An intelligent mineral water processing process optimization method and system
By improving the prototype network model and the pelican optimization algorithm, the problems of high annotation cost and local optima in the traditional mineral water treatment process optimization are solved, and efficient and adaptive mineral water treatment process optimization is achieved.
Patent Information
- Application Number
- CN202610749623.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-28
- Publication Date
- 2026-07-28
AI Technical Summary
Traditional mineral water treatment process optimization methods rely on a large number of labeled process parameter samples, which leads to high labeling costs and susceptibility to class imbalance and insufficient samples. The optimization effect is not ideal, and it is easy to get stuck in local optima in high-dimensional parameter space, making it difficult to quickly adjust parameters to cope with fluctuations in raw water quality.
We employ an improved prototype network model for data processing, utilize labeled and unlabeled data for learning, and combine it with the Pelican optimization algorithm for parameter optimization. This approach is highly adaptive and can optimize parameters in real time in dynamic environments.
It reduces class imbalance sensitivity, improves optimization accuracy when samples are insufficient, enhances the model's generalization ability and optimization effect, and can quickly cope with the complexity and variability of water quality data.
Smart Images
Figure CN122472281A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent optimization technology for mineral water treatment processes, specifically to an intelligent method and system for optimizing mineral water treatment processes. Background Technology
[0002] Mineral water treatment process optimization refers to the timely identification of inefficiencies or water quality abnormalities in the mineral water production process through multi-dimensional assessment of water quality parameters, treatment equipment operation status, raw water quality characteristics, and environmental conditions. This allows for the provision of intelligent control strategies and intervention measures, thereby not only helping to improve the stability and safety of mineral water production, reduce energy and chemical consumption, and promote the greening and intelligentization of the production process, but also driving quality improvement and resource utilization efficiency in the mineral water industry.
[0003] However, traditional mineral water treatment process optimization methods rely on a large number of labeled process parameter samples, resulting in high labeling costs and susceptibility to class imbalance and insufficient samples, leading to unsatisfactory optimization results. Traditional mineral water treatment process optimization methods also require a large amount of computational resources when optimizing process parameters, are prone to getting trapped in local optima in high-dimensional parameter spaces, and are difficult to quickly adjust parameters to cope with fluctuations in raw water quality, resulting in the inability to fully explore effective parameter combinations. Summary of the Invention
[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an intelligent method and system for optimizing mineral water treatment processes. Traditional mineral water treatment process optimization methods rely on a large number of labeled process parameter samples, leading to high labeling costs and susceptibility to class imbalance and insufficient samples, resulting in unsatisfactory optimization effects. This solution creatively employs an improved prototype network model for mineral water treatment process optimization. This model effectively utilizes labeled process parameter data and a large amount of unlabeled process parameter data, reducing sensitivity to class imbalance and improving optimization accuracy when samples are insufficient. Furthermore, it effectively integrates data features from different sources, enhancing the model's generalization ability and optimization effect. Traditional mineral water treatment process optimization methods require significant computational resources for process parameter optimization, are prone to getting trapped in local optima in high-dimensional parameter spaces, and struggle to quickly adjust parameters to cope with fluctuations in raw water quality, preventing the full exploration of effective parameter combinations. This solution creatively employs the Pelican Optimization Algorithm for process parameter optimization. This algorithm is highly adaptive, converges quickly, and can optimize parameters in real-time in dynamic production environments, thereby improving the optimization effect of the mineral water treatment process and effectively addressing the complexity and variability of water quality data.
[0005] The technical solution adopted by this invention is as follows: This invention provides an intelligent method for optimizing mineral water treatment processes, which includes the following steps:
[0006] Step S1: Acquisition of raw data;
[0007] Step S2: Preliminary data processing;
[0008] Step S3: Construction of process optimization model;
[0009] Step S4: Process parameter optimization;
[0010] Step S5: Optimization of mineral water treatment process.
[0011] Furthermore, in step S1, the raw data acquisition is used to acquire multi-dimensional raw information reflecting the state and results of the mineral water treatment process. Specifically, it involves obtaining a raw mineral water treatment dataset through data collection. The raw mineral water treatment dataset specifically includes raw water quality data, treatment equipment operation data, process control parameters, environmental data, and effluent water quality data.
[0012] Further, in step S2, the preliminary data processing is used to perform preliminary processing on the collected raw data, transforming the raw data into a regular and model-readable format, specifically including the following steps:
[0013] Step S21: Data cleaning, used to remove noise and invalid information from the original data to ensure data quality. Specifically, missing and duplicate values in the original dataset are removed by removing mineral water to obtain a cleaner coarsely processed dataset.
[0014] Step S22: Data labeling, used to provide necessary supervision signals for model training, while retaining some unlabeled data for semi-supervised learning. Specifically, unlabeled datasets are obtained by randomly sampling a portion of the coarsely processed dataset, and the remaining data are labeled as high-quality or in a state to be optimized according to the processing effect, resulting in labeled datasets with labels.
[0015] Step S23: Data encoding, which is used to convert non-numerical or categorical data into numerical vectors so that they can be processed by mathematical models. Specifically, one-hot encoding is used to encode the unlabeled dataset and the labeled dataset respectively to obtain the corresponding unlabeled encoded dataset and labeled encoded dataset.
[0016] Step S24: Data normalization, used to eliminate differences in units and orders of magnitude between different feature parameters. Specifically, the min-max method is used to scale the encoded data to a specific range to obtain a scaled unlabeled normalized dataset and a labeled normalized dataset.
[0017] Step S25: Dataset partitioning, used to obtain training data and test data, to simulate the model learning with a small number of samples and evaluate its ability to recognize new samples. Specifically, the labeled normalized dataset is partitioned to obtain a processing state support set for model training reference and a processing state query set for model validation.
[0018] By integrating and executing the aforementioned data cleaning, data labeling, data encoding, data normalization, and dataset partitioning steps, the final output includes the unlabeled normalized dataset, processing state support set, and processing state query set required for model construction.
[0019] Further, in step S3, the process optimization model construction is used to construct a model required for intelligent identification of the mineral water treatment process status. Specifically, it constructs an improved prototype network model as the process optimization model. The improved prototype network model uses a small amount of labeled data and a large amount of unlabeled data for effective learning, distinguishes between high-quality and unoptimized treatment states, and outputs the mineral water treatment process status prediction results.
[0020] The process optimization model construction specifically includes the following steps:
[0021] Step S31: Preliminary feature extraction, used to automatically learn and extract deep feature representations that are meaningful for distinguishing processing states from the preliminarily processed data. Specifically, the convolutional neural network module performs feature transformation on the unlabeled normalized dataset, the processing state support set, and the processing state query set respectively to obtain the corresponding unlabeled feature set, support feature set, and query feature set.
[0022] Step S32: Calculate the state prototype, which is used to calculate a representative feature center point for each type of processing state as a standard reference for that type. Specifically, by calculating the average value of the feature vectors of all samples in each type, the state prototype representing the typical features of that type of processing state is obtained.
[0023] Step S33: Design a semi-supervised learning method to enhance the model's learning ability when labeled data is limited, utilizing a large amount of unlabeled data. The steps include:
[0024] Step S331: Calculate transportation costs to quantify the cost of allocating unlabeled data to each existing processed state category, providing a basis for pseudo-label allocation. Specifically, this involves defining the differences between the features of the unlabeled data and each state prototype by calculating the cosine similarity between them, and then calculating the transportation costs.
[0025] Step S332: Assign pseudo-labels to automatically generate reliable category labels for unlabeled data. Specifically, by comprehensively considering the optimal assignment results based on optimal transportation theory and the confidence level of model prediction, a pseudo-label is dynamically assigned to each unlabeled feature.
[0026] Step S333: Correct the state prototype, which is used to update the feature representation of the original state prototype with newly introduced unlabeled data with pseudo-labels, so that it is closer to the real distribution of the data. Specifically, through a weighted average operation, the feature information of the original state prototype and the unlabeled data are fused to obtain a more representative updated state prototype after correction.
[0027] Step S334: Design a semi-supervised learning loss function to introduce the use of unlabeled data in the model training objective, driving the model to learn a feature space that can allocate unlabeled data to each processing state category at the lowest cost. Specifically, the loss term is constructed by minimizing the total transport cost between unlabeled features and state prototypes.
[0028] Step S34: Obtain the model output to drive the model to perform processing state category classification prediction. Specifically, calculate the similarity between the model input features and the updated state prototypes of each processing state category after correction, and process them through the softmax activation function to obtain the processing state category prediction probability distribution, which serves as the state prediction result of the mineral water processing process.
[0029] Step S35: Design a model loss function to guide the overall optimization direction of model training. Specifically, design a model compactness loss and a model uniformity loss, and combine them with semi-supervised learning loss to form a comprehensive total model loss function.
[0030] Step S36: Model construction and training, specifically, through the preliminary feature extraction, the computation of the state prototype, the design of semi-supervised learning, the acquisition of model output and the design of model loss function, the improved prototype network model is constructed by integrating these steps. The model is trained and its performance is verified based on the support feature set and the query feature set to obtain the improved prototype network model as the process optimization model.
[0031] Further, in step S4, the process parameter optimization is used to automatically search for the optimal combination of key process parameters after the model identifies the state to be optimized. Specifically, the Pelican optimization algorithm is used to perform optimization search to obtain the optimized combination of key process parameters. The combination of key process parameters specifically includes processing equipment operating data and process control parameters.
[0032] The process parameter optimization specifically includes the following steps:
[0033] Step S41: Algorithm initialization, used to set the starting search point, stopping conditions, and evaluation criteria for the Pelican optimization algorithm. The steps include:
[0034] Step S411: Initialize the search space to define the range and representation of the key process parameters to be optimized, providing the starting point for the algorithm's search. Specifically, this is achieved by constructing a set of pelican units, where the position vector of each pelican unit represents a combination of key process parameters, thereby initializing the pelican unit position matrix.
[0035] Step S412: Determine the iteration stopping condition, which is used to control the runtime and convergence criterion of the optimization algorithm. Specifically, the maximum number of iterations and the desired fitness threshold are preset as the conditions for the algorithm to terminate the search.
[0036] Step S413: Determine the fitness, which is used to quantitatively evaluate the merits of each set of key process parameter combinations. Specifically, by defining a fitness calculation function, the key process parameter combinations represented by the pelican unit are input and the fitness value is calculated. The fitness calculation function is defined by an independent multilayer perceptron.
[0037] Step S42: Unit position update, used to drive the pelican unit to move to a better area in the search space, includes the following steps:
[0038] Step S421: Design the approach-prey stage to guide the pelican unit to perform a local fine search in the direction of the currently known optimal solution. Specifically, the position adjustment is determined by comparing the fitness value of each pelican unit with that of the current global optimal unit.
[0039] Step S422: Design the surface predation stage to enhance the algorithm's global exploration capability in the later stages of the search and avoid getting stuck in local optima. Specifically, this is achieved by making the pelican unit randomly perturb its current position. The perturbation amplitude gradually decreases as the iteration progresses, thus achieving a balance from global exploration to local exploitation.
[0040] Step S43: Global optimal position update, used to record and update the currently found optimal parameter solution after each iteration. Specifically, by comparing the fitness values of all pelican units, the position of the pelican unit with the highest fitness is selected as the new global optimal position, and the pelican unit with the highest fitness is selected as the new global optimal unit.
[0041] Step S44: Algorithm Iteration Update, used to execute the complete Pelican optimization algorithm process, and finally output the optimized key process parameter combination. Specifically, by repeatedly executing the search and update process until the preset iteration stopping condition is reached, the final global optimal position is obtained, and the optimized key process parameter combination is obtained.
[0042] Furthermore, in step S5, the optimization of the mineral water treatment process specifically involves collecting current data as input to the process optimization model, determining the state of the mineral water treatment process, and if it is determined to be a non-optimizable state, then the process optimization model is used to continue monitoring; if it is determined to be an optimization state, then the Pelican optimization algorithm is used to perform optimization search, output the final optimized combination of key process parameters, and comprehensively adjust the actual production process based on this.
[0043] The present invention provides an intelligent mineral water treatment process optimization system, including a raw data acquisition module, a preliminary data processing module, a process optimization model construction module, a process parameter optimization module, and a mineral water treatment process optimization module;
[0044] The raw data acquisition module is used to collect raw data, obtain a raw dataset of mineral water treatment by collecting raw data, and send the raw dataset of mineral water treatment to the preliminary data processing module.
[0045] The preliminary data processing module is used for preliminary data processing. Through preliminary data processing, an unlabeled normalized dataset, a processing state support set, and a processing state query set are obtained, and the unlabeled normalized dataset, the processing state support set, and the processing state query set are sent to the process optimization model construction module.
[0046] The process optimization model construction module is used to construct a process optimization model. By constructing an improved prototype network model, a process optimization model is obtained, and the process optimization model is sent to the mineral water treatment process optimization module.
[0047] The process parameter optimization module is used for process parameter optimization. It automatically searches for the optimal combination of key process parameters by designing the Pelican Optimization Algorithm, obtains the optimized combination of key process parameters, and sends the algorithm logic to the mineral water treatment process optimization module.
[0048] The mineral water treatment process optimization module is used to optimize the mineral water treatment process. By using the process optimization model combined with the Pelican optimization algorithm, the mineral water treatment process is optimized to obtain the final optimized combination of key process parameters.
[0049] The beneficial effects achieved by the present invention using the above solution are as follows:
[0050] (1) In view of the technical problems of traditional mineral water treatment process optimization methods, which rely on a large number of labeled process parameter samples, resulting in high labeling costs and susceptibility to class imbalance and insufficient samples, and unsatisfactory optimization results, this solution creatively adopts an improved prototype network model for mineral water treatment process optimization. It can effectively utilize labeled process parameter data and a large number of unlabeled process parameter data, reduce class imbalance sensitivity, improve optimization accuracy when samples are insufficient, and effectively integrate data features from different sources, thereby improving the model's generalization ability and optimization effect.
[0051] (2) In view of the technical problems of traditional mineral water treatment process optimization methods, such as the need for a lot of computing resources when optimizing process parameters, the tendency to get stuck in local optima in high-dimensional parameter space, and the difficulty in quickly adjusting parameters to cope with fluctuations in raw water quality, resulting in the inability to fully explore effective parameter combinations, this solution creatively adopts the Pelican optimization algorithm for process parameter optimization. It has strong adaptability and fast convergence speed, and can optimize parameters in real time in dynamic production environment, thereby improving the optimization effect of mineral water treatment process and effectively coping with the complexity and variability of water quality data. Attached Figure Description
[0052] Figure 1 A flowchart illustrating an intelligent mineral water treatment process optimization method provided by the present invention;
[0053] Figure 2 A schematic diagram of a module for an intelligent mineral water treatment process optimization system provided by the present invention;
[0054] Figure 3 This is a flowchart illustrating the preliminary data processing in step S2.
[0055] Figure 4 A flowchart illustrating the process of optimizing the model for step S3;
[0056] Figure 5 A flowchart illustrating the process parameter optimization for step S4.
[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0058] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0059] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0060] Example 1, see Figure 1 This invention provides an intelligent method for optimizing mineral water treatment processes, which includes the following steps:
[0061] Step S1: Acquisition of raw data;
[0062] Step S2: Preliminary data processing;
[0063] Step S3: Construction of process optimization model;
[0064] Step S4: Process parameter optimization;
[0065] Step S5: Optimization of mineral water treatment process.
[0066] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the raw data acquisition is used to obtain multi-dimensional raw information reflecting the state and results of the mineral water treatment process. Specifically, it involves obtaining a raw dataset of mineral water treatment through data collection. The raw dataset of mineral water treatment specifically includes raw water quality data, treatment equipment operation data, process control parameters, environmental data, and effluent water quality data. The raw water quality data specifically includes raw water turbidity, raw water pH value, raw water total hardness, raw water mineralization, and raw water microbial indicators. The treatment equipment operation data specifically includes filtration equipment pressure, membrane flux, backwashing frequency, ozone generator power, and ultraviolet intensity. The process control parameters specifically include flocculant dosage, disinfectant dosage, pH adjustment dosage, filtration cycle, and rinsing intensity. The environmental data specifically includes ambient temperature, ambient humidity, and production period. The effluent water quality data specifically includes effluent turbidity, effluent pH value, residual chlorine content, bromate content, and microbial compliance status.
[0067] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the preliminary data processing is used to perform preliminary processing on the collected raw data, transforming the raw data into a regular and model-readable format. Specifically, it includes the following steps:
[0068] Step S21: Data cleaning, used to remove noise and invalid information from the original data to ensure data quality. Specifically, missing and duplicate values in the original dataset are removed by removing mineral water to obtain a cleaner coarsely processed dataset.
[0069] Step S22: Data labeling, used to provide necessary supervision signals for model training, while retaining some unlabeled data for semi-supervised learning. Specifically, unlabeled datasets are obtained by randomly sampling a portion of the coarsely processed dataset, and the remaining data are labeled as high-quality or in a state to be optimized according to the processing effect, resulting in labeled datasets with labels.
[0070] Step S23: Data encoding, which is used to convert non-numerical or categorical data into numerical vectors so that they can be processed by mathematical models. Specifically, one-hot encoding is used to encode the unlabeled dataset and the labeled dataset respectively to obtain the corresponding unlabeled encoded dataset and labeled encoded dataset.
[0071] Step S24: Data normalization, used to eliminate differences in units and orders of magnitude between different feature parameters. Specifically, the min-max method is used to scale the encoded data to a specific range to obtain a scaled unlabeled normalized dataset and a labeled normalized dataset.
[0072] Step S25: Dataset partitioning, used to obtain training data and test data, to simulate the model learning with a small number of samples and evaluate its ability to recognize new samples. Specifically, the labeled normalized dataset is partitioned to obtain a processing state support set for model training reference and a processing state query set for model validation.
[0073] By integrating and executing the aforementioned data cleaning, data labeling, data encoding, data normalization, and dataset partitioning steps, the final output includes the unlabeled normalized dataset, processing state support set, and processing state query set required for model construction.
[0074] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the process optimization model is constructed to build a model required for intelligent identification of the state of mineral water treatment process. Specifically, an improved prototype network model is constructed as the process optimization model. The improved prototype network model uses a small amount of labeled data and a large amount of unlabeled data for effective learning, distinguishes between high-quality and unoptimized treatment states, and outputs the prediction result of the state of mineral water treatment process.
[0075] The process optimization model construction specifically includes the following steps:
[0076] Step S31: Preliminary feature extraction, used to automatically learn and extract deep feature representations that are meaningful for distinguishing processing states from the preliminarily processed data. Specifically, the convolutional neural network module performs feature transformation on the unlabeled normalized dataset, the processing state support set, and the processing state query set respectively to obtain the corresponding unlabeled feature set, support feature set, and query feature set.
[0077] Step S32: Calculate the state prototype, which is used to calculate a representative feature center point for each type of processing state, serving as a standard reference for that category. Specifically, the state prototype representing the typical characteristics of that type of processing state is obtained by calculating the average of the feature vectors of all samples in each category. The formula used is as follows:
[0078] ;
[0079] In the formula, This represents the state prototype for the m-th processing state category. This represents the total number of samples in the m-th processing state category. Represents the m-th processing state category. Feature vectors of each sample;
[0080] Step S33: Design a semi-supervised learning method to enhance the model's learning ability when labeled data is limited, utilizing a large amount of unlabeled data. The steps include:
[0081] Step S331: Calculate the transportation cost, which is used to quantify the cost of assigning unlabeled data to each existing processed state category, providing a basis for pseudo-label assignment. Specifically, this involves defining the differences between the features of the unlabeled data and each state prototype by calculating the cosine similarity between them, and then calculating the transportation cost using the following formula:
[0082] ;
[0083] In the formula, Represents the state prototype of the m-th processing state category to the n-th processing state category. The transportation cost of an unlabeled feature This indicates the calculation of cosine similarity. This represents the state prototype of the m-th processing state category after normalization. Represents the normalized i-th There are 1 unlabeled feature, where T represents the transpose operation;
[0084] Step S332: Assign pseudo-labels to automatically generate reliable category labels for unlabeled data. Specifically, by comprehensively considering the optimal assignment results based on optimal transportation theory and the confidence level of model predictions, a pseudo-label is dynamically assigned to each unlabeled feature. The formula used is as follows:
[0085] ;
[0086] In the formula, This represents a pseudo-label for unlabeled features. This indicates pseudo-labels based on semi-supervised learning. Represents the allocation matrix. Let M represent the set of all assignment matrices, and M represent the total number of sample processing state categories. This represents the total number of unlabeled features. For the allocation matrix The element represents the first element. The degree to which each unlabeled feature is assigned to the m-th processing state category. The pseudo-labels representing the model calculations, Indicates the first The probability that an unlabeled feature belongs to the m-th processing state category. Indicates the confidence threshold. This indicates taking the value of the parameter that minimizes the expression. This indicates that the value of the parameter that makes the expression the largest is taken.
[0087] Step S333: Correct the state prototype, which uses newly introduced unlabeled data with pseudo-labels to update the feature representation of the original state prototype, making it closer to the true distribution of the data. Specifically, a weighted average operation is used to fuse the feature information of the original state prototype with the unlabeled data to obtain a more representative updated state prototype. The formula used is as follows:
[0088] ;
[0089] In the formula, This represents the updated state prototype for the m-th processing state category after modification. This represents the weighting parameter used to control the proportion of unlabeled features involved. Indicates the first One unlabeled feature;
[0090] Step S334: Design a semi-supervised learning loss function to introduce the use of unlabeled data into the model training objective, driving the model to learn a feature space that allows unlabeled data to be allocated to each processing state category at the lowest cost. Specifically, the loss term is constructed by minimizing the total transport cost between unlabeled features and state prototypes, using the following formula:
[0091] ;
[0092] In the formula, This represents the loss value for semi-supervised learning;
[0093] Step S34: Obtain the model output, which drives the model to predict the processing state category. Specifically, this involves calculating the similarity between the model input features and the updated state prototypes of each processing state category after correction, and then processing the results using the softmax activation function to obtain the probability distribution of the processing state category prediction. This distribution serves as the state prediction result for the mineral water processing process. The formula used is as follows:
[0094] ;
[0095] In the formula, The model represents the first The output of the model input features is the probability distribution of the processed state category prediction. Indicates the output weights. Indicates the output bias term. This represents the updated state prototype for the m-th processing state category after normalization. Represents the normalized i-th Each model input feature;
[0096] Step S35: Design the model loss function to guide the overall optimization direction of model training. Specifically, this involves designing model compactness loss and model uniformity loss, and combining them with semi-supervised learning loss to construct a comprehensive total model loss function. The formula used is as follows:
[0097] ;
[0098] In the formula, This represents the model tightness loss value, where A represents the total number of input features to the model. The model predicts the first The probability that a model input feature belongs to the m-th processing state category. This represents the model uniformity loss value. Represents the center of the state prototype. This indicates the calculation of the L2 norm. Indicates the corrected number The prototype for updating the state for each processing state category. This represents the total loss value of the model. This represents the weighting parameter used to control the proportion of semi-supervised learning loss involved;
[0099] Step S36: Model construction and training, specifically, through the preliminary feature extraction, the computation of the state prototype, the design of semi-supervised learning, the acquisition of model output and the design of model loss function, the improved prototype network model is constructed by integrating these steps. The model is trained and its performance is verified based on the support feature set and the query feature set to obtain the improved prototype network model as the process optimization model.
[0100] By performing the above operations, this solution addresses the technical problems of traditional mineral water treatment process optimization methods, which rely on a large number of labeled process parameter samples, resulting in high labeling costs and susceptibility to class imbalance and insufficient samples, leading to unsatisfactory optimization results. It creatively employs an improved prototype network model for mineral water treatment process optimization. This model effectively utilizes labeled process parameter data and a large amount of unlabeled process parameter data, reduces sensitivity to class imbalance, improves optimization accuracy when samples are insufficient, and effectively integrates data features from different sources, enhancing the model's generalization ability and optimization effect.
[0101] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the process parameter optimization is used to automatically search for the optimal combination of key process parameters after the model identifies the state to be optimized. Specifically, the Pelican optimization algorithm is used to perform optimization search to obtain the optimized combination of key process parameters. The combination of key process parameters specifically includes processing equipment operation data and process control parameters.
[0102] The process parameter optimization specifically includes the following steps:
[0103] Step S41: Algorithm initialization, used to set the starting search point, stopping conditions, and evaluation criteria for the Pelican optimization algorithm. The steps include:
[0104] Step S411: Initialize the search space to define the range and representation of the key process parameters to be optimized, providing the starting point for the algorithm's search. Specifically, this is achieved by constructing a set of pelican units, where the position vector of each pelican unit represents a combination of key process parameters, thereby initializing the pelican unit position matrix.
[0105] Step S412: Determine the iteration stopping condition, which is used to control the runtime and convergence criterion of the optimization algorithm. Specifically, the maximum number of iterations and the desired fitness threshold are preset as the conditions for the algorithm to terminate the search.
[0106] Step S413: Determine the fitness, which is used to quantitatively evaluate the merits of each set of key process parameter combinations. Specifically, by defining a fitness calculation function, the key process parameter combinations represented by the pelican unit are input and the fitness value is calculated. The fitness calculation function is defined by an independent multilayer perceptron.
[0107] Step S42: Unit position update, used to drive the pelican unit to move to a better area in the search space, includes the following steps:
[0108] Step S421: Design the approach-to-prey phase, which guides the pelican units to perform a fine-grained local search towards the currently known optimal solution. Specifically, the position adjustment is determined by comparing the fitness value of each pelican unit with that of the current global optimal unit, using the following formula:
[0109] ;
[0110] In the formula, This represents the position of the i-th pelican unit in the j-th dimension during the prey-approaching stage. and This represents a random value within the range [0,1] that takes different values. This indicates the prey's position, which is the position of the current globally optimal unit. This represents a parameter used to control the speed at which a pelican approaches its prey. This represents the fitness value of the prey. This represents the fitness value of the i-th pelican unit;
[0111] Step S422: Design the surface foraging stage to enhance the algorithm's global exploration capability in the later stages of the search and avoid getting trapped in local optima. Specifically, this is achieved by randomly perturbing the pelican units near their current positions, with the perturbation amplitude gradually decreasing as iterations proceed. This balances the shift from global exploration to local exploitation, and the formula used is as follows:
[0112] ;
[0113] In the formula, Let R represent the position of the i-th pelican unit in the j-th dimension during the surface feeding phase, d represent the current iteration number, and D represent the total number of iterations. This represents a random value within the range [0,1].
[0114] Step S43: Global optimal position update, used to record and update the currently found optimal parameter solution after each iteration. Specifically, by comparing the fitness values of all pelican units, the position of the pelican unit with the highest fitness is selected as the new global optimal position, and the pelican unit with the highest fitness is selected as the new global optimal unit.
[0115] Step S44: Algorithm Iteration Update, used to execute the complete Pelican optimization algorithm process, and finally output the optimized key process parameter combination. Specifically, by repeatedly executing the search and update process until the preset iteration stopping condition is reached, the final global optimal position is obtained, and the optimized key process parameter combination is obtained.
[0116] By performing the above operations, this solution addresses the technical problems of traditional mineral water treatment process optimization methods, which require a large amount of computational resources for process parameter optimization, are prone to getting trapped in local optima in high-dimensional parameter spaces, and are difficult to quickly adjust parameters to cope with fluctuations in raw water quality, thus failing to fully explore effective parameter combinations. This solution creatively adopts the Pelican Optimization Algorithm for process parameter optimization, which has strong adaptability and fast convergence speed. It can optimize parameters in real time in dynamic production environments, thereby improving the optimization effect of the mineral water treatment process and effectively dealing with the complexity and variability of water quality data.
[0117] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The optimization of the mineral water treatment process specifically involves collecting current data as input to the process optimization model, judging the state of the mineral water treatment process, and if it is judged to be a non-optimizable state, then the process optimization model is used to continue monitoring. If it is judged to be an optimization state, then the Pelican optimization algorithm is used to search for optimization, output the final optimized combination of key process parameters, and comprehensively adjust the actual production process based on this.
[0118] Example 7, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides an intelligent mineral water treatment process optimization system, including a raw data acquisition module, a preliminary data processing module, a process optimization model construction module, a process parameter optimization module, and a mineral water treatment process optimization module.
[0119] The raw data acquisition module is used to collect raw data, obtain a raw dataset of mineral water treatment by collecting raw data, and send the raw dataset of mineral water treatment to the preliminary data processing module.
[0120] The preliminary data processing module is used for preliminary data processing. Through preliminary data processing, an unlabeled normalized dataset, a processing state support set, and a processing state query set are obtained, and the unlabeled normalized dataset, the processing state support set, and the processing state query set are sent to the process optimization model construction module.
[0121] The process optimization model construction module is used to construct a process optimization model. By constructing an improved prototype network model, a process optimization model is obtained, and the process optimization model is sent to the mineral water treatment process optimization module.
[0122] The process parameter optimization module is used for process parameter optimization. It automatically searches for the optimal combination of key process parameters by designing the Pelican Optimization Algorithm, obtains the optimized combination of key process parameters, and sends the algorithm logic to the mineral water treatment process optimization module.
[0123] The mineral water treatment process optimization module is used to optimize the mineral water treatment process. By using the process optimization model combined with the Pelican optimization algorithm, the mineral water treatment process is optimized to obtain the final optimized combination of key process parameters.
[0124] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.
[0126] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. An intelligent method for optimizing mineral water treatment processes, characterized in that: The method includes the following steps: S1: Raw data acquisition. By collecting data, the raw dataset of mineral water treatment is obtained. The raw dataset of mineral water treatment specifically includes raw water quality data, treatment equipment operation data, process control parameters, environmental data, and effluent water quality data. S2: Preliminary data processing, which involves performing preliminary processing on the collected raw data to obtain an unlabeled normalized dataset, a processing status support set, and a processing status query set; S3: Process optimization model construction, used to build a model required for intelligent identification of the state of mineral water treatment process, specifically to build an improved prototype network model as the process optimization model. The improved prototype network model uses a small amount of labeled data and a large amount of unlabeled data for effective learning, distinguishes between high-quality and unoptimized treatment states, and outputs the prediction results of the state of mineral water treatment process. S4: Process parameter optimization, used to automatically search for the optimal combination of key process parameters after the model identifies the state to be optimized. Specifically, the Pelican optimization algorithm is used to search for the optimal combination of key process parameters. The combination of key process parameters specifically includes processing equipment operating data and process control parameters. S5: Mineral water treatment process optimization, specifically, collecting current data as input to the process optimization model, judging the state of the mineral water treatment process, if it is judged to be a non-optimizable state, then the process optimization model is used to continue monitoring, if it is judged to be an optimization state, then the Pelican optimization algorithm is used to search for optimization, outputting the final optimized combination of key process parameters, and based on this, comprehensively adjusting the actual production process.
2. The intelligent mineral water treatment process optimization method according to claim 1, characterized in that: The process optimization model construction specifically includes the following steps: Step S31: Preliminary feature extraction, used to automatically learn and extract deep feature representations that are meaningful for distinguishing processing states from the preliminarily processed data. Specifically, the convolutional neural network module performs feature transformation on the unlabeled normalized dataset, the processing state support set, and the processing state query set respectively to obtain the corresponding unlabeled feature set, support feature set, and query feature set. Step S32: Calculate the state prototype, which is used to calculate a representative feature center point for each type of processing state as a standard reference for that type. Specifically, by calculating the average value of the feature vectors of all samples in each type, the state prototype representing the typical features of that type of processing state is obtained. Step S33: Design a semi-supervised learning method to enhance the model's learning ability when labeled data is limited, utilizing a large amount of unlabeled data. The steps include: Step S331: Calculate transportation costs to quantify the cost of allocating unlabeled data to each existing processed state category, providing a basis for pseudo-label allocation. Specifically, this involves defining the differences between the features of the unlabeled data and each state prototype by calculating the cosine similarity between them, and then calculating the transportation costs. Step S332: Assign pseudo-labels to automatically generate reliable category labels for unlabeled data. Specifically, by comprehensively considering the optimal assignment results based on optimal transportation theory and the confidence level of model prediction, a pseudo-label is dynamically assigned to each unlabeled feature. Step S333: Correct the state prototype, which is used to update the feature representation of the original state prototype with newly introduced unlabeled data with pseudo-labels, so that it is closer to the real distribution of the data. Specifically, through a weighted average operation, the feature information of the original state prototype and the unlabeled data are fused to obtain a more representative updated state prototype after correction. Step S334: Design a semi-supervised learning loss function to introduce the use of unlabeled data in the model training objective, driving the model to learn a feature space that can allocate unlabeled data to each processing state category at the lowest cost. Specifically, the loss term is constructed by minimizing the total transport cost between unlabeled features and state prototypes. Step S34: Obtain the model output to drive the model to perform processing state category classification prediction. Specifically, calculate the similarity between the model input features and the updated state prototypes of each processing state category after correction, and process them through the softmax activation function to obtain the processing state category prediction probability distribution, which serves as the state prediction result of the mineral water processing process. Step S35: Design a model loss function to guide the overall optimization direction of model training. Specifically, design a model compactness loss and a model uniformity loss, and combine them with semi-supervised learning loss to form a comprehensive total model loss function. Step S36: Model construction and training, specifically, through the preliminary feature extraction, the computation of the state prototype, the design of semi-supervised learning, the acquisition of model output and the design of model loss function, the improved prototype network model is constructed by integrating these steps. The model is trained and its performance is verified based on the support feature set and the query feature set to obtain the improved prototype network model as the process optimization model.
3. The intelligent mineral water treatment process optimization method according to claim 1, characterized in that: The process parameter optimization specifically includes the following steps: Step S41: Algorithm initialization, used to set the starting search point, stopping conditions, and evaluation criteria for the Pelican optimization algorithm. The steps include: Step S411: Initialize the search space to define the range and representation of the key process parameters to be optimized, providing the starting point for the algorithm's search. Specifically, this is achieved by constructing a set of pelican units, where the position vector of each pelican unit represents a combination of key process parameters, thereby initializing the pelican unit position matrix. Step S412: Determine the iteration stopping condition, which is used to control the runtime and convergence criterion of the optimization algorithm. Specifically, the maximum number of iterations and the desired fitness threshold are preset as the conditions for the algorithm to terminate the search. Step S413: Determine the fitness, which is used to quantitatively evaluate the merits of each set of key process parameter combinations. Specifically, by defining a fitness calculation function, the key process parameter combinations represented by the pelican unit are input and the fitness value is calculated. The fitness calculation function is defined by an independent multilayer perceptron. Step S42: Unit position update, used to drive the pelican unit to move to a better area in the search space, includes the following steps: Step S421: Design the approach-prey stage to guide the pelican unit to perform a local fine search in the direction of the currently known optimal solution. Specifically, the position adjustment is determined by comparing the fitness value of each pelican unit with that of the current global optimal unit. Step S422: Design the surface predation stage to enhance the algorithm's global exploration capability in the later stages of the search and avoid getting stuck in local optima. Specifically, this is achieved by making the pelican unit randomly perturb its current position. The perturbation amplitude gradually decreases as the iteration progresses, thus achieving a balance from global exploration to local exploitation. Step S43: Global optimal position update, used to record and update the currently found optimal parameter solution after each iteration. Specifically, by comparing the fitness values of all pelican units, the position of the pelican unit with the highest fitness is selected as the new global optimal position, and the pelican unit with the highest fitness is selected as the new global optimal unit. Step S44: Algorithm Iteration Update, used to execute the complete Pelican optimization algorithm process, and finally output the optimized key process parameter combination. Specifically, by repeatedly executing the search and update process until the preset iteration stopping condition is reached, the final global optimal position is obtained, and the optimized key process parameter combination is obtained.
4. The intelligent mineral water treatment process optimization method according to claim 1, characterized in that: The raw water quality data specifically includes raw water turbidity, raw water pH value, raw water total hardness, raw water mineralization, and raw water microbial indicators. The treatment equipment operation data specifically includes filtration equipment pressure, membrane flux, backwashing frequency, ozone generator power, and ultraviolet intensity. The process control parameters specifically include flocculant dosage, disinfectant dosage, pH adjustment dosage, filtration cycle, and rinsing intensity. The environmental data specifically includes ambient temperature, ambient humidity, and production period. The effluent water quality data specifically includes effluent turbidity, effluent pH value, residual chlorine content, bromate content, and microbial compliance status.
5. The intelligent mineral water treatment process optimization method according to claim 1, characterized in that: The preliminary data processing specifically includes the following steps: Step S21: Data cleaning, used to remove noise and invalid information from the original data to ensure data quality. Specifically, missing and duplicate values in the original dataset are removed by removing mineral water to obtain a cleaner coarsely processed dataset. Step S22: Data labeling, used to provide necessary supervision signals for model training, while retaining some unlabeled data for semi-supervised learning. Specifically, unlabeled datasets are obtained by randomly sampling a portion of the coarsely processed dataset, and the remaining data are labeled as high-quality or in a state to be optimized according to the processing effect, resulting in labeled datasets with labels. Step S23: Data encoding, which is used to convert non-numerical or categorical data into numerical vectors so that they can be processed by mathematical models. Specifically, one-hot encoding is used to encode the unlabeled dataset and the labeled dataset respectively to obtain the corresponding unlabeled encoded dataset and labeled encoded dataset. Step S24: Data normalization, used to eliminate differences in units and orders of magnitude between different feature parameters. Specifically, the min-max method is used to scale the encoded data to a specific range to obtain a scaled unlabeled normalized dataset and a labeled normalized dataset. Step S25: Dataset partitioning, used to obtain training data and test data, to simulate the model learning with a small number of samples and evaluate its ability to recognize new samples. Specifically, the labeled normalized dataset is partitioned to obtain a processing state support set for model training reference and a processing state query set for model validation. By integrating and executing the aforementioned data cleaning, data labeling, data encoding, data normalization, and dataset partitioning steps, the final output includes the unlabeled normalized dataset, processing state support set, and processing state query set required for model construction.
6. An intelligent mineral water treatment process optimization system, used to implement the intelligent mineral water treatment process optimization method as described in any one of claims 1-5, characterized in that: It includes a raw data acquisition module, a preliminary data processing module, a process optimization model construction module, a process parameter optimization module, and a mineral water treatment process optimization module.
7. The intelligent mineral water treatment process optimization system according to claim 6, characterized in that: The raw data acquisition module is used to collect raw data, obtain a raw dataset of mineral water treatment by collecting raw data, and send the raw dataset of mineral water treatment to the preliminary data processing module. The preliminary data processing module is used for preliminary data processing. Through preliminary data processing, an unlabeled normalized dataset, a processing state support set, and a processing state query set are obtained, and the unlabeled normalized dataset, the processing state support set, and the processing state query set are sent to the process optimization model construction module. The process optimization model construction module is used to construct a process optimization model. By constructing an improved prototype network model, a process optimization model is obtained, and the process optimization model is sent to the mineral water treatment process optimization module. The process parameter optimization module is used for process parameter optimization. It automatically searches for the optimal combination of key process parameters by designing the Pelican Optimization Algorithm, obtains the optimized combination of key process parameters, and sends the algorithm logic to the mineral water treatment process optimization module. The mineral water treatment process optimization module is used to optimize the mineral water treatment process. By using the process optimization model combined with the Pelican optimization algorithm, the mineral water treatment process is optimized to obtain the final optimized combination of key process parameters.