Intelligent monitoring method and system for edible oil production
By dynamically adjusting the leaching process parameters through online detection and deep learning models, the problem of fluctuations in residual oil content and solvent residue caused by unstable raw material quality was solved, thereby improving the stability and efficiency of edible oil production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI JINLONG CEREALS & OILS CO LTD
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-19
AI Technical Summary
In current edible oil production, the unstable quality of raw materials leads to large fluctuations in the residual oil rate of the meal during the leaching process, unstable control of solvent residue, decreased production line stability, and a lack of dynamic matching mechanism.
By conducting online detection of pre-pressed oil meal to obtain quality feature vectors, and using deep learning models and optimization algorithms to dynamically adjust leaching process parameters, including solvent ratio, leaching time and temperature, precise matching is achieved, reducing fluctuations in residual oil content and solvent residue.
It achieves stable control of solvent residue, reduces fluctuations in residual oil content in meal, improves oil yield and product stability, and ensures efficient operation of the production line.
Smart Images
Figure CN122239619A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edible oil production control technology, specifically to an intelligent monitoring method and system for edible oil production. Background Technology
[0002] In the edible oil industry, the oil extraction process from vegetable oilseeds typically employs a combined "pre-pressing + solvent extraction" method. In the pre-pressing stage, a portion of the oil is extracted through mechanical pressing, and the resulting oil meal enters the solvent extraction system. Organic solvents are used to extract the remaining oil from the meal, thus ensuring the full utilization of oil resources. This process is of great significance for improving the economic efficiency of enterprises and reducing resource consumption.
[0003] However, with the diversification of oilseed crop sources and the variation in planting conditions, different batches of raw materials exhibit significant fluctuations in oil content, moisture content, protein content, and fatty acid composition. This instability in raw material quality directly impacts the pre-processing and pressing stages, resulting in considerable variation in the quality of the oilseed meal obtained after pressing. In modern, large-scale, continuous production models, this instability in material quality directly affects subsequent leaching effects, making it difficult to maintain long-term stability in the production process.
[0004] Currently, most production enterprises still rely on experience parameters or periodic sampling results as the basis for process adjustment. Key parameters such as leaching temperature, solvent ratio and leaching time are usually set according to predetermined ranges, lacking a dynamic matching mechanism for changes in material quality. This leads to large fluctuations in residual oil rate in meal or fluctuations in solvent residue in crude oil. At the same time, the stability of the production line decreases, and it is difficult to maintain the production process within the ideal economic and technical indicator range for a long time. Summary of the Invention
[0005] The purpose of this invention is to solve the problems mentioned in the background art, such as the large fluctuations in residual oil content, unstable control of solvent residue, and decreased stability of production line caused by fluctuations in material quality in existing leaching processes. Therefore, this invention proposes an intelligent monitoring method and system for edible oil production.
[0006] A first aspect of this invention provides an intelligent monitoring method for edible oil production, the method comprising: The oil meal obtained after pre-pressing by a press is subjected to online detection to obtain the quality feature vector of the oil meal; the quality feature vector includes at least oil content, moisture content, protein content and oleic acid content; The quality feature vector is compared with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample. If a match is found, the optimal combination of variables for the standard sample is taken as the target variable combination; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature. If there is a mismatch, a preset optimization algorithm is invoked to optimize the variable combination and obtain the target variable combination. The optimization algorithm obtains the predicted value of residual oil content and solvent content by inputting different variable combinations and the quality feature vector into a pre-trained deep learning model. The model prediction results are used to iteratively optimize and find the target variable combination that makes the solvent content meet the qualified standard and the residual oil content of the meal reach the lowest level. The control parameters of the leachate are adjusted in real time based on the combination of target variables.
[0007] By implementing this technical solution, dynamic adaptive adjustment of leaching process parameters can be achieved, enabling precise matching of solvent ratio, leaching temperature, and leaching time with material quality. Under the premise of ensuring that the solvent residue meets the qualified standards, the residual oil rate of the meal and its fluctuation can be effectively reduced, the oil yield and product stability can be improved, and the production line can be guaranteed to operate continuously and efficiently.
[0008] Optionally, comparing the quality feature vector with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample includes: Calculate the Euclidean distance between the quality feature vector and the quality feature vector in each standard sample; and use the standard sample with the smallest Euclidean distance as the reference sample. If the minimum Euclidean distance is not greater than a preset distance threshold, then the oilseed meal is determined to match the reference sample; The process of setting the standard sample set and distance threshold includes: For any given historical sample, denote it as the target sample; Calculate the Euclidean distance between the target sample and the quality feature vectors of other historical samples, sort them in ascending order of distance, and obtain the first sample set; For the first sample set, the variable combinations of historical samples and the target sample are sequentially checked one by one until they are inconsistent with the variable combinations of the target sample, thus obtaining K samples of the same type. If K is greater than the preset quantity threshold, the target sample is used as the standard sample; and the Euclidean distance between it and the Kth sample of the same type is used as the distance threshold reference value. The minimum value among all standard sample distance threshold reference values is taken as the distance threshold.
[0009] By implementing this technical solution, standard samples are screened based on the distribution density of historical data and distance thresholds are set, which improves the accuracy and robustness of matching and discrimination, avoids false matching, and enhances system stability.
[0010] Optionally, the optimization algorithm employs a two-stage optimization mechanism combining global search and local fine-grained search, and the optimization process includes: Step 1: Initialize the particle population; each particle corresponds to a combination of variables to be optimized; and calculate the fitness of each particle to determine the individual optimal position and the global optimal position. Step 2: Based on the inherent update mechanism of the particle swarm optimization algorithm, update the particle velocity and particle position to obtain a new particle population, which serves as the wolf pack. Step 3: Calculate the fitness of each gray wolf in the current wolf pack, and determine the alpha wolf, Wolves and delta wolves; based on the inherent update mechanism of the gray wolf optimization algorithm, the positions of ordinary wolves are updated to obtain a new wolf pack, which is used as a particle swarm; Step 4: Calculate the fitness of each particle and update the individual optimal position and the global optimal position; Step 5: Determine whether the preset stopping condition has been met. If not, return to step 2; otherwise, terminate the iteration and output the variable combination corresponding to the current global optimal position.
[0011] By implementing this technical solution, the global exploration capability of the particle swarm optimization algorithm and the local exploitation capability of the gray wolf optimization algorithm are fully utilized, thereby improving the global optimality and convergence accuracy of variable combination optimization and avoiding getting trapped in local optima.
[0012] Optionally, the population can be initialized using a combination of variables from several historical samples; specifically: Calculate the Euclidean distance between the quality feature vector and the quality feature vector in each historical sample; sort the samples by distance from smallest to largest to obtain the second sample set; M particle positions are generated by selecting the variable combinations of the first M historical samples in the second sample set; Within the constraints of the variables, N particle positions are randomly generated, resulting in a total of M+N particles.
[0013] By implementing this technical solution, the population is initialized by combining historical experience information, which improves the convergence speed of the algorithm. At the same time, random particles are introduced to maintain population diversity and enhance global search capabilities, thus balancing computational efficiency and the quality of the optimal solution.
[0014] Optionally, the fitness calculation formula is as follows: ; Where fitness is the desired fitness level; R pred This is the predicted residual oil content in the meal; It is the penalty coefficient; Penalty is the penalty for exceeding the limit; S pred This is the predicted solvent residue; S limit This is the acceptable standard for solvent residue.
[0015] By implementing this technical solution, a fitness function is constructed that integrates the goal of minimizing residual oil content in the meal with the safety constraint of solvent residue, thereby achieving unified optimization of multiple objectives and ensuring the synergistic improvement of production quality and safety indicators.
[0016] A second aspect of this invention provides an intelligent monitoring system for edible oil production, the system comprising: The data monitoring module is used to perform online detection on the oil meal obtained after pre-pressing by the press and to obtain the quality feature vector of the oil meal; the quality feature vector includes at least oil content, moisture content, protein content and oleic acid content; The feature matching module is used to compare the quality feature vector with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample. The matching determination module is used to, if a match is found, take the optimal combination of variables of the standard sample as the target variable combination; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature; The optimization determination module is used to call a preset optimization algorithm to optimize the variable combination if there is no match, so as to obtain the target variable combination. The optimization algorithm obtains the predicted value of residual oil content and solvent content by inputting different variable combinations and the quality feature vector into a pre-trained deep learning model. The model prediction results are used to iteratively optimize in order to find the target variable combination that makes the solvent content meet the qualified standard and the residual oil content of the meal reach the minimum. The control execution module is used to adjust the control parameters of the leachate in real time according to the combination of target variables.
[0017] Optionally, the system further includes a stable region modeling module; wherein: The stable region modeling module is used to: calculate the Euclidean distance between the target sample and the quality feature vectors of other historical samples for any given historical sample, and sort them in ascending order of distance to obtain a first sample set; sequentially determine whether the variable combinations of historical samples and the target sample are consistent until they are inconsistent, thus obtaining K similar samples; if K is greater than a preset quantity threshold, then the target sample is used as a standard sample; and the Euclidean distance between the target sample and the Kth similar sample is used as a distance threshold reference value; the minimum value among all the distance threshold reference values of the standard samples is taken as the distance threshold. The feature matching module is used to calculate the Euclidean distance between the quality feature vector of the oil meal and the quality feature vector in each standard sample; the standard sample corresponding to the smallest Euclidean distance is used as the reference sample; if the smallest Euclidean distance is not greater than a preset distance threshold, the oil meal is determined to match the reference sample.
[0018] Optionally, the optimization determination module employs a two-stage optimization mechanism combining global search and local fine-grained search; the optimization determination module includes: The initialization module is used to initialize the particle population; each particle corresponds to a combination of variables to be optimized; and the fitness of each particle is calculated to determine the individual optimal position and the global optimal position. The global exploration module is used to update particle velocity and particle position according to the inherent update mechanism of the particle swarm optimization algorithm, so as to obtain a new particle population, which is the wolf pack. The local exploration module is used to calculate the fitness of each gray wolf in the current wolf pack and determine the alpha wolf. Wolves and delta wolves; based on the inherent update mechanism of the gray wolf optimization algorithm, the positions of ordinary wolves are updated to obtain a new wolf pack, which is used as a particle swarm; The optimal solution update module is used to calculate the fitness of each particle and update the individual optimal position and the global optimal position; The iteration control module is used to determine whether the preset stopping condition has been met. If not, it returns to step two; otherwise, it terminates the iteration and outputs the variable combination corresponding to the current global optimal position.
[0019] Optionally, the initialization module initializes the population using a combination of variables from several historical samples; the initialization module includes: The similarity sorting module is used to calculate the Euclidean distance between the quality feature vector and the quality feature vector in each historical sample; sort them in ascending order of distance to obtain the second sample set; The optimal solution utilization module is used to select the variable combination of the first M historical samples in the second sample set to generate M particle positions; The random generation module is used to randomly generate N particle positions within the constraints of variables, generating a total of M+N particles.
[0020] Optionally, the fitness calculation formula is as follows: ; Where fitness is the desired fitness level; R pred This is the predicted residual oil content in the meal; It is the penalty coefficient; Penalty is the penalty for exceeding the limit; S pred This is the predicted solvent residue; S limit This is the acceptable standard for solvent residue. Attached Figure Description
[0021] Figure 1 A flowchart of an intelligent monitoring method for edible oil production provided in an embodiment of the present invention; Figure 2This is an architecture diagram of an intelligent monitoring system for edible oil production provided in an embodiment of the present invention. Detailed Implementation
[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention is provided in conjunction with the accompanying drawings and preferred embodiments.
[0023] This invention provides an intelligent monitoring method for edible oil production. See also... Figure 1 , Figure 1 A flowchart illustrating an intelligent monitoring method for edible oil production provided in an embodiment of the present invention. The method includes the following steps: S101, online detection is performed on the oil meal obtained after pre-pressing by the press to obtain the quality feature vector of the oil meal.
[0024] S102, compare the quality feature vector with the pre-stored standard sample set to determine whether the oilseed meal matches the standard sample.
[0025] S103, if a match is found, the optimal combination of variables in the standard sample is taken as the target variable combination.
[0026] S104 If there is no match, the preset optimization algorithm is called to optimize the variable combination and obtain the target variable combination.
[0027] S105, adjusts the control parameters of the leaching unit in real time according to the combination of target variables.
[0028] The quality feature vector includes at least oil content, moisture content, protein content, and oleic acid content; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature.
[0029] Based on the intelligent monitoring method for edible oil production provided by the embodiments of the present invention, a quality feature vector is constructed by online detection of pre-pressed oil meal, and combined with standard sample matching and deep learning model, dynamic adaptive adjustment of leaching parameters is achieved. This enables precise matching of solvent ratio, leaching time and temperature with material quality, reduces fluctuations in residual oil rate of meal while ensuring that solvent residue meets the standards, improves oil yield and product stability, reduces human intervention, ensures efficient operation of the production line, and achieves synergistic improvement of economic benefits and quality and safety.
[0030] In one implementation, online detection can be performed using a near-infrared spectroscopy analyzer, and the average value of detection data from multiple time windows can be taken as the quality data for the current batch. This effectively reduces errors caused by instantaneous noise, equipment vibration, and uneven flow conditions, making the quality data more stable and reliable. The quality data is then normalized to generate a quality feature vector.
[0031] In one embodiment, the database includes a historical sample set and a standard sample set. The process of determining whether oilseed meal matches the standard sample includes: Step 1: Calculate the Euclidean distance between the current oilseed meal quality feature vector and the quality feature vector in each historical sample; if there is a case where the distance is 0, then the oilseed meal is determined to match the historical sample.
[0032] Step 2: Calculate the Euclidean distance between the current oilseed meal quality feature vector and the quality feature vector in each standard sample; take the standard sample with the smallest Euclidean distance as the reference sample; if the smallest Euclidean distance is not greater than the preset distance threshold, then determine that the oilseed meal matches the reference sample.
[0033] In one implementation, the standard sample is a sample with a stable neighborhood, and the specific selection process includes: Step 1: Calculate the Euclidean distance between the target sample and the quality feature vectors of other historical samples, sort them in ascending order of distance to obtain the first sample set; the target sample is any historical sample.
[0034] Step 2: For the first sample set, sequentially determine whether the variable combinations of the historical samples and the target samples are consistent, until the variable combinations are inconsistent with the target samples, and then stop to obtain K samples of the same type.
[0035] Step 3: If K is greater than the preset quantity threshold, then the target sample is used as the standard sample.
[0036] Step 4: Use the Euclidean distance between the target sample and its Kth similar sample as a reference value for the distance threshold.
[0037] Step 5: Take the minimum value among all the distance threshold reference values of the standard samples as the distance threshold.
[0038] This implementation method selects standard samples based on historical data distribution density and local neighborhood relationships, ensuring that the standard samples reflect the stable state under real production conditions, thus avoiding misclassification of isolated abnormal data as standard samples. Simultaneously, it improves the reliability of standard sample selection by setting a threshold for the number of similar samples to assess the stability of target samples. The minimum value among the distance threshold reference values is selected as the final distance threshold, strictly constraining the matching decision boundary, thereby reducing the probability of false matching and enhancing the system's stability and robustness in industrial online monitoring environments.
[0039] In one implementation, the quantity threshold is essentially the minimum number of samples required to ensure statistical stability and avoid mismatches caused by sparse neighborhood samples. The specific value can be set by technical personnel based on the historical sample size and distribution density, for example, set to 15.
[0040] In one embodiment, a two-stage optimization algorithm combining global search and local fine-grained search is proposed. The optimization process includes: Step 1: Initialize the particle population. Each particle corresponds to a combination of variables to be optimized; and calculate the fitness of each particle to determine the individual optimal position and the global optimal position.
[0041] Step two: Based on the inherent update mechanism of the particle swarm optimization algorithm, update the particle velocity and particle position to obtain a new particle population, which serves as the wolf pack.
[0042] Step 3: Calculate the fitness of each gray wolf in the current wolf pack, and determine the alpha wolf, Wolves and delta wolves; based on the inherent update mechanism of the gray wolf optimization algorithm, the positions of ordinary wolves are updated to obtain a new wolf pack, which is used as a particle swarm.
[0043] Step 4: Calculate the fitness of each particle and update the individual optimal position and the global optimal position.
[0044] Step 5: Determine whether the preset stopping condition has been met. If not, return to step 2; otherwise, terminate the iteration and output the variable combination corresponding to the current global optimal position.
[0045] This embodiment constructs a two-stage optimization mechanism that combines global search and local fine-grained search. This mechanism can fully leverage the global exploration capability of the particle swarm optimization algorithm and the local exploitation capability of the gray wolf optimization algorithm, enabling the variable combination optimization process to have both strong search breadth and improved convergence accuracy. By alternately updating the population state between the two optimization algorithms, the problem of premature convergence or getting trapped in local optima by a single optimization algorithm can be effectively avoided, thereby improving the global optimality of the target variable combination.
[0046] In one implementation, the population is initialized using a combination of variables from several historical samples. Specifically, this includes: Step 1: Sort the current oilseed meal quality feature vectors and the quality feature vectors of each historical sample in ascending order of distance to obtain the second sample set.
[0047] Step 2: Select the variable combinations of the first M historical samples in the second sample set to generate M particle positions.
[0048] Step 3: Randomly generate N particle positions within the constraints of the variables, generating a total of M+N particles.
[0049] This implementation combines historical experience information, selecting the M most similar historical sample variable combinations as the initial particle positions. This makes the population initialization more closely resemble real production conditions, thereby improving the algorithm's convergence speed. Simultaneously, N particle positions are randomly generated, introducing randomness to ensure population diversity, enhance global search capabilities, and reduce the risk of the algorithm getting trapped in local optima. By combining experience-driven deterministic initialization with a stochastic exploration mechanism, the optimization algorithm achieves both high computational efficiency and the ability to find the global optimum.
[0050] In one implementation, the particle swarm size M+N can be set to 30, including 5 combinations of variables from historical samples and 25 combinations of randomly generated variables.
[0051] In one implementation, the constraint range of the variables can be adjusted according to the actual production process conditions, and should generally meet the safety operation requirements and process feasibility requirements of the edible oil production equipment. For example, the solvent ratio can be set between 0.8 and 1.5, the leaching time can be set between 60 and 120 minutes, and the leaching temperature can be set between 40℃ and 60℃.
[0052] In one implementation, the optimization objective is to find the combination of variables that minimizes solvent residue and reduces residual oil content in the meal. The fitness calculation formula is as follows: ; Where fitness is the desired fitness level; R pred This is the predicted residual oil content in the meal; This is the penalty coefficient, taking a larger value, such as 20; Penalty is the penalty for exceeding the limit; S pred This is the predicted solvent residue; S limit This is the acceptable standard for solvent residue.
[0053] This fitness function integrates the goal of minimizing residual oil content in the meal with the safety constraint of solvent residue, so that the optimization process not only pursues the reduction of residual oil content, but also ensures that the solvent residue meets the qualified standard, thereby achieving unified control of production quality and safety indicators.
[0054] In one implementation, different combinations of variables and quality feature vectors are input into a pre-trained deep learning model to obtain predicted values for residual oil content and solvent residue. The deep learning model can be a prediction model based on a multi-layer neural network structure, such as a multilayer perceptron. The model training process is supervised learning based on historical actual production data and laboratory sample data. During training, the actually measured residual oil content and solvent residue are used as the true labels. The backpropagation algorithm is used to continuously adjust the network weight parameters, optimizing model performance by minimizing the error function between the predicted values and the true labels.
[0055] This implementation method, by introducing a deep learning prediction model, can model the nonlinear mapping relationship between complex process parameters and production quality indicators, improve prediction accuracy, and thus provide data support for the real-time optimization of leaching unit control parameters.
[0056] In one implementation, the iteration stopping condition includes reaching the maximum number of iterations (e.g., 100) or the change in the global optimal fitness being less than a change threshold (e.g., 10) for several consecutive generations (e.g., 10 generations). -3 ).
[0057] This invention provides an intelligent monitoring system for edible oil production. See also... Figure 2 , Figure 2 This is an architecture diagram of an intelligent monitoring system for edible oil production provided in an embodiment of the present invention. The system includes: The data monitoring module is used to perform online detection on the oil meal obtained after pre-pressing by the press and to obtain the quality feature vector of the oil meal.
[0058] The feature matching module is used to compare the quality feature vector with the pre-stored standard sample set to determine whether the oilseed meal matches the standard sample.
[0059] The matching determination module is used to determine the optimal combination of variables of the standard sample as the target variable combination if a match is found.
[0060] The optimization determination module is used to call a preset optimization algorithm to optimize the variable combination if there is no match, so as to obtain the target variable combination.
[0061] The control execution module is used to adjust the control parameters of the leacher in real time according to the combination of target variables.
[0062] The quality feature vector includes at least oil content, moisture content, protein content, and oleic acid content; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature. The optimization algorithm inputs different variable combinations and the quality feature vector into a pre-trained deep learning model to obtain predicted values for residual oil content and solvent residue. Iterative optimization is then performed based on the model's prediction results to find the target variable combination that minimizes solvent residue and achieves the lowest possible residual oil content.
[0063] Based on the intelligent monitoring system for edible oil production provided by the embodiments of the present invention, the system performs online detection of pre-pressed oil meal, constructs a quality feature vector, and combines standard sample matching and deep learning models to achieve dynamic adaptive adjustment of leaching parameters. This enables precise matching of solvent ratio, leaching time and temperature with material quality, reduces fluctuations in residual oil rate of meal while ensuring that solvent residue meets standards, improves oil yield and product stability, reduces manual intervention, ensures efficient operation of the production line, and achieves synergistic improvement in economic benefits and quality safety.
[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention are within the scope of the claims of the present invention.
Claims
1. A smart monitoring method for edible oil production, characterized in that, The method includes: The oil meal obtained after pre-pressing by a press is subjected to online detection to obtain the quality feature vector of the oil meal; the quality feature vector includes at least oil content, moisture content, protein content and oleic acid content; The quality feature vector is compared with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample. If a match is found, the optimal combination of variables for the standard sample is taken as the target variable combination; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature. If there is a mismatch, a preset optimization algorithm is invoked to optimize the variable combination and obtain the target variable combination. The optimization algorithm obtains the predicted value of residual oil content and solvent content by inputting different variable combinations and the quality feature vector into a pre-trained deep learning model. The model prediction results are used to iteratively optimize and find the target variable combination that makes the solvent content meet the qualified standard and the residual oil content of the meal reach the lowest level. The control parameters of the leachate are adjusted in real time based on the combination of target variables.
2. The intelligent monitoring method for edible oil production according to claim 1, characterized in that, The step of comparing the quality feature vector with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample includes: Calculate the Euclidean distance between the quality feature vector and the quality feature vector in each standard sample; and use the standard sample with the smallest Euclidean distance as the reference sample. If the minimum Euclidean distance is not greater than a preset distance threshold, then the oilseed meal is determined to match the reference sample; The process of setting the standard sample set and distance threshold includes: For any given historical sample, denote it as the target sample; Calculate the Euclidean distance between the target sample and the quality feature vectors of other historical samples, sort them in ascending order of distance, and obtain the first sample set; For the first sample set, the variable combinations of historical samples and the target sample are sequentially checked one by one until they are inconsistent with the variable combinations of the target sample, thus obtaining K samples of the same type. If K is greater than the preset quantity threshold, the target sample is used as the standard sample; and the Euclidean distance between it and the Kth sample of the same type is used as the distance threshold reference value. The minimum value among all standard sample distance threshold reference values is taken as the distance threshold.
3. The intelligent monitoring method for edible oil production according to claim 1, characterized in that, The optimization algorithm employs a two-stage optimization mechanism combining global search and local fine-grained search. The optimization process includes: Step 1: Initialize the particle population; each particle corresponds to a combination of variables to be optimized; and calculate the fitness of each particle to determine the individual optimal position and the global optimal position. Step 2: Based on the inherent update mechanism of the particle swarm optimization algorithm, update the particle velocity and particle position to obtain a new particle population, which serves as the wolf pack. Step 3: Calculate the fitness of each gray wolf in the current wolf pack, and determine the alpha wolf, Wolves and delta wolves; based on the inherent update mechanism of the gray wolf optimization algorithm, the positions of ordinary wolves are updated to obtain a new wolf pack, which is used as a particle swarm; Step 4: Calculate the fitness of each particle and update the individual optimal position and the global optimal position; Step 5: Determine whether the preset stopping condition has been met. If not, return to step 2; otherwise, terminate the iteration and output the variable combination corresponding to the current global optimal position.
4. The intelligent monitoring method for edible oil production according to claim 3, characterized in that, The population is initialized using a combination of variables from several historical samples; specifically: Calculate the Euclidean distance between the quality feature vector and the quality feature vector in each historical sample; sort the samples by distance from smallest to largest to obtain the second sample set; M particle positions are generated by selecting the variable combinations of the first M historical samples in the second sample set; Within the constraints of the variables, N particle positions are randomly generated, resulting in a total of M+N particles.
5. The intelligent monitoring method for edible oil production according to claim 3, characterized in that, The formula for calculating fitness is: ； Where fitness is the desired fitness level; R pred This is the predicted residual oil content in the meal; It is the penalty coefficient; Penalty is the penalty for exceeding the limit; S pred This is the predicted solvent residue; S limit This is the acceptable standard for solvent residue.
6. An intelligent monitoring system for edible oil production, characterized in that, The system includes: The data monitoring module is used to perform online detection on the oil meal obtained after pre-pressing by the press and to obtain the quality feature vector of the oil meal; the quality feature vector includes at least oil content, moisture content, protein content and oleic acid content; The feature matching module is used to compare the quality feature vector with a pre-stored standard sample set to determine whether the oilseed meal matches the standard sample. The matching determination module is used to, if a match is found, take the optimal combination of variables of the standard sample as the target variable combination; the target variable combination includes at least one of solvent ratio, leaching time, and leaching temperature; The optimization determination module is used to call a preset optimization algorithm to optimize the variable combination if there is no match, so as to obtain the target variable combination. The optimization algorithm obtains the predicted value of residual oil content and solvent content by inputting different variable combinations and the quality feature vector into a pre-trained deep learning model. The model prediction results are used to iteratively optimize in order to find the target variable combination that makes the solvent content meet the qualified standard and the residual oil content of the meal reach the minimum. The control execution module is used to adjust the control parameters of the leachate in real time according to the combination of target variables.
7. The intelligent monitoring system for edible oil production according to claim 6, characterized in that, The system also includes a stable region modeling module; wherein: The stable region modeling module is used to: calculate the Euclidean distance between the target sample and the quality feature vectors of other historical samples for any given historical sample, and sort them in ascending order of distance to obtain a first sample set; sequentially determine whether the variable combinations of historical samples and the target sample are consistent until they are inconsistent, thus obtaining K similar samples; if K is greater than a preset quantity threshold, then the target sample is used as a standard sample; and the Euclidean distance between the target sample and the Kth similar sample is used as a distance threshold reference value; the minimum value among all the distance threshold reference values of the standard samples is taken as the distance threshold. The feature matching module is used to calculate the Euclidean distance between the quality feature vector of the oil meal and the quality feature vector in each standard sample; the standard sample corresponding to the smallest Euclidean distance is used as the reference sample; if the smallest Euclidean distance is not greater than a preset distance threshold, the oil meal is determined to match the reference sample.
8. The intelligent monitoring system for edible oil production according to claim 6, characterized in that, The optimization determination module employs a two-stage optimization mechanism combining global search and local fine-grained search; the optimization determination module includes: The initialization module is used to initialize the particle population; each particle corresponds to a combination of variables to be optimized; and the fitness of each particle is calculated to determine the individual optimal position and the global optimal position. The global exploration module is used to update particle velocity and particle position according to the inherent update mechanism of the particle swarm optimization algorithm, so as to obtain a new particle population, which is the wolf pack. The local exploration module is used to calculate the fitness of each gray wolf in the current wolf pack and determine the alpha wolf. Wolves and delta wolves; based on the inherent update mechanism of the gray wolf optimization algorithm, the positions of ordinary wolves are updated to obtain a new wolf pack, which is used as a particle swarm; The optimal solution update module is used to calculate the fitness of each particle and update the individual optimal position and the global optimal position; The iteration control module is used to determine whether the preset stopping condition has been met. If not, it returns to step two; otherwise, it terminates the iteration and outputs the variable combination corresponding to the current global optimal position.
9. The intelligent monitoring system for edible oil production according to claim 8, characterized in that, The initialization module initializes the population using a combination of variables from several historical samples; the initialization module includes: The similarity sorting module is used to calculate the Euclidean distance between the quality feature vector and the quality feature vector in each historical sample; sort them in ascending order of distance to obtain the second sample set; The optimal solution utilization module is used to select the variable combination of the first M historical samples in the second sample set to generate M particle positions; The random generation module is used to randomly generate N particle positions within the constraints of variables, generating a total of M+N particles.
10. The intelligent monitoring system for edible oil production according to claim 8, characterized in that, The formula for calculating fitness is: ； Where fitness is the desired fitness level; R pred This is the predicted residual oil content in the meal; It is the penalty coefficient; Penalty is the penalty for exceeding the limit; S pred This is the predicted solvent residue; S limit This is the acceptable standard for solvent residue.