Intelligent control method and system for squid formaldehyde removal process, equipment and medium

Abstract

The invention discloses an intelligent control method and system for a squid formaldehyde removal process, equipment and a medium. The method comprises the following steps: acquiring index parameters of a solution in the process of removing formaldehyde from squids, and generating an original data set containing a timestamp; normalizing the original data set to generate a normalized process data set which comprises a timestamp and a normalized field; index parameters are extracted from the normalized process data set to serve as feature vectors, the formaldehyde content at the corresponding time point serves as a label, sequence data are intercepted according to a preset time window, and a time sequence training data set is generated; constructing a long short-term memory network model, wherein the long short-term memory network model comprises a bidirectional LSTM layer and the like; according to the method, the time sequence dependence characteristics of process parameters are captured through a long short-term memory network (LSTM) to realize dynamic parameter adjustment, a data-driven real-time control system for formaldehyde removal of squids is finally formed, and the static regulation and control limitation of a traditional process is broken through.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent squid formaldehyde removal, and in particular to intelligent control of a squid formaldehyde removal process. Background Art

[0002] Squid and other aquatic products are prone to formaldehyde during processing, mainly due to the degradation of trimethylamine oxide and pollution in the processing process. Formaldehyde residues pose a serious threat to human health. my country's "National Food Safety Standard" clearly stipulates that the formaldehyde content in aquatic products shall not exceed 0.1 mg / kg. The inventors of this case have previously conducted in-depth research on the squid formaldehyde removal process (Zhang Cifeng, Sun Huamiao, Chen Jiannan, et al. Orthogonal test method to optimize the process conditions for formaldehyde removal in Indian Ocean squid [J]. Fujian Agricultural Science and Technology, 2024, 55(7): 69-73.) However, the existing squid formaldehyde removal process still has the following technical bottlenecks:

[0003] 1. Process parameters rely on experience-based control and lack dynamic adaptability

[0004] Traditional formaldehyde removal processes often rely on fixed parameters (such as pH and reaction temperature) and manual adjustments. However, formaldehyde production is dynamically influenced by multiple factors, including raw material quality. Fixed parameter models can easily lead to undertreatment (exceeding the specified residue limit) or overtreatment (damaging squid quality).

[0005] 2. Lack of real-time monitoring and feedback mechanisms

[0006] Existing technologies typically use offline detection (such as spectrophotometry) to assess formaldehyde content. This is subject to detection lag (requiring process termination and sampling), making it difficult to adjust process parameters in a timely manner. While some studies have attempted to introduce online sensors, these have not addressed the issue of multi-parameter temporal correlation modeling, resulting in insufficient prediction accuracy.

[0007] 3. Difficulty in collaborative optimization of complex process parameters

[0008] The formaldehyde removal process involves complex nonlinear relationships among parameters such as pH, temperature, and flow rate. Previous studies have employed PID control or rule-based strategies, but these strategies struggle to capture the dynamic temporal dependencies between these parameters (e.g., the delayed effect of temperature changes on formaldehyde release rate). Furthermore, parameter adjustments lack a quantitative basis, making them prone to falling into local optima.

[0009] 4. Insufficient model generalization ability

[0010] Although existing machine learning methods (such as support vector machines and random forests) can be used for formaldehyde prediction, they do not fully consider the temporal characteristics of process data and have limited ability to model long-term fluctuation patterns (such as the impact of periodic pH changes on reaction rate).

[0011] Based on the above-mentioned deficiencies, the core issues that need to be urgently addressed in this field include:

[0012] How to establish a dynamic time series mapping model between squid formaldehyde removal process parameters and formaldehyde content to achieve accurate prediction under multi-parameter coupling;

[0013] How to build a real-time closed-loop control mechanism to dynamically optimize process parameters based on prediction results, balancing processing efficiency and product quality;

[0014] How to design an interference-resistant incremental learning algorithm to adapt to model generalization requirements under complex working conditions such as raw material differences and environmental fluctuations. Summary of the Invention

[0015] To address the above problems, the present invention proposes an intelligent optimization method that integrates time series modeling and closed-loop control. The method captures the time series dependence characteristics of process parameters through a long short-term memory network (LSTM), combines the EWMA control chart with the PID algorithm to achieve dynamic parameter adjustment, and ultimately forms a data-driven squid formaldehyde removal real-time control system, breaking through the static control limitations of traditional processes.

[0016] According to one aspect of the present invention, there is provided an intelligent control method for a squid formaldehyde removal process, the method comprising:

[0017] Collect the indicator parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp;

[0018] Normalize the original data set to generate a normalized process data set, which includes timestamps and normalized fields;

[0019] Extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set;

[0020] Constructing a long short-term memory network model, the long short-term memory network model including a bidirectional LSTM layer; obtaining feature vectors from a time series training dataset, training the model to learn the temporal dependency between process parameters and formaldehyde content, and obtaining a prediction model;

[0021] The formaldehyde content in the squid formaldehyde removal process is predicted based on this prediction model, and the process parameters of the squid formaldehyde removal are adjusted in real time based on the prediction results.

[0022] In the above-mentioned technical solution, during the squid formaldehyde removal process, parameter indicators are collected from the solution at each stage, generating a time-stamped raw data set. Real-time monitoring of these parameters allows for a detailed record of all changes throughout the formaldehyde removal process. The time stamps provide the foundation for subsequent analysis of the temporal evolution of these parameters, ensuring the temporal coherence and integrity of the data, which is crucial for accurately grasping the dynamics of the formaldehyde removal process. The raw data set is then normalized to produce a normalized process data set containing timestamps and normalization fields. The advantage of normalization is that it eliminates differences in the dimensions and numerical ranges of different parameter indicators, bringing the data into a consistent scale. This prevents undue dominance or suppression of model training due to overly large or small values ​​in some parameters during subsequent model training and analysis. This allows the model to more objectively and accurately identify the inherent relationships between process parameters and formaldehyde content, thereby improving the model's robustness and accuracy. The time series training dataset is generated by selecting indicator parameters from the normalized process dataset as feature vectors, using the formaldehyde content at the corresponding time point as a label, and intercepting the sequence data according to a preset time window. This approach fully aligns with the temporal nature of the formaldehyde removal process. The defined time window captures the impact of fluctuations in various process parameters on formaldehyde content within a specific time period, providing the model with sufficient temporal information to help it learn the complex dynamic dependencies between process parameters and formaldehyde content, laying a solid data foundation for subsequent accurate formaldehyde content prediction and effective adjustment of process parameters.

[0023] A long-short-term memory network model incorporating a bidirectional LSTM layer was constructed and trained using feature vectors from a time series training dataset. This approach aims to enable the model to understand the temporal dependencies between process parameters and formaldehyde content, thereby constructing a predictive model. The bidirectional LSTM layer's strength lies in its ability to simultaneously account for past and future temporal information within a specific time window, aligning with the dynamic continuity and mutual influence of the formaldehyde removal process. Compared to unidirectional LSTM models and traditional time series models, it is more capable of comprehensively capturing the potential impact of process parameter changes on formaldehyde content, offering superior modeling and prediction capabilities, facilitating accurate prediction of formaldehyde content trends during squid formaldehyde removal.

[0024] The generated prediction model is used to predict formaldehyde levels during the squid formaldehyde removal process, and process parameters are adjusted in real time based on the prediction results. This real-time prediction provides early insight into potential fluctuations in formaldehyde levels, enabling rapid and precise optimization and adjustment of process parameters. This results in efficient progress and intelligent, refined control of the squid formaldehyde removal process, potentially yielding significant advantages in improving formaldehyde removal efficiency, ensuring squid quality, and reducing production costs.

[0025] In some embodiments, the parameter values ​​of the solution during the squid formaldehyde removal process are collected to generate a raw data set including a timestamp, including:

[0026] The pH value, reaction temperature, processing time, and formaldehyde content of the solution during the squid formaldehyde removal process are collected by sensors at a preset frequency to generate a first data set containing a timestamp. Each record contains a timestamp, pH value, reaction temperature, processing time, and formaldehyde content. The data integrity is checked from the first data set, and records with pH values ​​outside the range of 4 to 10 and formaldehyde contents exceeding the range of ±3 times the standard deviation of the historical data mean are eliminated to generate a second data set, which is used as the original data set.

[0027] In this technical solution, during the squid formaldehyde removal process, sensors collect key solution parameters at a preset frequency, including pH, reaction temperature, treatment time, and formaldehyde content, generating a time-stamped first dataset. This multi-dimensional data collection comprehensively records the factors influencing the formaldehyde removal process and its state changes. Timestamps ensure data chronology, and the preset frequency ensures data continuity and integrity, providing systematic, regular, and representative raw material for subsequent data processing and analysis.

[0028] Performing an integrity check on the first dataset is critical for ensuring data quality. Complete, unmissing data is the foundation of reliable analysis and prevents model training bias or erroneous conclusions caused by missing data. Records with pH values ​​outside the range of 4 to 10 are eliminated. This setting is based on the acceptable pH range for the solution in the formaldehyde removal process. Excessively low or high pH values ​​may be caused by abnormal conditions such as sensor failure and measurement error. Eliminating these abnormal data improves data accuracy and validity, ensuring that the remaining data more accurately reflects the impact of pH values ​​within the normal process range on formaldehyde removal.

[0029] Statistical methods were used to identify and eliminate outliers by eliminating records with formaldehyde levels outside the range of ±3 standard deviations from the historical mean. Three standard deviations represent a reasonable range for normal data fluctuations; values ​​outside this range are likely anomalous data points caused by measurement error or abnormal process interference. Eliminating these outliers better reflects the normal variation in formaldehyde levels during the actual formaldehyde removal process, facilitating the construction of accurate and practical prediction models, while preventing the model from being misled by abnormal data and producing illogical predictions.

[0030] The second dataset generated through the aforementioned data collection and cleaning process serves as the raw data set, providing a high-quality, reliable data foundation for model training and analysis in the intelligent squid formaldehyde removal control method. This is a key prerequisite for the effective operation and precise regulation of the intelligent control method, and is of great significance for improving the intelligent control level of the squid formaldehyde removal process and the stability and reliability of the formaldehyde removal effect. In summary, this data collection and cleaning method, through multi-dimensional, high-frequency data collection and strict data quality control, ensures the comprehensiveness, accuracy, and representativeness of the data, providing solid data support for the implementation of the intelligent squid formaldehyde removal control method.

[0031] In some embodiments, the index parameters are extracted from the normalized process data set as feature vectors, the formaldehyde content at the corresponding time point is used as a label, and the sequence data is intercepted according to a preset time window to generate a time series training data set, including:

[0032] Obtaining normalized pH, reaction temperature, and processing time from the process dataset to generate an initial feature vector, and extracting the formaldehyde content at the corresponding time point as a label to obtain a first training dataset containing a timestamp;

[0033] If there are missing values ​​in the feature vector of the first training data set, the missing values ​​are interpolated using the mean filling method to generate the second training data set;

[0034] The mutual information between each feature in the second training data set and the formaldehyde content is calculated using a time series feature selection algorithm, and features with a mutual information greater than 0.3 are selected to generate an optimized feature vector to obtain a time series training data set.

[0035] In this technical solution, the normalized pH, reaction temperature, and treatment time are extracted from the normalized process data set to form the initial feature vector. The formaldehyde content at the corresponding time point is used as a label to generate a first training data set with a timestamp. This process ensures the temporal nature of the data, enabling the model to capture the impact of process parameters on formaldehyde content over time and providing the foundational data structure for building an effective prediction model.

[0036] If there are missing values ​​in the first training dataset, we use mean filling to interpolate and generate the second training dataset. Mean filling is a simple and commonly used method for handling missing values. It replaces missing values ​​with the mean of the known data, thereby preserving the statistical characteristics of the dataset to a certain extent, avoiding insufficient model training samples or bias caused by missing data, and ensuring the smooth progress of the model training process.

[0037] Furthermore, a time series feature selection algorithm was used to calculate the mutual information between each feature in the second training dataset and the formaldehyde content. Features with mutual information greater than 0.3 were selected to generate optimized feature vectors, resulting in a time series training dataset. This step aims to screen out features with strong correlation and contribution to formaldehyde content prediction, while eliminating features with weaker correlations. This simplifies model input, reduces model complexity, and improves the model's predictive performance and generalization capabilities. This allows the model to more accurately focus on the temporal dependency between key features and formaldehyde content, thereby enhancing the efficiency and accuracy of the intelligent control method.

[0038] In summary, this process, through steps such as initial feature extraction, missing value processing, and feature selection optimization, generates a high-quality time series training dataset from a normalized process dataset. This provides a strong foundation for building an effective formaldehyde content prediction model for the squid formaldehyde removal process, thereby facilitating precise and intelligent control of the squid formaldehyde removal process. This series of operations ensures the temporal sequence, integrity, and relevance of the data, providing a solid foundation for model training and prediction, and is a key step in achieving intelligent and precise control of the squid formaldehyde removal process.

[0039] In some embodiments, a long short-term memory network model is constructed, wherein the long short-term memory network model includes a bidirectional LSTM layer; a feature vector is obtained from a time series training dataset, and a model is trained to learn the temporal dependency between process parameters and formaldehyde content to obtain a prediction model, including:

[0040] Obtain time series data including process parameters and formaldehyde content from the time series training dataset, extract feature vectors through preprocessing, and obtain quantitative representation;

[0041] A long short-term memory network model is used to train the model based on the feature vector to obtain an initial model; if the error of the initial model in predicting the formaldehyde content exceeds a first preset threshold, the weights of the long short-term memory layer are adjusted through back propagation, and the network model is updated until a model with an error lower than the first preset threshold is obtained, which is used as the prediction model.

[0042] In this technical solution, time series data, including process parameters and formaldehyde content, is extracted from the time series training dataset to form the data foundation for model training, ensuring data integrity and temporal consistency. Preprocessing extracts feature vectors and converts them into quantitative representations. This step transforms the raw data into a format that the model can recognize and process, facilitating subsequent effective training and enabling the model to learn patterns based on these quantitative features.

[0043] A Long Short-Term Memory (LSTM) model was trained based on feature vectors to generate an initial model. As a specialized recurrent neural network (RNN) architecture, LSTM, through its unique memory units and gating mechanism, effectively processes and predicts time series data, resolving the vanishing and exploding gradient issues that plague traditional RNNs when processing long sequences. In the squid formaldehyde removal process, which exhibits significant temporal dependencies, the LSTM model effectively captures the impact of process parameters on formaldehyde content over time. Initial model training is a fundamental step in the overall model building process, providing a starting point for subsequent optimization.

[0044] If the formaldehyde content error predicted by the initial model exceeds a first preset threshold, the LSTM layer weights are adjusted using a backpropagation algorithm, and the network model is updated until a model with an error below the first preset threshold is obtained, which is then used as the final prediction model. By setting the error threshold as the evaluation criterion, the model's prediction accuracy is ensured to meet practical application requirements. The backpropagation algorithm automatically adjusts the network weight parameters based on the prediction error, allowing the model to continuously learn and correct its own prediction bias, gradually improving the accuracy of formaldehyde content predictions. The resulting prediction model can, given given process parameters, relatively accurately predict the changing trend of formaldehyde content during the squid formaldehyde removal process, providing a reliable basis for subsequent real-time adjustment of process parameters and achieving intelligent and precise control of the squid formaldehyde removal process.

[0045] Overall, this process, through data preprocessing, initial model training, and optimization steps based on error feedback, has built an LSTM prediction model that can effectively learn the temporal dependency between process parameters and formaldehyde content, providing core technical support for the intelligent control of the squid formaldehyde removal process.

[0046] In some embodiments, the formaldehyde content of the squid formaldehyde removal process is predicted based on the prediction model, and the process parameters of the squid formaldehyde removal are adjusted in real time based on the prediction result, including:

[0047] The pH value, reaction temperature, processing time, and formaldehyde content of the solution during the formaldehyde removal process are collected in real time and input into the optimized training model to predict the formaldehyde content value;

[0048] If the predicted value is higher than a second preset threshold, the process parameter adjustment amount is calculated using a proportional integral differential algorithm based on the deviation between the predicted value and the target value to generate a process parameter adjustment set, which includes a pH target value, a reaction temperature target value, and a processing time adjustment amount;

[0049] Adjust the squid formaldehyde removal process parameters based on the process parameter adjustment set.

[0050] In the above technical solution, during the squid formaldehyde removal process, real-time data on the solution's pH, reaction temperature, processing time, and formaldehyde content are collected and fed into an optimized, trained model to predict formaldehyde content. This step enables real-time monitoring of the formaldehyde removal process. By feeding the collected current process parameter data into the trained prediction model, a predicted formaldehyde content value can be quickly obtained, providing a real-time basis for subsequent process parameter adjustments.

[0051] If the predicted value is higher than the second preset threshold, the process parameter adjustment amount is calculated based on the deviation between the predicted value and the target value through the proportional integral differential (PID) algorithm. The PID algorithm is a proportional (P), integral (I) and differential (D) control algorithm based on deviation, and is widely used in industrial process control. The proportional term adjusts the control amount proportionally according to the size of the current deviation, the integral term adjusts the control amount according to the cumulative history of the deviation to eliminate static error, and the differential term adjusts the control amount according to the rate of change of the deviation to predict system trends and suppress overshoot. Through the PID algorithm, the pH target value, reaction temperature target value and processing time adjustment amount can be calculated, and a process parameter adjustment set containing these values ​​can be generated. This adjustment based on the PID algorithm can provide quantitative and precise guidance for subsequent specific parameter adjustments.

[0052] Adjusting the squid formaldehyde removal process parameters based on the process parameter adjustment set is a key step in applying the previously predicted and calculated results to the actual production process. By adjusting key process parameters such as pH, reaction temperature, and treatment time in real time, the reaction conditions and kinetics of the formaldehyde removal process can be effectively influenced, thereby optimizing the formaldehyde removal effect. This real-time feedback and adjustment mechanism ensures that the formaldehyde removal process always operates within a relatively ideal process parameter range, improving formaldehyde removal efficiency and ensuring the quality of the squid product.

[0053] In some embodiments, adjusting squid formaldehyde removal process parameters based on the process parameter adjustment set further includes: collecting new solution pH, reaction temperature, processing time, and formaldehyde content to generate a third data set including a timestamp;

[0054] The third data set is input into the prediction model to predict the real-time formaldehyde value, and the process stability is determined based on the EWMA control chart; the smoothing coefficient λ of the EWMA control chart is 0.2, and the control limit is ±3 times the standard deviation;

[0055] If it is stable, the final process parameter set is generated, including the pH target value, reaction temperature target value and processing time adjustment.

[0056] In this technical solution, key process parameters such as pH, reaction temperature, and treatment time are adjusted in real time based on a process parameter adjustment set to optimize the reaction conditions and kinetics of the formaldehyde removal process, thereby improving formaldehyde removal efficiency. This step directly applies the previously predicted and calculated results to the actual production process, achieving dynamic optimization of process parameters.

[0057] Under the adjusted process conditions, the new solution pH, reaction temperature, treatment time, and formaldehyde content are collected in real time, generating a third data set with a timestamp. This step not only enables further monitoring of the adjusted formaldehyde removal process but also provides detailed data support for subsequent model validation and further adjustments by recording the new process parameter data.

[0058] The third data set is fed into the prediction model to predict real-time formaldehyde values. The EWMA control chart is then used to determine process stability. As an effective quality control tool, the EWMA control chart can promptly identify subtle process changes and stability issues, ensuring that adjusted process parameters maintain optimal formaldehyde removal process stability. By predicting new formaldehyde levels in real time and analyzing the EWMA control chart, it is possible to accurately determine whether the process has reached a stable state.

[0059] If the formaldehyde removal process is deemed stable, a final set of process parameters is generated, including target pH values, target reaction temperature, and process time adjustments. This final set of process parameters can be used to guide subsequent production operations, ensuring the efficient and stable performance of the squid formaldehyde removal process.

[0060] In some embodiments, the third data set is input into a prediction model to predict real-time formaldehyde values. The process stability is determined based on an EWMA control chart. If the process is stable, a final process parameter set is generated, including a pH target value, a reaction temperature target value, and a processing time adjustment, and further comprising:

[0061] If unstable, then:

[0062] A grid search method is used to adjust the value ranges of solution pH, reaction temperature, and processing time to generate new process parameter combinations and update the feature data set.

[0063] The formaldehyde content is repeatedly predicted using the updated feature data set, and the process stability is determined again based on the EWMA control chart. If stable, the final pH target value, reaction temperature target value, and treatment time adjustment are output to obtain the final process parameter set.

[0064] If not, return and repeat the above steps.

[0065] In the above technical solution, the squid formaldehyde removal process employs an exhaustive grid search method. By traversing a preset parameter grid of solution pH, reaction temperature, and treatment time, the team seeks the optimal process parameter combination for formaldehyde removal. If the process is unstable, the grid search adjusts the value ranges of these three parameters to generate a new process parameter combination. This new combination is then used to update the feature dataset, providing a new data foundation for subsequent predictions and adjustments, helping to gradually explore even more optimal process parameter combinations. Based on the updated feature dataset, the prediction model is again used to predict formaldehyde content, and the process stability is determined using an EWMA control chart. If the process is stable, the final target pH, reaction temperature, and treatment time adjustment are output to obtain the final process parameter set. If the process is still unstable, the process is repeated, adjusting the process parameters until the optimal combination is found, ensuring stable and reliable formaldehyde removal. Grid search explores all possible combinations within the predefined parameter space, ensuring that no optimal configuration is missed, thus improving formaldehyde removal effectiveness.

[0066] According to another aspect of the present invention, an intelligent control system for the formaldehyde removal process of squid is provided, based on the above method; the system comprises:

[0067] The acquisition module is used to collect the index parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp;

[0068] The preprocessing module is used to normalize the original data set and generate a normalized process data set, which includes timestamps and normalized fields;

[0069] The data module is used to extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set;

[0070] A neural network module is used to construct a long short-term memory network model, wherein the long short-term memory network model includes a bidirectional LSTM layer; obtain feature vectors from a time series training data set, train the model to learn the temporal dependency between process parameters and formaldehyde content, and obtain a prediction model;

[0071] The prediction module is used to predict the formaldehyde content in the squid formaldehyde removal process based on the prediction model, and adjust the process parameters of the squid formaldehyde removal in real time based on the prediction results.

[0072] In the above technical solution, in order to better use the above method, the present application proposes an intelligent control system for the squid formaldehyde removal process, and each module corresponds to each step of the above method. The specific principles have been described above and will not be repeated here.

[0073] According to yet another aspect of the present invention, there is provided an intelligent control device for a squid formaldehyde removal process, comprising: at least one processor and a memory communicatively connected to the at least one processor;

[0074] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the above method.

[0075] In the above technical solution, in order to better run and process the method, the above method is stored in a memory and a processor is used to execute the stored method. It should be noted that the principle and effect of each step have been described above and will not be further explained here.

[0076] According to another aspect of the present invention, a computer-readable storage medium is provided, storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0077] In the above technical solution, in order to better run and use the method, the above method is stored in a computer-readable storage medium and implemented by a processor. It should be noted that the principle and effect of each step have been described above and will not be further explained here. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0079] Figure 1 This is a flow chart of an embodiment of an intelligent control method for squid formaldehyde removal process according to the present invention;

[0080] Figure 2 The present invention is a schematic structural diagram of an intelligent control system for a squid formaldehyde removal process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The present invention will be described in further detail below with reference to the accompanying drawings and examples. It is particularly noted that the following examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Similarly, the following examples are only some embodiments of the present invention and are not intended to be exhaustive. All other embodiments obtained by those of ordinary skill in the art without creative effort are intended to fall within the scope of protection of the present invention.

[0082] Example 1

[0083] See also Figure 1 , an intelligent control method for a squid formaldehyde removal process, the method comprising:

[0084] S1, collect the index parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp;

[0085] In this embodiment, S1, collecting the index parameters of the solution during the squid formaldehyde removal process to generate a raw data set including a timestamp, including:

[0086] The pH value, reaction temperature, processing time, and formaldehyde content of the solution during the squid formaldehyde removal process are collected by sensors at a preset frequency to generate a first data set containing a timestamp. Each record contains a timestamp, pH value, reaction temperature, processing time, and formaldehyde content. The data integrity is checked from the first data set, and records with pH values ​​outside the range of 4 to 10 and formaldehyde contents exceeding the range of ±3 times the standard deviation of the historical data mean are eliminated to generate a second data set, which is used as the original data set.

[0087] For example, sensors acquire minute-by-minute data on pH, reaction temperature, processing time, and formaldehyde content, generating a raw data set with a timestamp. The pH and reaction temperature in the raw data set are filtered using a preset threshold range to identify anomalous data points, resulting in a filtered data set. Specifically, the sensors collect data once a minute, with a timestamp format of YYYY-MM-DD HH:MM:SS. For example, at 2023-10-15 14:05:00, the pH is recorded as 7.2, the reaction temperature is 25.3°C, the processing time is 15 minutes, and the formaldehyde content is 12.5 mg / L. The raw data is transmitted via an IoT terminal to a cloud database and stored using the time series database InfluxDB, indexed using the timestamp as the primary key. During the data preprocessing phase, a sliding window algorithm is used to detect outliers, with a window size of 5. Data points are marked as outliers if they deviate from the mean within the window by more than 2 standard deviations.

[0088] S2. Normalize the original data set to generate a normalized process data set, which includes a timestamp and normalized fields;

[0089] For example, four key fields, pH, reaction temperature, processing time, and formaldehyde content, are first extracted from the complete process data set. For example, a record has a pH of 8.2, a reaction temperature of 65.3°C, a processing time of 120 minutes, and a formaldehyde content of 0.45 mg / L. Statistics are calculated for each field. For example, if the minimum pH value is 6.5 and the maximum is 9.8, the normalized value of 8.2 is (8.2 - 6.5) / (9.8 - 6.5) = 0.515. Similarly, for the temperature field, if the minimum value is 50.0°C and the maximum value is 80.0°C, 65.3°C is normalized to (65.3 - 50.0) / (80.0 - 50.0) = 0.510. For the processing time field, if the minimum value is 90 minutes and the maximum value is 150 minutes, 120 minutes is normalized to (120 - 90) / (150 - 90) = 0.500. Next, the mean and standard deviation of each field are calculated. For example, if the mean formaldehyde content is 0.40 mg / L and the standard deviation is 0.05 mg / L, then the range of three times the standard deviation is 0.25 to 0.55 mg / L. Records outside this range are discarded. Finally, a new dataset is generated containing timestamps and normalized fields. For example, a record with a timestamp of 2023-05-01 14:30:00 has a normalized pH of 0.515, a reaction temperature of 0.510, a processing time of 0.500, and a formaldehyde content of 0.450.

[0090] S3. Extracting index parameters from the normalized process data set as feature vectors, using the formaldehyde content at the corresponding time point as a label, intercepting sequence data according to a preset time window, and generating a time series training data set;

[0091] In this embodiment, S3, extracting index parameters from the normalized process data set as feature vectors, using the formaldehyde content at the corresponding time point as a label, intercepting sequence data according to a preset time window, and generating a time series training data set, includes:

[0092] S31, obtaining normalized pH, reaction temperature, and processing time from the process data set, generating an initial feature vector, and extracting the formaldehyde content at the corresponding time point as a label to obtain a first training data set including a timestamp;

[0093] S32. If there are missing values ​​in the feature vectors of the first training data set, interpolate the missing values ​​using a mean filling method to generate a second training data set;

[0094] S33. Calculate the mutual information between each feature in the second training data set and the formaldehyde content using a time series feature selection algorithm, select features with a mutual information greater than 0.3 to generate an optimized feature vector, and obtain a time series training data set.

[0095] Exemplarily, S31: Generate a first training data set including a timestamp

[0096] Assume that the process data of a formaldehyde production plant is as follows (after normalization, the time interval is 1 hour):

[0097] Timestamp pH (X1) Reaction temperature (X2) Processing time (X3) Formaldehyde content (Y) 2023-01-01 08:00 0.5 0.8 0.6 0.72 2023-01-01 09:00 0.6 0.7 0.5 0.65 2023-01-01 10:00 NaN 0.9 0.7 0.81 2023-01-01 11:00 0.4 0.6 NaN 0.58

[0098] S32: Missing value interpolation to generate the second training data set

[0099] Calculate the mean of each feature:

[0100] Average pH = (0.5 + 0.6 + 0.4) / 3 = 0.5

[0101] Mean processing time = (0.6 + 0.5 + 0.7) / 3 = 0.6

[0102] Fill in missing values. The data after filling is as follows:

[0103] Timestamp pH (X1) Reaction temperature (X2) Processing time (X3) Formaldehyde content (Y) 2023-01-01 08:00 0.5 0.8 0.6 0.72 2023-01-01 09:00 0.6 0.7 0.5 0.65 2023-01-01 10:00 0.5 0.9 0.7 0.81 2023-01-01 11:00 0.4 0.6 0.6 0.58

[0104] S33: Feature selection to generate optimized feature vectors

[0105] Calculate the mutual information (hypothetical result):

[0106] Mutual information between reaction temperature (X2) and formaldehyde content = 0.45

[0107] Mutual information between treatment time (X3) and formaldehyde content = 0.35

[0108] Mutual information between pH (X1) and formaldehyde content = 0.55

[0109] Filter features: Only retain features with mutual information > 0.3 (X1, X2, X3)

[0110] Generate sequence data by time window. Assuming the preset time window is 30 minutes, convert the data into sequence input;

[0111] S4. Constructing a long short-term memory network model, wherein the long short-term memory network model includes a bidirectional LSTM layer; obtaining feature vectors from the time series training data set, training the model to learn the temporal dependency between process parameters and formaldehyde content, and obtaining a prediction model;

[0112] In this embodiment, S4, constructing a long short-term memory network model, wherein the long short-term memory network model includes a bidirectional LSTM layer; obtaining feature vectors from a time series training dataset, training the model to learn the temporal dependency between process parameters and formaldehyde content, and obtaining a prediction model, including:

[0113] S41, obtaining time series data including process parameters and formaldehyde content from a time series training data set, extracting feature vectors through preprocessing, and obtaining a quantitative representation;

[0114] S42. Use a long short-term memory network model and train the model based on the feature vector to obtain an initial model; if the error of the initial model in predicting the formaldehyde content exceeds a first preset threshold, adjust the weight of the long short-term memory layer through back propagation and update the network model until a model with an error lower than the first preset threshold is obtained, which is used as the prediction model.

[0115] Exemplarily, time series data containing process parameters and formaldehyde content is obtained from a time series training dataset, and feature vectors are extracted through preprocessing to obtain a quantitative representation. A long short-term memory network is used to construct a network model comprising an input layer, two long short-term memory layers with 64 hidden units each, and a fully connected output layer. The model is trained based on the feature vectors to obtain an initial model. If the error in the formaldehyde content predicted by the initial model exceeds a preset threshold, the weights of the long short-term memory layer are adjusted through backpropagation, and the network model is updated to obtain an optimized model. Based on the optimized model, temporal dependency features are extracted from the time series data to generate a temporal relationship representation between the process parameters and the formaldehyde content. The temporal relationship representation is processed through the fully connected output layer, and the formaldehyde content is predicted to obtain a predicted result. If the deviation between the predicted result and the actual formaldehyde content exceeds a preset threshold, the parameters of the long short-term memory layer are further optimized through a gradient descent algorithm to obtain a fine-tuned model. The fine-tuned model is used to process new time series data, generate process parameter adjustment suggestions, and obtain optimized parameters.

[0116] Specifically, when constructing the long short-term memory network model, the input layer structure needs to be designed first. The input layer receives time series data, such as process parameters (temperature, pressure, humidity) and formaldehyde content data collected every hour. The time step is set to 24 steps, representing 24 hours of historical data. The input data is standardized using the Z-score method, with the mean temperature of 25°C and the standard deviation of 3°C, the mean pressure of 101kPa and the standard deviation of 5kPa, the mean humidity of 60% and the standard deviation of 10%, and the mean formaldehyde content of 0.1mg / m 3 , standard deviation is 0.02mg / m 3. Then, two LSTM hidden layers are constructed, each containing 64 hidden units. The tanh activation function is used, the dropout rate is set to 0.2 to prevent overfitting, the initial weights use the Xavier normal distribution, and the bias is initialized to 0. The output shape of the first layer LSTM is (batch size, time step, 64), and the second layer LSTM returns the complete sequence output. The fully connected output layer uses a linear activation function with an output dimension of 1, corresponding to the predicted value of formaldehyde content. The mean square error loss function is used during training, and the optimizer selects Adam. The initial learning rate is 0.001, the batch size is 32, and the number of training rounds is 100. The early stopping mechanism is used during training, and the training is terminated when the validation set loss does not decrease for 5 consecutive rounds. After the model training is completed, the test set is used to evaluate the performance, and the calculated root mean square error is 0.015mg / m 3 , coefficient of determination R 2 It reaches 0.92, indicating that the model can effectively capture the temporal dependence between process parameters and formaldehyde content.

[0117] S5. Predicting the formaldehyde content in the squid formaldehyde removal process based on the prediction model, and adjusting the process parameters of the squid formaldehyde removal in real time based on the prediction result.

[0118] In this embodiment, S5, predicting the formaldehyde content in the squid formaldehyde removal process based on the prediction model, and adjusting the squid formaldehyde removal process parameters in real time based on the prediction result, includes:

[0119] S51, collecting the pH value, reaction temperature, processing time, and formaldehyde content of the solution during the formaldehyde removal process in real time, and inputting the data into the optimized training model to predict the formaldehyde content value;

[0120] S52: If the predicted value is higher than a second preset threshold, the process parameter adjustment amount is calculated using a proportional integral differential algorithm based on the deviation between the predicted value and the target value to generate a process parameter adjustment set, including a pH target value, a reaction temperature target value, and a processing time adjustment amount;

[0121] S53. Adjust the squid formaldehyde removal process parameters based on the process parameter adjustment set.

[0122] For example, in the squid formaldehyde removal process, the proportional integral differential (PID) algorithm dynamically adjusts process parameters (such as reaction temperature, processing time, etc.) by quantifying the deviation between the predicted value and the target value. The specific calculation process is as follows:

[0123] 1. PID algorithm core formula

[0124] The process parameter adjustment is composed of three parts:

[0125]

[0126] Where: e(t) (deviation between formaldehyde target value and predicted value); k p : Proportional coefficient (directly responds to the current deviation); k i =k p / T i : Integral coefficient (eliminating historical cumulative deviation); k d =k p ·T d : differential coefficient (suppresses the trend of future deviation changes); T i : Integration time, T d : Differentiation time.

[0127] 2. Parameter adjustment calculation steps in squid formaldehyde removal process

[0128] Step 1: Deviation Calculation

[0129] Obtain the formaldehyde content prediction value (such as 0.08mg / kg) in real time and compare it with the target threshold (such as 0.1mg / kg): e(t) = 0.1-0.08 = 0.02mg / kg

[0130] Step 2: PID component calculation

[0131] Assume preset parameters (calibrated through process experiments): k p =0.8,T i =5min,T d =0 (no differential effect), integral coefficient k i =0.8 / 5=0.16min -1

[0132] Component calculation:

[0133] 1. Proportional term (P): directly responds to the current deviation; P = 0.8 × 0.02 = 0.016;

[0134] 2. Integral term (I): accumulated deviations over the past 5 minutes (assuming the integral sum of historical deviations);

[0135] I = 0.16 × 0.05 = 0.008;

[0136] 3. Differential term (D): T d =0, so D = 0;

[0137] Total adjustment: Δu(t) = 0.016 + 0.008 + 0 = 0.024

[0138] 3. Mapping rules from adjustment amount to process parameters

[0139] According to the process parameter sensitivity experiment, the conversion relationship between the adjustment amount Δu(t) and the specific parameters is established:

[0140] 1. Reaction temperature adjustment

[0141] Experiments show that for every 1°C increase in temperature, the formaldehyde degradation rate increases by 15% (effective in the pH range of 6-8)

[0142] ΔT=Δu(t)×K T (K T =2.5℃ / unit adjustment)

[0143] Example:

[0144] ΔT=0.024×2.5=0.06℃

[0145] Control logic: If the predicted value is lower than the target, the temperature is increased to accelerate the decomposition of formaldehyde.

[0146] Other indicators are the same as above and will not be explained in detail.

[0147] 4. Anti-interference design and parameter calibration

[0148] 1. Noise filtering: Use sliding average filtering (window length = 10 sampling points) on the sensor data to avoid sudden change interference.

[0149] 2. Integral anti-saturation: Set the upper limit of the integral term (such as |I|≤0.1) to prevent long-term deviation accumulation from causing overshoot.

[0150] 3. Parameter calibration method

[0151] Step response method: In a steady-state process, suddenly increase the formaldehyde concentration, observe the PID response curve, and adjust k p 、T i To overshoot <5%;

[0152] Ziegler-Nichols method: The critical proportionality method is used to determine the initial parameters, and then fine-tuned according to the process characteristics.

[0153] 5. Practical Application Examples

[0154] Scenario: During the processing of a batch of squid, the model predicted formaldehyde values ​​were below the target for three consecutive times (0.08 → 0.07 → 0.06 mg / kg).

[0155] PID response:

[0156] Cumulative deviation

[0157] Calculated adjustment Δu(t) = 0.8 × 0.04 + 0.16 × 0.09 = 0.032 + 0.0144 = 0.0464

[0158] Parameter adjustment:

[0159] Temperature increase: 0.0464 × 2.5 = 0.116 ° C; flow rate increase: 0.0464 × 10 = 0.464 mL / min; pH adjustment: 0.0464 × 0.5 = 0.023 pH

[0160] Effect: After adjustment, the predicted formaldehyde value in the next cycle returns to 0.09 mg / kg, avoiding the deterioration of meat quality caused by excessive processing.

[0161] 6. Comparative advantages over traditional methods

[0162]

[0163] Through quantitative control of the PID algorithm, this solution reduces energy consumption by 12% (avoiding ineffective heating) while ensuring that formaldehyde meets the standard, and improves the squid quality qualification rate to 98%.

[0164] In this embodiment, S53, adjusting the squid formaldehyde removal process parameters based on the process parameter adjustment set, further includes:

[0165] S54, collecting new solution pH, reaction temperature, processing time, and formaldehyde content, and generating a third data set including a timestamp;

[0166] S55, inputting the third data set into the prediction model to predict the real-time formaldehyde value, and determining the process stability based on the EWMA control chart; the smoothing coefficient λ of the EWMA control chart is 0.2, and the control limit is ±3 times the standard deviation;

[0167] S56. If stable, generate a final process parameter set, including a pH target value, a reaction temperature target value, and a processing time adjustment amount.

[0168] For example, the pH value, reaction temperature, processing time, and formaldehyde content of the new solution are collected to generate a third data set including a timestamp.

[0169] Example data collection record:

[0170] Timestamp pH Reaction temperature (℃) Processing time (min) Formaldehyde content (mg / L) 2025-05-14 08:00 7.2 35 45 0.12 2025-05-14 08:05 7.3 36 50 0.10 2025-05-14 08:10 7.1 34 48 0.11 2025-05-14 08:15 7.2 35 47 0.10 2025-05-14 08:20 7.3 36 50 0.09

[0171] Instructions for generating the third data set: During the squid formaldehyde removal process, the pH value, reaction temperature, processing time, and formaldehyde content of the solution are collected in real time every 5 minutes through sensors, and the corresponding timestamps are recorded to form the third data set, which provides a data basis for subsequent predictions and process stability determination.

[0172] Based on the third data set input into the prediction model, the real-time formaldehyde content prediction value is obtained as follows:

[0173] Timestamp Predicted formaldehyde content (mg / L) 2025-05-14 08:00 0.11 2025-05-14 08:05 0.09 2025-05-14 08:10 0.10 2025-05-14 08:15 0.09 2025-05-14 08:20 0.08

[0174] The smoothing coefficient λ = 0.2 is used to calculate the EWMA value at each time point according to the formula. For example, the initial EWMA value is 0.11, the first value of the predicted formaldehyde content, the EWMA at the second time point = 0.2 × 0.09 + (1-0.2) × 0.11 = 0.105, and so on.

[0175] Control limit determination: Calculate the historical standard deviation of the predicted formaldehyde content. Assuming it is 0.01, the control limits are the mean ± 3 × 0.01. Assuming the mean is 0.10, the control limits are 0.07 to 0.13.

[0176] Stable state determination: Compare each calculated EWMA value with the control limit. If all EWMA values ​​are within the control limit and there is no abnormal fluctuation, the process is determined to be stable.

[0177] Generation of the final process parameter set under stable conditions: Based on the EWMA control chart, if the EWMA values ​​calculated above are all within the control limits and the trend is stable, the final process parameter set is generated as follows:

[0178] Process parameters Target value / adjustment amount pH target value 7.2 Reaction temperature target value (℃) 35 Processing time adjustment (min) +2

[0179] The target pH and reaction temperature values ​​were determined based on the optimal parameters under stable process conditions. The process time adjustment represents an additional 2 minutes to ensure the stability and reliability of formaldehyde removal. This final set of process parameters will serve as the standard for subsequent squid formaldehyde removal processes, guiding production operations and ensuring stable and consistent product quality.

[0180] In this embodiment, S55, the third data set is input into the prediction model to predict the real-time formaldehyde value, and the process stability state is determined based on the EWMA control chart; the smoothing coefficient λ of the EWMA control chart is 0.2, and the control limits are ±3 times the standard deviation, and further includes:

[0181] If it is unstable, then: use a grid search method to adjust the value ranges of solution pH, reaction temperature, and processing time to generate a new process parameter combination and update the feature data set;

[0182] S58. Repeat the formaldehyde content prediction using the updated feature data set, and re-determine the process stability based on the EWMA control chart. If stable, output the final pH target value, reaction temperature target value, and treatment time adjustment amount to obtain the final process parameter set.

[0183] S59: If not, return to repeat the above steps.

[0184] Exemplary grid search method: pre-set the value range and step size of the solution pH (such as 6.0-9.0), reaction temperature (such as 30℃-60℃), and processing time (such as 30min-60min) to generate all possible parameter combinations. For example, the pH step size is 0.5, the reaction temperature step size is 5℃, and the processing time step size is 5min, which will generate a series of parameter combinations. Generate new process parameter combinations: for example, pH 7.0, reaction temperature 40℃, and processing time 45min. Update the feature data set: add the new process parameter combination to the feature data set for subsequent prediction and adjustment.

[0185] The updated feature data set is input into the prediction model to obtain a new formaldehyde content prediction value. The process stability state is re-determined based on the EWMA control chart: using a smoothing coefficient λ = 0.2 and control limits of ±3 times the standard deviation, the EWMA value is recalculated and the process stability state is determined. The specific calculation is as follows:

[0186] Assume that the formaldehyde content prediction value sequence corresponding to the updated feature dataset is 0.08, 0.09, 0.07, 0.08, and 0.06 (unit: mg / L). The calculated mean is 0.076 and the standard deviation is 0.011. The control limits are 0.076 ± 3 × 0.011, that is, 0.043 to 0.109. Calculate the EWMA value at each time point:

[0187] The EWMA value at the first time point is 0.08.

[0188] The EWMA value at the second time point = 0.2 × 0.09 + 0.8 × 0.08 = 0.082.

[0189] The EWMA value at the third time point = 0.2 × 0.07 + 0.8 × 0.082 = 0.0796.

[0190] The EWMA value at the fourth time point = 0.2 × 0.08 + 0.8 × 0.0796 = 0.07968.

[0191] The EWMA value at the fifth time point = 0.2 × 0.06 + 0.8 × 0.07968 = 0.07374.

[0192] If all EWMA values ​​are within the control limits and there are no abnormal fluctuations, the process is considered stable.

[0193] If it is stable, the final process parameter set is output: for example, the pH target value is 8.5, the reaction temperature target value is 50°C, and the processing time adjustment amount is +5min.

[0194] S59: If not, return to repeat the above steps.

[0195] If unstable, continue to adjust: If the process is still unstable, re-enter step S57 and continue to use the grid search method to adjust the process parameters until a suitable parameter combination is found to stabilize the process.

[0196] Example 2

[0197] See also Figure 2 , an intelligent control system for the squid formaldehyde removal process, based on the above method; the system includes:

[0198] The acquisition module is used to collect the index parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp;

[0199] The preprocessing module is used to normalize the original data set and generate a normalized process data set, which includes timestamps and normalized fields;

[0200] The data module is used to extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set;

[0201] A neural network module is used to construct a long short-term memory network model, wherein the long short-term memory network model includes a bidirectional LSTM layer; obtain feature vectors from a time series training data set, train the model to learn the temporal dependency between process parameters and formaldehyde content, and obtain a prediction model;

[0202] The prediction module is used to predict the formaldehyde content in the squid formaldehyde removal process based on the prediction model, and adjust the process parameters of the squid formaldehyde removal in real time based on the prediction results.

[0203] In the above technical solution, in order to better use the method described in one of the embodiments, the present application proposes an intelligent control system for the squid formaldehyde removal process, and each module corresponds to each step of the above method. The specific principles have been described above and will not be repeated here.

[0204] Embodiment 3

[0205] An intelligent control device for a squid formaldehyde removal process, comprising: at least one processor and a memory communicatively connected to the at least one processor;

[0206] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in one of the embodiments.

[0207] In the above technical solution, in order to better run and process the method described in one of the embodiments, the above method is stored in a memory and executed by a processor. It should be noted that the principle and effect of each step have been described above and will not be further explained here.

[0208] Example 4

[0209] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method described in one of the embodiments.

[0210] In the above technical solution, in order to better run and use the method described in one of the embodiments, the above method is stored in a computer-readable storage medium and implemented using a processor. It should be noted that the principles and effects of each step have been described above and will not be further explained here.

[0211] The above descriptions are only some embodiments of the present invention and do not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent control method for squid formaldehyde removal process, characterized in that: The method comprises: Collect the indicator parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp; Normalize the original data set to generate a normalized process data set, which includes timestamps and normalized fields; Extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set; Constructing a long short-term memory network model, the long short-term memory network model including a bidirectional LSTM layer; obtaining feature vectors from a time series training dataset, training the model to learn the temporal dependency between process parameters and formaldehyde content, and obtaining a prediction model; The formaldehyde content in the squid formaldehyde removal process is predicted based on this prediction model, and the process parameters of the squid formaldehyde removal are adjusted in real time based on the prediction results.

2. The intelligent control method for squid formaldehyde removal process according to claim 1, characterized in that: Collect the indicator parameters of the solution during the squid formaldehyde removal process and generate a raw data set with timestamps, including: The pH value, reaction temperature, processing time, and formaldehyde content of the solution during the squid formaldehyde removal process are collected by sensors at a preset frequency to generate a first data set containing a timestamp. Each record contains a timestamp, pH value, reaction temperature, processing time, and formaldehyde content. The data integrity is checked from the first data set, and records with pH values ​​outside the range of 4 to 10 and formaldehyde contents exceeding the range of ±3 times the standard deviation of the historical data mean are eliminated to generate a second data set, which is used as the original data set.

3. The intelligent control method for squid formaldehyde removal process according to claim 1, characterized in that: Extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set, including: Obtaining normalized pH, reaction temperature, and processing time from the process dataset to generate an initial feature vector, and extracting the formaldehyde content at the corresponding time point as a label to obtain a first training dataset containing a timestamp; If there are missing values ​​in the feature vector of the first training data set, the missing values ​​are interpolated using the mean filling method to generate the second training data set; The mutual information between each feature in the second training data set and the formaldehyde content is calculated using a time series feature selection algorithm, and features with a mutual information greater than 0.3 are selected to generate an optimized feature vector to obtain a time series training data set.

4. The intelligent control method for squid formaldehyde removal process according to claim 1, characterized in that: A long short-term memory network model is constructed, wherein the long short-term memory network model includes a bidirectional LSTM layer; a feature vector is obtained from a time series training data set, and a model is trained to learn the temporal dependency between process parameters and formaldehyde content to obtain a prediction model, including: Obtain time series data including process parameters and formaldehyde content from the time series training dataset, extract feature vectors through preprocessing, and obtain quantitative representation; A long short-term memory network model is used to train the model based on the feature vector to obtain an initial model; if the error of the initial model in predicting the formaldehyde content exceeds a first preset threshold, the weights of the long short-term memory layer are adjusted through back propagation, and the network model is updated until a model with an error lower than the first preset threshold is obtained, which is used as the prediction model.

5. The intelligent control method for squid formaldehyde removal process according to claim 1, characterized in that: Based on this prediction model, the formaldehyde content in the squid formaldehyde removal process is predicted, and the process parameters of the squid formaldehyde removal process are adjusted in real time based on the prediction results, including: The pH value, reaction temperature, processing time, and formaldehyde content of the solution during the formaldehyde removal process are collected in real time and input into the optimized training model to predict the formaldehyde content value; If the predicted value is higher than a second preset threshold, the process parameter adjustment amount is calculated using a proportional integral differential algorithm based on the deviation between the predicted value and the target value to generate a process parameter adjustment set, which includes a pH target value, a reaction temperature target value, and a processing time adjustment amount; Adjust the squid formaldehyde removal process parameters based on the process parameter adjustment set.

6. The intelligent control method for squid formaldehyde removal process according to claim 5, characterized in that: Adjust the squid formaldehyde removal process parameters based on the process parameter adjustment set, and then also include: Collect new solution pH, reaction temperature, processing time, and formaldehyde content to generate a third data set containing a timestamp; The third data set is input into the prediction model to predict the real-time formaldehyde value, and the process stability is determined based on the EWMA control chart; the smoothing coefficient λ of the EWMA control chart is 0.2, and the control limit is ±3 times the standard deviation; If it is stable, the final process parameter set is generated, including the pH target value, reaction temperature target value and processing time adjustment.

7. The intelligent control method for squid formaldehyde removal process according to claim 6, characterized in that: The third data set is input into the prediction model to predict the real-time formaldehyde value. The process stability is determined based on the EWMA control chart. If stable, the final process parameter set is generated, including the pH target value, the reaction temperature target value, and the processing time adjustment amount, and also includes: If unstable, then: A grid search method is used to adjust the value ranges of solution pH, reaction temperature, and processing time to generate new process parameter combinations and update the feature data set. Repeatedly predict the formaldehyde content using the updated feature data set to determine whether the predicted value is lower than the preset threshold for five consecutive times. If so, output the final pH target value, reaction temperature target value, and treatment time adjustment amount to obtain the final process parameter set; If not, return and repeat the above steps.

8. An intelligent control system for the squid formaldehyde removal process, characterized in that: Based on the method according to any one of claims 1 to 7; the system comprises: The acquisition module is used to collect the indicator parameters of the solution during the squid formaldehyde removal process and generate a raw data set including a timestamp; The preprocessing module is used to normalize the original data set and generate a normalized process data set, which includes a timestamp and normalized fields; The data module is used to extract indicator parameters from the normalized process data set as feature vectors, use the formaldehyde content at the corresponding time point as a label, intercept the sequence data according to the preset time window, and generate a time series training data set; A neural network module is used to construct a long short-term memory network model, wherein the long short-term memory network model includes a bidirectional LSTM layer; obtain feature vectors from a time series training data set, train the model to learn the temporal dependency between process parameters and formaldehyde content, and obtain a prediction model; The prediction module is used to predict the formaldehyde content in the squid formaldehyde removal process based on the prediction model, and adjust the process parameters of the squid formaldehyde removal in real time based on the prediction results.

9. An intelligent control device for the squid formaldehyde removal process, characterized in that: include: at least one processor and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images