteristics in combination with the dynamic association map; Construct a hierarchical and progressive intelligent controller including a global planning layer and a local execution layer, optimize the current control strategy in real time by combining the dynamic programming algorithm, adaptively update the control parameters according to the changes of the dynamic association graph, establish the constraint conditions for parameter adjustment based on the co-variation characteristics, and apply the constraint conditions to obtain the optimized control strategy; Convert the optimized control strategy into control instructions to adjust the process parameters, collect the adjusted process parameters for online learning, continuously optimize the dynamic association graph, and form a closed-loop optimization control.

[0005] In an alternative embodiment, Collect the process parameters in the production process of the washing composition, generate the time-series data stream of the process parameters and perform data preprocessing to obtain the standard process parameter data including: Collect the process parameters in the production process of the washing composition, where the process parameters include the reaction kettle temperature, reaction kettle pressure, pH value of the reaction solution, stirring speed, reaction kettle liquid level, and raw material addition rate, and record them at preset time intervals to form the time-series data stream of the process parameters; Perform data preprocessing on the time-series data stream of the process parameters, identify and remove abnormal data by setting the parameter fluctuation threshold, fill in the missing data using the interpolation algorithm, and verify the validity of the data based on the physical constraint relationship of the process parameters to generate the preprocessed process parameter data; Perform standardization processing on the preprocessed process parameter data, convert the process parameters with different dimensions into a unified numerical range to obtain the standard process parameter data.

[0006] In an alternative embodiment, Input the standard process parameter data into the cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through the multi-scale feature extraction layer, fuse the feature information at different scales by combining the residual connection structure, and perform priority sorting on the features through the self-attention mechanism to obtain the multi-dimensional feature representation and construct the dynamic association graph of the process parameters including: Construct a cascaded autoencoder deep learning network using a symmetric encoder-decoder structure, input the standard process parameter data into the cascaded autoencoder deep learning network, and generate the encoded feature matrix by sequentially performing convolution operation, batch normalization processing, and activation function operation through multiple cascaded encoding units; Extract features from the encoded feature matrix through multiple convolutional kernels with different sizes, use the convolutional kernel with the smallest size to extract the reference feature matrix, and use other convolutional kernels to extract the time-series feature matrix group; Perform feature mapping transformation on each time series feature matrix in the time series feature matrix group to obtain a transformed feature matrix group, and perform a residual superposition operation on the transformed feature matrices in the transformed feature matrix group and the reference feature matrix to generate a fused feature matrix; Map the fused feature matrix into a query matrix, a key matrix, and a value matrix, calculate the product of the query matrix and the transpose of the key matrix to obtain an attention score matrix, and multiply the attention score matrix by the value matrix after performing normalization processing to obtain an attention feature matrix; Extract the feature vectors corresponding to the process parameters in the attention feature matrix, calculate the ratio of the inner product of any two feature vectors to the product of the modulus lengths to obtain the correlation strength between the process parameters, set the process parameters as vertices, the correlation relationship between the process parameters as edges, and the correlation strength as the edge weight to construct a dynamic correlation graph of the process parameters.

[0007] In an alternative embodiment, Extracting the feature vectors corresponding to the process parameters in the attention feature matrix and calculating the ratio of the inner product of any two feature vectors to the product of the modulus lengths to obtain the correlation strength between the process parameters includes: Extract the feature vectors corresponding to the process parameters from the attention feature matrix, and input the feature vectors corresponding to the process parameters into a neural differential dynamics model. The neural differential dynamics model maps the feature vectors corresponding to the process parameters into state vectors through a differential equation, and the differential equation consists of the time derivative of the state vector and a neural network; Use an adaptive Runge-Kutta integrator to adjust the integration step according to the rate of change of the state vector to solve the differential equation and obtain the time evolution characteristics of the feature vectors corresponding to the process parameters; Construct a group of nonlinear oscillators according to the time evolution characteristics. Each process parameter corresponds to a nonlinear oscillator. The initial state of the nonlinear oscillator is set according to the feature vector corresponding to the process parameter. The frequency of the nonlinear oscillator characterizes the change characteristics of the process parameter, and the amplitude of the nonlinear oscillator characterizes the importance of the process parameter; Establish a coupling network between the nonlinear oscillators, connect the nonlinear oscillators through a nonlinear coupling equation and analyze the synchronization state, and calculate the correlation strength between any two process parameters according to the correlation strength calculation formula. The correlation strength calculation formula combines the synchronization state of the nonlinear oscillator, the previously obtained state trajectory similarity, and the ratio of the inner product of any two feature vectors to the product of the modulus lengths through a weight factor.

[0008] In an alternative embodiment, Group the process parameters based on the dynamic association graph, generate process parameter groups, obtain the parameter change trends through online monitoring, and calculate the co-variation characteristics in combination with the dynamic association graph, including: Determine the association threshold according to the connection relationship of the process parameters in the dynamic association graph and the distribution characteristics of the pre-calculated association strength, and divide the process parameters with an association strength greater than the association threshold into the same process parameter group; Collect the real-time data of each process parameter in the process parameter group through an online sensor, process the real-time data through a sliding time window, and extract the time series change trend characteristics of the process parameters; Map the time series change trend characteristics to the process parameter nodes corresponding to the dynamic association graph, and calculate the change consistency of each parameter in the process parameter group based on the connection relationship in the dynamic association graph to obtain the co-variation characteristics of the process parameter group.

[0009] In an alternative embodiment, Construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, optimize the current control strategy in real time in combination with the dynamic programming algorithm, adaptively update the control parameters according to the change of the dynamic association graph, establish the constraint conditions for parameter adjustment based on the co-variation characteristics, and apply the constraint conditions to obtain the optimized control strategy, including: Obtain the historical data of the process parameters, construct the system state vector and the control input vector according to the historical data, and construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer. The global planning layer determines the global objective function including the system state deviation term, the control input constraint term, and the system energy consumption term based on the system state vector and the control input vector, and the local execution layer constructs a parameter constraint matrix based on the process parameter connection relationship of the dynamic association graph; Establish a system discrete state equation according to the system state vector, construct a value function with a discount factor based on the system discrete state equation and the global objective function, obtain the feedback gain matrix through iterative optimization of the value function, and construct the first control strategy according to the feedback gain matrix; Adaptively update the control parameters according to the change of the dynamic association graph, calculate the update amount of the control gain matrix, which is related to the gradient of the global objective function, the learning rate, and the parameter constraint matrix, adaptively adjust the control parameters in the first control strategy according to the update amount to obtain the second control strategy, and establish the constraint conditions for parameter adjustment based on the co-variation characteristics; Predict the system state sequence in the future time domain according to the second control strategy and the system discrete state equation, and apply the constraint conditions to obtain the optimized control strategy.

[0010] In an alternative embodiment, Based on the system state vector, a system discrete state equation is established. Based on the system discrete state equation and the global objective function, a value function with a discount factor is constructed. By iteratively optimizing the value function, a feedback gain matrix is obtained. Constructing a first regulation strategy based on the feedback gain matrix includes: Obtain the system state vector, map the system state vector through an encoder network to obtain a hidden state vector, reconstruct the system state vector through a decoder network with the hidden state vector and calculate the reconstruction loss; Construct a causal graph structure based on historical data and calculate the mutual information and information entropy between nodes. Calculate the causal strength weight according to the mutual information and the information entropy. Based on the hidden state vector, establish a system discrete state equation. Calculate a causal correction term according to the causal strength weight and combine it with the system discrete state equation to obtain a causally enhanced system discrete state equation; Based on the system discrete state equation and the global objective function, construct an explicit value function and an implicit value function. Calculate an adaptive weight factor according to the state prediction error. Use the adaptive weight factor to perform a weighted combination of the explicit value function and the implicit value function to obtain a value function with a discount factor; Use the Bellman equation and the temporal difference algorithm to update the explicit value function and the implicit value function respectively. By iteratively optimizing the value function with the discount factor, a feedback gain matrix is obtained. Calculate the causal effect of the control input and correct the feedback gain matrix according to the causal effect. Combine the corrected feedback gain matrix with the causal correction term to construct a first regulation strategy.

[0011] In a second aspect of the embodiments of the present invention, a production regulation system for a washing composition based on multi-parameter intelligent analysis is provided, including: A first unit for collecting process parameters in the production process of the washing composition, generating a time-series data stream of process parameters and performing data preprocessing to obtain standard process parameter data; A second unit for inputting the standard process parameter data into a cascaded autoencoder deep learning network, performing hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fusing feature information at different scales by combining a residual connection structure, and performing priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; A third unit for grouping process parameters based on the dynamic association map, generating process parameter groups and obtaining parameter change trends through online monitoring, and calculating co-varying features by combining the dynamic association map; The fourth unit is used to construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, optimize the current control strategy in real time in combination with the dynamic programming algorithm, adaptively update the control parameters according to the changes of the dynamic association map, establish constraint conditions for parameter adjustment based on the co-variation characteristics, and apply the constraint conditions to obtain an optimized control strategy; The fifth unit is used to convert the optimized control strategy into control instructions to adjust process parameters, collect the adjusted process parameters for online learning, continuously optimize the dynamic association map, and form a closed-loop optimization control.

[0012] In the present invention, by adopting a cascaded autoencoder deep learning network combined with a multi-scale feature extraction layer and a residual connection structure, it is possible to adaptively extract and analyze the features of process parameters, accurately capture the complex dynamic association relationships between parameters, significantly improve the prediction accuracy of parameter changes in the production process of washing compositions, make production control more accurate and effective, combine global planning with local execution, realize real-time optimization of control strategies through the dynamic programming algorithm, and can adaptively update control parameters according to the dynamic association map of process parameters, solve the problems of response lag and insufficient control of traditional control methods in the face of multi-parameter co-variation, improve the stability and consistency of the production process, continuously accumulate experience and optimize control strategies through continuous online learning and dynamic association map update, adapt to changes in production conditions and fluctuations in raw material characteristics, reduce production energy consumption and material consumption, improve the stability of product quality, and realize the intelligentization and greening of the production process of washing compositions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic flow chart of the production control method of washing compositions based on multi-parameter intelligent analysis according to an embodiment of the present invention; Figure 2 It is an analysis diagram of the phase space trajectory of a non-linear oscillator of the production control method of washing compositions based on multi-parameter intelligent analysis according to an embodiment of the present invention; Figure 3 It is a comparison diagram of the adaptive ability of the dynamic association map of the production control method of washing compositions based on multi-parameter intelligent analysis according to an embodiment of the present invention; Figure 4 It is a heat map of the causal intensity weights between key process parameters of the production control method of washing compositions based on multi-parameter intelligent analysis according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0016] Figure 1 It is a schematic flowchart of a method for regulating the production of a washing composition based on multi-parameter intelligent analysis according to an embodiment of the present invention. As Figure 1 shown, the method includes: Collect process parameters during the production of the washing composition, generate a time-series data stream of process parameters and perform data preprocessing to obtain standard process parameter data; Input the standard process parameter data into a cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fuse feature information at different scales by combining a residual connection structure, and perform priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; Group the process parameters based on the dynamic association map, generate process parameter groups and obtain the parameter change trend through online monitoring, and calculate the co-variation features in combination with the dynamic association map; Construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, optimize the current regulation strategy in real time by combining a dynamic programming algorithm, adaptively update the control parameters according to the changes in the dynamic association map, establish constraint conditions for parameter adjustment based on the co-variation features, and apply the constraint conditions to obtain an optimized regulation strategy; Convert the optimized regulation strategy into a control instruction to adjust the process parameters, collect the adjusted process parameters for online learning, continuously optimize the dynamic association map, and form a closed-loop optimization control.

[0017] In an alternative embodiment, Collecting process parameters during the production of the washing composition, generating a time-series data stream of process parameters and performing data preprocessing to obtain standard process parameter data includes: Collecting process parameters in the production process of the detergent composition, the process parameters including reactor temperature, reactor pressure, reaction liquid pH value, stirring speed, reactor liquid level, and raw material addition rate, and recording them at preset time intervals to form a process parameter time series data stream; Performing data preprocessing on the process parameter time series data stream, identifying and removing abnormal data by setting parameter fluctuation thresholds, filling missing data with an interpolation algorithm, and verifying the validity of the data based on the physical constraint relationship of the process parameters to generate preprocessed process parameter data; The pre-processed process parameter data is standardized, and process parameters of different dimensions are converted into a unified numerical range to obtain standard process parameter data.

[0018] The distributed data collector collects process parameters such as reactor temperature, reactor pressure, reaction liquid pH value, stirring speed, reactor liquid level and raw material addition rate during the production process of the detergent composition. The data is collected continuously at a frequency of once every 10 seconds according to the preset time interval, and the collected data is sorted in the order of timestamps to form a process parameter time series data stream.

[0019] Abnormal data is identified and processed for the collected process parameter time series data stream. By setting the allowable fluctuation range for each process parameter, the data exceeding the fluctuation threshold is marked as abnormal data. For temperature parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 5°C; for pressure parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 0.2MPa; for pH parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 0.5; for stirring speed parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 50rpm; for liquid level parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 10%; for raw material addition rate parameters, the fluctuation threshold is set to the absolute value of the difference between two adjacent sampling values not exceeding 20%. The identified abnormal data is removed from the data stream.

[0020] Fill the missing data in the process parameter time series data stream after removing abnormal data. Use the cubic spline interpolation algorithm to interpolate the valid data before and after the missing data point to obtain the filling value that conforms to the parameter change trend. During the interpolation calculation, select three valid data points before and after the missing point as the interpolation reference points, and obtain the parameter estimation value of the missing point position by constructing a cubic spline curve.

[0021] Verify the validity of the filled data based on the physical constraint relationships between process parameters. There is a positive correlation between temperature and pressure, a corresponding relationship between pH value and raw material addition rate, and an integral relationship between liquid level and raw material addition rate. By verifying whether the physical constraint relationships are satisfied, ensure the rationality of the filled data and generate the preprocessed process parameter data.

[0022] Perform standardization processing on the preprocessed process parameter data to uniformly convert parameters with different dimensions into the numerical range of 0 - 1. Adopt the maximum - minimum normalization method, using the historical extreme values of the process parameters as the normalization benchmark. Through standardization processing, eliminate the dimensional differences between different process parameters and obtain the standard process parameter data.

[0023] Exemplarily, during the production process of a certain batch of washing composition, it is collected that the reactor temperature fluctuates between 75 - 85 °C, the pressure fluctuates between 0.15 - 0.25 MPa, the pH value fluctuates between 8.5 - 9.5, the stirring speed fluctuates between 400 - 600 rpm, the liquid level fluctuates between 55% - 75%, and the raw material addition rate fluctuates between 2 - 4 kg / min. After abnormal data identification, it is found that the temperature data has abnormal values of 92 °C at the 127th sampling and 69 °C at the 245th sampling; the pressure data has an abnormal value of 0.35 MPa at the 156th sampling. After removing these abnormal data points, there are data missing at the 127th, 156th, and 245th sampling positions respectively. Adopt the cubic spline interpolation algorithm to interpolate based on the data before and after the missing points, and obtain the estimated temperature value at the 127th sampling as 83 °C, the estimated pressure value at the 156th sampling as 0.22 MPa, and the estimated temperature value at the 245th sampling as 77 °C. Through the verification of the physical constraint relationship, confirm that the correlation between the interpolation result and other parameters conforms to the physical law. Perform standardization processing on all parameters to obtain the standard process parameter data within a unified range.

[0024] In this embodiment, the method of distributed data acquisition and recording at fixed time intervals is adopted, which ensures the continuity and integrity of the process parameter acquisition process, provides a high - quality raw data basis for subsequent data processing, and performs abnormal identification based on the differential fluctuation threshold set according to the physical characteristics of the process parameters, avoiding misjudgment that may be caused by using a fixed threshold. The setting of the differential fluctuation threshold fully considers the dynamic characteristics and variation laws of different process parameters, improves the accuracy and reliability of abnormal data identification, and uses the cubic spline interpolation algorithm to fill the missing data. Compared with simple linear interpolation, it can better reflect the dynamic change trend of the parameters.

[0025] In an alternative embodiment, Input the standard process parameter data into a cascaded auto - encoding deep learning network. Perform hierarchical feature extraction on the standard process parameter data through a multi - scale feature extraction layer, fuse the feature information at different scales by combining with a residual connection structure, and perform priority sorting on the features through a self - attention mechanism to obtain a multi - dimensional feature representation and construct a dynamic association graph of process parameters, including: Construct a cascaded auto - encoding deep learning network using a symmetric encoder - decoder structure. Input the standard process parameter data into the cascaded auto - encoding deep learning network, and generate an encoded feature matrix by successively performing convolution operations, batch normalization processing, and activation function operations through multiple cascaded encoding units; Extract features from the encoded feature matrix through multiple convolutional kernels of different sizes. Use the convolutional kernel with the smallest size to extract a reference feature matrix, and use convolutional kernels of other sizes to extract a group of temporal feature matrices; Perform feature mapping transformation on each temporal feature matrix in the group of temporal feature matrices to obtain a group of transformed feature matrices, and perform residual superposition operation on the transformed feature matrices in the group of transformed feature matrices and the reference feature matrix to generate a fused feature matrix; Map the fused feature matrix into a query matrix, a key matrix, and a value matrix. Calculate the product of the query matrix and the transpose of the key matrix to obtain an attention score matrix. After performing normalization processing on the attention score matrix, multiply it with the value matrix to obtain an attention feature matrix; Extract the feature vectors corresponding to the process parameters in the attention feature matrix, calculate the ratio of the inner product of any two feature vectors to the product of their norms to obtain the association strength between process parameters. Set the process parameters as vertices, the association relationship between process parameters as edges, and the association strength as edge weights to construct a dynamic association graph of process parameters.

[0026] Construct a cascaded auto - encoding deep learning network structure, which includes three layers of encoding units and corresponding three layers of decoding units. Each encoding unit consists of a convolutional layer, a batch normalization layer, and an activation function layer. In the first - layer encoding unit, perform feature extraction on the input process parameter data using a convolutional operation with a stride of 1, and set the number of convolutional kernels to 64. After eliminating data distribution deviation through the batch normalization layer, introduce non - linear characteristics through the activation function. The second - layer encoding unit reduces the number of convolutional kernels to 32, and the third - layer encoding unit further reduces it to 16, gradually reducing the feature dimension and extracting more abstract feature representations.

[0027] Multi-scale feature extraction is performed on the encoding feature matrix, and convolution kernels of sizes 3×3, 5×5, and 7×7 are selected respectively. The 3×3 convolution kernel is used to extract local fine features to obtain a benchmark feature matrix that characterizes short-term correlation; the 5×5 convolution kernel is used to extract medium-time scale features to capture the changing rules of process parameters over a longer period of time; the 7×7 convolution kernel is used to extract long-term correlation features to obtain the evolution pattern of process parameters over a longer time span. The feature matrices extracted by the 5×5 and 7×7 convolution kernels are combined into a temporal feature matrix group.

[0028] For each feature matrix in the time series feature matrix group, a learnable feature mapping layer is used to convert it to the same feature space as the benchmark feature matrix. The feature mapping process is implemented using a fully connected layer, and the output dimension of the mapping layer is consistent with the benchmark feature matrix. The residual superposition operation of element-by-element addition is performed on the mapped feature matrix and the benchmark feature matrix to fuse the feature information on multiple time scales into a unified feature representation.

[0029] The fused feature matrix is mapped into a query matrix, a key matrix, and a value matrix through three fully connected layers with the same structure but independent parameters. The query matrix represents the feature query information at the current moment, the key matrix stores the feature index information at each moment, and the value matrix contains the actual feature content. The matrix product of the transposed query matrix and the key matrix is calculated to obtain the attention score matrix that describes the degree of correlation of process parameters at different moments. The attention score matrix is normalized row by row so that the sum of the scores of each row is 1, and then multiplied with the value matrix to obtain the attention feature matrix.

[0030] Extract the feature vector corresponding to each process parameter from the attention feature matrix, and the vector dimension is the same as the number of hidden units in the attention layer. For any two process parameters, calculate the inner product of their corresponding feature vectors and divide it by the product of the module lengths of the two vectors to obtain the correlation strength coefficient ranging from -1 to 1. A positive correlation strength indicates a positive correlation, and a negative correlation indicates a negative correlation. The larger the absolute value, the stronger the correlation. Finally, a graph structure is constructed with process parameters as vertices, and edges are connected between parameter pairs with correlation strengths greater than the preset threshold. The weight of the edge is set to the corresponding correlation strength to form a complete dynamic correlation map.

[0031] Exemplarily, during the production process of a certain batch of washing compositions, the standardized data of six process parameters are input into a cascaded autoencoder network. Through three layers of encoding units, the first layer uses 64 3×3 convolutional kernels, the second layer uses 32 5×5 convolutional kernels, and the third layer uses 16 7×7 convolutional kernels to extract features respectively. Taking the feature matrix extracted by the 3×3 convolutional kernel as a benchmark, the other two feature matrices are mapped through a fully connected layer with a dimension of 256 and then superimposed with the benchmark feature matrix. After calculation by the attention mechanism, a feature vector with a dimension of 128 is extracted, and the correlation strength between process parameters is calculated: the correlation strength between the reactor temperature and pressure is 0.85, showing a strong positive correlation; the correlation strength between temperature and pH value is 0.72, showing a significant positive correlation; the correlation strength between pressure and stirring speed is 0.65, showing a moderate positive correlation; the correlation strength between pH value and raw material addition rate is 0.78, showing a strong positive correlation; while the correlation strength between the liquid level and other parameters is generally lower than 0.5, indicating that its correlation is relatively weak. Setting the correlation strength threshold to 0.6, all edges with a correlation strength greater than 0.6 are retained in the constructed dynamic correlation map, intuitively showing the main influence relationships between process parameters.

[0032] In this embodiment, through the layer-by-layer processing of the multi-layer encoding unit, the deep extraction of process parameter features is realized. The combination of convolutional operation, batch normalization processing and activation function in each layer of the encoding unit effectively improves the accuracy and robustness of feature extraction, avoiding the problem that traditional single-layer feature extraction methods are prone to losing important information. By introducing the attention mechanism, the adaptive capture of the dynamic correlation relationship between process parameters is realized. The method of calculating the correlation strength based on the inner product and modulus ratio of feature vectors can not only reflect the direction of the correlation between parameters, but also accurately quantify the degree of correlation, providing an accurate numerical basis for process parameter optimization.

[0033] In an alternative embodiment, extracting the feature vectors corresponding to the process parameters in the attention feature matrix and calculating the ratio of the inner product to the modulus product of any two feature vectors to obtain the correlation strength between process parameters includes: extracting the feature vectors corresponding to the process parameters from the attention feature matrix and inputting the feature vectors corresponding to the process parameters into a neural differential dynamics model, which maps the feature vectors corresponding to the process parameters into state vectors through a differential equation composed of the time derivative of the state vector and a neural network; using an adaptive Runge-Kutta integrator to adjust the integration step according to the change rate of the state vector to solve the differential equation, and obtaining the time evolution characteristics of the feature vectors corresponding to the process parameters; Construct a set of nonlinear oscillators according to the described time-evolution characteristics. Each process parameter corresponds to a nonlinear oscillator. The initial state of the nonlinear oscillator is set according to the eigenvector corresponding to the process parameter. The frequency of the nonlinear oscillator characterizes the variation characteristics of the process parameter, and the amplitude of the nonlinear oscillator characterizes the importance degree of the process parameter. Establish a coupling network among the nonlinear oscillators, connect the nonlinear oscillators through a nonlinear coupling equation and analyze the synchronization state. Calculate the correlation strength between any two process parameters according to the correlation strength calculation formula. The correlation strength calculation formula combines the synchronization state of the nonlinear oscillator, the pre-obtained similarity degree of the state trajectories, and the ratio of the inner product to the product of the norms of any two eigenvectors through a weight factor.

[0034] Separate the eigenvector corresponding to each process parameter from the attention feature matrix through a feature extraction operation. The eigenvector is extracted row by row from the matrix through a slicing operation, and the dimension of each vector is the same as the number of hidden units in the attention layer. Construct a neural differential dynamics model, which consists of a three-layer fully connected neural network. The number of neurons in the input layer is the same as the dimension of the eigenvector. The hidden layer uses a nonlinear activation function, and the output layer generates the rate of change of the state vector. The weight matrix of the neural network is optimized through the backpropagation algorithm, and the loss function includes a state prediction error term and a regularization term.

[0035] Implement an adaptive Runge-Kutta integrator for solving differential equations. The integrator uses the fourth-order Runge-Kutta formula as the basic algorithm and calculates four intermediate state values at each time step. Dynamically adjust the size of the integration step by estimating the local truncation error, that is, the difference between the numerical solutions at two adjacent integration step sizes. When the truncation error exceeds the preset threshold, halve the current step size; when the truncation error is much smaller than the threshold, appropriately increase the step size. Record the change trajectory of the state vector over time during the integration process, and set the sampling interval according to actual needs to ensure capturing the key features of the state evolution.

[0036] Based on the obtained time-evolution characteristics, construct a corresponding nonlinear oscillator for each process parameter. The oscillator uses an improved Van der Pol model, which includes a nonlinear damping term and a driving term. When initializing the oscillator, map the eigenvector through dimensionality reduction to obtain the initial position and velocity of the oscillator. The natural frequency of the oscillator is determined through Fourier analysis, and the main frequency components in the time-evolution characteristics are selected. The oscillation amplitude is comprehensively determined according to the norm of the eigenvector and the importance weight of the process parameter in the production process.

[0037] Implement a coupling mechanism in the oscillator network and use a non - linear function to describe the interaction between oscillators. When initializing the coupling strength matrix, set the initial value according to the degree of parameter correlation calculated previously. A stronger correlation corresponds to a larger coupling strength. Obtain the dynamic evolution of the oscillator network by solving the coupled differential equations. Calculate the phase synchronization index in real - time. The phase synchronization index is based on the statistical distribution characteristics of the phase difference and is used to quantify the synchronization degree between oscillators.

[0038] Calculate the correlation strength between process parameters through multi - index fusion. Calculate the synchronization characteristics of oscillators, including the phase synchronization index and the frequency entanglement degree. For the phase synchronization index, statistically analyze the degree of phase locking within a fixed - time window; for the frequency entanglement degree, analyze the peak characteristics of the cross - correlation function. Secondly, evaluate the similarity degree of state trajectories and calculate the normalized distance between trajectories using the dynamic time warping algorithm. Segment the trajectories by time window, find the optimal time alignment method within each window, and cumulatively calculate the distance metric after alignment. Calculate the cosine similarity of eigenvectors, which is the ratio of the inner product of vectors to the product of their norms. Combine these three types of indicators through adaptive weights, and the weight coefficients are dynamically adjusted according to the verification effect in historical data.

[0039] Exemplarily, during the production process of a certain batch of washing compositions, extract 128 - dimensional eigenvectors corresponding to six process parameters from the attention feature matrix. Input the eigenvectors into a three - layer neural differential dynamics model with the number of neurons in the hidden layers being 256, 128, and 64 respectively, and output a 32 - dimensional state vector. Set the initial step size of the integrator to 0.01, the truncation error threshold to 0.0001, and the actual step size after adaptive adjustment fluctuates between 0.005 and 0.02. Calculate the state evolution sequence at 2000 time points in total.

[0040] Construct six non - linear oscillators. Among them, the oscillator corresponding to the temperature parameter has an initial amplitude of 1.2, a main frequency of 0.52 Hz, and a sub - harmonic frequency of 0.26 Hz; the oscillator corresponding to the pressure parameter has an initial amplitude of 1.1 and a main frequency of 0.48 Hz; the oscillator corresponding to the pH value parameter has an initial amplitude of 0.9 and a main frequency of 0.35 Hz. Analyze the synchronization behavior within a 100 - second time window. The phase synchronization index between temperature and pressure is 0.95, and the frequency entanglement degree is 0.88; the phase synchronization index between temperature and pH value is 0.82, and the frequency entanglement degree is 0.75.

[0041] The state trajectory similarity analysis uses a 10-second sliding window with an overlap rate of 50%. The normalized distance of the temperature and pressure trajectories is 0.15, and the normalized distance of the temperature and pH value trajectories is 0.28. The cosine similarity calculation of the feature vectors shows that the similarity between temperature and pressure is 0.91, and the similarity between temperature and pH value is 0.83. With the synchronous feature weight set to 0.4, the trajectory similarity weight set to 0.35, and the vector similarity weight set to 0.25, the calculated correlation strength between temperature and pressure is 0.92, and the correlation strength between temperature and pH value is 0.81.

[0042] In this embodiment, through the neural differential dynamics model, the feature vectors of the process parameters are mapped to a continuous state space, achieving accurate modeling of the dynamic characteristics of the parameters. By using an adaptive Runge-Kutta integrator to solve the differential equations, the accuracy and efficiency of numerical calculations are significantly improved. Introducing a non-linear oscillator network to describe the coupling relationship between process parameters can more accurately characterize the non-linear dynamic interaction between parameters. The calculation of the correlation strength integrates information from three levels: the oscillator synchronization state, the state trajectory similarity, and the feature vector correlation. In the prior art, the analysis of the correlation relationship between process parameters usually adopts simple statistical correlation methods or neural network models with fixed structures, ignoring the dynamic coupling characteristics between parameters and being difficult to capture the variation laws of parameters at different time scales. Most traditional methods are based on static data analysis and cannot effectively reflect the complex non-linear interaction relationship between process parameters. This embodiment overcomes the limitations of traditional discrete modeling methods, can accurately describe the continuous change process of process parameters, optimizes the utilization efficiency of computing resources while ensuring the calculation accuracy, and intuitively reflects the change characteristics and importance of process parameters through the frequency and amplitude characteristics of the oscillator, providing a new analysis perspective for process parameter optimization and more in-depth theoretical guidance for process optimization.

[0043] Figure 2This is the phase space trajectory analysis diagram of the washing composition production regulation method based on multi-parameter intelligent analysis in the embodiments of the present invention, showing the trajectory evolution of the non-linear oscillator corresponding to four groups of key process parameters in the phase space. The horizontal axis represents the oscillator position, and the vertical axis represents the speed. The trajectory shape and convergence characteristics reflect the correlation pattern between the parameters. The trajectory of the temperature-pressure parameter pair (solid square) forms a regular limit cycle in the phase space, with a diameter of 4.8 units and a period stable at 6.2 seconds, showing strong synchronization characteristics, and the corresponding phase synchronization index is 0.95. The trajectory of the temperature-pH value parameter pair (triangle) presents an elliptical shape, with the ratio of the major axis to the minor axis being 1.8:1 and a period of 7.3 seconds, and the phase synchronization index is 0.82. The trajectory of the pressure-rotation speed parameter pair (diamond) forms a distorted ring, with a period of 5.8 seconds, and the phase synchronization index is 0.84. The trajectory of the pH value-additive concentration parameter pair (cross) is the most irregular, showing quasi-periodic characteristics, and the phase synchronization index is only 0.65. Through the analysis of the geometric characteristics of the phase space trajectory, the stability and strength of the correlation between the parameters can be quantitatively evaluated. Compared with the traditional correlation analysis method, this technical solution can capture the complex non-linear relationship between the parameters more comprehensively by observing the dynamic behavior of the oscillator. The experimental results show that the consistency between the convergence characteristics of the phase space trajectory and the parameter correlation strength in the actual production process reaches 92.7%, which is much higher than 76.3% of the traditional method. In particular, the trajectory of the temperature-pressure parameter pair presents a highly regular limit cycle, which highly coincides with its high correlation strength (0.92); while the irregular trajectory of the pH value-additive concentration parameter pair intuitively reflects its low correlation strength (0.65). This visualization method based on non-linear dynamics not only provides a quantitative measurement of the correlation strength, but also demonstrates the stability, periodicity and predictability of the parameter coupling system, providing a new theoretical basis and methodological support for process parameter optimization.

[0044] In an alternative embodiment, Based on the dynamic correlation map, the process parameters are grouped to generate process parameter groups, and the parameter change trends are obtained through online monitoring. The co-variation characteristics calculated in combination with the dynamic correlation map include: Determine the correlation threshold according to the connection relationship of the process parameters in the dynamic correlation map and the distribution characteristics of the pre-calculated correlation strength, and divide the process parameters with the correlation strength greater than the correlation threshold into the same process parameter group; Collect the real-time data of each process parameter in the process parameter group through an online sensor, process the real-time data through a sliding time window, and extract the time series change trend characteristics of the process parameters; Map the time-series change trend features to the process parameter nodes corresponding to the dynamic association graph, and calculate the change consistency of each parameter in the process parameter group based on the connection relationships in the dynamic association graph to obtain the co-variation features of the process parameter group.

[0045] Perform statistical analysis on the distribution of association strengths in the dynamic association graph, count the number of connection edges in each strength interval, and draw a distribution histogram of the association strengths. By analyzing the distribution characteristics, identify the main clustering intervals of the association strengths. Set candidate thresholds at the boundaries of the clustering intervals, and use the silhouette coefficient to evaluate the rationality of parameter grouping under different thresholds. Select the candidate value with the largest silhouette coefficient as the final association threshold. According to the determined association threshold, divide the process parameters with association strengths higher than the threshold into the same group to form multiple process parameter groups with strong association relationships.

[0046] Real-time collect process parameter data through an online sensor, and set the collection frequency according to the change characteristics of the process parameters. Process the collected data using a sliding time window, determine the window length according to the characteristic time scale of the process, and set the window overlap rate to half of the window length. In each time window, perform noise reduction preprocessing on the data sequence to remove the influence of abnormal fluctuations and measurement noise. Extract the statistical features of the data within the window, including mean, variance, skewness, kurtosis, etc. At the same time, calculate the first-order difference and second-order difference of the data to analyze the change rate and acceleration characteristics of the parameters.

[0047] Perform trend analysis on the processed data sequence, and use the piecewise linear fitting method to identify the local change trends of the data. By comparing the trend features of adjacent time windows, determine the persistence and mutability of parameter changes. Combine wavelet transform to perform multi-scale decomposition on the data, and extract the change patterns at different time scales. Convert the extracted trend features into feature vectors as the dynamic attributes of the process parameter nodes.

[0048] Update the node attributes in the dynamic association graph, and map the time-series feature vectors to the corresponding parameter nodes. Based on the connection relationships in the graph, analyze the similarity of the feature vectors of adjacent nodes to evaluate the degree of parameter change consistency. Pass node information within the local neighborhood through graph convolution operations, considering the influence of directly connected nodes and indirectly connected nodes. Calculate the overall coordination of the parameters within each process parameter group to obtain a co-variation index reflecting the dynamic characteristics of the parameter group.

[0049] Exemplarily, during the production process of a certain batch of washing compositions, the dynamic correlation map contains six process parameter nodes. By analyzing the distribution of correlation strengths, it is found that there is an obvious clustering phenomenon when the strength values are between 0.7 and 0.9. After evaluation by the silhouette coefficient, the correlation threshold is set to 0.75, and the six parameters are divided into two parameter groups: the first group includes temperature, pressure, and pH value, and the second group includes stirring speed, liquid level, and flow rate.

[0050] Online sensors with a sampling frequency of 10 Hz are used to collect data, a sliding time window of 60 seconds is set, and the window overlap rate is 30 seconds. Median filtering is used to remove outliers within the window, and the statistical characteristics and change trends of the data are calculated. Analysis shows that the temperature and pressure in the first group show a significant positive correlation change trend, and the consistency of the change rates reaches 90%. The change of the pH value lags slightly, and the consistency with the changes of temperature and pressure is 85%. In the second group, the change trends of the stirring speed and flow rate are similar, with a consistency of 80%, while the change of the liquid level is relatively independent.

[0051] After mapping the trend characteristics to the correlation map, the overall co-variation index of the first group is calculated to be 0.88, showing strong co-variation characteristics; the co-variation index of the second group is 0.75, and the co-variation is relatively weak.

[0052] In this embodiment, the method of determining the correlation threshold through the dynamic correlation map and the correlation strength distribution characteristics realizes the adaptive grouping of process parameters. Through data preprocessing and feature extraction within the window, the interference of data noise is effectively eliminated, and the accuracy of time series feature extraction is improved. Through means such as statistical feature analysis, differential feature extraction, and multi-scale decomposition, the change laws of process parameters are comprehensively characterized.

[0053] In an alternative embodiment, A hierarchical progressive intelligent controller including a global planning layer and a local execution layer is constructed, the current regulation strategy is optimized in real time by combining the dynamic programming algorithm, and the control parameters are adaptively updated according to the changes of the dynamic correlation map. Based on the co-variation characteristics, constraint conditions for parameter adjustment are established, and the optimized regulation strategy obtained by applying the constraint conditions includes: Obtain the historical data of process parameters, construct a system state vector and a control input vector according to the historical data, and construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer. The global planning layer determines a global objective function including a system state deviation term, a control input constraint term, and a system energy consumption term based on the system state vector and the control input vector. The local execution layer constructs a parameter constraint matrix based on the process parameter connection relationship of the dynamic correlation map; Establish a system discrete state equation based on the system state vector, construct a value function with a discount factor based on the system discrete state equation and the global objective function, obtain a feedback gain matrix through iterative optimization of the value function, and construct a first control strategy according to the feedback gain matrix; Adaptively update the control parameters according to the change of the dynamic association graph, calculate the update amount of the control gain matrix, where the update amount is related to the gradient of the global objective function, the learning rate, and the parameter constraint matrix, adaptively adjust the control parameters in the first control strategy according to the update amount to obtain a second control strategy, and establish a constraint condition for parameter adjustment based on the co-variation characteristics; Predict the system state sequence in the future time domain according to the second control strategy and the system discrete state equation, and apply the constraint condition to obtain an optimized control strategy.

[0054] Preprocess the historical data of process parameters. Use sliding median filtering to remove outliers, and use wavelet transform to denoise the data. Standardize the preprocessed data to eliminate the dimension difference. Discretize the continuous data sequence through time window segmentation to construct a system state vector. The state vector includes the current value, target value deviation, change rate, and change acceleration of process parameters. At the same time, extract the control input data, including the operation amount and its change amount of each actuator, and construct a control input vector.

[0055] Construct an objective function in the global planning layer. The state deviation term of the objective function adopts a quadratic function form, and different weights are assigned to the deviations of different process parameters. The control input constraint term includes input amplitude limit and change rate limit, and is added to the objective function in the form of soft constraint. The system energy consumption term considers the power consumption characteristics of the actuator and establishes a mapping relationship between energy consumption and control input. In the local execution layer, construct a parameter constraint matrix based on the degree centrality and edge weight of the parameter nodes in the dynamic association graph. The element value of the constraint matrix reflects the coupling strength and influence direction between parameters.

[0056] Establish a system discrete state equation, including a state transition matrix, a control input matrix, and a disturbance term. Fit the historical data by the least squares method to identify the system matrix in the state equation. Introduce a time-decaying discount factor to construct a value function, and the discount factor decays exponentially with the time interval. Use the policy iteration algorithm to solve the value function, and each iteration includes two steps: policy evaluation and policy improvement. In the policy evaluation stage, fix the current policy to calculate the state value; in the policy improvement stage, update the control policy based on the Bellman optimality principle. The iteration process continues until the value function converges or reaches the maximum number of iterations. Calculate the feedback gain matrix according to the optimal value function, and this matrix maps the state deviation to the optimal control quantity.

[0057] Calculate the change rate of the parameter correlation strength in the real-time computed dynamic correlation graph. When the change rate exceeds the preset threshold, trigger the adaptive update mechanism of the control parameters. Use the stochastic gradient descent method to calculate the update amount of the control gain matrix. During the update process, the gradient of the objective function is approximately calculated by forward difference, and the learning rate is adaptively adjusted according to the gradient norm. Transform the parameter constraint matrix into a projection operator, and project the updated control parameters to ensure that the constraint conditions are met. Adjust the control parameters in the first regulation strategy according to the update amount to obtain the second regulation strategy.

[0058] Based on the co-variation characteristics of the process parameter group, establish the parameter adjustment constraint conditions. The constraint conditions include the ratio constraint of the parameter change rate, the change synchronization constraint, and the extreme value constraint. Transform the constraint conditions into the constraint form of a convex optimization problem. In the prediction time domain, perform state recursion based on the second regulation strategy and the discrete state equation. Use the moving horizon optimization method to minimize the prediction error under the premise of meeting the constraint conditions. Calculate the optimal control sequence through a convex optimization solver to obtain the optimized regulation strategy.

[0059] Exemplarily, during the production process of a certain batch of washing compositions, 2000 hours of historical process parameter data were collected with a sampling period of 1 second. Use a sliding window with a length of 60 to preprocess the data and extract the state characteristics of six parameters: temperature, pressure, pH value, stirring speed, liquid level, and flow rate. Construct a 96-dimensional state vector, including 16-dimensional original values, 16-dimensional target deviations, 32-dimensional first-order differences, and 32-dimensional second-order differences. The control input vector includes the adjustment amounts of 12 actuators.

[0060] In the global objective function, the deviation weights for temperature and pressure are set to 1, the deviation weight for pH value is set to 0.8, and the deviation weights for other parameters are set to 0.6. The control input constraint limits the adjustment range of the actuator within plus or minus 30%, and the change rate is limited within 2% per second. The energy consumption term uses a quadratic function to describe the power consumption characteristics of the actuator. The coupling coefficient between temperature and pressure in the constraint matrix is 0.9, and the coupling coefficient between temperature and pH value is 0.75.

[0061] The policy iteration uses a discount factor of 0.95, the maximum number of iterations is set to 1000, and the convergence threshold is set to 0.001. Adaptive update is triggered when the correlation strength change rate exceeds 0.05 per hour, and the initial value of the learning rate is set to 0.01. The co-variation constraint requires that the ratio of the change rates of temperature and pressure be maintained between 0.9 and 1.1, the phase difference does not exceed 10 seconds, and the extreme value deviation does not exceed 5%. In the 5-minute prediction time domain, state prediction and optimization calculations are performed every 10 seconds, and finally an optimized regulation strategy that meets various constraints is obtained.

[0062] In this embodiment, through the systematic preprocessing and feature extraction of historical process parameter data, high-dimensional system state vectors and control input vectors are constructed. An architecture design of a hierarchical progressive intelligent controller is adopted to achieve the organic combination of global planning and local execution. Based on the control strategy optimization method of value function iteration, by introducing a time-decaying discount factor, the balance between short-term control effects and long-term stability is achieved. The application of the policy iteration algorithm ensures the convergence of the control strategy, and the construction of the feedback gain matrix provides a reliable guarantee for the precise regulation of process parameters.

[0063] Figure 3 This is a comparison chart of the adaptive ability of the dynamic correlation map of the washing composition production regulation method based on multi-parameter intelligent analysis in the embodiments of the present invention, showing the comparison of the adaptive ability of the dynamic correlation map of this technical solution with three existing methods (Bayesian network method, association rule mining, and traditional correlation analysis) at six different stages of the production process.

[0064] In the initial stage, the parameter association recognition accuracy rate of this technical solution is 78.5%. Although it is ahead of the Bayesian network method (75.3%), association rule mining (70.2%), and traditional correlation analysis (65.8%), the advantage is not obvious. As the production process progresses to the batching stage, this technical solution reaches 93.2%, the Bayesian network method is 86.4%, association rule mining is 78.5%, and traditional correlation analysis is only 67.3%. In the reaction stage, the recognition accuracy rate of this technical solution climbs to 107.8%, far higher than the Bayesian network method (97.6%), association rule mining (84.3%), and traditional correlation analysis (70.5%). In the cooling stage, although this solution decreases, it still maintains a high level of 98.3%, while the other methods drop to 89.2%, 82.7%, and 68.9% respectively. Especially in the pressure release stage where the process conditions change drastically, the recognition accuracy rate of this technical solution is as high as 111.5%, which is 18 percentage points higher than the second-place Bayesian network method (93.5%), and 31 and 41.8 percentage points higher than association rule mining (80.5%) and traditional correlation analysis (69.7%) respectively. In the final product extraction stage, this solution still maintains a high accuracy rate of 99.7%, while the Bayesian network method is 90.8%, association rule mining is 82.3%, and traditional correlation analysis is 70.2%. Among them, the parameter association recognition accuracy rate in this technical solution adopts a relative benchmark evaluation index, that is, the accuracy rate in the chart is relative to the parameter association model under the pre-set standard working conditions (as the 100% benchmark point), and the accuracy rate = (number of correctly recognized associations / number of associations recognized by the benchmark model) × (prediction accuracy of association strength / benchmark accuracy) × (prediction accuracy of association change trend / benchmark accuracy) × 100%.

[0065] This technical solution can calculate the change rate of the parameter correlation strength in the dynamic correlation graph in real time and trigger the adaptive update mechanism of the control parameters according to the change. In contrast, the traditional correlation analysis method performs the worst throughout the production process, and the recognition accuracy rate is always between 65.8% and 70.5%, unable to effectively capture the complex non-linear relationship between process parameters. The significant advantage of this technical solution in the adaptive recognition ability of the parameter correlation relationship under dynamic working conditions, especially in the stage of rapid change of process conditions, the average recognition accuracy rate has increased by 28.7%.

[0066] In an alternative embodiment, Based on the system state vector, establish a system discrete state equation. Based on the system discrete state equation and the global objective function, construct a value function with a discount factor. Through iterative optimization of the value function, obtain a feedback gain matrix. The first regulation strategy constructed according to the feedback gain matrix includes: Obtain the system state vector, map the system state vector through an encoder network to obtain a hidden state vector, reconstruct the system state vector through a decoder network with the hidden state vector and calculate the reconstruction loss; Construct a causal graph structure based on historical data and calculate the mutual information and information entropy between nodes. Calculate the causal strength weight according to the mutual information and the information entropy. Based on the hidden state vector, establish a system discrete state equation. Calculate the causal correction term according to the causal strength weight and combine it with the system discrete state equation to obtain a causally enhanced system discrete state equation; Based on the system discrete state equation and the global objective function, construct an explicit value function and an implicit value function. Calculate the adaptive weight factor according to the state prediction error. Use the adaptive weight factor to perform a weighted combination of the explicit value function and the implicit value function to obtain a value function with a discount factor; Use the Bellman equation and the temporal difference algorithm to update the explicit value function and the implicit value function respectively. Through iterative optimization of the value function with a discount factor, obtain a feedback gain matrix. Calculate the causal effect of the control input and correct the feedback gain matrix according to the causal effect. Combine the corrected feedback gain matrix with the causal correction term to construct the first regulation strategy.

[0067] Construct a deep encoder-decoder network structure. The encoder consists of four encoding modules connected in series. Each module contains a fully connected layer, a batch normalization layer, and a non-linear activation layer. The weight matrix of the fully connected layer is initialized by a normal distribution, and the bias term is initialized to zero. The batch normalization layer calculates the mean and variance of the features, normalizes the features, and introduces learnable scaling and translation parameters. The activation layer uses the rectified linear unit function to introduce non-linearity. Each layer of the encoder gradually reduces the feature dimension and compresses the original state vector into a hidden state vector. The decoder adopts a structure that is mirror-symmetric to the encoder and gradually restores the feature dimension layer by layer. The network parameters are optimized by the backpropagation algorithm. The mean squared error is used as the reconstruction loss function, and the stochastic gradient descent method with an adaptive learning rate is used for parameter update.

[0068] When constructing the causal graph structure, the historical data is segmented according to a fixed time window. For the data within each time window, the mutual information between pairwise process parameters is calculated. The marginal distribution and joint distribution of the parameters are constructed by the kernel density estimation method, and the mutual information value is calculated based on the distribution function. At the same time, the information entropy of each parameter is calculated, and the probability distribution is estimated by the histogram method and the entropy value is calculated. The mutual information and information entropy are normalized to obtain the causal strength weights with values ranging from 0 to 1. Based on the hidden state vector, a discrete state equation is constructed, and the causal strength weights are transformed into correction terms of the state transition matrix. The direction of the correction term is determined by the causal relationship, and the amplitude is controlled by the weight magnitude. The corrected state transition matrix is substituted into the state equation to obtain the causality-enhanced state equation.

[0069] Design the structures of the explicit value function and the implicit value function respectively. The explicit value function adopts the form of a quadratic function, and the coefficient matrix is estimated by the least squares method. The implicit value function is approximated by a multi-layer perceptron network, which contains multiple hidden layers, and each layer uses a different number of neurons. The state prediction error is calculated based on the historical data, and the error sequence is smoothed by exponential smoothing to obtain the smoothed error. An adaptive weight update rule is designed. When the smoothed error is lower than the lower threshold, the weight of the explicit value function increases linearly; when the smoothed error is higher than the upper threshold, the weight of the explicit value function decays exponentially. The sum of the weight factors always remains 1 to ensure the stability of the combined value function. A discount factor that decays exponentially with time is introduced to construct a value function expression considering long-term rewards.

[0070] Optimize the explicit value function and the implicit value function in parallel. The explicit value function iterates through the Bellman equation, calculating the value estimate and optimal control quantity for the current state in each iteration and updating the value function parameters. The implicit value function is trained using the temporal difference algorithm, calculating the temporal difference error of continuous state transition samples and adjusting the network weights through error backpropagation. The optimization processes of the two value functions are carried out synchronously, sharing the transition samples in the experience replay buffer. Perform policy iteration on the combined value function, alternating between policy evaluation and policy improvement during the iteration process. In the policy evaluation phase, fix the current policy to calculate the state value, and in the policy improvement phase, update the policy parameters based on the latest value estimate.

[0071] Calculate the causal effect of the control input through the perturbation analysis method. Apply a small perturbation to each control input, record the response changes of the state variables, and estimate the causal effect intensity through the slope of the response curve. Convert the causal effect intensity into a correction coefficient for the gain matrix and perform an element-wise correction on the feedback gain matrix. Finally, combine the corrected gain matrix with the causal correction term in the state equation to construct a complete regulation strategy.

[0072] Exemplarily, during the production process of a certain batch of washing compositions, a 96-dimensional state vector is collected, including the original values, target deviations, change rates, and acceleration information of six process parameters. A four-layer encoder network is constructed, with the number of neurons in each layer being 96, 64, 48, and 32 in sequence, and optimized using mini-batch stochastic gradient descent with a batch size of 128. After 5000 rounds of training, the reconstruction loss is reduced from the initial 0.82 to 0.043.

[0073] Based on 3000 hours of historical data, construct a causal graph using a sliding time window with a length of 300 seconds. The mutual information between temperature and pressure calculated through kernel density estimation is 0.85, the information entropy of temperature is 0.92, and the information entropy of pressure is 0.88. After normalization, the causal intensity weight of temperature on pressure is 0.78, and the causal intensity weight of pressure on temperature is 0.65. Add the causal relationship to the state equation and correct the corresponding elements in the state transition matrix.

[0074] Construct a neural network with a 64-48-32 structure as an implicit value function approximator, using the rectified linear unit as the activation function. Set the prediction error threshold interval to 5% to 10%. When the error is less than 5%, the weight of the explicit value function increases by 0.02 in each iteration; when the error is greater than 10%, the weight decay coefficient is 0.95. The policy iteration uses a discount factor of 0.95. After 800 rounds of iteration, the value function converges to below 0.001.

[0075] A perturbation with an amplitude of 1% is applied to the control input. By analyzing the state response curve, it is found that the causal effect of the control input on temperature is 0.82, and the causal effect on pressure is 0.71. Accordingly, the corresponding element values in the feedback gain matrix are corrected. The temperature loop gain is increased by 15%, and the pressure loop gain is increased by 10%. While ensuring the control accuracy, the regulation strategy fully considers the causal relationship between parameters.

[0076] In this embodiment, an encoder-decoder network structure is used to perform dimensionality reduction mapping and reconstruction on the system state. The essential features of the state vector are extracted through multi-layer non-linear transformation. The causal graph structure and information theory method are introduced to analyze the causal relationship between process parameters. A hybrid optimization architecture combining explicit value function and implicit value function is used, and at the same time, the proportion of the two value functions is dynamically adjusted through an adaptive weight mechanism, which not only ensures the real-time response ability of the control strategy but also improves the long-term stability of the strategy. In the prior art, intelligent control methods mainly rely on pure data-driven models, only focusing on the correlation between process parameters, ignoring the deep causal relationship between parameters, and it is difficult to accurately grasp the mechanism of action between parameters, resulting in insufficient adaptability and robustness of the control strategy. Traditional value function optimization methods often adopt a single function form, and it is difficult to balance the real-time performance and long-term stability of control when facing complex process. The control strategy of this embodiment can accurately identify and utilize the causal relationship between process parameters, significantly improving the control accuracy. The adaptive optimization mechanism enables the system to have stronger environmental adaptability and anti-interference ability. The control correction based on causal effect ensures the rationality and reliability of the regulation strategy, providing strong technical support for the intelligent and precise control of the process.

[0077] Figure 4 This is the causal strength weight heat map between key process parameters of the washing composition production regulation method based on multi-parameter intelligent analysis in the embodiment of the present invention, showing the causal strength weight matrix between six key process parameters, calculated through a 300-second sliding time window based on 3000 hours of historical data. The darker areas indicate stronger causal relationships.

[0078] According to Figure 4 As shown, the causal strength weight of temperature on pressure is 0.78, while the causal strength of pressure on temperature is 0.65, indicating an asymmetric bidirectional causal relationship. The causal relationship between flow rate and pH value is relatively significant. The causal strength of flow rate on pH value is 0.71, while the causal strength of pH value on flow rate is 0.68. There is a strong mutual causal relationship between viscosity and stirring speed. The weight of viscosity on stirring speed is 0.73, and the weight of stirring speed on viscosity is 0.75.

[0079] Compared with traditional correlation analysis methods (such as Pearson correlation coefficient), the causal strength weight calculated by combining mutual information and information entropy in this technical solution can accurately capture the non-linear causal relationship between parameters, avoiding the false causal inferences that may be caused by relying solely on statistical correlation. This causal strength weight matrix provides a reliable quantitative basis for the subsequent correction of the state equation, enabling the system model to more accurately reflect the complex causal interactions between actual process parameters.

[0080] In the second aspect of the embodiments of the present invention, a production regulation system for washing compositions based on multi-parameter intelligent analysis is provided, including: A first unit for collecting process parameters during the production of washing compositions, generating a time-series data stream of process parameters and performing data preprocessing to obtain standard process parameter data; A second unit for inputting the standard process parameter data into a cascaded autoencoder deep learning network, performing hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fusing feature information at different scales by combining a residual connection structure, and performing priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; A third unit for grouping process parameters based on the dynamic association map, generating process parameter groups and obtaining the parameter change trend through online monitoring, and calculating the co-variation features in combination with the dynamic association map; A fourth unit for constructing a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, optimizing the current regulation strategy in real time by combining a dynamic programming algorithm, adaptively updating control parameters according to the changes in the dynamic association map, establishing constraint conditions for parameter adjustment based on the co-variation features, and applying the constraint conditions to obtain an optimized regulation strategy; A fifth unit for converting the optimized regulation strategy into a control instruction to adjust process parameters, collecting the adjusted process parameters for online learning, continuously optimizing the dynamic association map, and forming a closed-loop optimization control.

[0081] The present invention can be a method, a device, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions for performing various aspects of the present invention uploaded thereon.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for regulating the production of a washing composition based on multi-parameter intelligent analysis, characterized in that, Including: Collect process parameters during the production process of the washing composition, generate a time-series data stream of process parameters and perform data preprocessing to obtain standard process parameter data; Input the standard process parameter data into a cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fuse feature information at different scales by combining a residual connection structure, and perform priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; Group the process parameters based on the dynamic association map, generate process parameter groups and obtain the parameter change trend through online monitoring, and calculate the co-variation characteristics in combination with the dynamic association map; Construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, optimize the current control strategy in real time in combination with the dynamic programming algorithm, adaptively update the control parameters according to the changes in the dynamic association map, establish constraint conditions for parameter adjustment based on the co-variation characteristics, and apply the constraint conditions to obtain an optimized control strategy; Convert the optimized control strategy into a control instruction to adjust the process parameters, collect the adjusted process parameters for online learning, continuously optimize the dynamic association map, and form a closed-loop optimization control.

2. The method according to claim 1, wherein Collect process parameters during the production process of the washing composition, generate a time-series data stream of process parameters and perform data preprocessing to obtain standard process parameter data, including: Collect process parameters during the production process of the washing composition, where the process parameters include the reaction kettle temperature, reaction kettle pressure, pH value of the reaction solution, stirring speed, reaction kettle liquid level, and raw material addition rate, and record them at preset time intervals to form a time-series data stream of process parameters; Perform data preprocessing on the time-series data stream of process parameters, identify and remove abnormal data by setting a parameter fluctuation threshold, fill in missing data using an interpolation algorithm, and verify the validity of the data based on the physical constraint relationship of the process parameters to generate preprocessed process parameter data; Perform standardization processing on the preprocessed process parameter data, convert process parameters with different dimensions to a unified numerical range to obtain standard process parameter data.

3. The method according to claim 1, wherein Input the standard process parameter data into a cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fuse feature information at different scales by combining a residual connection structure, and perform priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters, including: Construct a cascaded autoencoder deep learning network using a symmetric encoder-decoder structure, input the standard process parameter data into the cascaded autoencoder deep learning network, and generate an encoded feature matrix by sequentially performing convolution operations, batch normalization processing, and activation function operations through multiple cascaded encoding units; Perform feature extraction on the encoded feature matrix through multiple convolution kernels of different sizes, use the convolution kernel with the smallest size to extract the reference feature matrix, and use other size convolution kernels to extract a group of time-series feature matrices; Perform feature mapping transformation on each temporal feature matrix in the temporal feature matrix group to obtain a transformed feature matrix group, and perform residual superposition operation on the transformed feature matrices in the transformed feature matrix group and the reference feature matrix to generate a fused feature matrix; Map the fused feature matrix to a query matrix, a key matrix, and a value matrix, calculate the product of the query matrix and the transpose of the key matrix to obtain an attention score matrix, perform normalization processing on the attention score matrix, and then multiply it by the value matrix to obtain an attention feature matrix; Extract the feature vectors corresponding to the process parameters in the attention feature matrix, calculate the ratio of the inner product of any two feature vectors to the product of the modulus lengths to obtain the correlation strength between the process parameters, set the process parameters as vertices, the correlation relationship between the process parameters as edges, and the correlation strength as the edge weights to construct a dynamic correlation graph of the process parameters.

4. The method according to claim 3, characterized in that, Extracting the feature vectors corresponding to the process parameters in the attention feature matrix and calculating the ratio of the inner product of any two feature vectors to the product of the modulus lengths to obtain the correlation strength between the process parameters includes: Extract the feature vectors corresponding to the process parameters from the attention feature matrix, and input the feature vectors corresponding to the process parameters into a neural differential dynamics model. The neural differential dynamics model maps the feature vectors corresponding to the process parameters to state vectors through a differential equation, and the differential equation consists of the time derivative of the state vector and a neural network; Use an adaptive Runge-Kutta integrator to adjust the integration step size according to the change rate of the state vector to solve the differential equation and obtain the time evolution characteristics of the feature vectors corresponding to the process parameters; Construct a group of nonlinear oscillators according to the time evolution characteristics. Each process parameter corresponds to a nonlinear oscillator. The initial state of the nonlinear oscillator is set according to the feature vector corresponding to the process parameter. The frequency of the nonlinear oscillator characterizes the change characteristics of the process parameter, and the amplitude of the nonlinear oscillator characterizes the importance of the process parameter; Establish a coupling network between the nonlinear oscillators, connect the nonlinear oscillators through a nonlinear coupling equation and analyze the synchronization state, and calculate the correlation strength between any two process parameters according to the correlation strength calculation formula. The correlation strength calculation formula combines the synchronization state of the nonlinear oscillators, the pre-obtained state trajectory similarity, and the ratio of the inner product of any two feature vectors to the product of the modulus lengths through weight factors.

5. The method according to claim 1, wherein Group the process parameters based on the dynamic correlation graph, generate process parameter groups and obtain the parameter change trend through online monitoring. Combining the dynamic correlation graph to calculate the co-variation characteristics includes: Determine the correlation threshold according to the connection relationship of the process parameters in the dynamic correlation graph and the distribution characteristics of the pre-calculated correlation strength, and divide the process parameters with a correlation strength greater than the correlation threshold into the same process parameter group; Collect the real-time data of each process parameter in the process parameter group through an online sensor, process the real-time data through a sliding time window, and extract the temporal change trend characteristics of the process parameters. Map the time series change trend features to the process parameter nodes corresponding to the dynamic association graph, calculate the change consistency of each parameter in the process parameter group based on the connection relationship in the dynamic association graph, and obtain the co-variation features of the process parameter group.

6. The method according to claim 1, characterized in that Construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, combine the dynamic programming algorithm to optimize the current control strategy in real time, adaptively update the control parameters according to the changes of the dynamic association graph, establish the constraint conditions for parameter adjustment based on the co-variation features, and apply the constraint conditions to obtain the optimized control strategy, including: Obtain the historical data of process parameters, construct a system state vector and a control input vector according to the historical data, and construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer. The global planning layer determines a global objective function including a system state deviation term, a control input constraint term, and a system energy consumption term based on the system state vector and the control input vector. The local execution layer constructs a parameter constraint matrix based on the process parameter connection relationship of the dynamic association graph; Establish a system discrete state equation according to the system state vector, construct a value function with a discount factor based on the system discrete state equation and the global objective function, obtain a feedback gain matrix through iterative optimization of the value function, and construct a first control strategy according to the feedback gain matrix; Adaptively update the control parameters according to the changes of the dynamic association graph, calculate the update amount of the control gain matrix, the update amount is related to the gradient of the global objective function, the learning rate, and the parameter constraint matrix, adaptively adjust the control parameters in the first control strategy according to the update amount to obtain a second control strategy, and establish the constraint conditions for parameter adjustment based on the co-variation features; Predict the system state sequence in the future time domain according to the second control strategy and the system discrete state equation, and apply the constraint conditions to obtain the optimized control strategy.

7. The method according to claim 6, characterized in that Establish a system discrete state equation according to the system state vector, construct a value function with a discount factor based on the system discrete state equation and the global objective function, obtain a feedback gain matrix through iterative optimization of the value function, and construct a first control strategy according to the feedback gain matrix, including: Obtain the system state vector, map the system state vector through an encoder network to obtain a hidden state vector, reconstruct the system state vector through a decoder network and calculate the reconstruction loss; Construct a causal graph structure based on historical data and calculate the mutual information and information entropy between nodes, calculate the causal strength weight according to the mutual information and the information entropy, establish a system discrete state equation according to the hidden state vector, calculate a causal correction term based on the causal strength weight and combine it with the system discrete state equation to obtain a causally enhanced system discrete state equation; Construct an explicit value function and an implicit value function based on the discrete state equation of the system and the global objective function, calculate an adaptive weight factor according to the state prediction error, and use the adaptive weight factor to perform weighted combination on the explicit value function and the implicit value function to obtain a value function with a discount factor; Use the Bellman equation and the temporal difference algorithm to update the explicit value function and the implicit value function respectively, obtain a feedback gain matrix through iterative optimization of the value function with a discount factor, calculate the causal effect of the control input and correct the feedback gain matrix according to the causal effect, and combine the corrected feedback gain matrix with the causal correction term to construct a first regulation strategy.

8. A production control system for washing compositions based on multi-parameter intelligent analysis, which is used to implement the method described in any one of the foregoing claims 1-7, characterized in that, Including: The first unit is used to collect process parameters in the production process of the washing composition, generate a time-series data stream of process parameters and perform data preprocessing to obtain standard process parameter data; The second unit is used to input the standard process parameter data into a cascaded autoencoder deep learning network, perform hierarchical feature extraction on the standard process parameter data through a multi-scale feature extraction layer, fuse feature information at different scales by combining a residual connection structure, and perform priority sorting on the features through a self-attention mechanism to obtain a multi-dimensional feature representation and construct a dynamic association map of process parameters; The third unit is used to group the process parameters based on the dynamic association map, generate process parameter groups and obtain the parameter change trend through online monitoring, and calculate the co-variation characteristics by combining the dynamic association map; The fourth unit is used to construct a hierarchical progressive intelligent controller including a global planning layer and a local execution layer, perform real-time optimization of the current regulation strategy by combining the dynamic programming algorithm, adaptively update the control parameters according to the change of the dynamic association map, establish constraint conditions for parameter adjustment based on the co-variation characteristics, and apply the constraint conditions to obtain an optimized regulation strategy; The fifth unit is used to convert the optimized regulation strategy into a control instruction to adjust the process parameters, collect the adjusted process parameters for online learning, continuously optimize the dynamic association map, and form a closed-loop optimization control.

Images