A smart method and system for predicting the effect of mineral water treatment

Abstract

This invention discloses an intelligent method and system for predicting the treatment effect of mineral water. The method includes constructing a basic dataset, building a response vector, generating a stable sample set, dual-channel prediction, and outputting the treatment effect level. This invention relates to the field of intelligent water treatment technology, specifically to an intelligent method and system for predicting the treatment effect of mineral water. This solution performs state compression, trajectory deflection analysis, and stability score calculation on the response vector to select stable sample segments and extract their central features as a stability reference. This enhances the ability to capture stable treatment patterns, reduces noise fluctuation interference, and improves the representativeness and reliability of the predicted input. Furthermore, a dual-channel prediction mechanism is constructed, modeling stable reference information and offset correction information separately, and adaptively fusing them with the sample stability and trajectory deflection levels. This improves the accuracy, robustness, and adaptability of the treatment effect prediction.

Description

Technical Field

[0001] This invention relates to the field of intelligent water treatment technology, specifically to an intelligent method and system for predicting the treatment effect of mineral water. Background Technology

[0002] As drinking water safety requirements continue to rise, mineral water treatment is gradually shifting from relying on manual experience and single-indicator judgments to data-driven, intelligent monitoring and prediction. Current treatment systems typically integrate raw water quality, process operating parameters, and effluent test results to coordinate the control of processes such as filtration, disinfection, aeration, and mineral retention. They also utilize sensor data acquisition, online monitoring, and intelligent analysis to improve treatment efficiency and stability. This allows for earlier identification of water quality trends, more accurate assessment of treatment effects, and provides forward-looking data for process adjustments, thereby achieving a higher level of automation and refined management.

[0003] However, existing technologies for predicting the effects of mineral water treatment still have significant shortcomings. On the one hand, the raw water quality, process parameters, and effluent results often come from complex sources and have inconsistent sampling frequencies. There is noise, outliers, and temporal misalignment among the data, resulting in unstable model input. On the other hand, traditional methods often focus on single static features or simple temporal correlations, making it difficult to simultaneously characterize regular changes under stable operating conditions and offset corrections under fluctuating operating conditions. Therefore, they are not adaptable to local disturbances, slow drifts, and periodic anomalies, and the prediction results are easily affected by noise. There is still room for improvement in accuracy, robustness, and generalization ability. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent method and system for predicting the treatment effect of mineral water. Addressing the problems of large sample state fluctuations, difficulty in accurately identifying stable operating conditions, and insufficient effective reference segments leading to unreliable predictions in existing mineral water treatment effect prediction methods, this solution performs state compression, trajectory deflection analysis, and stability score calculation on the response vector to select stable sample segments and extract their central features as stable references. Furthermore, addressing the difficulty in simultaneously considering the regularity fitting under stable operating conditions and the offset correction under abnormal fluctuations in existing mineral water treatment effect prediction methods, resulting in prediction results being easily affected by local disturbances and insufficient generalization ability, this solution constructs a dual-channel prediction mechanism. It models stable sample reference information and offset correction information separately, and then adaptively fuses them together with the sample stability degree and trajectory deflection degree. This allows the model to accurately depict stable treatment patterns while promptly correcting errors caused by abnormal offsets.

[0005] The technical solution adopted in this invention is as follows: An intelligent method for predicting the treatment effect of mineral water, the method comprising the following steps:

[0006] Step S1: Construct a basic dataset by collecting information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleaning and anomaly repair, and finally forming a standardized basic dataset that can be directly used for subsequent modeling.

[0007] Step S2: Construct a response vector by integrating raw water quality information, process operation information, and historical treatment effect information into a unified response vector, while supplementing the trend of state changes and the correlation information between variables;

[0008] Step S3: Generate a stable sample set. Extract stable state features from the response vector during the processing. First, convert the samples into state representations that can reflect the processing trend, and then select stable sample segments.

[0009] Step S4: Dual-channel prediction. A dual-channel prediction mechanism is established. One channel focuses on using stable sample information for stable reference prediction, while the other channel focuses on analyzing the correction effects caused by sample offset and fluctuation. Finally, the two results are fused to obtain the prediction result of the mineral water treatment effect.

[0010] Step S5: Output the treatment effect level. Summarize the final prediction results into a comprehensive score, and output the corresponding treatment effect level according to the range in which the score falls, forming the final result of the mineral water treatment effect.

[0011] Furthermore, in step S1, constructing the basic dataset specifically includes the following steps:

[0012] Step S11: Data collection. Collect basic data on the mineral water treatment process and unify data from different sources and frequencies to the same time scale.

[0013] Step S12: Outlier identification. Identify outlier samples in the collected data and perform robust repair on them.

[0014] Further, in step S2, the construction of the response vector specifically includes the following steps:

[0015] Step S21: Form a unified response vector by integrating the raw water quality status, process status, and historical treatment effect detection vectors into a unified response vector;

[0016] Step S22: Construct an offset response vector. Extract a stable benchmark from historical samples and construct an offset response vector based on it. First, assign a stable weight to each sample according to the stability of the sample in the fluctuation of the operating conditions, and then use these weights to calculate the stable benchmark vector. Subsequently, extract the difference between each sample and the benchmark vector, and introduce the product relationship between the sample and the benchmark to form the offset response vector.

[0017] Further, in step S3, generating a stable sample set specifically includes the following steps:

[0018] Step S31: Generate a triaxial state feature vector, converting the offset response vector of each sample into a triaxial feature vector that can characterize the processing state; compress the offset response vector into three main components through three sets of directional projections, which are used to characterize the impurity removal trend, mineral retention trend and operational stability trend, respectively.

[0019] Step S32: Calculate the trajectory deflection degree to determine whether the changes of the sample in the feature state space are stable and to identify whether there are obvious fluctuations or deflections; first calculate the state change amplitude between adjacent samples, and then calculate the trajectory deflection degree through the three-point difference method; then, combine the two to form a local stability index.

[0020] Step S33: Construct a stability score by combining the mineral retention during the processing with the degree of local stability to form a score index for screening stable sample segments; extract the components related to mineral retention and impurity control from the response vector and compare them with the target level to calculate the consistency score; then multiply the consistency score by the degree of local stability to obtain the stability score.

[0021] Step S34: Generate stable sample segments by setting continuous samples with high stability scores as stable sample segments.

[0022] Further, in step S4, the dual-channel prediction specifically includes the following steps:

[0023] Step S41: Construct dual-channel neural network input features by combining the offset response vector obtained in step S2, the three-axis state feature vector obtained in step S3, and the stable sample segment information into dual-channel neural network input features;

[0024] Step S42: Stable prediction. Input the stable channel input features obtained in step S41 into the stable reference neural network, extract the stable processing mode features through multi-layer nonlinear mapping, and obtain the stable reference prediction result.

[0025] Step S43: Correct the prediction. Input the offset channel input features obtained in step S41 into the offset correction neural network. First, extract the nonlinear expression of the offset features, then generate the correction term, and superimpose it with the stable reference prediction result to obtain the offset correction prediction result.

[0026] Step S44: Fuse the dual-channel prediction results. Based on the stability of the sample and the degree of trajectory deflection, adaptively fuse the stable reference prediction results and the offset correction prediction results to form the final processing effect prediction vector.

[0027] Step S45: Joint optimization. During the training phase, the stable reference prediction result, the offset correction prediction result, and the final fusion result are jointly optimized.

[0028] Furthermore, in step S5, the output processing effect level specifically includes the following steps:

[0029] Step S51: Calculate the comprehensive processing effect score. The components of the final processing effect prediction vector are weighted and summed according to preset weights to obtain the comprehensive processing effect score.

[0030] Step S52: Output the processing effect level, mapping the overall processing effect score to a specific processing effect level, thereby giving the final processing evaluation result.

[0031] The present invention provides an intelligent mineral water treatment effect prediction system, including a basic dataset construction module, a response vector construction module, a stable sample set generation module, a dual-channel prediction module, and an output treatment effect level module;

[0032] The module for building the basic dataset collects information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleans and repairs anomalies, and finally forms a standardized basic dataset that can be directly used for subsequent modeling. The data is then sent to the module for building the response vector.

[0033] The response vector construction module receives data sent by the basic dataset construction module, integrates raw water quality information, process operation information, and historical treatment effect information into a unified response vector, and supplements the state change trend and the correlation information between variables, and sends the data to the stable sample set generation module.

[0034] The module for generating a stable sample set receives data sent by the module for constructing a response vector. Based on the stable state features extracted during the processing of the response vector, the module first converts the samples into a state representation that can reflect the processing trend, then selects stable sample segments, and sends the data to the dual-channel prediction module.

[0035] The dual-channel prediction module receives data sent by the stable sample set generation module, establishes a dual-channel prediction mechanism, one channel focuses on using stable sample information for stable reference prediction, and the other channel focuses on analyzing the correction effect caused by sample offset and fluctuation. Finally, the two results are fused to obtain the mineral water treatment effect prediction result, and the data is sent to the output treatment effect level module.

[0036] The output processing effect level module receives data sent by the dual-channel prediction module, summarizes the final prediction results into a comprehensive score, and outputs the corresponding processing effect level according to the range in which the score falls, thus forming the final result of the mineral water processing effect.

[0037] The beneficial effects achieved by the present invention using the above solution are as follows:

[0038] (1) In view of the problems in the existing mineral water treatment effect prediction methods, such as large fluctuations in sample state, difficulty in accurately identifying stable working conditions, and insufficient effective reference segments, which lead to unreliable prediction basis, this scheme performs state compression, trajectory deflection analysis and stability score calculation on the response vector, selects stable sample segments, and extracts their central features as stable references, thereby enhancing the model's ability to capture the stable treatment pattern, reducing the interference of noise fluctuations on subsequent predictions, and improving the representativeness and credibility of the prediction input.

[0039] (2) In view of the problem that existing mineral water treatment effect prediction methods are difficult to simultaneously take into account the regular fitting under stable working conditions and the offset correction under abnormal fluctuations, resulting in the prediction results being easily affected by local disturbances and having insufficient generalization ability, this scheme constructs a dual-channel prediction mechanism, models stable sample reference information and offset correction information separately, and then combines the sample stability degree and trajectory deflection degree for adaptive fusion, so that the model can not only maintain the accurate characterization of stable treatment mode, but also correct the error caused by abnormal offset in time, thereby improving the accuracy, robustness and adaptability of treatment effect prediction. Attached Figure Description

[0040] Figure 1 A schematic diagram illustrating an intelligent method for predicting the treatment effect of mineral water provided by the present invention;

[0041] Figure 2 A schematic diagram of an intelligent mineral water treatment effect prediction system provided by the present invention;

[0042] Figure 3 This is a schematic diagram of step S1;

[0043] Figure 4 This is a schematic diagram of step S2;

[0044] Figure 5 This is a schematic diagram of step S3;

[0045] Figure 6 This is a schematic diagram of step S4;

[0046] Figure 7 This is a schematic diagram of step S5.

[0047] 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

[0048] 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.

[0049] 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.

[0050] Example 1, see Figure 1 This invention provides an intelligent method for predicting the treatment effect of mineral water, which includes the following steps:

[0051] Step S1: Construct a basic dataset by collecting information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleaning and anomaly repair, and finally forming a standardized basic dataset that can be directly used for subsequent modeling.

[0052] Step S2: Construct a response vector by integrating raw water quality information, process operation information, and historical treatment effect information into a unified response vector, while supplementing the trend of state changes and the correlation information between variables;

[0053] Step S3: Generate a stable sample set. Extract stable state features from the response vector during the processing. First, convert the samples into state representations that can reflect the processing trend, and then select stable sample segments.

[0054] Step S4: Dual-channel prediction. A dual-channel prediction mechanism is established. One channel focuses on using stable sample information for stable reference prediction, while the other channel focuses on analyzing the correction effects caused by sample offset and fluctuation. Finally, the two results are fused to obtain the prediction result of the mineral water treatment effect.

[0055] Step S5: Output the treatment effect level. Summarize the final prediction results into a comprehensive score, and output the corresponding treatment effect level according to the range in which the score falls, forming the final result of the mineral water treatment effect.

[0056] Example 2, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S1, the construction of the basic dataset specifically includes the following steps:

[0057] Step S11: Data Acquisition. Basic data is collected around the mineral water treatment process, unifying data from different sources and frequencies to the same time scale. Specifically, raw water quality data, treatment process operation data, and historical treatment effect monitoring data are collected. Raw water quality data includes turbidity, pH, conductivity, hardness, mineral ion concentration, microbial indicators, and organic matter indicators. Treatment process operation data includes filtration pressure, flow rate, temperature, aeration intensity, disinfectant dosage, and equipment operating time. Historical treatment effect monitoring data includes effluent turbidity, mineral retention level, and microbial control effectiveness. Resampling and alignment are performed using a unified time point as a benchmark, ensuring that each unified time point corresponds to a complete set of treatment status records, thus forming the standard time-series sample required for subsequent modeling, as shown below:

[0058] ;

[0059] in, Indicates the first One original record; Indicates the total number of original records; Indicates alignment to the first Sample records after a unified time point; Indicates the first A unified time point; Indicates the first The original sampling time of each record; Indicates the first The original record for the first Time weights for a unified time node; Indicates the time smoothing bandwidth. Represents an exponential function with the natural constant as its base;

[0060] Step S12: Outlier identification. Outlier samples in the collected data are identified and robustly repaired. Specifically, the deviation between the sample and the statistical center of the neighborhood is calculated using the temporal neighborhood as a reference. When the deviation is significantly excessive, robust statistical results within the neighborhood are used for repair, and the repair strength is adjusted according to the degree of deviation, as shown below:

[0061] ;

[0062] in, Indicates the first The original sample; the temporal neighborhood is defined as: ; Indicates the first The temporal neighborhood corresponding to each sample and They represent the first Article and Section The sampling time of each sample; Indicates the length of the neighborhood time window; and This indicates the batch number or processing stage number to which the sample belongs; A vector representing the median of the neighborhood samples; This represents the median absolute deviation of the neighborhood samples; This indicates taking the L1 norm; Indicates the degree of sample deviation; Indicates the sample confidence weight; This indicates the repaired sample; Indicates sample The mean vector of all vectors in the neighborhood of ; This represents a minimal constant to prevent the denominator from being zero.

[0063] Example 3, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S2, the construction of the response vector specifically includes the following steps:

[0064] Step S21: Form a unified response vector by integrating the raw water quality status, process status, and historical treatment effect detection vectors into a unified response vector. Specifically, the restored raw water quality indicators, restored process operating parameters, and historical treatment effect detection vectors are concatenated into a single vector, composed of effluent turbidity, mineral retention level, and microbial control effect. Furthermore, a change characteristic is introduced to characterize the changing trend of the state during the mineral water treatment process, as shown below:

[0065] ;

[0066] in, Indicates the first Feature vector of the original water quality after remediation for each sample; Indicates the first The post-repair process operation vector for each sample; This indicates the amount of change in the raw water quality status; Indicates the amount of change in process status; This represents the interaction between raw water quality and the process; This represents the historical processing effect detection vector; Operators that allow a matrix to be expanded into a vector by its columns; Indicates transpose;

[0067] Step S22: Construct the offset response vector. Extract a stable benchmark from historical samples and construct the offset response vector based on it. In practice, first assign a stable weight to each sample according to its stability during operating condition fluctuations, and then use these weights to calculate the stable benchmark vector. Subsequently, extract the difference between each sample and the benchmark vector, and introduce the product relationship between the sample and the benchmark to form the offset response vector. The vector can simultaneously reflect absolute and relative changes, describing slow drift, local disturbances, and continuous deviations in mineral water treatment, as shown below:

[0068] ;

[0069] in, Represents the stable reference vector; Indicates the first Stable weights for each sample; Indicates the fluctuation scale parameter; Represents the offset response vector; This represents element-wise multiplication; This represents the L2 norm.

[0070] Example 4, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S3, generating a stable sample set specifically includes the following steps:

[0071] Step S31: Generate a triaxial state feature vector, converting the offset response vector of each sample into a triaxial feature vector that can characterize the processing state; in specific implementation, the offset response vector is compressed into three main components through three sets of directional projections, which are used to characterize the impurity removal trend, mineral retention trend, and operational stability trend, respectively, as shown below:

[0072] ;

[0073] in, Indicates the first The sample at the th Components on each state axis; Indicates the first The three-axis state feature vector of each sample; Indicates the first One projection direction vector; Indicates the first One projection bias term; Represents the hyperbolic tangent function; Choose 1, 2, or 3;

[0074] Step S32: Calculate the trajectory deflection degree to determine whether the changes of the sample in the feature state space are stable and to identify whether there are obvious fluctuations or deflections. In practice, first calculate the amplitude of state changes between adjacent samples, and then calculate the trajectory deflection degree using a three-point difference method. Subsequently, combine the two to form a local stability index, as shown below:

[0075] ;

[0076] in, Indicates the first The magnitude of state change for each sample; Indicates the first The trajectory deflection of each sample; Indicates the first The degree of local stability of each sample; and Indicates the stability adjustment coefficient; , , These represent the state feature vectors of three adjacent samples respectively;

[0077] Step S33: Construct a stability score by combining mineral retention during the processing with the degree of local stability to form a scoring index for screening stable sample segments. Specifically, components related to mineral retention and impurity control are extracted from the response vector and compared with the target level to calculate a consistency score. This consistency score is then multiplied by the degree of local stability to obtain the stability score, as shown below:

[0078] ;

[0079] in, Indicates the first Mineral retention index of each sample; Indicates the first Impurity control indicators for each sample; Representing vectors The One component; This represents a set of feature indexes related to mineral retention. This represents a set of feature indices related to impurity control; Indicates the retention criteria for the target mineral; Indicates the target impurity control baseline; and These represent the permissible fluctuation scales for mineral retention and impurity control, respectively. Indicates the consistency score; Indicates a stable score;

[0080] Step S34: Generate stable sample segments by setting consecutive samples with high stability scores as stable sample segments. In specific implementation, when the stability score of a sample segment is higher than a preset threshold, it is merged into a single stable sample segment, as shown below:

[0081] ;

[0082] in, Indicates the first The sample set in a stable sample segment; Belongs to the A segment; Indicates the stable scoring threshold; Indicates the first The center vector of a stable sample segment; Indicates the sample segment number.

[0083] By performing the above operations, this solution addresses the problems in existing mineral water treatment effect prediction methods, such as large fluctuations in sample states, difficulty in accurately identifying stable operating conditions, and insufficient effective reference segments, which lead to unreliable prediction basis. This solution performs state compression, trajectory deflection analysis, and stability score calculation on the response vector to select stable sample segments and extract their central features as stable references. This enhances the model's ability to capture stable processing patterns, reduces the interference of noise fluctuations on subsequent predictions, and improves the representativeness and credibility of the prediction input.

[0084] Example 5, see Figure 1 and Figure 6 This embodiment is based on the above embodiment. In step S4, the dual-channel prediction specifically includes the following steps:

[0085] Step S41: Construct dual-channel neural network input features by combining the offset response vector obtained in step S2, the three-axis state feature vector obtained in step S3, and the stable sample segment information into dual-channel neural network input features; in specific implementation, for the first... For each sample, first offset it from the response vector. Three-axis state feature vector Local stability and trajectory deflection Combined into an offset channel input vector, and simultaneously compared with the center vector of each stable sample segment. The relationships between these elements are used to construct stable channel input information, as shown below:

[0086] ;

[0087] in, Indicates stable channel input characteristics; This represents the stable context vector obtained by weighting the center vectors of stable sample segments; Indicates the first The nth sample pair Attention weights for a stable sample segment; This indicates the score for the unnormalized match; Represents the stable channel attention projection vector; Represents the stable channel feature mapping matrix; Represents the stable channel bias vector; Represents the hyperbolic tangent function; This represents the vector concatenation operator; Indicates the input features of the offset channel; Indicates the first The offset response vector of each sample; Show the first The degree of local stability of each sample; Indicates the first The trajectory deflection of each sample; This represents the total number of stable sample segments;

[0088] Step S42: Stable prediction, input the stable channel obtained in step S41 into the feature. In the input stable reference neural network, stable processing mode features are extracted through multi-layer nonlinear mapping to obtain the stable reference prediction result, as shown below:

[0089] ;

[0090] in, Indicates the first Stable reference prediction results for each sample; and These represent the weight matrices of the first and second layers of the stable channel, respectively. and These represent the bias vectors of the first and second layers of the stable channel, respectively. Represents a linear rectified function;

[0091] Step S43: Correct the prediction, and input the offset channel obtained in step S41 into the feature. The input offset correction neural network first extracts the nonlinear representation of the offset features, then generates a correction term, which is then superimposed with the stable reference prediction result to obtain the offset correction prediction result, as shown below:

[0092] ;

[0093] in, Indicates the first The offset correction amount corresponding to each sample; Indicates the first Offset correction prediction results for each sample; and These represent the weight matrices of the first and second layers of the offset channel, respectively. and These represent the offset vectors of the first and second layers of the offset channel, respectively;

[0094] Step S44: Fuse the dual-channel prediction results. Based on the stability of the samples and the degree of trajectory deflection, adaptively fuse the stable reference prediction results and the offset correction prediction results to form the final processing effect prediction vector. In specific implementation, first, based on the stability score of the samples... and trajectory deflection Calculate the offset channel fusion coefficient, and then perform a weighted fusion of the prediction results from the two channels, as shown below: , ;in, Indicates the first Offset channel fusion coefficients for each sample; Indicates the first The final processing effect prediction vector for each sample;

[0095] Step S45: Joint optimization. During the training phase, the stable reference prediction result, the offset correction prediction result, and the final fusion result are jointly optimized, as shown below:

[0096] ;

[0097] in, Represents the joint training loss function; This represents the total number of historical samples; Indicates the first The detection vector of the true processing effect of each sample; Indicates the first The final predicted vector for each sample; This represents the set of all network parameters for a stable channel; This represents the set of all network parameters for the offset channel; and These represent the regularization coefficients for the stable channel and the offset channel, respectively. This represents the F-norm.

[0098] By performing the above operations, this solution addresses the problem in existing mineral water treatment effect prediction methods that struggle to simultaneously consider both the regularity fitting under stable operating conditions and the offset correction under abnormal fluctuations, leading to prediction results being easily affected by local disturbances and insufficient generalization ability. This solution constructs a dual-channel prediction mechanism, modeling stable sample reference information and offset correction information separately, and then adaptively fusing them together with the sample stability degree and trajectory deflection degree. This allows the model to accurately depict stable treatment patterns while promptly correcting errors caused by abnormal offsets, thereby improving the accuracy, robustness, and adaptability of treatment effect prediction.

[0099] Example 6, see Figure 1 and Figure 7 This embodiment is based on the above embodiment. In step S5, the output processing effect level specifically includes the following steps:

[0100] Step S51: Calculate the overall processing effect score. The components of the final processing effect prediction vector are weighted and summed according to preset weights to obtain the overall processing effect score, expressed as: ; Indicates the first The overall processing effect score of each sample; Represents the final processing effect prediction vector The first in One component; These correspond to the effects of effluent turbidity control, mineral retention, and microbial control, respectively. Indicates the first The rating weights of each component are given, and the following conditions are met: ;

[0101] Step S52: Output the processing effect level, mapping the overall processing effect score to a specific processing effect level, thereby giving the final processing evaluation result, as shown below:

[0102] ;

[0103] in, Indicates the first The treatment effect level of each sample; , and This represents the level thresholds arranged from high to low, and satisfies... Level 0 indicates the best processing effect, level 1 indicates a good processing effect, level 2 indicates a moderate processing effect, and level 3 indicates a poor processing effect.

[0104] Example 7, see Figure 2Based on the above embodiments, this embodiment provides an intelligent mineral water treatment effect prediction system, including a basic dataset construction module, a response vector construction module, a stable sample set generation module, a dual-channel prediction module, and an output treatment effect level module.

[0105] The module for building the basic dataset collects information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleans and repairs anomalies, and finally forms a standardized basic dataset that can be directly used for subsequent modeling. The data is then sent to the module for building the response vector.

[0106] The response vector construction module receives data sent by the basic dataset construction module, integrates raw water quality information, process operation information, and historical treatment effect information into a unified response vector, and supplements the state change trend and the correlation information between variables, and sends the data to the stable sample set generation module.

[0107] The module for generating a stable sample set receives data sent by the module for constructing a response vector. Based on the stable state features extracted during the processing of the response vector, the module first converts the samples into a state representation that can reflect the processing trend, then selects stable sample segments, and sends the data to the dual-channel prediction module.

[0108] The dual-channel prediction module receives data sent by the stable sample set generation module, establishes a dual-channel prediction mechanism, one channel focuses on using stable sample information for stable reference prediction, and the other channel focuses on analyzing the correction effect caused by sample offset and fluctuation. Finally, the two results are fused to obtain the mineral water treatment effect prediction result, and the data is sent to the output treatment effect level module.

[0109] The output processing effect level module receives data sent by the dual-channel prediction module, summarizes the final prediction results into a comprehensive score, and outputs the corresponding processing effect level according to the range in which the score falls, thus forming the final result of the mineral water processing effect.

[0110] It should be noted that, in this document, 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.

[0111] 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.

[0112] 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 predicting the treatment effect of mineral water, characterized in that: The method includes the following steps: Step S1: Construct a basic dataset by collecting information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleaning and anomaly repair, and finally forming a standardized basic dataset that can be directly used for subsequent modeling. Step S2: Construct a response vector by integrating raw water quality information, process operation information, and historical treatment effect information into a unified response vector, while supplementing the trend of state changes and the correlation information between variables; Step S3: Generate a stable sample set. Extract stable state features from the response vector during the processing. First, convert the samples into state representations that can reflect the processing trend, and then select stable sample segments. Step S4: Dual-channel prediction. A dual-channel prediction mechanism is established. One channel focuses on using stable sample information for stable reference prediction, while the other channel focuses on analyzing the correction effects caused by sample offset and fluctuation. Finally, the two results are fused to obtain the prediction result of the mineral water treatment effect. Step S5: Output the treatment effect level. Summarize the final prediction results into a comprehensive score, and output the corresponding treatment effect level according to the range in which the score falls, forming the final result of the mineral water treatment effect.

2. The intelligent method for predicting the treatment effect of mineral water according to claim 1, characterized in that: In step S3, generating a stable sample set specifically includes the following steps: Step S31: Generate a triaxial state feature vector, converting the offset response vector of each sample into a triaxial feature vector that can characterize the processing state; compress the offset response vector into three main components through three sets of directional projections, which are used to characterize the impurity removal trend, mineral retention trend and operational stability trend, respectively. Step S32: Calculate the trajectory deflection degree to determine whether the changes of the sample in the feature state space are stable and to identify whether there are obvious fluctuations or deflections; first calculate the state change amplitude between adjacent samples, and then calculate the trajectory deflection degree through the three-point difference method; then, combine the two to form a local stability index. Step S33: Construct a stability score by combining the mineral retention during the processing with the degree of local stability to form a score index for screening stable sample segments; extract the components related to mineral retention and impurity control from the response vector and compare them with the target level to calculate the consistency score; then multiply the consistency score by the degree of local stability to obtain the stability score. Step S34: Generate stable sample segments by setting continuous samples with high stability scores as stable sample segments.

3. The intelligent method for predicting the treatment effect of mineral water according to claim 1, characterized in that: In step S4, the dual-channel prediction specifically includes the following steps: Step S41: Construct dual-channel neural network input features by combining the offset response vector obtained in step S2, the three-axis state feature vector obtained in step S3, and the stable sample segment information into dual-channel neural network input features; Step S42: Stable prediction. Input the stable channel input features obtained in step S41 into the stable reference neural network, extract the stable processing mode features through multi-layer nonlinear mapping, and obtain the stable reference prediction result. Step S43: Correct the prediction. Input the offset channel input features obtained in step S41 into the offset correction neural network. First, extract the nonlinear expression of the offset features, then generate the correction term, and superimpose it with the stable reference prediction result to obtain the offset correction prediction result. Step S44: Fuse the dual-channel prediction results. Based on the stability of the sample and the degree of trajectory deflection, adaptively fuse the stable reference prediction results and the offset correction prediction results to form the final processing effect prediction vector. Step S45: Joint optimization. During the training phase, the stable reference prediction result, the offset correction prediction result, and the final fusion result are jointly optimized.

4. The intelligent method for predicting the effect of mineral water treatment according to claim 1, characterized in that: In step S5, the output processing effect level specifically includes the following steps: Step S51: Calculate the comprehensive processing effect score. The components of the final processing effect prediction vector are weighted and summed according to preset weights to obtain the comprehensive processing effect score. Step S52: Output the processing effect level, mapping the overall processing effect score to a specific processing effect level, thereby giving the final processing evaluation result.

5. The intelligent method for predicting the effect of mineral water treatment according to claim 1, characterized in that: In step S2, constructing the response vector specifically includes the following steps: Step S21: Form a unified response vector by integrating the raw water quality status, process status, and historical treatment effect detection vectors into a unified response vector; Step S22: Construct an offset response vector. Extract a stable benchmark from historical samples and construct an offset response vector based on it. First, assign a stable weight to each sample according to the stability of the sample in the fluctuation of the operating conditions, and then use these weights to calculate the stable benchmark vector. Subsequently, extract the difference between each sample and the benchmark vector, and introduce the product relationship between the sample and the benchmark to form the offset response vector.

6. The intelligent method for predicting the effect of mineral water treatment according to claim 1, characterized in that: In step S1, constructing the basic dataset specifically includes the following steps: Step S11: Data collection. Collect basic data on the mineral water treatment process and unify data from different sources and frequencies to the same time scale. Step S12: Outlier identification. Identify outlier samples in the collected data and perform robust repair on them.

7. An intelligent mineral water treatment effect prediction system, used to implement the intelligent mineral water treatment effect prediction method as described in any one of claims 1-6, characterized in that: It includes modules for building a basic dataset, building response vectors, generating a stable sample set, dual-channel prediction, and outputting the processing effect level.

8. The intelligent mineral water treatment effect prediction system according to claim 7, characterized in that: The module for building the basic dataset collects information on raw water quality, process operation status, and treatment effect involved in the mineral water treatment process, cleans and repairs anomalies, and finally forms a standardized basic dataset that can be directly used for subsequent modeling. The data is then sent to the module for building the response vector. The response vector construction module receives data sent by the basic dataset construction module, integrates raw water quality information, process operation information, and historical treatment effect information into a unified response vector, and supplements the state change trend and the correlation information between variables, and sends the data to the stable sample set generation module. The module for generating a stable sample set receives data sent by the module for constructing a response vector. Based on the stable state features extracted during the processing of the response vector, the module first converts the samples into a state representation that can reflect the processing trend, then selects stable sample segments, and sends the data to the dual-channel prediction module. The dual-channel prediction module receives data sent by the stable sample set generation module, establishes a dual-channel prediction mechanism, one channel focuses on using stable sample information for stable reference prediction, and the other channel focuses on analyzing the correction effect caused by sample offset and fluctuation. Finally, the two results are fused to obtain the mineral water treatment effect prediction result, and the data is sent to the output treatment effect level module. The output processing effect level module receives data sent by the dual-channel prediction module, summarizes the final prediction results into a comprehensive score, and outputs the corresponding processing effect level according to the range in which the score falls, thus forming the final result of the mineral water processing effect.

Images