Intelligent data analysis system for traditional Chinese medicine concentration

Abstract

The invention provides an intelligent data analysis system for traditional Chinese medicine concentration. The intelligent data analysis system comprises a data acquisition module used for acquiring temperature, concentration and flow velocity data in real time; the data preprocessing module carries out noise filtering and standardization processing on the collected data; a multi-dimensional feature extraction module extracts time-frequency domain joint features; the dynamic parameter modeling module constructs a time-varying parameter concentration model; the optimization control decision module generates a control instruction; the real-time feedback adjusting module adjusts parameters of the concentration equipment according to the instruction; and the result output module outputs a control signal and a visual analysis result. According to the system, through cooperative work of all the modules, intelligent monitoring and optimal control over the traditional Chinese medicine concentration process are achieved, and the concentration efficiency and quality are improved. The concentration efficiency and the product quality stability can be improved, and the energy consumption and the manual intervention cost can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of a traditional Chinese medicine concentration process, and more particularly, to an intelligent data analysis system for traditional Chinese medicine concentration. Background Art

[0002] In the production of Traditional Chinese Medicine (TCM), concentration is a critical step, and its quality directly impacts the efficacy and stability of TCM preparations. Traditional TCM concentration processes rely heavily on manual experience and fixed parameter control, making it difficult to accurately monitor and adjust key parameters such as temperature, concentration, and flow rate during the concentration process in real time. Furthermore, existing technologies have limited capabilities for extracting and modeling dynamic features of the concentration process, making it impossible to effectively address complex and changing production environments. This results in low concentration efficiency, large fluctuations in product quality, and difficulty in achieving intelligent and refined production.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the existing technology: traditional methods are unable to monitor and control the dynamic changes of the concentration process in real time and accurately, lack the ability to deeply mine and model multi-dimensional features, and are difficult to meet the needs of the traditional Chinese medicine concentration process for efficient, stable and intelligent control. Summary of the Invention

[0004] The present invention provides an intelligent data analysis system for traditional Chinese medicine concentration, comprising:

[0005] Data acquisition module, data preprocessing module, multi-dimensional feature extraction module, dynamic parameter modeling module, optimization control decision module, real-time feedback adjustment module and result output module;

[0006] The data acquisition module is used to collect temperature, concentration and flow rate data in real time during the concentration process of traditional Chinese medicine;

[0007] The data preprocessing module performs noise filtering and standardization on the output of the data acquisition module;

[0008] The multi-dimensional feature extraction module extracts time-frequency domain joint features based on the output of the data preprocessing module;

[0009] The dynamic parameter modeling module constructs a time-varying parameter concentration model based on the features of the multi-dimensional feature extraction module;

[0010] The optimization control decision module generates control instructions based on the model of the dynamic parameter modeling module;

[0011] The real-time feedback adjustment module adjusts the concentration equipment parameters according to the instructions of the optimization control decision module;

[0012] The result output module outputs control signals and visual analysis results.

[0013] Furthermore, the multi-dimensional feature extraction module includes:

[0014] The time-frequency decomposition unit uses a custom multi-scale weighted fusion algorithm to decompose the preprocessed data;

[0015] Feature fusion unit, which performs nonlinear correlation calculation on the decomposed sub-signals;

[0016] Dimensionality reduction unit, which screens key features through a dynamic weight allocation algorithm;

[0017] The formula of the custom multi-scale weighted fusion algorithm is:

[0018]

[0019] Among them, S(t) is the fused feature sequence, w k is the dynamic weight of the k-th scale, x(τ) is the preprocessed data, φ k is a custom basis function, Δ is the time window length, and N is the total number of scales.

[0020] Furthermore, the dynamic weight allocation algorithm satisfies:

[0021]

[0022] in, is the variance of the k-th scale data, μ k is the k-th scale mean, μ ref is the reference mean, and λ is the adjustment factor.

[0023] Furthermore, the dynamic parameter modeling module constructs a model through the following steps:

[0024] Step S41: constructing a hidden Markov chain state transition equation based on the feature sequence;

[0025] Step S42: using a sliding window mechanism to update the state transition probability matrix;

[0026] Step S43: Calculate the dynamic equation of the concentration process through a bidirectional recurrent neural network:

[0027]

[0028] Where C(t) is the real-time concentration, T(t) is the temperature, F(t) is the flow rate, and α and β are time-varying parameters.

[0029] Furthermore, the time-varying parameter is updated by the following formula:

[0030]

[0031] Among them, α0 is the initial parameter, γ is the learning rate, τ is the integral variable, and L is the length of the backtracking time;

[0032] β(t)=β0·exp(-η·|C(t)-C target |)

[0033] Among them, β0 is the initial parameter, η is the adjustment factor, C target is the target concentration.

[0034] Furthermore, the optimization control decision module includes:

[0035] The objective function building block defines the optimization goal of the concentration process:

[0036]

[0037] Among them, J is the objective function value, T is the total number of time steps, ω1, ω2, ω3 are weight factors, C(t) is the real-time concentration, C target is the target concentration, ΔT(t) is the temperature change rate;

[0038] Constraint generation unit to set the physical boundaries of temperature and flow rate;

[0039] The dynamic programming solving unit uses an improved tabu search algorithm to solve the optimal control sequence.

[0040] Furthermore, the improved tabu search algorithm includes the following operations:

[0041] Sub-step S531: Introduce Gaussian perturbation term ∈~N(0,σ(t)) when generating neighborhood solution 2 ), where σ(t)=σ0·e -kt ;

[0042] Where ∈ is the Gaussian perturbation term, σ(t) is the time-dependent standard deviation, σ0 is the initial standard deviation, and k is the attenuation coefficient;

[0043] Sub-step S532: storing the hash code value of the historical solution in the taboo table;

[0044] Sub-step S533: Dynamically adjust the acceptance probability threshold using a simulated annealing mechanism.

[0045] Furthermore, the real-time feedback adjustment module includes:

[0046] Error calculation unit, real-time comparison of the deviation between actual concentration and model prediction value;

[0047] The parameter adjustment unit dynamically modifies the control instructions according to the following formula:

[0048]

[0049] Among them, F adj (t) is the adjusted control instruction, F(t) is the original control instruction, K p , K i is the proportional integral coefficient, e(t) is the concentration error at time t, and m is the length of the integration window;

[0050] The stability detection unit triggers model reconstruction when the error does not decrease after n consecutive adjustments.

[0051] Furthermore, the proportional integral coefficient is adaptively adjusted by the following formula:

[0052]

[0053] Among them, K p0 is the initial proportional coefficient, ζ is the adjustment constant;

[0054] K i (t) = K i0 ·tanh(ξ·dτ)

[0055] Among them, K i0 is the initial integral coefficient, ξ is the adjustment constant, and T0 is the integral time constant.

[0056] Furthermore, the result output module includes:

[0057] 3D visualization unit, synchronously displaying concentration field, temperature gradient and flow velocity distribution;

[0058] Parameter storage unit, recording optimized control instructions and actual response data by timestamp;

[0059] The control signal generation unit converts discrete instructions into analog signal sequences that can be executed by the device.

[0060] According to the above-mentioned embodiment of the present invention, there are at least the following beneficial effects: the intelligent data analysis system of the present invention can realize real-time monitoring and precise control of the Chinese medicine concentration process. The key parameters such as temperature, concentration and flow rate are acquired in real time by the data acquisition module, and combined with the data preprocessing and multi-dimensional feature extraction module, it can effectively remove noise and extract joint features in the time-frequency domain, providing high-quality data for subsequent modeling. The dynamic parameter modeling module constructs a time-varying parameter concentration model based on the hidden Markov chain and the bidirectional recurrent neural network, which can accurately reflect the dynamic characteristics of the concentration process and provide a reliable basis for optimization control. The optimization control decision module adopts an improved taboo search algorithm, combined with the objective function and constraints, which can efficiently solve the optimal control sequence and realize the refined regulation of the concentration process, thereby improving the concentration efficiency and product quality stability.

[0061] In addition, the real-time feedback regulation module of the present invention can compare the deviation between the actual concentration and the model prediction value in real time, and adjust the control instructions according to the dynamic correction formula to ensure the stability of the concentration process. The stability detection unit can trigger model reconstruction when the error has not decreased after multiple consecutive adjustments, which can further enhance the adaptability and robustness of the system. The result output module synchronously displays the concentration field, temperature gradient and flow rate distribution through a three-dimensional visualization unit, providing operators with intuitive production process information, and at the same time records the optimized control instructions and actual response data by timestamp to facilitate subsequent analysis and tracing. Overall, the present invention can improve the intelligence level of the traditional Chinese medicine concentration process, reduce energy consumption and manual intervention costs, and promote the modernization of traditional Chinese medicine production. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily apparent by reading the following detailed description with reference to the accompanying drawings, in which several embodiments of the present invention are shown by way of example and not limitation, in which:

[0063] Figure 1 A schematic diagram of the structure of an intelligent data analysis system for traditional Chinese medicine concentration provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0064] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0065] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0066] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0067] Reference below Figure 1 , Figure 1 This is a schematic diagram of the structure of an intelligent data analysis system for Chinese medicine concentration provided by one embodiment of the present invention. Figure 1 As shown, an intelligent data analysis system 100 for Chinese medicine concentration includes:

[0068] Data acquisition module 101, data preprocessing module 102, multi-dimensional feature extraction module 103, dynamic parameter modeling module 104, optimization control decision module 105, real-time feedback adjustment module 106 and result output module 107;

[0069] The data acquisition module is used to collect temperature, concentration and flow rate data in real time during the concentration process of traditional Chinese medicine;

[0070] The data preprocessing module performs noise filtering and standardization on the output of the data acquisition module;

[0071] The multi-dimensional feature extraction module extracts time-frequency domain joint features based on the output of the data preprocessing module;

[0072] The dynamic parameter modeling module constructs a time-varying parameter concentration model based on the features of the multi-dimensional feature extraction module;

[0073] The optimization control decision module generates control instructions based on the model of the dynamic parameter modeling module;

[0074] The real-time feedback adjustment module adjusts the concentration equipment parameters according to the instructions of the optimization control decision module;

[0075] The result output module outputs control signals and visual analysis results.

[0076] It should be noted that the intelligent data analysis system of the present invention is used for optimizing the control of the Chinese medicine concentration process. Its core lies in achieving refined management of the concentration process through the collaborative work of multiple modules. The data acquisition module is responsible for real-time acquisition of key parameters in the Chinese medicine concentration process, including temperature, concentration and flow rate. These parameters are important factors affecting the concentration effect. Among them, temperature reflects the heating conditions, concentration represents the ratio of solute to solvent, and flow rate determines the residence time of the material in the equipment. The role of the data preprocessing module is to process the collected raw data, remove noise and standardize it to ensure the accuracy of subsequent analysis. The multidimensional feature extraction module extracts joint time-frequency domain features based on the preprocessed data. This process can reveal the dynamic information hidden in the data and provide a richer feature basis for modeling. The dynamic parameter modeling module uses the extracted features to construct a mathematical model of the concentration process. The model can reflect the dynamic change law of the concentration process and provide theoretical support for optimization control. The optimization control decision module generates control instructions based on the model to adjust the operating parameters of the concentration equipment to achieve the purpose of optimizing the concentration process. The real-time feedback control module adjusts equipment parameters according to control instructions and makes dynamic corrections based on actual operating conditions to ensure the stability and efficiency of the concentration process. The result output module is responsible for outputting control signals and visualizing analysis results, allowing operators to monitor and analyze the entire concentration process.

[0077] Specifically, the data acquisition module uses a sensor network to monitor temperature, concentration, and flow rate data in real time during the TCM concentration process. Temperature sensors, such as thermocouples or thermistors, measure temperature changes within the heating kettle. Concentration sensors, based on optical or conductivity principles, detect the solute concentration of the concentrate in real time. Flow rate sensors, such as flow meters, measure the flow rate of the material in the pipeline. Data collected by these sensors is transmitted to the data preprocessing module at a certain sampling frequency (e.g., once per second). This module filters the collected data, for example, using a low-pass filter to remove high-frequency noise. It also normalizes the data to a range between [0, 1] to facilitate processing by subsequent modules. The time-frequency decomposition unit in the multidimensional feature extraction module uses a custom multiscale weighted fusion algorithm to decompose the preprocessed data into multiple sub-signals, each representing feature information at a different scale. The feature fusion unit calculates nonlinear correlations among these sub-signals to further extract the inherent connections between features. The dimensionality reduction unit uses a dynamic weight allocation algorithm to select key features, reduce data dimensionality, and improve the model's computational efficiency. The dynamic parameter modeling module constructs a hidden Markov chain state transition equation based on the extracted feature sequence and dynamically updates the state transition probability matrix through a sliding window mechanism to adapt to the dynamic changes in the concentration process. A bidirectional recurrent neural network is used to calculate the dynamic equations of the concentration process, where time-varying parameters are updated using specific formulas to reflect the real-time changes in the concentration process.

[0078] Preferably, sensors in the data acquisition module should have high precision and reliability. For example, the accuracy of temperature sensors should reach ±0.1°C, the accuracy of concentration sensors should reach ±0.01%, and the accuracy of flow rate sensors should reach ±1%. The noise filtering algorithm in the data preprocessing module can use a wavelet transform combined with threshold processing to more effectively remove noise and retain useful data information. In the multidimensional feature extraction module, the dynamic weights in the customized multi-scale weighted fusion algorithm can be adaptively adjusted based on the variance and mean of the data to better reflect the importance of features at different scales. In the dynamic parameter modeling module, the state transition equation of the hidden Markov chain can be optimized using Bayesian optimization methods to improve the model's prediction accuracy. The bidirectional recurrent neural network structure can adopt a long short-term memory network (LSTM) or a gated recurrent unit (GRU) to enhance the memory capacity and anti-forgetting properties of time series data. The improved tabu search algorithm in the optimization control decision module can further improve the efficiency and accuracy of solving the optimal control sequence by incorporating various heuristic strategies, such as genetic algorithms or particle swarm optimization algorithms.

[0079] In some embodiments, the multi-dimensional feature extraction module includes:

[0080] The time-frequency decomposition unit uses a custom multi-scale weighted fusion algorithm to decompose the preprocessed data;

[0081] Feature fusion unit, which performs nonlinear correlation calculation on the decomposed sub-signals;

[0082] Dimensionality reduction unit, which screens key features through a dynamic weight allocation algorithm;

[0083] The formula of the custom multi-scale weighted fusion algorithm is:

[0084]

[0085] Among them, S(t) is the fused feature sequence, w k is the dynamic weight of the k-th scale, x(τ) is the preprocessed data, φ k is a custom basis function, Δ is the time window length, and N is the total number of scales.

[0086] It should be noted that the multi-dimensional feature extraction module is one of the core components of this system, and its main function is to extract joint time-frequency domain features from preprocessed data. This module includes a time-frequency decomposition unit, a feature fusion unit, and a dimensionality reduction unit. The time-frequency decomposition unit uses a custom multi-scale weighted fusion algorithm to decompose the preprocessed data to obtain feature information at different scales. The feature fusion unit performs nonlinear correlation calculations on the decomposed sub-signals to further explore the intrinsic connections between features. The dimensionality reduction unit uses a dynamic weight allocation algorithm to screen key features, thereby reducing data dimensions and improving the efficiency of subsequent modeling. The formula for the custom multi-scale weighted fusion algorithm is:

[0087]

[0088] Among them, F(t) is the fused feature sequence, α k is the dynamic weight of the k-th scale, f k (x) is the preprocessed data, g k (x) is a custom basis function, τ is the time window length, and N is the total number of scales. These parameters and algorithms are designed to extract more representative and discriminative features through multi-scale analysis and dynamic weight adjustment, providing high-quality input for subsequent dynamic parameter modeling.

[0089] Specifically, the time-frequency decomposition unit in the multi-dimensional feature extraction module decomposes the pre-processed data into multiple sub-signals through a custom multi-scale weighted fusion algorithm. The multi-scale here refers to analyzing the data at different time scales to capture the characteristics of the data in different frequency ranges. For example, a lower scale may focus on high-frequency information (such as rapidly changing noise or transient signals), while a higher scale focuses on low-frequency information (such as long-term trends or slowly changing signals). Custom basis function g k (x) can be a wavelet function or other function suitable for time series analysis, which is used to project data at different scales. The time window length τ is a key parameter, which determines the time range of the data analyzed at each moment. If τ is set too large, the features may be too smooth and detail information may be lost; if τ is set too small, sufficient contextual information may not be captured. The total number of scales N determines the level of refinement of the analysis and is usually selected based on the complexity of the data and computing resources. The feature fusion unit performs nonlinear correlation calculations on the decomposed sub-signals. The purpose of this step is to explore the intrinsic connections between different sub-signals, such as by calculating the mutual information or correlation coefficient between the sub-signals, so as to obtain a more comprehensive feature representation. The dimensionality reduction unit screens key features through a dynamic weight allocation algorithm. The calculation formula for dynamic weights is:

[0090]

[0091] in, is the variance of the k-th scale data, μ k is the mean of the kth scale, μ ref is the reference mean, and β is the adjustment factor. This formula dynamically adjusts the weights based on the difference between the variance and the mean, ensuring that key features occupy a more important position in subsequent modeling.

[0092] Preferably, the custom basis function g k(x) Morlet wavelet or other wavelet functions suitable for TCM concentration data can be selected. These functions have good localization characteristics in time-frequency analysis and can effectively capture local features of the data. The time window length τ can be adjusted according to the dynamic characteristics of the actual concentration process. For example, in the early stage of concentration, when the data changes rapidly, a smaller τ can be set to capture transient information; in the later stage of concentration, when the data changes slowly, τ can be appropriately increased to smooth noise. For the dynamic weight allocation algorithm, the adjustment factor β can be optimized according to actual needs. For example, the optimal β value can be selected through cross-validation to ensure the rationality and effectiveness of the weight allocation. In addition, the feature fusion unit can introduce deep learning methods, such as convolutional neural networks (CNNs), to perform feature fusion on the decomposed sub-signals to further enhance the feature representation capability. In addition to the dynamic weight allocation algorithm, the dimensionality reduction unit can also use principal component analysis (PCA) or other dimensionality reduction techniques to reduce computational complexity and improve the generalization ability of the model.

[0093] In some embodiments, the dynamic weight allocation algorithm satisfies:

[0094]

[0095] in, is the variance of the k-th scale data, μ k is the k-th scale mean, μ ref is the reference mean, and λ is the adjustment factor.

[0096] It should be noted that the dynamic weight allocation algorithm is an important step in the multi-dimensional feature extraction module for screening key features. Its core is to dynamically adjust the importance of different features in subsequent modeling by calculating the weight of each scale feature. The formula of this algorithm is:

[0097]

[0098] Among them, α k is the dynamic weight of the k-th scale, is the variance of the k-th scale data, μ k is the mean of the kth scale, μ ref is the reference mean, and β is the adjustment factor. This formula dynamically adjusts weights based on the difference between the variance and mean, ensuring that key features play a more prominent role in subsequent modeling. In this way, the system can dynamically adjust feature weights based on the real-time characteristics of the data, thereby improving the adaptability and accuracy of feature extraction.

[0099] Specifically, the core of the dynamic weight allocation algorithm is to measure the importance of features at different scales by the difference between variance and mean. It reflects the degree of fluctuation of the data at the kth scale. The larger the variance, the more significant the data change at this scale, and it may contain more useful information. The mean μ k It represents the average level of the data on this scale. ref is a preset benchmark value used to measure the deviation between the current mean and the ideal state. The adjustment factor β is a parameter used to control the amplitude of weight adjustment. The larger its value, the more sensitive the weight is to the mean deviation. In practical applications, μ ref The weight allocation can be set based on historical data or target concentrations, for example, taking the average of historical data or the expected value of the target concentration. The value of the adjustment factor β can be adjusted through experiments or optimization algorithms (such as genetic algorithms) to ensure the rationality and effectiveness of the weight allocation. In addition, the dynamic weight allocation algorithm can not only screen out key features but also suppress the influence of noise features to a certain extent, thereby improving the accuracy and stability of subsequent modeling.

[0100] Preferably, in order to further optimize the dynamic weight allocation algorithm, an adaptive adjustment mechanism can be introduced. For example, the adjustment factor β can be dynamically adjusted according to the real-time changes of the data instead of being fixed. Specifically, when the data fluctuates greatly, the value of β can be appropriately reduced to enhance the sensitivity to high-frequency features; when the data tends to be stable, the value of β can be increased to highlight the importance of low-frequency features. In addition, the reference mean μ ref It can also be dynamically updated based on real-time data, for example using a sliding window average, to better adapt to the dynamic changes in the concentration process. As an alternative, the dynamic weight allocation algorithm can be combined with other feature selection methods, such as those based on information entropy, to further improve the accuracy and efficiency of feature selection. Through these optimizations and alternatives, the system can more flexibly respond to feature extraction requirements in different scenarios, thereby enhancing the intelligence of the entire traditional Chinese medicine concentration process.

[0101] In some embodiments, the dynamic parameter modeling module constructs a model by:

[0102] Step S41: constructing a hidden Markov chain state transition equation based on the feature sequence;

[0103] Step S42: using a sliding window mechanism to update the state transition probability matrix;

[0104] Step S43: Calculate the dynamic equation of the concentration process through a bidirectional recurrent neural network:

[0105]

[0106] Where C(t) is the real-time concentration, T(t) is the temperature, F(t) is the flow rate, and α and β are time-varying parameters.

[0107] It should be noted that the dynamic parameter modeling module is a key link in this system's implementation of refined control over the TCM concentration process. This module achieves dynamic modeling of the concentration process by constructing a hidden Markov chain state transition equation, using a sliding window mechanism to update the state transition probability matrix, and combining it with a bidirectional recurrent neural network to calculate the dynamic equations of the concentration process. The hidden Markov chain is a mathematical model used to describe system state transitions, capable of capturing the transition patterns between different states during the concentration process. The sliding window mechanism is used to dynamically update the state transition probability matrix to adapt to the dynamic changes in the concentration process. The bidirectional recurrent neural network is a neural network structure capable of processing time series data and effectively modeling the time-varying characteristics of the concentration process. Through these technical means, the dynamic parameter modeling module can provide accurate model support for the optimization control decision module, thereby achieving efficient control of the TCM concentration process.

[0108] Specifically, the implementation process of the dynamic parameter modeling module includes the following steps: First, a hidden Markov chain state transition equation is constructed based on the feature sequence. The core of this step is to divide the concentration process into several discrete states and describe the transition relationship between these states through the state transition equation. The state transition equation can be expressed as:

[0109]

[0110] Among them, s t Indicates the current state, s t-1 represents the previous state, and A is the state transition probability matrix. Secondly, a sliding window mechanism is used to update the state transition probability matrix. The size of the sliding window can be set according to the dynamic characteristics of the concentration process. For example, the window size can be set to 10 time steps to capture the short-term dynamic changes of the concentration process. Finally, a bidirectional recurrent neural network is used to calculate the dynamic equation of the concentration process:

[0111]

[0112] Where C(t) is the real-time concentration, T(t) is the temperature, F(t) is the flow rate, and k1 and k2 are time-varying parameters. These parameters are dynamically adjusted using a specific update formula to reflect the real-time changes in the concentration process.

[0113] Preferably, in order to further improve the accuracy and adaptability of the dynamic parameter modeling module, the above-mentioned operation steps can be refined or alternatives can be introduced. For example, when constructing a hidden Markov chain, the state transition probability matrix can be tuned by the Bayesian optimization method to improve the prediction accuracy of the model. The sliding window mechanism can dynamically adjust the window size according to the dynamic characteristics of the concentration process. For example, in the early stage of concentration, when the data changes rapidly, a smaller window can be set to capture transient information; and in the later stage of concentration, when the data changes slowly, the window can be appropriately increased to smooth the noise. For a bidirectional recurrent neural network, a long short-term memory network (LSTM) or a gated recurrent unit (GRU) can be used as a network structure to enhance the memory capacity and anti-forgetting characteristics of time series data. In addition, the update formula of the time-varying parameters can be optimized by introducing a machine learning algorithm (such as the gradient descent method) to further improve the efficiency and accuracy of the parameter update. Through these optimization measures, the dynamic parameter modeling module can more accurately reflect the dynamic characteristics of the Chinese medicine concentration process and provide more reliable model support for subsequent optimization control.

[0114] In some embodiments, the time-varying parameter is updated by the following formula:

[0115]

[0116] Among them, α0 is the initial parameter, γ is the learning rate, τ is the integral variable, and L is the length of the backtracking time;

[0117] β(t)=β0·exp(-η·|C(t)-C target |)

[0118] Among them, β0 is the initial parameter, η is the adjustment factor, C target is the target concentration.

[0119] It should be noted that the update mechanism for time-varying parameters is a key technology in the dynamic parameter modeling module, used to adapt to the dynamic changes in the traditional Chinese medicine concentration process. These time-varying parameters can be dynamically adjusted based on real-time data from the concentration process, ensuring the accuracy and adaptability of the model. Specifically, the update formula for these time-varying parameters consists of two parts: one for updating temperature-related parameters and the other for updating concentration-related parameters. By introducing parameters such as the learning rate, adjustment factor, and lookback time length, these formulas can dynamically adjust model parameters based on real-time data to better reflect the dynamic characteristics of the concentration process. This dynamic update mechanism enables the system to maintain efficient and stable operation in complex production environments.

[0120] Specifically, the update formula for time-varying parameters is as follows: For temperature-related parameters, the update formula is

[0121]

[0122] Among them, k 10 is the initial parameter, η is the learning rate, τ is the backtracking time length, and L is the loss function. The learning rate η determines the step size of the parameter update and is usually set to a small positive number, such as 0.01. The backtracking time length τ represents the time range considered when updating the parameters. It can be set according to the dynamic characteristics of the concentration process, such as taking 10 time steps. For parameters related to concentration, the update formula is

[0123] k2(t)=k 20 ·exp(-γ·|C(t)-C target |)

[0124] Among them, k 20 is the initial parameter, γ is the adjustment factor, C(t) is the real-time concentration, C target is the target concentration. The adjustment factor γ controls the parameter's sensitivity to concentration deviations and is typically adjusted based on experimental data. This update mechanism dynamically adjusts parameters based on the deviation between the real-time concentration and the target concentration, ensuring the stability and efficiency of the concentration process.

[0125] Preferably, in order to further optimize the update mechanism of time-varying parameters, the above-mentioned operation steps can be refined or alternatives can be introduced. For example, the learning rate η can adopt an adaptive adjustment strategy, such as dynamically adjusting the learning rate according to the gradient change of the loss function to improve the efficiency and stability of parameter updates. The backtracking time length τ can be dynamically adjusted according to the dynamic characteristics of the concentration process. For example, in the early stage of concentration, when the data changes rapidly, a shorter backtracking time length can be set to quickly respond to changes; and in the later stage of concentration, when the data changes slowly, the backtracking time length can be appropriately increased to smooth the noise. For the adjustment factor γ, a machine learning algorithm (such as a genetic algorithm or Bayesian optimization) can be introduced for optimization to ensure its adaptability under different working conditions. In addition, a regularization term can be introduced into the loss function to prevent overfitting during the parameter update process. Through these optimization measures, the update mechanism of time-varying parameters can more accurately reflect the dynamic characteristics of the traditional Chinese medicine concentration process, thereby providing more reliable model support for the optimization control of the system.

[0126] In some embodiments, the optimization control decision module includes:

[0127] The objective function building block defines the optimization goal of the concentration process:

[0128]

[0129] Among them, J is the objective function value, T is the total number of time steps, ω1, ω2, ω3 are weight factors, C(t) is the real-time concentration, C targetis the target concentration, ΔT(t) is the temperature change rate;

[0130] Constraint generation unit to set the physical boundaries of temperature and flow rate;

[0131] The dynamic programming solving unit uses an improved tabu search algorithm to solve the optimal control sequence.

[0132] It should be noted that the optimization control decision module is the core component of this system for achieving efficient optimization of the traditional Chinese medicine concentration process. This module achieves refined control of the concentration process by constructing an objective function, generating constraints, and using an improved tabu search algorithm to solve the optimal control sequence. The objective function is used to quantify the optimization goals of the concentration process, such as minimizing energy consumption, maximizing concentration efficiency, or ensuring product quality stability; the constraints define the feasible range of physical parameters (such as temperature and flow rate) in the concentration process; the improved tabu search algorithm is a heuristic optimization algorithm that avoids falling into local optimal solutions by introducing tabu tables and dynamic adjustments of neighborhood solutions during the search process, thereby efficiently solving the global optimal control strategy. Through these technical means, the optimization control decision module can provide scientific and reasonable control instructions for the traditional Chinese medicine concentration process, ensuring the efficiency and stability of the production process.

[0133] Specifically, the objective function construction unit of the optimization control decision module defines the optimization objective of the concentration process, which is in the form of

[0134]

[0135] Among them, J is the objective function value, T is the total number of time steps, w1, w2, w3 are weight factors, C(t) is the real-time concentration, C target is the target concentration, T′(t) is the temperature change rate, and F(t) is the flow rate. The weight factor is used to balance the priorities of different optimization objectives. For example, if you pay more attention to the precise control of concentration, you can increase the value of w1. The constraint generation unit defines the boundary conditions of physical parameters such as temperature and flow rate. For example, the temperature range can be set to 60°C to 100°C, and the flow rate range can be set to 0.5L / min to 2L / min. The improved tabu search algorithm stores the hash code values of historical solutions in the taboo table and introduces Gaussian perturbation terms and simulated annealing mechanisms to dynamically adjust the acceptance probability threshold during the search process, thereby avoiding local optimal solutions and improving global search capabilities.

[0136] Preferably, in order to further improve the performance of the optimization control decision module, the above-mentioned operation steps can be refined or alternatives can be introduced. For example, the weight factors w1, w2, and w3 can be optimized and adjusted by experimental data or machine learning algorithms (such as genetic algorithms) to ensure the adaptability of the objective function under different working conditions. For the constraints, a dynamic constraint adjustment mechanism can be introduced to dynamically adjust the boundary range of the physical parameters according to the real-time state of the concentration process, such as appropriately relaxing the temperature range in the early stage of concentration to speed up the concentration speed, and tightening the temperature range in the later stage of concentration to ensure product quality. For the improved taboo search algorithm, a variety of heuristic strategies can be introduced, such as particle swarm optimization (PSO) or ant colony algorithm, combined with the taboo search algorithm, to further improve the solution efficiency and global optimization ability. In addition, a multi-objective optimization method can also be introduced to consider multiple optimization targets such as energy consumption, efficiency and product quality at the same time, thereby providing a more comprehensive and efficient optimization control strategy for the traditional Chinese medicine concentration process.

[0137] In some embodiments, the improved tabu search algorithm includes the following operations:

[0138] Sub-step S531: Introduce Gaussian perturbation term ∈~N(0,σ(t)) when generating neighborhood solution 2 ), where σ(t)=σ0·e -kt ;

[0139] Where ∈ is the Gaussian perturbation term, σ(t) is the time-dependent standard deviation, σ0 is the initial standard deviation, and k is the attenuation coefficient;

[0140] Sub-step S532: storing the hash code value of the historical solution in the taboo table;

[0141] Sub-step S533: Dynamically adjust the acceptance probability threshold using a simulated annealing mechanism.

[0142] It should be noted that the improved tabu search algorithm is a key technology for efficiently solving optimal control sequences in the optimization control decision module. By introducing a Gaussian perturbation term, a hash code storage mechanism for the tabu table, and a simulated annealing mechanism, the algorithm dynamically adjusts the acceptance probability threshold during the search process, thereby avoiding local optima while improving global search capabilities. The Gaussian perturbation term introduces randomness when generating neighborhood solutions, increasing search diversity; the tabu table stores hash codes of historical solutions to record explored solution spaces and avoid repeated searches; and the simulated annealing mechanism dynamically adjusts the acceptance probability threshold, allowing for acceptance of inferior solutions with a certain probability, thereby escaping local optima. These improvements make the tabu search algorithm more efficient and robust in complex optimization problems.

[0143] Specifically, the improved tabu search algorithm includes the following key steps: First, a Gaussian perturbation term is introduced when generating neighborhood solutions, which is in the form of

[0144] ∈N(0,σ(t) 2 )

[0145] Among them, ∈ is the Gaussian disturbance term, σ(t) is the time-related standard deviation, and its expression is

[0146] σ(t)=σ0·e -kt

[0147] σ0 is the initial standard deviation, and k is the decay coefficient, which is used to control the decay rate of the standard deviation over time. The initial standard deviation σ0 can be set according to the scale and complexity of the problem, such as 0.1 or 0.2, and the decay coefficient κ can be adjusted according to the convergence speed of the optimization process, usually around 0.01. Secondly, the hash code values of historical solutions are stored in the taboo table. The hash code can be used to efficiently record and retrieve the explored solution space to avoid repeated searches. Finally, the simulated annealing mechanism is used to dynamically adjust the acceptance probability threshold, allowing poor solutions to be accepted with a certain probability, thereby increasing the diversity of the search and avoiding local optimality. The acceptance probability of the simulated annealing mechanism can be calculated by the Metropolis criterion, which is in the form of

[0148]

[0149] Where ΔE is the cost change of the solution, k B is the Boltzmann constant, and T is the temperature parameter, which is used to control the size of the acceptance probability.

[0150] Preferably, in order to further improve the performance of the improved tabu search algorithm, the above-mentioned operation steps can be refined or alternatives can be introduced. For example, the initial standard deviation σ0 of the Gaussian perturbation term can be dynamically adjusted according to the complexity of the problem. For complex optimization problems, σ0 can be appropriately increased to increase the diversity of the search; for simple optimization problems, σ0 can be reduced to speed up the convergence rate. The decay coefficient κ can be optimized through experiments or machine learning methods to ensure that the decay rate of the standard deviation matches the convergence rate of the optimization process. For the storage mechanism of the tabu table, an adaptive adjustment strategy can be introduced to dynamically adjust the length of the tabu table according to the progress of the search process to avoid falling into the taboo state too early or reducing the search efficiency due to the excessive length of the tabu table.

[0151] Furthermore, the temperature parameter T in the simulated annealing mechanism can adopt an adaptive cooling strategy. For example, the temperature can be dynamically adjusted based on the cost of the current solution. This maintains a high acceptance probability in the early stages of optimization to explore the solution space, while lowering the acceptance probability in the later stages of optimization to improve search accuracy. Through these optimization measures, the improved tabu search algorithm can demonstrate greater efficiency and robustness in optimizing the complex traditional Chinese medicine concentration process, providing a better control strategy for the system.

[0152] In some embodiments, the real-time feedback adjustment module includes:

[0153] Error calculation unit, real-time comparison of the deviation between actual concentration and model prediction value;

[0154] The parameter adjustment unit dynamically modifies the control instructions according to the following formula:

[0155]

[0156] Among them, F adj (t) is the adjusted control instruction, F(t) is the original control instruction, K p , K i is the proportional integral coefficient, e(t) is the concentration error at time t, and m is the length of the integration window;

[0157] The stability detection unit triggers model reconstruction when the error does not decrease after n consecutive adjustments.

[0158] It should be noted that the real-time feedback control module is a crucial component of this system's stable operation of the TCM concentration process. This module uses an error calculation unit to compare the deviation between the actual concentration and the model's predicted value in real time and dynamically adjusts control instructions based on this deviation to ensure the stability of the concentration process. The stability detection unit monitors the change in error after multiple consecutive adjustments. If the error persists, it triggers a model reconstruction, thereby ensuring the long-term stability and adaptability of the system. This module's design enables the system to adjust control strategies in real time within a dynamically changing production environment, ensuring the efficient and stable operation of the concentration process.

[0159] Specifically, the core of the real-time feedback regulation module lies in the error calculation unit and the parameter adjustment unit. The error calculation unit calculates the concentration error by comparing the deviation between the actual concentration and the model prediction value.

[0160] ∈(t)=C actual (t)-C predicted (t)

[0161] Among them, C actual (t) is the actual measured concentration, C predicted (t) is the concentration predicted by the model. The parameter adjustment unit dynamically corrects the control instruction according to the concentration error, and its formula is

[0162]

[0163] Among them, u(t) is the original control instruction, K p and K i They are proportional coefficient and integral coefficient respectively, which are used to adjust the dynamic response of the control instruction.p Determines the direct impact of the error on the control command and is usually set to a small positive number, such as 0.1; the integral coefficient K i It is used to eliminate the cumulative effect of errors and is usually set in the range of about 0.01. The stability detection unit monitors the change in error after n consecutive adjustments. If the error does not decrease, it triggers model reconstruction to ensure the long-term stability of the system.

[0164] Preferably, in order to further improve the performance of the real-time feedback adjustment module, the above operation steps can be refined or alternative solutions can be introduced. For example, the proportional coefficient K p and the integral coefficient K i It can be optimized by adaptive adjustment strategy, such as dynamically adjusting K according to the size of the error p and K i When the error is large, K can be appropriately increased. p To speed up the response; when the error is small, K can be reduced p To improve control accuracy. Integral coefficient K i It can be adjusted dynamically according to the accumulated error to avoid system oscillation caused by excessive integration.

[0165] Furthermore, the trigger conditions in the stability detection unit can be adjusted based on actual production needs. For example, a threshold for error change can be set, triggering model reconstruction only when the error change is less than this threshold, thereby reducing unnecessary model updates. As an alternative, a fuzzy controller or a neural network controller can be introduced to further improve the system's adaptability and dynamic response capabilities. Through these optimization measures, the real-time feedback regulation module can operate more efficiently in complex production environments, ensuring the stability and efficiency of the traditional Chinese medicine concentration process.

[0166] In some embodiments, the proportional-integral coefficient is adaptively adjusted by the following formula:

[0167]

[0168] Among them, K p0 is the initial proportional coefficient, ζ is the adjustment constant;

[0169] K i (t) = K i0 ·tanh(ξ·dτ)

[0170] Among them, K i0 is the initial integral coefficient, ξ is the adjustment constant, and T0 is the integral time constant.

[0171] It should be noted that the adaptive adjustment mechanism of the proportional-integral coefficient is a key technology used in the real-time feedback control module to optimize the control command adjustment process. By dynamically adjusting the proportional and integral coefficients, the system can flexibly adjust the response characteristics of the control command based on the magnitude and changing trend of the real-time error, thereby improving the system's stability and adaptability. The proportional coefficient is used to directly adjust the impact of the error on the control command, while the integral coefficient is used to eliminate the cumulative effect of the error. This adaptive adjustment mechanism enables the system to automatically optimize the control strategy under different operating conditions, ensuring the efficient and stable operation of the traditional Chinese medicine concentration process.

[0172] Specifically, the adaptive adjustment formula of the proportional integral coefficient is:

[0173]

[0174] K i (t) = K i0 tanh(μ∈(t))

[0175] Among them, K p0 and K i0 are the initial proportional coefficient and initial integral coefficient respectively, λ and μ are adjustment constants, and ∈(t) is the concentration error at time t. Initial proportional coefficient K p0 It is usually set according to the initial response characteristics of the system, for example, it is set to about 0.1 to ensure that the system has an appropriate response speed in the initial stage. The adjustment constant λ is used to control the sensitivity of the proportional coefficient to the change of error, and its value is usually between 0.1 and 1. The initial integral coefficient K i0 The adjustment constant μ is set based on the system's steady-state error characteristics, for example, around 0.01 to eliminate the cumulative effect of errors. The adjustment constant μ controls the magnitude of the integral coefficient's change with error, and its value is typically between 0.01 and 0.1. This dynamic adjustment mechanism enables the system to respond quickly to large errors and fine-tune when they are small, thereby achieving precise control of the traditional Chinese medicine concentration process.

[0176] Preferably, in order to further optimize the adaptive adjustment mechanism of the proportional integral coefficient, the above operation steps can be refined or alternative solutions can be introduced. For example, the initial proportional coefficient K p0 and the initial integration coefficient K i0 Optimization and adjustment can be performed using experimental data or machine learning methods to ensure the system's adaptability under different operating conditions. The control constants λ and μ can be adaptively adjusted based on the system's dynamic characteristics. For example, if the system responds too quickly, λ can be appropriately increased to suppress oscillations, while if the system responds too slowly, λ can be appropriately reduced to speed up the response.

[0177] Furthermore, a multi-parameter adaptive adjustment mechanism, such as simultaneous adjustment of the proportional, integral, and differential (PID) coefficients, can be introduced to further improve the system's control performance. As an alternative, data-driven adaptive control methods can be introduced, such as using neural networks or fuzzy logic controllers to dynamically adjust the proportional-integral coefficients based on real-time data, thereby achieving a more flexible control strategy. Through these optimization measures, the adaptive adjustment mechanism of the proportional-integral coefficients can better adapt to the dynamic changes in the traditional Chinese medicine concentration process, ensuring the efficient and stable operation of the system.

[0178] In some embodiments, the result output module includes:

[0179] 3D visualization unit, synchronously displaying concentration field, temperature gradient and flow velocity distribution;

[0180] Parameter storage unit, recording optimized control instructions and actual response data by timestamp;

[0181] The control signal generation unit converts discrete instructions into analog signal sequences that can be executed by the device.

[0182] It should be noted that the result output module is a key component of the system for displaying and recording the optimization control process and results. It uses a three-dimensional visualization unit to simultaneously display the concentration field, temperature gradient, and flow velocity distribution, providing operators with intuitive information about the production process. The parameter storage unit records optimization control instructions and actual response data by timestamp, facilitating subsequent analysis and traceability. The control signal generation unit converts discrete instructions into analog signal sequences executable by the equipment, enabling precise control of the concentration equipment. This module's design enables the system to not only monitor and optimize the concentration process in real time but also provide support for subsequent process improvements and data analysis, ensuring the efficiency and traceability of the TCM concentration process.

[0183] Specifically, the three-dimensional visualization unit of the result output module displays the distribution of key parameters in the concentration process through a graphical interface. The concentration field shows the change in solute concentration in space, the temperature gradient reflects the heat distribution during the heating process, and the flow rate distribution shows the flow of materials in the pipeline. These visualization information are implemented through software tools (such as MATLAB or Python's Matplotlib library), which can help operators quickly understand the dynamic changes in the concentration process. The parameter storage unit records the optimized control instructions and actual response data through a database or log file, and the timestamp is used to mark the time sequence of the data for subsequent analysis. The control signal generation unit converts the optimized discrete control instructions into analog signals, such as converting digital signals into voltage or current signals through a digital-to-analog converter (DAC) to drive the actuator of the concentration equipment. The implementation of these functions ensures the accuracy and reliability of the system output, and also provides data support for the optimization of the production process.

[0184] Preferably, in order to further enhance the functionality and user experience of the result output module, the above-mentioned operation steps can be refined or alternatives can be introduced. For example, the three-dimensional visualization unit can introduce interactive functions, allowing operators to gain in-depth understanding of the details of the concentration process by zooming, rotating, and selecting specific areas. The parameter storage unit can adopt a distributed database or cloud storage technology to improve the reliability and access speed of data storage, while supporting multi-user access and data sharing. The control signal generation unit can introduce a redundant design, such as ensuring the reliability of the signal through dual-channel output, or introducing a fault detection and switching mechanism to deal with equipment failure or signal interference.

[0185] Furthermore, intelligent diagnostic functionality can be introduced to automatically analyze abnormalities in the production process based on stored data and provide improvement suggestions. Through these optimization measures, the result output module can provide more comprehensive support for the traditional Chinese medicine concentration process, improving the overall system performance and user experience.

[0186] The above-mentioned embodiments of the present invention have the following beneficial effects: the present invention obtains the temperature, concentration and flow rate data of the traditional Chinese medicine concentration process in real time through the data acquisition module, and uses the data preprocessing module to perform noise filtering and standardization processing, which can effectively remove the interference in the data and provide high-quality input for subsequent analysis. The multi-dimensional feature extraction module adopts a customized multi-scale weighted fusion algorithm and a dynamic weight allocation algorithm, which can accurately extract the joint features of the time and frequency domains and screen key features, thereby providing more representative data support for the dynamic parameter modeling module. The dynamic parameter modeling module constructs a time-varying parameter concentration model based on a hidden Markov chain and a bidirectional recurrent neural network. Combined with the objective function of the optimization control decision module and the improved taboo search algorithm, it can efficiently solve the optimal control sequence, realize the refined regulation of the concentration process, and improve the concentration efficiency and product quality stability. The real-time feedback adjustment module can adjust the concentration equipment parameters in real time through error calculation and dynamic correction control instructions, combined with the stability detection unit, to ensure the stability and adaptability of the process and further enhance the robustness of the system.

[0187] In addition, the result output module of the present invention synchronously displays the concentration field, temperature gradient and flow rate distribution through a three-dimensional visualization unit, providing operators with intuitive production process information, facilitating real-time monitoring and decision-making. The parameter storage unit records the optimization control instructions and actual response data by timestamp, which is convenient for subsequent analysis and tracing, and provides data support for the continuous optimization of the production process. The adaptive adjustment mechanism of the proportional integral coefficient can further optimize the generation of control instructions, ensuring that the system can maintain efficient operation under different working conditions. Overall, the various modules of the present invention work together to improve the intelligence level of the traditional Chinese medicine concentration process, reduce energy consumption and manual intervention costs, and promote the modernization of traditional Chinese medicine production.

[0188] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a terminal device such as a computer, a server, a mobile phone, or a tablet.

[0189] The above descriptions are merely some preferred embodiments of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also encompass other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept. For example, a technical solution formed by mutually replacing the above-mentioned features with (but not limited to) technical features having similar functions disclosed in the embodiments of the present invention.

Claims

1. An intelligent data analysis system for traditional Chinese medicine concentration, characterized in that: Includes the following modules: Data acquisition module, data preprocessing module, multi-dimensional feature extraction module, dynamic parameter modeling module, optimization control decision module, real-time feedback adjustment module and result output module; The data acquisition module is used to collect temperature, concentration and flow rate data in real time during the concentration process of traditional Chinese medicine; The data preprocessing module performs noise filtering and standardization on the output of the data acquisition module; The multi-dimensional feature extraction module extracts time-frequency domain joint features based on the output of the data preprocessing module; The dynamic parameter modeling module constructs a time-varying parameter concentration model based on the features of the multi-dimensional feature extraction module; The optimization control decision module generates control instructions based on the model of the dynamic parameter modeling module; The real-time feedback adjustment module adjusts the concentration equipment parameters according to the instructions of the optimization control decision module; The result output module outputs control signals and visual analysis results.

2. The system according to claim 1, characterized in that The multi-dimensional feature extraction module includes: The time-frequency decomposition unit uses a custom multi-scale weighted fusion algorithm to decompose the preprocessed data; Feature fusion unit, which performs nonlinear correlation calculation on the decomposed sub-signals; Dimensionality reduction unit, which screens key features through a dynamic weight allocation algorithm; The formula of the custom multi-scale weighted fusion algorithm is: Among them, S(t) is the fused feature sequence, w k is the dynamic weight of the k-th scale, x(τ) is the preprocessed data, φ k is a custom basis function, Δ is the time window length, and N is the total number of scales.

3. The system according to claim 2, characterized in that The dynamic weight allocation algorithm satisfies: in, is the variance of the k-th scale data, μ k is the k-th scale mean, μ ref is the reference mean, λ is the adjustment factor, w k is a dynamic weight.

4. The system according to claim 3, characterized in that The dynamic parameter modeling module builds the model through the following steps: Step S41: constructing a hidden Markov chain state transition equation based on the feature sequence; Step S42: using a sliding window mechanism to update the state transition probability matrix; Step S43: Calculate the dynamic equation of the concentration process through a bidirectional recurrent neural network: Where C(t) is the real-time concentration, T(t) is the temperature, F(t) is the flow rate, and α and β are time-varying parameters.

5. The system according to claim 4, characterized in that The time-varying parameters are updated by the following formula: Among them, α0 is the initial parameter, γ is the learning rate, τ is the integral variable, L is the length of the backtracking time, and α(T) is the time-varying parameter β(t)=β0·exp(-η·|C(t)-C target |) Among them, β0 is the initial parameter, η is the adjustment factor, C target is the target concentration.

6. The system according to claim 5, characterized in that The optimization control decision module includes: The objective function building block defines the optimization goal of the concentration process: Among them, J is the objective function value, T is the total number of time steps, ω1, ω2, ω3 are weight factors, C(t) is the real-time concentration, C target is the target concentration, ΔT(t) is the temperature change rate; Constraint generation unit to set the physical boundaries of temperature and flow rate; The dynamic programming solving unit uses an improved tabu search algorithm to solve the optimal control sequence.

7. The system according to claim 6, characterized in that The improved tabu search algorithm includes the following operations: Sub-step S531: Introduce Gaussian perturbation term ∈~N(0,σ(t)) when generating neighborhood solution 2 ), where σ(t)=σ0·e -kt ; Where ∈ is the Gaussian perturbation term, σ(t) is the time-dependent standard deviation, σ0 is the initial standard deviation, and k is the attenuation coefficient; Sub-step S532: storing the hash code value of the historical solution in the taboo table; Sub-step S533: Dynamically adjust the acceptance probability threshold using a simulated annealing mechanism.

8. The system according to claim 7, characterized in that The real-time feedback adjustment module includes: Error calculation unit, real-time comparison of the deviation between actual concentration and model prediction value; The parameter adjustment unit dynamically modifies the control instructions according to the following formula: Among them, F adj (t) is the adjusted control instruction, F(t) is the original control instruction, K p , K i is the proportional integral coefficient, e(t) is the concentration error at time t, and m is the length of the integration window; The stability detection unit triggers model reconstruction when the error does not decrease after n consecutive adjustments.

9. The system according to claim 8, characterized in that The proportional integral coefficient is adaptively adjusted by the following formula: Among them, K p0 is the initial proportional coefficient, ζ is the adjustment constant; K i (t)=K i0 ·tanh(ξ·dτ) Among them, K i0 is the initial integral coefficient, ξ is the adjustment constant, and T0 is the integral time constant.

10. The system according to claim 9, characterized in that The result output module includes: 3D visualization unit, synchronously displaying concentration field, temperature gradient and flow velocity distribution; Parameter storage unit, recording optimized control instructions and actual response data by timestamp; The control signal generation unit converts discrete instructions into analog signal sequences that can be executed by the device.

Images