Real-time collection and optimization system for durian peel extraction process parameters

Abstract

The present application relates to the technical field of intelligent control of natural product extraction, in particular to a real-time collection and optimization system for durian peel extraction process parameters. It comprises a multi-dimensional sensing collection unit, a data transmission preprocessing unit, an intelligent parameter optimization unit, a closed-loop execution control unit, and a feedback verification and working condition control unit. The present application adopts a same-source double-response differential technique before and after feeding, which can online decouple and identify the characteristics of raw material batches and the drift of equipment reference, rely on material coupling batch fingerprint to correct process mapping relationship, improve the parameter adaptation ability of different raw material batches, and reduce the fluctuation of extraction indicators between batches. At the same time, through the material additional time lag, the process sequence time lag is aligned and reconstructed, and combined with the credible adjustment domain of multi-factor dynamic calculation, adaptive optimization is carried out, which not only avoids adjustment overshoot, but also prevents the optimization amplitude from being conservative, improves the parameter adjustment precision and production line operation efficiency, and adapts to the real-time optimization demand of complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for natural product extraction, and more specifically, to a system for real-time acquisition and optimization of process parameters for durian peel extraction. Background Technology

[0002] High-value utilization of by-products is an important development direction for the deep processing industry of agricultural products. Extraction of active ingredients from durian peel is a key link in extending the durian industrial chain and increasing product added value. Currently, extraction production lines generally adopt a common hardware architecture of "sensor acquisition - industrial control computing - actuator". Precise control of process parameters is the core requirement to ensure extraction efficiency, stabilize product quality, and control production energy consumption.

[0003] Currently, the mainstream technical solutions for large-scale applications in the industry can be divided into two categories. One category is a single-loop closed-loop control solution with fixed process parameters. This solution uses a programmable logic controller (PLC) as the control core, collecting basic operating data such as temperature, liquid level, and extraction time through field sensors, and performing routine adjustments according to preset process thresholds. This type of solution has a mature architecture, stable operation, and low deployment cost, making it a common configuration for small and medium-sized extraction production lines. The other category is a batch offline correction parameter adjustment solution. This solution is based on a basic automatic control system with an offline quality inspection process. Process personnel manually adjust the process parameters for the next batch based on the quality inspection results of each batch of finished products. This type of solution can adapt to fluctuations in raw material properties to a certain extent, has a low implementation threshold, and is mostly used in extraction production lines with multiple varieties and small batches.

[0004] However, existing solutions still suffer from common technical bottlenecks that restrict the level of refined control over the extraction process. Firstly, due to the combined effects of differences in raw material origin and maturity, as well as the aging of equipment over long-term operation, batch-to-batch fluctuations in raw material characteristics and equipment baseline drift often overlap, making it impossible to independently identify these two factors online. This results in the mapping model between process parameters and extraction indicators being difficult to adapt to batch differences, easily causing fluctuations in extraction yield and active ingredient content. Secondly, there are inherent lags in material wetting and active ingredient mass transfer during the extraction process, and the duration of these lags dynamically changes with the amount of material fed and the state of the material. Existing parameter optimization lacks a dynamic and reliable constraint mechanism, making the adjustment process prone to overshoot oscillations, or the optimization range being too conservative to fully utilize production line efficiency. These problems directly lead to poor batch-to-batch product quality consistency and insufficient energy consumption control precision, with particularly limited adaptability in production scenarios where raw material properties fluctuate significantly. Therefore, we propose a real-time acquisition and optimization system for durian peel extraction process parameters. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time acquisition and optimization system for durian peel extraction process parameters, in order to solve the problems mentioned in the background technology, such as the difficulty in independently identifying the coupling between raw material batch characteristics and equipment baseline drift, the dynamic changes in extraction process time lag, and the lack of dynamic and reliable constraints for parameter optimization.

[0006] To address the aforementioned technical problems, the present invention aims to provide a real-time acquisition and optimization system for durian peel extraction process parameters, comprising: A multi-dimensional sensing and acquisition unit is used to acquire on-site physical quantity signals during the durian peel extraction process, and to synchronously acquire and quantify the extraction process parameters and dynamic response data to obtain the original process data. The data transmission preprocessing unit receives the raw process data and performs noise reduction, standardization, and time synchronization processing on the raw process data to obtain real-time operating condition data. The intelligent parameter optimization unit receives the real-time operating condition data and the feedback deviation data output by the feedback verification and operating condition control unit. It employs a batch adaptive optimization technique that decouples equipment status from raw material characteristics to dynamically calibrate and optimize the mapping relationship between batch operating condition characteristics and process parameters in the extraction process, generating optimal process parameter instructions. A closed-loop execution control unit is used to receive the optimal process parameter instruction, perform dynamic adjustment and limited adjustment tests on the operating parameters of the extraction equipment, and output execution status feedback data. The feedback verification and operating condition control unit is used to receive the execution status feedback data, real-time operating condition data and externally input batch offline quality data, monitor and verify the parameter adjustment effect, equipment operating status and batch quality and determine anomalies, and generate feedback deviation data and system operation control signals.

[0007] As a further improvement to this technical solution, the multi-dimensional sensing acquisition unit includes a synchronous acquisition control module, a multi-dimensional sensing array module, and a signal quantization output module, wherein: The synchronous acquisition control module generates a synchronous acquisition trigger signal based on a preset sampling period to control the sampling time of all acquisition channels to remain synchronized; The multi-dimensional sensor array module senses the on-site physical quantity signals of the durian peel extraction process based on the synchronous acquisition trigger signal, and collects the analog sensor signals corresponding to the extraction process parameters and dynamic response data. The signal quantization output module performs analog-to-quantization conversion based on analog sensing signals and marks them with a unified time identifier to generate raw process data.

[0008] As a further improvement to this technical solution, the data transmission preprocessing unit includes a data receiving buffer module, a noise reduction filtering module, and a standardization synchronization module, wherein: The data receiving and caching module receives the original process data and completes data temporary storage and transmission validity verification. The noise reduction and filtering module filters out interference noise introduced during the acquisition and transmission process based on the original process data output by the data receiving and buffering module, and obtains the noise-reduced process data. The standardization and synchronization module performs range standardization and time axis synchronization adjustment based on the noise-reduced process data to generate real-time operating condition data.

[0009] As a further improvement to this technical solution, the intelligent parameter optimization unit includes a batch decoupling characterization module, a time delay alignment module, a process mapping correction module, and a trusted domain optimization module, wherein: The batch decoupling characterization module adopts the technique of dual response difference before and after feeding, and extracts the equipment reference drift characteristics and raw material batch characteristics in the process based on real-time working condition data, and outputs the material coupling batch fingerprint, equipment reference drift factor and material additional effect time delay. The time delay alignment module uses a time axis translation calibration technique based on the material's additional time delay to reconstruct the process parameter sequence through time delay alignment. The process mapping correction module uses a batch fingerprint-based feature conditional modulation technique to perform batch adaptation correction on the mapping relationship between process parameters and extracted indicators. The trusted domain optimization module employs a batch adaptive optimization technique with multi-factor dynamic constraints. Based on the aligned process parameter sequence, the mapping relationship between the corrected process parameters and the extracted indicators, the equipment baseline drift factor, and the feedback deviation data, it solves for the optimal combination of process parameters and generates the optimal process parameter instruction.

[0010] As a further improvement to this technical solution, the batch decoupling characterization module includes a response acquisition and calibration submodule, a material characteristic decoupling submodule, and an equipment drift calculation submodule, wherein: The response acquisition and calibration submodule is used to extract the equipment-medium reference response under steady-state condition before feeding and the equipment-medium-raw material coupling response under steady-state condition after feeding from the input real-time operating data. The two responses are then normalized to obtain the reference response value and coupling response value corresponding to the unit adjustment action of the j-th adjustment channel. At the same time, the action delay corresponding to each of the two responses is extracted. The material characteristic decoupling submodule calculates the material additional response corresponding to the unit solid load and constructs the material coupled batch fingerprint based on the baseline response value, coupled response value and the current batch of durian peel solid feed amount of the unit adjustment action; calculates the difference in action time delay between the two responses to obtain the material additional action time delay of each adjustment channel; calculates the raw material batch unfamiliarity based on the material coupled batch fingerprint and the calibrated set of historical qualified batch fingerprints; and generates the time delay identification credibility based on the confidence level of the response curve fitting. The additional material response per unit solid load is calculated using the following formula: In the formula, For the first Batch No. Additional response of material per unit solid load in each regulating channel; For the first Normalized unit action response value of the baseline response before batch feeding; For the first The normalized unit action response value of the coupled response after batch feeding; For the first Batch solid durian peel feed quantity; This is a very small positive constant whose value is much smaller than the amount of solid durian peel used in the current batch, used to avoid division by zero anomalies; subscript For batch dimension identification, subscript To adjust the channel dimension identifier; the device drift calculation submodule is used to compare the current batch reference response with the calibrated historical device reference response to generate a device reference drift factor.

[0011] As a further improvement to this technical solution, the process of reconstructing the time-delay alignment of the process parameter sequence of the time-delay alignment module includes the following steps: S32.1 Extract the material additional effect time delay corresponding to each adjustment channel from the output results of the batch decoupling characterization module; S32.2. Based on the time of the adjustment action, the historical process parameter sequence is shifted along the time axis to the corresponding channel's material additional action time delay, so that the control action and response effect correspond one-to-one in the time dimension, and the time delay aligned process parameter sequence is obtained.

[0012] As a further improvement to this technical solution, the batch adaptation and correction process of the mapping relationship between process parameters and extracted indicators in the process mapping correction module includes the following steps: S33.1 Based on the material coupled batch fingerprint, generate the feature gain coefficient and feature offset coefficient of the corresponding dimension through the preset linear mapping rule; S33.2 Using the aforementioned feature gain coefficient and feature offset coefficient, the intermediate feature quantity of the mapping model corresponding to the mapping relationship between process parameters and extracted indicators is conditionally modulated dimension by dimension to obtain batch-adaptive feature quantity. S33.3 Based on the corrected batch adaptation feature, output the mapping relationship between the process parameters and extraction indicators adapted to the current raw material batch.

[0013] As a further improvement to this technical solution, the trusted domain optimization module includes a trusted domain dynamic calculation submodule and a multi-objective constraint optimization submodule, wherein: The trusted domain dynamic calculation submodule is used to calculate the trusted adjustment domain scaling coefficient based on the unfamiliarity of the raw material batch, the equipment reference drift factor, and the recent prediction error and time lag identification trusted level within a preset statistical window, and to dynamically adjust the allowable fluctuation range of the optimization parameters; the trusted adjustment domain scaling coefficient is calculated by the following formula: In the formula, for The scaling factor of the reliable adjustment domain at any given time; The benchmark reliable adjustment domain coefficients; To assess the reliability of time delay identification, the subscript... Used as a time delay dimension identifier; for The unfamiliarity of each batch of raw materials; For the first Batch equipment reference drift factor; for The recent prediction error within the preset statistical window corresponding to the time point is obtained from the feedback deviation data statistics. These are preset positive weighting coefficients; the aforementioned confidence level, unfamiliarity level, drift factor, and prediction error are all normalized dimensionless parameters, with values ​​ranging from [value range missing]. ; Subscript For time dimension identification, subscript For batch dimension identification; The multi-objective constraint optimization submodule is used to solve for the optimal combination of process parameters and generate the optimal process parameter instruction within the adjusted credible adjustment domain and equipment safety boundary, based on the mapping relationship between the corrected process parameters and the extracted indicators.

[0014] As a further improvement to this technical solution, the closed-loop execution control unit includes an instruction parsing and adaptation module, a dynamic adjustment execution module, a restricted test control module, and an execution status feedback module, wherein: The instruction parsing and adaptation module receives the optimal process parameter instruction, parses it, and converts it into control parameters that can be identified and executed by each adjustment channel of the extraction equipment. The dynamic adjustment execution module performs real-time closed-loop adjustment of the operating parameters of the extraction equipment based on the control parameters, so that the corresponding process parameters track to the target value; The restricted test control module is used to issue test adjustment commands in the preset steady-state operating condition stage, control the equipment to perform small-amplitude restricted adjustment tests and collect corresponding dynamic response data; The execution status feedback module collects and summarizes the adjustment execution status, real-time operating parameters of the equipment, and limited test response data, generates execution status feedback data, and outputs it.

[0015] As a further improvement to this technical solution, the feedback verification and operating condition control unit includes a multi-source data alignment module, a regulation effect verification module, an operating condition quality judgment module, and a control signal output module, wherein: The multi-source data alignment module receives execution status feedback data, real-time operating condition data, and externally input batch offline quality data. Based on a unified time reference, it completes time sequence alignment and data dimension matching, and outputs the aligned multi-source verification dataset. The adjustment effect verification module compares the target set value and actual operating value of the process parameters, the predicted value of the extracted index and the offline quality detection value of the batch, based on the aligned multi-source verification dataset, calculates and generates feedback deviation data and outputs it to the intelligent parameter optimization unit. The working condition quality judgment module, based on the aligned multi-source verification dataset, monitors the operating status of the extraction equipment and the batch extraction quality in real time, identifies abnormal working conditions such as equipment drift exceeding limits and quality index deviation, and determines the abnormality level. The control signal output module generates a system operation control signal of the corresponding level based on the anomaly determination result, and outputs it to the corresponding execution unit to perform control actions.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention employs a dual-response differential technology before and after material feeding, which can decouple and identify the characteristics of raw material batches and the equipment baseline drift online. By relying on the material coupling batch fingerprint to correct the process mapping relationship, it improves the parameter adaptability of different raw material batches, reduces the fluctuation of extraction indicators between batches, and ensures stable product quality. 2. This invention completes the time delay alignment and reconstruction of the process sequence by using the time delay of the additional effect of materials, and carries out adaptive optimization by combining the reliable adjustment domain of multi-factor dynamic calculation. This avoids both adjustment overshoot and prevents the optimization range from being too conservative, thereby improving the parameter adjustment accuracy and production line operation efficiency, and thus adapting to the real-time optimization needs of complex working conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall system framework structure of the present invention; Figure 2 This is a schematic flowchart of the closed-loop optimization control method for process parameters in this invention. The meanings of the labels in the diagram are as follows: 1. Multi-dimensional sensing and acquisition unit; 11. Synchronous acquisition and control module; 12. Multi-dimensional sensor array module; 13. Signal quantization output module; 2. Data transmission preprocessing unit; 21. Data receiving buffer module; 22. Noise reduction and filtering module; 23. Standardization and synchronization module; 3. Intelligent parameter optimization unit; 31. Batch decoupling characterization module; 32. Time delay alignment module; 33. Process mapping correction module; 34. Trusted domain optimization module; 4. Closed-loop execution control unit; 41. Instruction parsing and adaptation module; 42. Dynamic adjustment execution module; 43. Restricted test control module; 44. Execution status feedback module; 5. Feedback verification and operating condition control unit; 51. Multi-source data alignment module; 52. Adjustment effect verification module; 53. Operating condition quality judgment module; 54. Control signal output module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figures 1-2 As shown, this embodiment provides a real-time acquisition and optimization system for durian peel extraction process parameters, used for real-time acquisition, optimization, and closed-loop control of process parameters during durian peel extraction. The system includes a multi-dimensional sensing acquisition unit 1, a data transmission preprocessing unit 2, an intelligent parameter optimization unit 3, a closed-loop execution control unit 4, and a feedback verification and operating condition management unit 5. The output of the multi-dimensional sensing acquisition unit 1 is connected to the input of the data transmission preprocessing unit 2; the output of the data transmission preprocessing unit 2 is connected to the inputs of both the intelligent parameter optimization unit 3 and the feedback verification and operating condition management unit 5; the output of the intelligent parameter optimization unit 3 is connected to the input of the closed-loop execution control unit 4; the output of the closed-loop execution control unit 4 is connected to the input of the feedback verification and operating condition management unit 5; and the feedback deviation data output of the feedback verification and operating condition management unit 5 is connected to the input of the intelligent parameter optimization unit 3, forming a closed loop.

[0020] The multi-dimensional sensing acquisition unit 1 is used to acquire on-site physical quantity signals during the durian peel extraction process, synchronously acquiring and quantizing extraction process parameters and dynamic response data to obtain raw process data. It includes a synchronous acquisition control module 11, a multi-dimensional sensing array module 12, and a signal quantization output module 13. Specifically: the synchronous acquisition control module 11 generates a synchronous acquisition trigger signal based on a preset sampling period, controlling the sampling time of all acquisition channels to remain synchronized. The preset sampling period is determined based on the dynamic response time constant of the extraction process and the data update rate requirements of the subsequent closed-loop control loop, ensuring that the sampling period is less than a preset proportion of the main time constant of the fastest-changing process parameter during extraction, thus guaranteeing that the acquired data reflects the dynamic changes of the extraction process. The synchronous acquisition control module 11 internally sets a unified clock reference, generates trigger pulses according to this clock reference and the preset sampling period, and sends the trigger pulses in parallel to each acquisition channel of the multi-dimensional sensing array module 12. Each acquisition channel starts sampling at the same moment upon receiving the trigger pulse, ensuring that the sampling time of different physical quantities remains consistent within a preset synchronization error range. The synchronous acquisition trigger signal can be transmitted in the form of a hard-wired level signal or a fieldbus synchronization message.

[0021] The multi-dimensional sensor array module 12 senses the on-site physical quantity signals during the durian peel extraction process based on a synchronous acquisition trigger signal, and acquires analog sensor signals corresponding to the extraction process parameters and dynamic response data. Specifically, the multi-dimensional sensor array module 12 consists of multiple acquisition channels, each corresponding to one or more sensing elements. The sensing elements are used to convert the physical quantities during the extraction process into analog electrical signals. For example, temperature can be measured using a platinum resistance thermometer or a thermocouple sensor, pressure can be measured using a pressure transmitter, flow rate can be measured using an electromagnetic flowmeter or a vortex flowmeter, and liquid level can be measured using a radar level gauge or a hydrostatic level gauge. In this embodiment, the analog sensor signals corresponding to the extraction process parameters can include physical quantity signals reflecting the operating status of the extraction process, such as extraction temperature, extraction pressure, solvent flow rate, and liquid level; the analog sensor signals corresponding to the dynamic response data can include response signals showing the changes of the above physical quantities over time after the adjustment is applied. The output terminals of each sensing element are connected to the corresponding input terminals of the signal quantization output module 13.

[0022] The signal quantization output module 13 performs analog-to-quantization conversion based on the analog sensing signal and marks it with a unified time identifier to generate raw process data.

[0023] Specifically, the signal quantization output module 13 receives the analog sensing signals output from each acquisition channel, first performs signal conditioning on each analog sensing signal, such as amplification, isolation, and anti-aliasing filtering, and then converts the conditioned analog signals into digital signals. The sampling time for the analog-to-quantization conversion is determined by the synchronous acquisition trigger signal generated by the synchronous acquisition control module 11.

[0024] After completing the analog-to-quantum conversion, the signal quantization output module 13 marks each sampled data with a time identifier based on the unified time reference provided by the synchronous acquisition control module 11, forming raw process data with channel identifier, sampled value, and time identifier. The raw process data is transmitted to the data transmission preprocessing unit 2 through the data interface. The data interface can be one or more of industrial Ethernet, fieldbus, or serial communication interface.

[0025] Through the above-mentioned synchronous acquisition triggering and unified time identification processing, the data of each channel in the raw process data output by the multi-dimensional sensing acquisition unit 1 has a consistent sampling time, which provides a data foundation with consistent time dimension for subsequent data transmission preprocessing unit 2 to perform noise reduction, standardization and time synchronization processing.

[0026] Data transmission preprocessing unit 2 is used to receive raw process data, perform noise reduction, standardization, and time synchronization processing on the raw process data to obtain real-time operating condition data; it includes a data receiving buffer module 21, a noise reduction filtering module 22, and a standardization synchronization module 23, wherein: The data receiving and buffering module 21 receives the raw process data and completes data temporary storage and transmission validity verification. Specifically, the data receiving and buffering module 21 receives the raw process data output by the multi-dimensional sensing and acquisition unit 1 through the data interface, and writes the received raw process data into the buffer area in the order of sampling time to buffer data fluctuations caused by the inconsistency between the acquisition rate and the subsequent processing rate. Transmission validity verification includes checking the integrity of the data frame, the legality of the channel identifier, the continuity of the time identifier, and the rationality of the sampled values. When a missing data frame, abnormal channel identifier, discontinuous time identifier, or sampled values ​​exceeding the normal range of the corresponding sensing element are detected, the data frame is determined to be invalid and discarded or marked as invalid data. For invalid data with a small number of consecutive missing frames, linear interpolation or filling with the average of preceding and following frames can be used to complete the data, avoiding data gaps affecting subsequent processing. Valid data frames that pass the verification are output to the noise reduction and filtering module 22. The buffer area can adopt a first-in-first-out queue structure.

[0027] The noise reduction and filtering module 22 filters out interference noise introduced during the acquisition and transmission process based on the raw process data output by the data receiving and buffering module 21, obtaining the noise-reduced process data. Specifically, the noise reduction and filtering module 22 performs digital filtering on the process data sequence of each acquisition channel. Based on the difference between the frequency of the extracted process signal change and the frequency of the interference noise, one or more of low-pass filtering, moving average filtering, or median filtering are selected. Low-pass filtering is used to suppress high-frequency random noise, moving average filtering is used to smooth short-term fluctuations, and median filtering is used to remove occasional pulse interference. The filtering parameters are determined according to the dynamic characteristics and noise level of the physical quantities in each channel. For physical quantities such as temperature, pressure, and liquid level that change relatively slowly, moving average filtering can be used; for physical quantities such as flow rate that change rapidly, low-pass filtering or lighter smoothing processing can be used. The filtered data from each channel constitutes the noise-reduced process data and is output to the standardization and synchronization module 23.

[0028] The standardization and synchronization module 23 performs range standardization processing and time axis synchronization and regularization based on the noise-reduced process data to generate real-time operating condition data.

[0029] Specifically, range standardization is a process that maps physical quantities with different dimensions and ranges in each channel to a unified dimensionless numerical range in order to eliminate the impact of differences in the dimensions and amplitudes of each physical quantity on subsequent optimization calculations.

[0030] Range standardization can be performed by linear mapping to the preset lower and upper engineering limits of each channel, so that the standardized values ​​fall within the preset range (e.g., mapped to the [0,1] or [-1,1] range); alternatively, a standardization method based on the mean and standard deviation can be used. The lower and upper engineering limits of each channel are preset according to the measurement range of the corresponding sensing element and the normal operating range of the extraction process.

[0031] Time axis synchronization normalization is used to eliminate minor timestamp deviations that may occur during the buffering and transmission of data from each channel. It resamples or interpolates and aligns the data from each channel using a unified time base, so that each channel has corresponding data values ​​at the same point in time.

[0032] After range standardization and time axis synchronization, real-time operating condition data with each channel standardized and arranged according to a unified time axis is generated and output to the intelligent parameter optimization unit 3 and the feedback verification and operating condition control unit 5.

[0033] Through the above processing, the data transmission preprocessing unit 2 converts the raw process data output by the multi-dimensional sensing acquisition unit 1 into real-time operating condition data with unified dimensions, time alignment, and controlled noise, providing a consistent data foundation for the batch decoupling, time delay alignment, and optimization calculation of the intelligent parameter optimization unit 3.

[0034] The intelligent parameter optimization unit 3 receives real-time operating condition data and feedback deviation data output by the feedback verification and operating condition control unit 5. It employs a batch adaptive optimization technique that decouples equipment status from raw material characteristics to dynamically calibrate and optimize the mapping relationship between batch operating condition characteristics and process parameters in the extraction process, generating optimal process parameter instructions. The intelligent parameter optimization unit 3 includes a batch decoupling characterization module 31, a time-delay alignment module 32, a process mapping correction module 33, and a trusted domain optimization module 34. Specifically, the batch decoupling characterization module 31 receives real-time operating condition data output by the data transmission preprocessing unit 2, and its output is connected to the time-delay alignment module 32, the process mapping correction module 33, and the trusted domain optimization module 34, respectively. The output of the time-delay alignment module 32 is connected to the trusted domain optimization module 34. The output of the process mapping correction module 33 is also connected to the trusted domain optimization module 34. The trusted domain optimization module 34 also receives feedback deviation data output by the feedback verification and operating condition control unit 5. The output of the trusted domain optimization module 34 is connected to the closed-loop execution control unit 4.

[0035] In this embodiment, the real-time operating condition data is multi-channel data after range standardization and time axis synchronization normalization by the data transmission preprocessing unit 2. The feedback deviation data includes the deviation between the target set value and the actual operating value of the process parameters, and the deviation between the predicted value of the extracted index and the offline quality detection value of the batch. The intelligent parameter optimization unit 3 uses the real-time operating condition data to separate the equipment reference drift characteristics and the raw material batch characteristics. Based on the separation results, it performs time-delay alignment on the historical process parameters, performs batch adaptation correction on the mapping relationship between process parameters and extracted indexes, and solves for the optimal combination of process parameters within the reliable adjustment domain and the equipment safety boundary to generate the optimal process parameter command.

[0036] The batch decoupling characterization module 31 employs a dual-response differential technique before and after material feeding. Based on real-time operating data, it separates and extracts the equipment baseline drift characteristics and raw material batch characteristics during the process, outputting the material coupling batch fingerprint, equipment baseline drift factor, and material additional effect time delay. The batch decoupling characterization module 31 includes a response acquisition and calibration submodule, a material characteristic decoupling submodule, and an equipment drift calculation submodule, wherein: The response acquisition and calibration submodule is used to extract the equipment-medium reference response under steady-state conditions before feeding and the equipment-medium-raw material coupling response under steady-state conditions after feeding from the input real-time operating data. The two responses are then normalized to obtain the reference response value and coupling response value corresponding to the unit adjustment action of the j-th adjustment channel. At the same time, the action delay corresponding to each of the two responses is extracted. Specifically, the response acquisition and calibration submodule receives real-time operating condition data, which includes a data sequence of process parameters and response physical quantities changing over time for each adjustment channel. In this embodiment, the adjustment channels include at least one of an extraction temperature adjustment channel, an extraction pressure adjustment channel, and a solvent flow rate adjustment channel. Each adjustment channel corresponds to an actuator, such as a regulating valve, a heater, or a pump.

[0037] The response acquisition and calibration submodule first identifies two stages from real-time operating data: the medium steady-state before feeding and the wetting steady-state after feeding. The determination rule for the medium steady-state before feeding is as follows: before feeding begins, the extraction equipment contains only the extraction medium. When the fluctuation amplitudes of the extraction temperature, extraction pressure, and liquid level within the preset steady-state determination time window are less than their respective preset steady-state thresholds, it is determined to be the medium steady-state before feeding. The determination rule for the wetting steady-state after feeding is as follows: after solid feeding is completed, when the fluctuation amplitudes of the extraction temperature, extraction pressure, and liquid level within the preset steady-state determination time window are less than their respective preset steady-state thresholds, and the deviations of basic process parameters such as solvent flow rate, medium temperature, and tank pressure from the baseline steady-state before feeding do not exceed the preset allowable range (for example, the deviation does not exceed ±2%), it is determined to be the wetting steady-state after feeding. If the deviation of the basic operating conditions between the two steady-state stages exceeds the allowable range, the step test data is invalid. The test is re-executed after the operating conditions are adjusted and stabilized to ensure consistency between the baseline operating conditions before and after feeding, eliminating the interference of system state changes on decoupled calculations. The aforementioned preset steady-state determination time window and preset steady-state threshold are pre-set based on the extraction process requirements and equipment dynamic characteristics. The setting principle is as follows: the steady-state threshold is taken as 1 / 3 to 1 / 2 of the allowable fluctuation range of the extraction process indicators, and not less than 3 times the measurement accuracy of the corresponding sensor, to ensure the engineering rationality of the steady-state determination. For example, the steady-state determination time window can be set to 60s, the extraction temperature steady-state threshold can be set to ±0.5℃, the extraction pressure steady-state threshold can be set to ±2kPa, and the liquid level steady-state threshold can be set to ±1cm. The above values ​​are examples and can be adjusted according to the specific equipment and process characteristics.

[0038] During the two steady-state phases, the response acquisition and calibration submodule sends a test adjustment request through the data interface between the intelligent parameter optimization unit 3 and the closed-loop execution control unit 4. The test adjustment request carries the adjustment channel identifier and step amplitude. Each adjustment channel performs a step test independently in sequence. After receiving the test adjustment request, the closed-loop execution control unit 4 pauses the normal adjustment command output of the corresponding adjustment channel and performs a step adjustment action of the preset amplitude as requested. During the test, the other non-test adjustment channels switch to output hold mode to maintain the current parameters constant, and resume normal adjustment of all channels after the test is completed. The response acquisition and calibration submodule records the response curve of the physical quantity of the response changing with time after the adjustment action is applied. The response curve obtained after applying a step adjustment action under the steady state of the medium before feeding is recorded as the equipment-medium reference response; the response curve obtained after applying a step adjustment action of the same amplitude under the wetting steady state after feeding is recorded as the equipment-medium-raw material coupled response. The amplitude of the step adjustment is a preset proportion of the actuator's normal adjustment range. For example, the step change in the valve opening is 5%-10% of the normal adjustment range to ensure that the response signal is identifiable and has minimal interference with the normal extraction process. After the step test is completed, the system waits for the corresponding channel response to return to the steady state before the test before proceeding to subsequent steps. From the time of material feeding until the batch adaptive optimization parameters are issued, the system defaults to operating with preset baseline process parameters. The step test is a small-amplitude, short-duration disturbance, and the system quickly returns to steady-state operation after the test, without significantly affecting the overall extraction quality. The response data during the test is marked as test data and is not included in subsequent feedback verification statistics.

[0039] For the Each adjustment channel has a response acquisition and calibration submodule. The submodule divides the change in the baseline response before material feeding by the step adjustment action of that channel to obtain the response value corresponding to a unit adjustment action before material feeding. Similarly, it divides the change in the coupled response after material feeding by the step adjustment action of that channel to obtain the response value corresponding to a unit adjustment action after material feeding. The change in response is defined as the difference between the steady-state value of the response curve after the step action and the steady-state value before the step action. The steady-state value is the average value of the response curve within the steady-state determination time window. Action normalization is performed to ensure comparability of responses from different adjustment channels. The normalized unit response value of the baseline response before feeding into each regulating channel is denoted as: , No. The normalized unit action response value of the coupled response after feeding through each regulating channel is denoted as: subscript Indicates the first Batch, Subscript Indicates the first There are several adjustment channels. The unit response value mentioned above is a dimensionless value. Simultaneously, the response acquisition and calibration submodule extracts the corresponding action delays of the pre-feeding baseline response and the post-feeding coupled response. The action delay is the time required from the moment the adjustment is applied until the change in the response curve reaches a preset proportion of the steady-state change of the channel response; this preset proportion is, for example, 2%. In this embodiment, the action delay is determined using a threshold method, and the action delay of the pre-feeding baseline response is denoted as... The time delay of the coupled response after feeding is denoted as . .

[0040] The material characteristic decoupling submodule calculates the additional material response corresponding to a unit solid load and constructs a material coupled batch fingerprint based on the baseline response value, coupled response value, and the current batch solid feed amount of durian peel. It calculates the difference in action time delay between the two responses to obtain the additional material action time delay for each adjustment channel. Based on the material coupled batch fingerprint and the calibrated set of historical qualified batch fingerprints, it calculates the raw material batch unfamiliarity. It generates time delay identification credibility based on the confidence level of the response curve fitting. Specifically, the material characteristic decoupling submodule receives the unit action response value of each adjustment channel output by the response acquisition and calibration submodule. , and the time delay of action And receive the current batch of durian peel solid feed. The amount of durian peel solid feed is the actual dry weight of the durian peels fed into the extraction equipment, in kg. The additional material response per unit solid load is calculated using the following formula: It should be noted that this calculation is based on the assumption of linear superposition of responses, and is applicable to steady-state conditions where the fluctuation range of process parameters is small and the extraction process is approximately linear, which matches the application scenario of steady-state testing before and after feeding.

[0041] In the formula, For the first Batch No. The additional response of the unit solid load material in the first regulating channel represents the response of a unit mass of durian peel solids to the first regulating channel. Additional contribution of each regulating channel unit regulating action response; For the first Normalized unit action response value of the baseline response before batch feeding; For the first The normalized unit action response value of the coupled response after batch feeding; For the first Batch solid durian peel input quantity; Subscript For batch dimension identification, subscript To adjust the channel dimension identifier; The value is a very small positive constant, much smaller than the amount of solid durian peel used in the current batch, used to avoid division by zero anomalies. In this embodiment... Pick ; will the first The additional material response per unit solid load for all regulating channels in a batch is arranged in channel order, constituting... Batch material coupled with batch fingerprint vector ,Right now ,in To adjust the total number of channels. Material-coupled batch fingerprints are used to characterize the impact of the current batch of durian peel raw material on the dynamic response of the extraction process.

[0042] Furthermore, the material characteristic decoupling submodule also calculates the time delay of the additional material effect in each adjustment channel. Material lag in each regulating channel This is the difference between the time delay of the coupled response after material feeding and the time delay of the reference response before material feeding, i.e.: Material lag effect refers to the change in the transmission delay of the regulating effect caused by the addition of raw materials, and the unit is 1. when When the value is positive, it indicates that the time for the adjustment effect to be transmitted to the response increases after the raw material is added; when... A negative value indicates a decrease in the transmission time of the regulating effect after the raw material is added. The material additional effect delay of each regulating channel constitutes the material additional effect delay vector. .

[0043] The calculated time lag of the material's additional effect needs to be verified for rationality: if the time lag is negative and its absolute value is less than the preset noise threshold (for example, 5 seconds), it is determined to be measurement noise, and the threshold is set accordingly. If the absolute value of the negative value exceeds the noise threshold, the step test is deemed to have failed. The time delay data of the corresponding channel of the most recent qualified batch is reused, and the test data is marked as invalid and not included in subsequent statistics. If multiple tests fail in a row and there is no available historical data, the preset default time delay parameters are used to run the test. At the same time, the system is marked as a low confidence state, and the range of the confidence adjustment domain is narrowed.

[0044] Furthermore, the unfamiliarity of raw material batches The calculation is based on material-coupled batch fingerprints. fingerprint set of calibrated historical qualified batches The degree of difference between them. Collection of fingerprints from historical qualified batches. is composed of material-coupled batch fingerprints corresponding to batches judged as qualified through offline quality inspection among a plurality of past batches, and is updated after the offline quality data of each batch is returned. For example, an upper limit of the historical set capacity can be set to retain the fingerprint data of the latest N qualified batches, and eliminate the batch data with the earliest time when the capacity is exceeded, so as to ensure calculation efficiency and data timeliness. When the historical qualified batch fingerprint set is empty, take , which indicates that there is no historical qualified reference for the current batch, and the maximum strangeness degree is adopted for processing. In the cold start scenario where the system is put into use for the first time or restarted after major equipment overhaul, no-load medium steady-state calibration is performed first: in the medium steady-state stage before feeding, the restricted test management and control module sequentially applies standard step adjustment signals to each adjustment channel, and collects to obtain the initial equipment reference response vector , completing equipment reference initialization. The first batch of production does not enable the adaptive optimization function, and operates with preset reference process parameters, and collects process response data synchronously; after the batch is judged as qualified by offline quality inspection, its material-coupled batch fingerprint is incorporated into the historical qualified batch fingerprint set , and the batch adaptive optimization calculation is officially started from the second batch. When the historical qualified batch fingerprint set is not empty, the adjustment channels are independently executed first standardization processing, where the mean and standard deviation of each channel are taken from the statistical values of the corresponding channels in the historical qualified batch fingerprint set; when the number of historical qualified batches is less than a preset threshold (exemplarily 3), the standardization processing is not performed temporarily, or a preset reference standard deviation is used to participate in the calculation, so as to avoid zero division abnormality caused by insufficient sample size, eliminate the dimension and magnitude differences between channels, and then calculate the current batch fingerprint and each sample in the historical qualified batch fingerprint set Euclidean distance, and take the minimum distance among them as the similarity measure between the current batch and the historical qualified batches, that is: In the formula, represents the vector 2-norm. The minimum distance is mapped to a dimensionless strangeness parameter in the 0-1 interval. In this embodiment, the following mapping is adopted: In the formula, is a preset similarity scale parameter. In the cold start phase, an initial empirical value can be preset (for example, it can be calibrated based on the fingerprint distance of laboratory test batches). After enough historical batches are accumulated, the calibration method is: take 1~2 times of the average value of the Euclidean distances between every two fingerprints of the historical qualified batches , so that the strangeness of the vast majority of qualified batches is in a relatively low interval, which is used to control the sensitivity of the strangeness to distance changes. This parameter is an example and can be adjusted according to the actual batch data distribution. When , , which indicates that the current batch fingerprint is completely consistent with a certain historical qualified batch; when is much greater than hour, A value close to 1 indicates a significant difference between the current batch and historically qualified batches. A higher level of unfamiliarity with the raw material batch indicates a greater difference in the characteristics of the current batch compared to historically qualified batches, and the allowable fluctuation range should be narrowed accordingly during subsequent optimization efforts.

[0045] Furthermore, the reliability of time delay identification Confidence scores are generated based on response curve fitting. In this embodiment, the response curves of each adjustment channel are fitted using a first-order inertial pure time-delay model, and the goodness-of-fit for each channel is obtained after fitting. The goodness-of-fit for each channel is truncated to a value greater than or equal to 0, and the truncated goodness-of-fit is denoted as . The overall time delay identification reliability is obtained by averaging the truncation goodness of fit for all adjustment channels. ,Right now: In the formula, To adjust the total number of channels. The value range is 0-1. The closer it is to 1, the better the response curve fitting effect is, which can indirectly characterize the reliability of the time delay identification result. This indicates a non-negative truncation of the goodness of fit: when the goodness of fit is negative, it means that the response curve fit is worse than the mean prediction, and the corresponding channel time delay identification result is completely unreliable, so it is treated as having a confidence level of 0. The time delay identification confidence level is output to the confidence domain optimization module 34 for subsequent calculation of the confidence adjustment domain scaling factor.

[0046] The equipment drift calculation submodule is used to compare the current batch baseline response with the calibrated historical equipment baseline response to generate the equipment baseline drift factor.

[0047] Specifically, the equipment drift calculation submodule receives the normalized unit action response value of the reference response before feeding each adjustment channel in the current batch, output by the response acquisition and calibration submodule. And read the calibrated historical device reference response vector. The historical equipment baseline response vector is obtained by applying a unit adjustment force to each adjustment channel under steady-state conditions of the medium before feeding materials, during the initial operation of the equipment or after maintenance calibration, and recording the response. This is achieved through repeated measurements and averaging the results. For example, three consecutive step tests can be performed, and the average value after discarding abnormal response data can be used as the baseline value. After the equipment completes a major overhaul, sensor calibration, or maintenance, the no-load baseline calibration must be performed again, and the historical equipment baseline response vector must be updated. For example, the preset calibration cycle can be set according to the equipment's operating time and production plan, such as once per quarter or after every 100 batches of production. The historical equipment baseline response is recalibrated and updated after each equipment maintenance or preset calibration cycle. The relative deviation of each adjustment channel Calculate using the following formula: In the formula, The first in the historical equipment reference response vector The reference unit response value of each adjustment channel is calculated. A weighted average of the relative deviations of each adjustment channel is taken, and the equipment reference drift factor is obtained after saturation processing. : In the formula, To adjust the total number of channels, For the first The preset drift weight coefficients of each adjustment channel satisfy... In a simplified implementation, the weights of each channel can be taken to be equal. By default, each adjustment channel has a similar impact on drift. Device reference drift factor. This is a normalized dimensionless parameter, ranging from 0 to 1. A larger value indicates a greater deviation between the equipment's baseline response before the current batch of material is fed and its historical calibration state, resulting in more significant equipment drift. For example, values ​​below 0.2 can be set to indicate slight drift. A value of 0.5 or higher indicates moderate drift, while a value above 0.5 indicates significant drift. Different drift levels correspond to different optimization constraint strengths. The equipment baseline drift factor is output to the trusted domain optimization module 34, which is used to dynamically adjust the trusted adjustment domain during optimization.

[0048] Through the processing of the above-mentioned response acquisition and calibration submodule, material characteristic decoupling submodule, and equipment drift calculation submodule, the batch decoupling characterization module 31 separates the material coupling batch fingerprint characterizing the characteristics of the raw material batch, the equipment reference drift factor characterizing the equipment state change, and the material additional effect time delay characterizing the time delay change caused by the material from the real-time operating data. This method assumes that the core characteristic differences between raw material batches can be characterized by the dynamic response difference of unit solid load, which is suitable for scenarios where the fluctuation of physical properties of solid raw materials such as durian peel extraction mainly affects the mass transfer response characteristics. If it is necessary to further improve the adaptation accuracy, measurable physical property dimensions such as raw material moisture content and particle size can be added to the batch fingerprint. This provides basic data for subsequent time delay alignment, process mapping correction, and trusted domain optimization.

[0049] The time-delay alignment module 32 employs a time-axis translation calibration technique based on the material-added time delay to reconstruct the process parameter sequence, eliminating response lag offsets caused by differences in raw material characteristics. This ensures that the control actions and response effects of different batches are comparable in the time dimension, providing aligned time-series data support for determining the optimization benchmark. Specifically, the time-delay alignment module 32 receives the material-added effect time delay vector output by the batch decoupling characterization module 31. The system reads historical process parameter sequences from the historical process parameter sequence memory. These historical process parameter sequences are data sequences showing the changes over time of the process parameter settings or adjustment commands executed before the current batch, such as the setpoint records of each adjustment channel during the extraction process of the previous batch. The time-delay alignment module 32 performs time-delay alignment reconstruction of the process parameter sequences, including steps S32.1 and S32.2.

[0050] S32.1 Extract the material additional effect time delay corresponding to each adjustment channel from the output of the batch decoupling characterization module 31; specifically, the time delay alignment module 32 reads the material additional effect time delay vector output by the batch decoupling characterization module 31. , obtained the The material additional effect time delay corresponding to each adjustment channel .

[0051] S32.2. Based on the time of the adjustment action, the historical process parameter sequence is shifted along the time axis to the corresponding channel's material additional action time delay, so that the control action and response effect correspond one-to-one in the time dimension, and the time delay aligned process parameter sequence is obtained.

[0052] In S32.2, for the first Historical process parameters of each adjustment channel Based on the moment when the adjustment action occurs, the process parameter sequence of this channel is shifted along the time axis. .like If the value is positive, the process parameter sequence will be shifted in the negative direction of the time axis. ,Right now This indicates that the actual effect of the adjustment is delayed relative to the timing of the adjustment action, taking into account the material's added time lag; if If the value is negative, the process parameter sequence will be shifted in the positive direction of the time axis. ,Right now The unified representation is: in The first time delay alignment A sequence of process parameters for each adjustment channel. After the above translation, the control action and response effect achieve a one-to-one correspondence in the time dimension, resulting in a time-delay aligned sequence of process parameters, which is then output to the trusted domain optimization module 34.

[0053] The process mapping correction module 33 employs a batch fingerprint-based feature conditional modulation technique to perform batch adaptation correction on the mapping relationship between process parameters and extracted indicators. Specifically, the process mapping correction module 33 receives the material coupling batch fingerprint output by the batch decoupling characterization module 31. The mapping relationship between process parameters and extraction indicators is batch-adapted and corrected. In this embodiment, the mapping relationship between process parameters and extraction indicators refers to a quantitative calculation relationship with process parameters as input and extraction indicators as output. This mapping relationship is implemented by a calculation model, which includes several intermediate mapping features. The intermediate mapping features are intermediate variables used to calculate the output of the extraction indicators after feature transformation of the input process parameters in the calculation model. In an optional implementation, the calculation model is implemented using a single hidden layer feedforward neural network, which includes an input layer, a hidden layer, and an output layer: the input layer receives... Each adjustment channel corresponds to a process parameter (such as extraction temperature, extraction pressure, solvent flow rate, etc.); the hidden layer obtains the mapped intermediate feature quantity by performing linear transformation and nonlinear activation on the input process parameters, with a dimension of . The output layer obtains predicted values ​​of extraction indicators (such as the extraction rate of effective components, the concentration of target products, etc.) by performing a linear transformation on the intermediate features of the mapping. This calculation model is only an example; other calculation models that can establish a quantitative relationship between process parameters and extraction indicators, such as multiple regression models or support vector regression models, can also be used.

[0054] The above calculation model is pre-trained based on a sample of process parameters and offline quality inspection values ​​from historical qualified batches after time-delay alignment. This ensures that the input time series benchmark for model training and optimization inference is consistent. The training objective is to minimize the mean square error between the extracted index prediction values ​​and the offline inspection values. The model parameters are solved using gradient descent or least squares methods. During the system's cold start phase, when there is no proprietary production data, laboratory pilot data, intermediate pilot data, or a general pre-trained model with the same raw materials and process scenarios can be used as the initial basic model. The model is then iteratively updated gradually as proprietary qualified batch data is accumulated. After accumulating a preset number of qualified batches, the basic model can be incrementally updated to ensure model adaptability.

[0055] Specifically, the process mapping correction module 33 performs batch adaptation correction of the mapping relationship between process parameters and extracted indicators, including steps S33.1 to S33.3: S33.1. Based on the material-coupled batch fingerprint, generate the feature gain coefficient and feature offset coefficient of the corresponding dimension through a preset linear mapping rule; In S33.1, let the material-coupled batch fingerprint... The dimension is The dimension of the intermediate feature quantity is The preset linear mapping rule is: In the formula, The feature gain coefficient vector has a dimension of . ; The feature offset coefficient vector has a dimension of ; and The given linear mapping matrix has dimensions of 1. ; and The preset offset vector has dimensions of 1. The aforementioned linear mapping matrix and offset vector are pre-trained based on historical batch data. During training, the material-coupled batch fingerprints of historical batches are used as input, with the goal of minimizing the prediction error of the extracted indicators on historical batches using the least squares method or gradient descent method. The dimensions of the feature gain coefficients and feature offset coefficients are consistent with the dimensions of the intermediate features in the mapping.

[0056] S33.2. Using feature gain coefficients and feature offset coefficients, the intermediate feature quantities of the mapping model corresponding to the mapping relationship between process parameters and extracted indicators are conditionally modulated dimension by dimension to obtain batch-adaptive feature quantities; In S33.2, the intermediate feature quantity vector of the design calculation model is... , dimension Dimension-by-dimensional conditional modulation is performed as follows: In the formula, To adapt the feature vector for the batch, the dimension is ; This represents element-wise multiplication, i.e., the characteristic gain coefficient vector. The Each element and the intermediate feature vector of the mapping The Multiply each element together, then add the feature offset coefficient vector. The Each element. This modulation method adapts the intermediate feature quantity of the computational model to the characteristics of the current raw material batch by performing gain scaling and offset adjustment on each dimension of the mapped intermediate feature quantity.

[0057] S33.3 Based on the corrected batch adaptation feature, output the mapping relationship between the process parameters and extraction indicators adapted to the current raw material batch.

[0058] In S33.3, the batch adaptation feature quantity obtained after modulation is... Replace the intermediate features in the original computational model The computational model then continues the calculation from batch-adapted feature quantities to extracted index outputs, thereby obtaining the mapping relationship between process parameters adapted to the current raw material batch and extracted indexes. The corrected mapping relationship is output to the confidence domain optimization module 34, which is used to predict extracted indexes based on candidate process parameters in subsequent optimization calculations.

[0059] The Trusted Domain Optimization Module 34 employs a batch adaptive optimization technique with multi-factor dynamic constraints. Based on the aligned process parameter sequence, the mapping relationship between the corrected process parameters and extracted indicators, the equipment baseline drift factor, and feedback deviation data, it solves for the optimal combination of process parameters and generates the optimal process parameter instruction. For example, the baseline optimal parameters for each batch can be calculated before each batch is fed and after the steady-state test is completed. During the extraction process, the parameters can be updated at preset intervals (e.g., every 10 minutes) according to real-time operating conditions, balancing optimization efficiency and process stability. Specifically, the Trusted Domain Optimization Module 34 includes a Trusted Domain Dynamic Calculation Submodule and a Multi-Objective Constraint Optimization Submodule. The Trusted Domain Dynamic Calculation Submodule calculates the Trusted Adjustment Domain Scaling Coefficient based on the unfamiliarity of the raw material batch, the equipment baseline drift factor, the recent prediction error within the preset statistical window, and the time lag identification credibility, dynamically adjusting the allowable fluctuation range of the optimization parameters. The Multi-Objective Constraint Optimization Submodule, within the adjusted Trusted Adjustment Domain and the equipment safety boundary, solves for the optimal combination of process parameters based on the mapping relationship between the corrected process parameters and extracted indicators, and generates the optimal process parameter instruction. The aligned process parameter sequence is used to determine the current reference value of the optimization parameters of each adjustment channel, ensuring that the reference value and the response effect correspond in time sequence, and avoiding the reference selection deviation caused by time delay differences.

[0060] Specifically, the Trust Domain Dynamic Calculation Submodule is used to calculate the Trust Adjustment Domain Scaling Coefficient based on the raw material batch unfamiliarity, equipment reference drift factor, recent prediction error and time lag within a preset statistical window to identify trustworthiness, and dynamically narrow the allowable fluctuation range of the optimization parameters according to the operating condition confidence level. Specifically, the Trust Domain Dynamic Calculation Submodule receives the raw material batch unfamiliarity output by the batch decoupling characterization module 31. Equipment reference drift factor Reliability of Time Delay Recognition It also receives the deviation between the predicted value of the extracted index and the offline quality detection value of the batch from the feedback deviation data output by the feedback verification and operating condition control unit 5. The trusted domain dynamic calculation submodule calculates the recent prediction error within a preset statistical window from the feedback deviation data. In this embodiment, the preset statistical window is the preset time period closest to the current time, such as the deviation between the predicted values ​​of the extracted indicators and the offline quality detection values ​​of the batches in the last 3 batches or the last 24 hours. Recent prediction error. Using mean absolute percentage error The calculation is performed and then normalized to map to the 0-1 interval. The specific normalization method is as follows: In the formula, The average absolute percentage error between the predicted value of the extracted indicator and the offline quality test value of the batch within the preset statistical window, expressed as a percentage. In this embodiment, the preset upper limit for error normalization is used. Take 10%, that is, when When it reaches or exceeds 10%, This indicates a large prediction error. The above... For example, the scaling factor can be adjusted according to the actual prediction accuracy requirements. The confidence adjustment domain scaling factor is calculated using the following formula: In the formula, for The scaling factor of the reliable adjustment domain at any given time; The reference reliable adjustment domain coefficient is a preset positive value, which is used in this embodiment. Set the value to 1.0; The reliability of time delay identification is represented by a value ranging from 0 to 1. for The batch unfamiliarity of raw materials, with a value range of 0-1; For the first The equipment reference drift factor for a batch ranges from 0 to 1. This is the recent prediction error within the statistical window corresponding to the current batch. It is updated on a batch-by-batch basis and recalculated once after each batch of offline quality data is returned. The value range is 0-1. The preset positive weighting coefficients are determined through sensitivity analysis of historical batches. The controlled variable method can be used to analyze the impact of each factor's fluctuation on the extracted indicators; the greater the impact, the higher the corresponding weight. For example, in this embodiment, all coefficients are set to 1.0. The above weighting coefficients are merely examples and can be adjusted according to the impact of each factor on process stability. (Subscript) For time dimension identification, subscript This is used as a batch dimension identifier.

[0061] It is worth noting that, in order to avoid the optimization function becoming rigid and ineffective due to excessive shrinkage of the trusted domain, a minimum lower limit for the scaling factor is set. (For example, (Take 0.2); calculated according to the above formula Less than At that time, according to The values ​​are selected to ensure the system retains basic optimization exploration space. For example, if the unfamiliarity of raw materials, equipment baseline drift factor, and recent prediction error are all stable within the preset acceptable threshold for three consecutive batches, the scaling factor can be gradually increased by 0.1 for each batch until it returns to the baseline value. This enables steady-state recovery of the trusted domain.

[0062] When the unfamiliarity of the raw material batch, the equipment reference drift factor, and the recent prediction error are greater, the denominator is larger and the scaling factor of the confidence adjustment domain is smaller. This indicates that the characteristics of the current batch are significantly different from those of the past, the equipment drift is obvious, or the prediction error is large. The allowable fluctuation range of the optimization parameters should be narrowed. When the confidence of the time delay identification is higher, the scaling factor is larger, indicating that the time delay identification result is reliable. The allowable fluctuation range of the optimization parameters can be appropriately widened.

[0063] Furthermore, the trusted domain dynamic calculation submodule adjusts the trusted domain scaling factor according to the trusted domain scaling factor. The allowable fluctuation range of the optimization parameters is dynamically adjusted. In this embodiment, the current reference value of the optimization parameters of each adjustment channel is determined based on the time-delay aligned process parameter sequence output by the time-delay alignment module 32. Specifically, the process parameter setting value at the end of the previous batch is taken as the initial reference value of the optimization parameters for the current batch; when there is a historical adjustment record that has been executed in the current batch, the value of the historical adjustment record closest to the current time is taken as the current reference value. Let the first... The preset baseline for each adjustment channel allows for a fluctuation half-width of [value missing]. Scaling factor after minimum lower limit verification Applying to the base fluctuation half-width, the adjusted first The allowable fluctuation range of each adjustment channel is .when When the value decreases to less than 1, the allowable fluctuation range is correspondingly reduced; when While maintaining the minimum lower limit, the allowable fluctuation range remains within the minimum safe exploration interval, avoiding complete locking of the optimization capability; when When the value is 0, the allowable fluctuation range degenerates to the current baseline value, and no further adjustments are made. When the allowable fluctuation range of all adjustment channels degenerates to the current baseline value, the optimization calculation is skipped, and the current baseline value is directly output as the optimal combination of process parameters. The adjusted reliable adjustment domain is output to the multi-objective constraint optimization submodule.

[0064] The multi-objective constraint optimization submodule is used to solve for the optimal combination of process parameters and generate the optimal process parameter instruction within the adjusted credible adjustment domain and equipment safety boundary, based on the mapping relationship between the corrected process parameters and the extracted indicators.

[0065] Specifically, the multi-objective constraint optimization submodule receives the adjusted reliable adjustment domain output by the reliable domain dynamic calculation submodule, as well as the equipment safety boundary. The equipment safety boundary represents the range of parameters allowed for the extracted equipment's operation, preset according to the equipment's factory parameters and operating specifications. For example, the safe range for extraction temperature is no more than the equipment's maximum temperature tolerance, and the safe range for extraction pressure is no more than the equipment's maximum pressure tolerance. The intersection of the equipment safety boundary and the adjusted reliable adjustment domain yields the actual optimization constraint domain. Let the first... The optimization parameters for each adjustment channel are: The actual optimization constraint domain is: In the formula, For the first The lower limit of the actual optimization constraint domain for each adjustment channel; For the first The upper limit of the actual optimization constraint domain for each adjustment channel; and The first The lower and upper limits of the equipment safety boundaries for each adjustment channel.

[0066] It is worth noting that if the calculated lower limit of the actual optimization constraint domain is greater than the upper limit, that is, when the constraint domain is empty, the midpoint of the device safety boundary or the current benchmark value is taken as the fixed optimization value of the channel, and the channel is marked as being in a boundary-constrained state and will not participate in adaptive optimization adjustment for the time being.

[0067] The multi-objective constraint optimization submodule solves for the optimal combination of process parameters within the optimization constraint domain based on the mapping relationship between the corrected process parameters and the extracted indices. In this embodiment, the optimization objective is to make the predicted values ​​of the extracted indices as close as possible to the target extracted indices, while satisfying the constraints of the equipment safety boundary and the reliable adjustment domain. Target extracted indices Pre-set based on production plan or process requirements, it can be updated before the start of each batch. The optimization problem is: The constraints are: In the formula, The objective function is... This is a vector of process parameter combinations, with dimension 1. ; The predicted value of the extracted index is calculated based on the mapping relationship between the corrected process parameters and the extracted index, and its dimension is consistent with the number of extracted indexes; Extract the index for the objective. Let the optimal solution to the above optimization problem be denoted as... This is the optimal combination of process parameters.

[0068] In an alternative implementation, the optimization objective may further include an energy consumption adjustment term to simultaneously balance the extraction index deviation and energy consumption adjustment. Before calculation, the extraction index deviation term and the energy consumption term need to be dimensionless. In this case, the objective function can be expanded to: in, This is the energy consumption cost function for the adjustment range of process parameters, used to approximately characterize the additional energy cost caused by the parameter deviating from the reference value. Numerically, it is positively correlated with the adjustment range. In this embodiment, it is defined as the sum of the weighted absolute deviations of the process parameters of each adjustment channel relative to the current reference value, i.e.: in: For the first The energy consumption weighting coefficient of each adjustment channel satisfies In simplified implementation, the energy consumption weights of each channel can be taken as equal values. . and ... .

[0069] Solving the above optimization problem can employ conventional constrained optimization algorithms in this field. When the mapping relationship is a linear or convex model, a sequential quadratic programming algorithm can be used; when the mapping relationship is a nonlinear, non-convex model, global optimization algorithms such as genetic algorithms and particle swarm optimization can be used to obtain the optimal combination of process parameters. Afterwards, The values ​​corresponding to each adjustment channel are converted into target setpoints for each adjustment channel, forming the optimal process parameter command. The optimal process parameter command is output to the closed-loop execution control unit 4 for dynamic adjustment of the operating parameters of the extraction equipment.

[0070] The closed-loop execution control unit 4 receives the optimal process parameter command, performs dynamic adjustment and limited adjustment tests on the operating parameters of the extraction equipment, and outputs execution status feedback data. Specifically, the optimal process parameter command is the target setpoint value of each adjustment channel generated by the intelligent parameter optimization unit 3, and the execution status feedback data is one of the input data for the feedback verification and operating condition control unit 5 to verify the adjustment effect and determine the operating condition quality. The closed-loop execution control unit 4 performs real-time closed-loop adjustment of the operating parameters of the extraction equipment according to the optimal process parameter command, and performs small-amplitude limited adjustment tests during the preset steady-state operating condition stage, summarizing and outputting the adjustment execution status, real-time operating parameters of the equipment, and limited test response data. This unit includes an instruction parsing and adaptation module 41, a dynamic adjustment execution module 42, a limited test control module 43, and an execution status feedback module 44. Specifically, the instruction parsing and adaptation module 41 receives the optimal process parameter command, parses it, and converts it into control parameters that can be identified and executed by each adjustment channel of the extraction equipment. Specifically, the instruction parsing and adaptation module 41 receives the optimal process parameter instruction output by the intelligent parameter optimization unit 3. The optimal process parameter instruction carries the target values ​​corresponding to each adjustment channel, such as the target extraction temperature, target extraction pressure, and target solvent flow rate. The instruction parsing and adaptation module 41 converts the target values ​​of each adjustment channel into control parameters that can be recognized and executed by the corresponding actuator according to a preset channel configuration table. The channel configuration table records the actuator type, control signal type, signal range, and conversion rules corresponding to each adjustment channel. For example, the extraction temperature adjustment channel corresponds to a heater power adjustment actuator, and the target temperature value is converted into a 4-20mA analog signal or the corresponding power percentage; the solvent flow rate adjustment channel corresponds to a regulating valve, and the target flow rate value is converted into a valve opening percentage. The converted control parameters are output to the dynamic adjustment execution module 42.

[0071] The dynamic adjustment execution module 42 adjusts the operating parameters of the extraction equipment in real time based on the control parameters, so that the corresponding process parameters are tracked to the target value. Specifically, the dynamic adjustment execution module 42 receives the control parameters of each adjustment channel output by the instruction parsing and adaptation module 41, and simultaneously acquires the actual values ​​of the corresponding process parameters for each adjustment channel in real time through field instruments or sensor transmitters, such as the actual values ​​of extraction temperature, extraction pressure, and solvent flow rate. The dynamic adjustment execution module 42, with the inner loop controller as its core, compares the target values ​​of the control parameters with the actual values ​​of the process parameters to obtain the adjustment deviation, and calculates the actuator drive signal according to the preset closed-loop control law.

[0072] In this embodiment, the closed-loop control law can be a proportional-integral-derivative (PID) control law, which calculates the actuator drive quantity based on the proportional, integral, and derivative terms of the adjustment deviation, so that the corresponding process parameter tracks to the target value. During the closed-loop adjustment process, the dynamic adjustment execution module 42 updates the actual value of the process parameter according to a preset sampling period and calculates the actuator drive signal in real time until the actual value of the process parameter enters the preset steady-state error band near the target value. The closed-loop adjustment process of each adjustment channel is carried out independently and does not interfere with each other.

[0073] During non-test periods, the actuator drive signal output by the dynamic adjustment execution module 42 is directly output to the extraction device actuator; during test periods, the actuator drive signal and the test disturbance signal generated by the restricted test control module 43 are superimposed and output to the extraction device actuator. After the test, it is restored to the actuator drive signal output only by the dynamic adjustment execution module 42.

[0074] The restricted test control module 43 is used to issue test adjustment commands during the preset steady-state operating condition stage, control the equipment to perform small-amplitude restricted adjustment tests and collect corresponding dynamic response data; during the execution of single-channel step test, the other non-test adjustment channels are switched to output hold mode, the closed-loop adjustment is paused, the parameters of the non-test channels are kept constant, and the interference of channel coupling on the response curve is eliminated; after the test is completed and the response returns to steady state, the normal closed-loop control of all channels is restored.

[0075] Specifically, the restricted test control module 43 triggers test adjustments during a preset steady-state operating condition phase. The steady-state operating condition phase is determined based on the fluctuation of the actual values ​​of the process parameters in each adjustment channel. For example, if, within a preset judgment time window, the fluctuation range of the actual values ​​of the process parameters in each adjustment channel relative to the target value is less than their respective preset steady-state thresholds, it is determined to be a steady-state operating condition phase. The preset judgment time window and preset steady-state thresholds can be preset according to the extraction process requirements; for example, the judgment time window could be 60 seconds, and the extraction temperature steady-state threshold could be [not specified]. The steady-state threshold for extracting pressure is The steady-state threshold for solvent flow rate is Full scale.

[0076] After determining that the system has entered a steady-state operating phase, the restricted test control module 43 selects the adjustment channel to be tested and controls the corresponding actuator to perform a small-amplitude restricted adjustment test by superimposing a test disturbance signal. The small-amplitude restricted adjustment test refers to superimposing a step or pulse disturbance signal with limited amplitude and short duration on the actuator's normal adjustment output. The amplitude of the disturbance signal does not exceed a preset proportion of the actuator's normal adjustment range, for example, not exceeding 10% of the normal adjustment range, and the test duration does not exceed the upper limit of the transient time allowed by the process.

[0077] During the application of the test disturbance signal, the restricted test control module 43 synchronously collects the dynamic response data corresponding to the adjustment channel, i.e., the response curve of the actual value of the process parameter changing over time. After the test is completed, the restricted test control module 43 removes the disturbance signal and restores the normal adjustment output before the test.

[0078] The adjustment command corresponding to the test disturbance signal, the test start time, the test end time, and the response data are marked as test data and sent to the response acquisition and calibration submodule of the execution status feedback module 44 and the intelligent parameter optimization unit 3. When the response acquisition and calibration submodule needs to perform a response acquisition test, it sends a test adjustment request to the restricted test control module 43, which uniformly schedules the execution. The test command is only issued after the steady-state operating condition is determined; execution is suspended before the steady state is reached to avoid conflict with normal adjustment commands.

[0079] The execution status feedback module 44 collects and summarizes the adjustment execution status, real-time operating parameters of the equipment, and restricted test response data, generates execution status feedback data, and outputs it.

[0080] Specifically, the execution status feedback module 44 collects the adjustment execution status of the dynamic adjustment execution module 42 in real time, including the target value of each adjustment channel, the current actual value, the actuator drive signal, the adjustment start time, the steady state entry time, the steady state error, and the overshoot.

[0081] Simultaneously, the execution status feedback module 44 collects real-time operating parameters of each adjustment channel device, that is, extracts the actual values ​​of the process parameters corresponding to each adjustment channel of the device. For the restricted adjustment test executed by the restricted test control module 43, the execution status feedback module 44 receives the test adjustment command content and the corresponding dynamic response data, and retains the test data marker to distinguish it from normal operation data.

[0082] The execution status feedback module 44 summarizes the above-mentioned adjustment execution status, real-time equipment operating parameters and restricted test response data according to a unified time axis, generates execution status feedback data, and outputs it to the feedback verification and operating condition control unit 5 for adjustment effect verification, equipment operating status monitoring and batch quality judgment.

[0083] Through the above-mentioned instruction parsing, dynamic adjustment execution, restricted test control and execution status feedback processing, the closed-loop execution control unit 4 can transform the optimal process parameter instructions generated by the intelligent parameter optimization unit 3 into the actual adjustment actions of each adjustment channel of the extraction equipment, and complete the restricted adjustment test in the steady-state operating condition stage, while providing complete execution status feedback data for the feedback verification and operating condition control unit 5.

[0084] The feedback verification and operating condition control unit 5 receives execution status feedback data, real-time operating condition data, and externally input batch offline quality data. It monitors and verifies the parameter adjustment effect, equipment operating status, and batch quality, and determines anomalies, generating feedback deviation data and system operation control signals. Specifically, the execution status feedback data consists of the adjustment execution status, real-time equipment operating parameters, and restricted test response data output by the closed-loop execution control unit 4; the real-time operating condition data is multi-channel data output by the data transmission preprocessing unit 2 after range standardization and time axis synchronization; the batch offline quality data is the offline detection results of the current batch of extracted products, such as effective component content and extraction rate; the extraction index prediction value is the predicted value calculated by the mapping relationship between the corrected process parameters and extraction indexes in the intelligent parameter optimization unit 3. The feedback verification and operating condition control unit 5 aligns, verifies, and determines the above data, generating feedback deviation data and system operation control signals to achieve closed-loop monitoring of the extraction process's operating status and batch quality.

[0085] The feedback verification and operating condition control unit 5 includes a multi-source data alignment module 51, a regulation effect verification module 52, an operating condition quality judgment module 53, and a control signal output module 54. Specifically, the multi-source data alignment module 51 receives execution status feedback data, real-time operating condition data, and externally input batch offline quality data. Based on a unified time reference, it completes time-series alignment and data dimension matching, outputting the aligned multi-source verification dataset. Specifically, the multi-source data alignment module 51 receives four data streams: execution status feedback data, real-time operating condition data, batch offline quality data, and extracted indicator prediction values. The execution status feedback data carries a unified time identifier, with a time granularity equal to the preset sampling period of the closed-loop execution control unit 4. The real-time operating condition data also carries a unified time identifier, with a time granularity equal to the preset sampling period of the data transmission preprocessing unit 2. The batch offline quality data is batch-level data obtained through offline detection after the batch ends, identified by the batch number. The extracted indicator prediction values ​​are calculated by the intelligent parameter optimization unit 3 during the optimization process and sent to the multi-source data alignment module 51 along with the calculation request for the feedback deviation data, or transmitted via a separate data channel.

[0086] The multi-source data alignment module 51 first aligns the execution status feedback data and real-time operating condition data according to a unified time base. Since the sampling periods of the two types of data may differ, the data with the shorter sampling period is used as the reference time axis during alignment. Interpolation or nearest-neighbor value is then performed on the data with the longer sampling period to ensure a correspondence between the two types of data at the same time point. Linear interpolation can be used; this embodiment uses linear interpolation as an example. For batch offline quality data, a globally unique batch ID is used as the matching basis. Each production batch is assigned a unique identifier, and process data, step test data, and offline detection data are all bound to the corresponding batch ID. The offline detection results corresponding to this batch are then dimensionally matched with the data within the corresponding time period in the execution status feedback data and real-time operating condition data to form associated data under the same batch. The extracted indicator prediction values ​​are dimensionally matched with the corresponding batch's multi-source data based on the batch number.

[0087] The resulting multi-source verification dataset after alignment is a collection of data arranged along a unified timeline and carrying batch numbers. It includes target set values ​​for process parameters, actual operating values ​​of process parameters, predicted values ​​of extracted indicators, offline quality inspection values ​​for the batch, adjustment execution status, and restricted test response markers. Restricted test response data retains test data markers in the dataset, which the adjustment effect verification module 52 and the operating condition quality judgment module 53 use to distinguish between test data and normal operating data during use.

[0088] The adjustment effect verification module 52 compares the target set value and actual operating value of the process parameters, the predicted value of the extracted index and the offline quality detection value of the batch, based on the aligned multi-source verification dataset, calculates and generates feedback deviation data and outputs it to the intelligent parameter optimization unit 3; Specifically, the adjustment effect verification module 52 extracts the target set value of the process parameters, the actual operating value of the process parameters, the predicted value of the extracted index and the offline quality detection value of the batch from the multi-source verification dataset.

[0089] When comparing the target setpoints and actual operating values ​​of the process parameters, the adjustment effect verification module 52 calculates the process parameter deviations of each adjustment channel within a preset statistical window. Process parameter deviation of each adjustment channel The difference between the target setpoint and the actual operating value for this channel, i.e.: In the formula, For the first The target setpoint for each adjustment channel. For the first The actual operating value of each adjustment channel The dimensions are consistent with the corresponding process parameters.

[0090] When comparing the predicted values ​​of the extracted indicators with the offline quality detection values ​​of the batch, the adjustment effect verification module 52, after the offline quality data of the current batch is returned, backtracks and matches the process parameters and predicted values ​​of the corresponding batch to calculate the deviation of the extracted indicators. When offline quality data is returned with a lag, only historical prediction error statistics and batch fingerprint database are updated, without interfering with the real-time control logic of the currently running batch. Extract indicator deviations. The calculation formula is as follows: In the formula, The predicted values ​​of the extracted indicators are calculated by the mapping relationship between the corrected process parameters and the extracted indicators in the intelligent parameter optimization unit 3. This refers to the offline quality inspection value for the batch. The dimensions are consistent with the extracted indicators.

[0091] The adjustment effect verification module 52 summarizes the above-mentioned process parameter deviations and extracted index deviations, generates feedback deviation data, and outputs the feedback deviation data to the intelligent parameter optimization unit 3 for the dynamic calculation submodule of the credibility domain to calculate the recent prediction error. Among them, the process parameter deviations are mainly used for adjustment effect verification, and the extracted index deviations are used for recent prediction error. The statistics show that the feedback deviation data includes the corresponding time stamp and batch number.

[0092] The operating condition quality judgment module 53, based on the aligned multi-source verification dataset, monitors the operating status of the extraction equipment and the batch extraction quality in real time, identifying abnormal operating conditions such as equipment operating parameters exceeding limits (reflecting the degree of equipment baseline drift) and quality index deviations, and determining the level of abnormality. Specifically, the operating condition quality judgment module 53 monitors the following two aspects based on the multi-source verification dataset: Monitoring the operating status of the extraction equipment includes: real-time monitoring of the actual operating values ​​of the process parameters of each adjustment channel to determine whether they exceed the equipment safety boundary; and continuous tracking of the process parameter deviations of each adjustment channel. When the process parameter deviation of a certain channel exceeds a preset deviation threshold within multiple consecutive sampling periods, the adjustment effect of that channel is determined to be abnormal. The equipment safety boundary is consistent with the equipment safety boundary used by the trusted domain optimization module 34.

[0093] Monitoring the quality of batch extraction includes comparing the batch's offline quality detection value with preset quality indicator standards after the offline quality data for the current batch is returned. If the batch's offline quality detection value deviates from the quality indicator standard by more than a preset deviation threshold, the batch is deemed to have deviated from its quality indicator.

[0094] Furthermore, the anomaly levels are divided into three levels based on the degree and duration of deviation, as follows: Level 1 is the alert level anomaly, indicating that the deviation of process parameters or quality indicators is minor and short-lived, not affecting the continued operation of the batch, and only needs to be recorded and continuously observed; Level 2 is the warning level anomaly, indicating that the deviation of process parameters or quality indicators is significant or prolonged, requiring adjustment of the optimization strategy or parameters for subsequent batches; Level 3 is the stop level anomaly, indicating that the process parameters exceed the equipment safety boundary, or the quality indicators deviate severely, causing the current batch to be unable to continue, requiring the extraction process to be stopped and the process transferred to the anomaly handling procedure.

[0095] The thresholds for classifying the above-mentioned anomalies can be preset according to extraction process requirements and quality control standards. For example, when the deviation of process parameters does not exceed ±5% of the target set value and the deviation of quality indicators does not exceed the standard value. When the process parameter deviation exceeds a certain level, it is classified as a warning level; However, it did not exceed the equipment safety boundary, or the quality indicators deviated from the limit. But not exceeding When the actual operating value of the process parameters exceeds the equipment safety boundary, or the quality indicators deviate from the specified level, it is judged as a warning level; When this occurs, it is determined to be a stop level. The above thresholds are examples and can be adjusted according to the specific process.

[0096] Finally, the working condition quality judgment module 53 outputs the abnormal judgment result to the control signal output module 54. The abnormal judgment result includes the abnormality type, abnormality level, the relevant adjustment channel or quality indicator, and the judgment basis.

[0097] The control signal output module 54 generates a system operation control signal of the corresponding level based on the anomaly determination result and outputs it to the corresponding execution unit to perform control actions. Specifically, the control signal output module 54 receives the anomaly determination result output by the operating condition quality determination module 53 and generates a corresponding system operation control signal according to the anomaly level.

[0098] When the anomaly level is at the alert level, the control signal output module 54 generates an alert signal and outputs it to the display and recording module of the feedback verification and operating condition control unit 5 or the host computer to remind the operator to pay attention to the current operating condition. When the anomaly level is at the warning level, the control signal output module 54 generates a warning signal and outputs it to the intelligent parameter optimization unit 3 to trigger the intelligent parameter optimization unit 3 to reduce the reliable adjustment range accordingly in the next batch optimization, or outputs it to the host computer to prompt the operator to adjust the subsequent process strategy. When the anomaly level is at the stop level, the control signal output module 54 generates a stop signal and outputs it to the closed-loop execution control unit 4 to trigger the stop process of the extraction equipment, such as turning off the heater, stopping the solvent feed, and releasing the pressure. Batches with stop-level anomalies are judged as invalid production batches, and their material coupling batch fingerprints are not included in the historical qualified batch fingerprint set. When restarting production after an abnormal shutdown, a pre-feeding no-load benchmark test must be performed first to update the equipment benchmark response before feeding materials to avoid drift calculation distortion caused by changes in equipment status.

[0099] Furthermore, the system operation control signals carry the anomaly type, anomaly level, target execution unit identifier, and control action content. Each execution unit executes the corresponding control action according to the control signal content.

[0100] Through the above-mentioned multi-source data alignment, adjustment effect verification, working condition quality judgment and control signal output processing, the feedback verification and working condition control unit 5 can monitor and verify the parameter adjustment effect, equipment operating status and batch quality of the extraction process, and return the feedback deviation data to the intelligent parameter optimization unit 3 to form a closed loop, while outputting system operation control signals to ensure the safe and stable operation of the extraction process.

[0101] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A system for real-time acquisition and optimization of process parameters for durian peel extraction, characterized in that, include: Multi-dimensional sensing acquisition unit (1) is used to acquire on-site physical quantity signals during the durian peel extraction process, and to synchronously acquire and quantize the extraction process parameters and dynamic response data to obtain the original process data. Data transmission preprocessing unit (2), the data transmission preprocessing unit (2) is used to receive the original process data, perform noise reduction, standardization and time synchronization processing on the original process data to obtain real-time operating condition data; The intelligent parameter optimization unit (3) is used to receive the real-time operating data and the feedback deviation data output by the feedback verification and operating condition control unit (5). It adopts the batch adaptive optimization technology of decoupling equipment status and raw material characteristics to dynamically calibrate and optimize the mapping relationship between batch operating characteristics and process parameters in the extraction process, and generate the optimal process parameter instruction. The closed-loop execution control unit (4) is used to receive the optimal process parameter instruction, dynamically adjust and limit the operating parameters of the extraction equipment, and output the execution status feedback data. Feedback verification and operating condition control unit (5) is used to receive the execution status feedback data, real-time operating condition data and externally input batch offline quality data, monitor and verify the parameter adjustment effect, equipment operating status and batch quality and determine the anomaly, and generate feedback deviation data and system operation control signals.

2. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 1, characterized in that, The multi-dimensional sensing acquisition unit (1) includes a synchronous acquisition control module (11), a multi-dimensional sensing array module (12), and a signal quantization output module (13), wherein: The synchronous acquisition control module (11) generates a synchronous acquisition trigger signal based on a preset sampling period to control the sampling time of all acquisition channels to remain synchronized; The multi-dimensional sensor array module (12) senses the on-site physical quantity signals of the durian peel extraction process based on the synchronous acquisition trigger signal, and collects the analog sensor signals corresponding to the extraction process parameters and dynamic response data. The signal quantization output module (13) performs analog-to-quantization conversion based on the analog sensing signal and marks it with a unified time identifier to generate the original process data.

3. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 1, characterized in that, The data transmission preprocessing unit (2) includes a data receiving buffer module (21), a noise reduction and filtering module (22), and a standardization and synchronization module (23), wherein: The data receiving and caching module (21) receives the original process data and completes data temporary storage and transmission validity verification; The noise reduction filtering module (22) filters out the interference noise introduced during the acquisition and transmission process based on the original process data output by the data receiving buffer module (21) to obtain the noise-reduced process data. The standardization and synchronization module (23) performs range standardization and time axis synchronization and regularization based on the noise reduction process data to generate real-time operating condition data.

4. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 1, characterized in that, The intelligent parameter optimization unit (3) includes a batch decoupling characterization module (31), a time delay alignment module (32), a process mapping correction module (33), and a trusted domain optimization module (34), wherein: The batch decoupling characterization module (31) adopts the technique of differentiating the same source double response before and after feeding. Based on the real-time working condition data, it separates and extracts the equipment reference drift characteristics and raw material batch characteristics during the process, and outputs the material coupling batch fingerprint, equipment reference drift factor and material additional effect time delay. The time delay alignment module (32) uses a time axis translation calibration technique based on the material's additional time delay to reconstruct the process parameter sequence through time delay alignment. The process mapping correction module (33) uses a batch fingerprint-based feature conditional modulation technique to perform batch adaptation correction on the mapping relationship between process parameters and extracted indicators. The trusted domain optimization module (34) adopts a batch adaptive optimization technique with multi-factor dynamic constraints. Based on the aligned process parameter sequence, the mapping relationship between the corrected process parameters and the extracted indicators, the equipment reference drift factor and the feedback deviation data, it solves the optimal process parameter combination and generates the optimal process parameter instruction.

5. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 4, characterized in that, The batch decoupling characterization module (31) includes a response acquisition and calibration submodule, a material characteristic decoupling submodule, and an equipment drift calculation submodule, wherein: The response acquisition and calibration submodule is used to extract the equipment-medium reference response under steady-state condition before feeding and the equipment-medium-raw material coupling response under steady-state condition after feeding from the input real-time operating data. The two responses are then normalized to obtain the reference response value and coupling response value corresponding to the unit adjustment action of the j-th adjustment channel. At the same time, the action delay corresponding to each of the two responses is extracted. The material characteristic decoupling submodule calculates the material additional response corresponding to the unit solid load and constructs the material coupled batch fingerprint based on the baseline response value, coupled response value and the current batch of durian peel solid feed amount of the unit adjustment action; calculates the difference in action time delay between the two responses to obtain the material additional action time delay of each adjustment channel; calculates the raw material batch unfamiliarity based on the material coupled batch fingerprint and the calibrated set of historical qualified batch fingerprints; and generates the time delay identification credibility based on the confidence level of the response curve fitting. The additional material response per unit solid load is calculated using the following formula: In the formula, For the first Batch No. Additional response of material per unit solid load in each regulating channel; For the first The normalized unit action response value of the baseline response before batch feeding; For the first The normalized unit action response value of the coupled response after batch feeding; For the first Batch solid durian peel feed quantity; n is a very small positive constant, much smaller than the amount of solid durian peel used in the current batch, used to avoid division by zero anomalies; subscript For batch dimension identification, subscript To adjust the channel dimension identifier; The equipment drift calculation submodule is used to compare the current batch baseline response with the calibrated historical equipment baseline response to generate the equipment baseline drift factor.

6. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 4, characterized in that, The process of reconstructing the time delay alignment sequence of the time delay alignment module (32) includes the following steps: S32.1 Extract the material additional effect delay corresponding to each adjustment channel from the output of the batch decoupling characterization module (31); S32.

2. Based on the time of the adjustment action, the historical process parameter sequence is shifted along the time axis to the corresponding channel's material additional action time delay, so that the control action and response effect correspond one-to-one in the time dimension, and the time delay aligned process parameter sequence is obtained.

7. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 4, characterized in that, The process of batch adaptation and correction of the mapping relationship between process parameters and extracted indicators in the process mapping correction module (33) includes the following steps: S33.1 Based on the material coupled batch fingerprint, generate the feature gain coefficient and feature offset coefficient of the corresponding dimension through the preset linear mapping rule; S33.2 Using the aforementioned feature gain coefficient and feature offset coefficient, the intermediate feature quantity of the mapping model corresponding to the mapping relationship between process parameters and extracted indicators is conditionally modulated dimension by dimension to obtain batch-adaptive feature quantity. S33.3 Based on the corrected batch adaptation feature, output the mapping relationship between the process parameters and extraction indicators adapted to the current raw material batch.

8. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 4, characterized in that, The trusted domain optimization module (34) includes a trusted domain dynamic calculation submodule and a multi-objective constraint optimization submodule, wherein: The trusted domain dynamic calculation submodule is used to calculate the trusted adjustment domain scaling coefficient based on the unfamiliarity of the raw material batch, the equipment reference drift factor, and the recent prediction error and time lag identification trusted level within a preset statistical window, and to dynamically adjust the allowable fluctuation range of the optimization parameters; the trusted adjustment domain scaling coefficient is calculated by the following formula: In the formula, for The scaling factor of the reliable adjustment domain at any given time; The benchmark reliable adjustment domain coefficients; To assess the reliability of time delay identification, the subscript... Used as a time delay dimension identifier; For the first The unfamiliarity of each batch of raw materials; For the first Batch equipment reference drift factor; for The recent prediction error within the preset statistical window corresponding to the time point is obtained from the feedback deviation data statistics. These are preset positive weighting coefficients; subscripts For time dimension identification, subscript For batch dimension identification; The multi-objective constraint optimization submodule is used to solve for the optimal combination of process parameters and generate the optimal process parameter instruction within the adjusted credible adjustment domain and equipment safety boundary, based on the mapping relationship between the corrected process parameters and the extracted indicators.

9. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 1, characterized in that, The closed-loop execution control unit (4) includes an instruction parsing and adaptation module (41), a dynamic adjustment execution module (42), a restricted test control module (43), and an execution status feedback module (44), wherein: The instruction parsing and adaptation module (41) receives the optimal process parameter instruction, parses and converts it into control parameters that can be identified and executed by each adjustment channel of the extraction equipment; The dynamic adjustment execution module (42) adjusts the operating parameters of the extraction equipment in real time based on the control parameters, so that the corresponding process parameters track to the target value. The restricted test control module (43) is used to issue test adjustment instructions in the preset steady-state working condition stage, control the equipment to perform small-amplitude restricted adjustment tests and collect corresponding dynamic response data; The execution status feedback module (44) collects and summarizes the adjustment execution status, real-time operating parameters of the equipment and restricted test response data, generates execution status feedback data and outputs it.

10. The real-time acquisition and optimization system for durian peel extraction process parameters according to claim 1, characterized in that, The feedback verification and operating condition control unit (5) includes a multi-source data alignment module (51), a regulation effect verification module (52), an operating condition quality judgment module (53), and a control signal output module (54), wherein: The multi-source data alignment module (51) receives execution status feedback data, real-time operating condition data and externally input batch offline quality data, completes time sequence alignment and data dimension matching based on a unified time reference, and outputs the aligned multi-source verification dataset. The adjustment effect verification module (52) compares the target set value and actual operating value of the process parameters, the predicted value of the extracted index and the offline quality detection value of the batch based on the aligned multi-source verification dataset, calculates and generates feedback deviation data and outputs it to the intelligent parameter optimization unit (3). The working condition quality judgment module (53) monitors the operating status of the extraction equipment and the batch extraction quality in real time based on the aligned multi-source verification dataset, identifies abnormal working conditions and determines the abnormality level. The control signal output module (54) generates a system operation control signal of the corresponding level based on the anomaly determination result and outputs it to the corresponding execution unit to perform control actions.

Images