A big data-based intelligent control system for range hood airflow
The intelligent control system for range hood airflow, which combines neural networks and optimization algorithms, solves the problems of blind airflow regulation and lag in traditional systems. It achieves precise and adaptive control of range hood airflow, improving the accuracy and stability of regulation.
Patent Information
- Application Number
- CN202611278043.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Traditional intelligent control systems for range hood airflow lack the ability to identify the real-time operation of the range hood, resulting in blind airflow regulation and lag in response. High-dimensional feature selection technology suffers from feature redundancy and limitations in noise optimization strategies, making it unable to effectively capture multi-scale temporal features. Furthermore, the methods for acquiring regulation strategies lack dynamic adaptability, leading to low accuracy, lag in response, and insufficient stability in the range hood regulation and monitoring results.
A control strategy combining neural networks and optimization algorithms is adopted. A multi-scale temporal adaptive bidirectional gated recurrent neural network is constructed, and the control optimization objective function is dynamically set. The optimal control strategy is generated through multi-source feature adaptive screening and particle swarm optimization algorithm to achieve adaptive intelligent control of the range hood's air volume.
It significantly improves the accuracy and timeliness of range hood airflow control, reduces energy consumption and noise, enhances user comfort, improves the adaptability and stability of control strategies, and achieves precise judgment and adaptive control of range hood airflow.
Smart Images

Figure CN123110643A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer control technology, specifically to an intelligent control system for the air volume of a range hood based on big data. Background Technology
[0002] A big data-based intelligent control system for range hood airflow refers to an intelligent control system that uses multi-source data acquisition and big data processing technology to perceive the actual operation of the range hood in real time during kitchen cooking, and automatically adjusts the airflow of the range hood through intelligent algorithms to achieve efficient smoke removal and energy consumption optimization. This ensures efficient smoke removal while effectively reducing energy consumption, reducing operating noise, and improving the user experience.
[0003] However, traditional intelligent airflow control systems for range hoods suffer from several problems. They rely on a single parameter of oil fume concentration to trigger fixed speed settings, lacking real-time monitoring of the range hood's operation. This results in significant blindness in airflow control, delayed response, and low operating efficiency. Furthermore, the high-dimensional feature selection techniques in traditional intelligent airflow control systems are limited by single optimization criteria and noise optimization strategies, leading to feature redundancy, residual noise features, and omission of key features. This results in low accuracy of the output results from the range hood control and monitoring model. Existing models for range hood control and monitoring also suffer from limitations in effectively capturing multi-scale temporal features of range hood operating data, weak temporal adaptability, and difficulty in fully utilizing bidirectional temporal information, leading to low accuracy and delayed response in the control and monitoring results. Finally, existing methods for acquiring control strategies suffer from a lack of dynamic adaptability in the objective function of the control optimization, resulting in poor adaptability of the acquired control strategies, large fluctuations in control effects, and insufficient stability. Summary of the Invention
[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a big data-based intelligent control system for range hood airflow. It addresses the technical problems of traditional intelligent control systems that rely on a single oil fume concentration parameter to trigger fixed speed settings, lacking real-time monitoring of the range hood's operation, resulting in significant blindness, delayed response, and low operating efficiency in airflow control. This solution innovatively combines precise monitoring of the range hood's control direction based on neural networks with dynamic generation of control strategies based on optimization algorithms, significantly improving the accuracy, timeliness, and reliability of airflow control. This addresses the shortcomings of traditional oil fume... The limitations of single optimization criteria and noise optimization strategies in high-dimensional feature selection technology for intelligent airflow control systems of range hoods easily lead to feature redundancy, residual noise features, and omission of key features, resulting in low accuracy of the output results of the range hood control and monitoring model. This solution innovatively constructs a dual-objective feature set adaptation function, integrating the two core objectives of monitoring result differentiation accuracy and feature simplification. Furthermore, it proposes an adaptive beneficial noise mechanism to dynamically adjust noise type and intensity, effectively eliminating redundant and noise features, reducing redundant load during model training, improving feature selection accuracy, and enhancing the overall performance of the range hood control and monitoring model. The reliability of the selected features ultimately improves the accuracy of the output results of the range hood control and monitoring model, enabling precise judgment of the direction of range hood airflow control. Addressing the technical problems of existing range hood control and monitoring models, such as ineffective capture of multi-scale temporal features of range hood operating data, weak temporal adaptive capability, and difficulty in fully utilizing bidirectional temporal information, leading to low accuracy and delayed response in the control and monitoring results, this solution innovatively improves the construction of the range hood control and monitoring model using a multi-scale temporal adaptive bidirectional gated recurrent neural network. This is achieved by introducing a multi-scale temporal feature extraction structure, combined with the bidirectional... The ability to capture temporal information accurately captures multi-scale temporal correlation features for judging the direction of range hood control, significantly improving the accuracy, timeliness, and stability of range hood control monitoring results. Addressing the technical problem in existing methods for acquiring control strategies—namely, the lack of dynamic adaptability in the control optimization objective function, leading to poor adaptability, large fluctuations in control effects, and insufficient stability—this solution innovatively sets the weight coefficients of each objective sub-item in the control optimization objective function based on real-time range hood control monitoring results. This achieves precise adaptation of the objective function to the real-time operating scenario, improving the adaptability and stability of the control strategy.
[0005] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent control system for range hood air volume based on big data, including a data acquisition module, a multi-source feature adaptive screening module, a range hood regulation and monitoring module, an intelligent regulation strategy generation module, and a range hood air volume intelligent control module;
[0006] The data acquisition module specifically obtains intelligent control optimization data for the range hood's airflow through data acquisition operations and data optimization processing.
[0007] The multi-source feature adaptive screening module first constructs an initial candidate feature weight vector set, then distinguishes between the dual objectives of accuracy and feature simplification by fusing monitoring results, constructs a feature set fit function, calculates the feature set fit value corresponding to each candidate weight vector, and then obtains the globally optimal feature weight vector through high-quality weight vector replication, dynamic optimization of beneficial noise based on an adaptive beneficial noise mechanism, weight set update, and iterative optimization of feature set fit at iteration termination. Finally, the optimal feature set for regulation and monitoring is screened based on the globally optimal feature weight vector to obtain the optimal feature set for regulation and monitoring.
[0008] The range hood control and monitoring module specifically involves first constructing a range hood control and monitoring model using an improved multi-scale temporal adaptive bidirectional gated recurrent neural network, then training the model, and finally performing real-time monitoring of the range hood control to obtain real-time range hood control and monitoring results.
[0009] The intelligent control strategy generation module first constructs a set of control parameter optimization boundaries based on real-time range hood control monitoring results and designs a control optimization objective function. Then, it uses a particle swarm optimization algorithm to iteratively search and optimize the control parameters to obtain the optimal control parameter combination. Finally, it outputs the optimal range hood airflow control strategy based on the optimal control parameter combination.
[0010] The intelligent airflow control module for the range hood specifically enables adaptive control and intelligent decision-making of the range hood's airflow based on real-time monitoring results and the optimal airflow control strategy.
[0011] Furthermore, the data acquisition module specifically acquires raw data for intelligent control of range hood airflow through data collection, and performs data optimization processing on the raw data to obtain optimized data for intelligent control of range hood airflow; the raw data for intelligent control of range hood airflow includes historical range hood operation scenario monitoring data and real-time range hood operation scenario monitoring data;
[0012] The historical and real-time range hood operation scenario monitoring data both include kitchen data, range hood operation data, cooking data, user interaction data, and external kitchen data; the historical range hood operation scenario monitoring data also includes historical range hood control monitoring results; the data optimization processing specifically includes data cleaning, data standardization, and data encoding.
[0013] Furthermore, the multi-source feature adaptive filtering module specifically includes the following steps:
[0014] Candidate weight vector initialization, specifically based on the total feature dimension D of the intelligent control optimization data for range hood airflow, defines the candidate feature weight vector as follows: D-dimensional continuous vectors within the interval, and N candidate feature weight vectors are randomly generated to construct an initial set of candidate feature weight vectors;
[0015] The feature set fit calculation specifically involves constructing a feature set fit function by fusing the range hood control monitoring results to differentiate between two objectives: accuracy and feature simplification. The feature set corresponding to each candidate weight vector is then quantitatively evaluated to obtain the feature set fit value. The formula used is as follows:
[0016] ;
[0017] In the formula, This represents the fitness value of the k-th candidate feature weight vector. Represented as a balance parameter, with a range of values. This is used to balance the discrimination accuracy and feature simplification of the monitoring results for range hood control. The error rate of the output result of the range hood control and monitoring model is represented by the feature set of the k-th candidate feature weight vector. This represents the weight of the j-th feature in the k-th candidate feature weight vector. Indicates the feature selection threshold. Indicates an indicator function, when it satisfies Returns 1 if the condition is met, otherwise returns 0.
[0018] Feature set fit iterative optimization specifically involves obtaining the globally optimal feature weight vector based on the initial candidate feature weight vector set through high-quality weight vector replication, dynamic optimization with beneficial noise, weight set update, and iteration termination. This includes the following steps:
[0019] The high-quality weight vector is selectively copied. Specifically, based on the fitness value of each candidate feature weight vector, the copy probability of each candidate feature weight vector is calculated through a roulette wheel selection mechanism. The copy number of each candidate feature weight vector is calculated according to the copy probability. Finally, the copy operation is performed on each candidate feature weight vector according to the copy number to generate a set of high-quality candidate feature weight copy vectors.
[0020] Beneficial noise dynamic optimization specifically involves adaptively adjusting the noise type and intensity through an adaptive beneficial noise mechanism. First, the noise intensity of the high-quality candidate feature weight replication vector is calculated. Then, the noise type is selected differentially based on its fitness value. Finally, an optimized vector of high-quality candidate feature weights is generated. The process includes the following steps:
[0021] The noise intensity is calculated specifically by calculating the fitness value of the high-quality candidate feature weight copy vector in the set of high-quality candidate feature weight copy vectors.
[0022] Differentiated noise type selection involves calculating the selection probability of Gaussian noise, Cauchy noise, and difference noise based on the fit value of the weight replication vector for each high-quality candidate feature, and then randomly generating... Random numbers are uniformly distributed across intervals. Noise types are selected based on the relationship between the random numbers and the selection probability. Finally, a d-dimensional noise vector is generated based on the selected noise type.
[0023] The generation of high-quality candidate feature weight optimization vectors involves injecting the noise vector and noise intensity into the high-quality candidate feature weight copy vector to generate an initial optimization vector. Then, abnormal weights are removed through boundary truncation to obtain the final high-quality candidate feature weight optimization vector.
[0024] The weight set is updated by merging the candidate feature weight vector set of the current iteration with the optimized high-quality candidate feature weight vector set to obtain a temporary merged set. All candidate feature weight vectors in the temporary merged set are then sorted in descending order according to their corresponding fitness values, and the top-ranked vectors are selected. The candidate feature weight vectors are used as the set of candidate feature weight vectors for the next iteration;
[0025] The iteration terminates when at least one candidate feature weight vector in the set of candidate feature weight vectors for the next iteration has a fitness value higher than the fitness threshold, or when the current iteration count reaches the maximum iteration count. In this case, the iteration terminates and the candidate feature weight vector with the highest fitness value is selected from the set of candidate feature weight vectors for the next iteration as the global optimal feature weight vector.
[0026] The optimal feature selection for regulation and monitoring is specifically based on the global optimal feature weight vector combined with the feature selection threshold. The threshold judgment is performed on each feature weight of the global optimal feature weight vector, and all features that meet the selection conditions are summarized to form the optimal feature set for regulation and monitoring.
[0027] Furthermore, the range hood control and monitoring module specifically includes the following steps:
[0028] A range hood control and monitoring model is constructed, specifically by using an improved multi-scale temporal adaptive bidirectional gated recurrent neural network; this includes the following steps:
[0029] The deep extraction of global features in time series involves first inputting the optimal feature set of regulation and monitoring, and then using a bidirectional gated recurrent neural network to extract high-dimensional global features from the dynamic time series features from both forward and reverse time series directions. This yields high-dimensional forward time series feature vectors and high-dimensional reverse time series feature vectors, which are then fused using element-level addition to obtain a high-dimensional global time series feature vector.
[0030] The multi-scale detailed feature extraction of time series is specifically achieved by constructing a two-layer nested bidirectional gated recurrent neural network. First, the high-dimensional time series global feature vector is input into the first layer of the bidirectional gated recurrent neural network to extract the short-to-medium-term dependency features of the time series data and output the intermediate-scale time series feature vector. Then, the intermediate-scale time series feature vector is input into the second layer of the bidirectional gated recurrent neural network to extract local detailed features and output the local intermediate feature vector.
[0031] Multi-scale temporal static feature fusion specifically involves fusing high-dimensional temporal global feature vectors, intermediate-scale temporal feature vectors, and local intermediate feature vectors through element-level addition, while simultaneously concatenating static features from the optimal feature set of regulation and monitoring to obtain a multi-scale fused feature vector.
[0032] The control direction is determined by first inputting a multi-scale fusion feature vector, mapping the feature dimensions through a fully connected layer, and then normalizing the output result using a softmax activation function to obtain the range hood control monitoring result.
[0033] The monitoring model training specifically involves using the optimal feature set of historical regulation and monitoring as the training data for the monitoring model to train the model and obtain the trained range hood regulation and monitoring model.
[0034] Real-time monitoring of range hood control involves inputting the optimal feature set of real-time control monitoring into the trained range hood control monitoring model to obtain the real-time range hood control monitoring results.
[0035] Furthermore, the intelligent control strategy generation module specifically includes the following steps:
[0036] Construct a set of optimization boundaries for control parameters. Specifically, based on the current real-time monitoring results of range hood control, retrieve the preset control parameter boundary table to confirm the upper and lower limits of each control parameter search, thus forming a set of parameter optimization boundaries.
[0037] The design of the regulation and optimization objective function involves taking the smoke exhaust compliance deviation, energy consumption optimization, and noise control as three sub-objectives of the regulation and optimization objective function. First, all sub-objectives are normalized to units, and then the weight coefficients of each sub-objective are set based on the real-time range hood regulation and monitoring results. Finally, the weighted sum of each sub-objective is performed to construct the regulation and optimization objective function.
[0038] The iterative search and optimization of control parameters specifically involves using a particle swarm optimization algorithm to dynamically optimize and search for the control parameters, obtaining the optimal combination of control parameters that best suits the current real-time monitoring results of the range hood; this includes the following steps:
[0039] Particle swarm initialization specifically involves using the control parameters as the individual particle position vectors in the optimization algorithm, and randomly generating M individual particle position vectors based on the parameter optimization boundary set, thereby completing the initialization of the particle swarm.
[0040] The particle fitness value is calculated by using the control optimization objective function as the particle fitness calculation function, thereby calculating the fitness value of each particle in the population.
[0041] Particle update involves first updating the particle velocity, and then updating the particle position based on the updated particle velocity.
[0042] The particle search terminates by first updating the optimal particle position, and then terminating the search and obtaining the global optimal particle position when the particle fitness value is higher than the fitness threshold or when the maximum number of iterations is reached; the global optimal particle position specifically refers to the optimal combination of control parameters.
[0043] The optimal control strategy output is specifically based on the optimal combination of control parameters, generating and outputting the optimal range hood airflow control strategy under the current real-time range hood control monitoring results.
[0044] Furthermore, the intelligent control module for range hood airflow first generates and judges an intelligent control strategy based on the real-time range hood control monitoring results. If the real-time range hood control monitoring results indicate no control or adjustment, the intelligent control strategy generation module outputs the corresponding optimal range hood airflow control strategy based on the real-time range hood control monitoring results. Finally, the dynamic control of the range hood airflow is completed according to the strategy. Otherwise, the current range hood operating parameters are maintained, realizing adaptive control and intelligent decision-making of the range hood airflow.
[0045] The beneficial effects achieved by the present invention using the above solution are as follows:
[0046] (1) In view of the technical problems in the traditional intelligent control system of range hood air volume, which relies on a single oil fume concentration parameter to trigger a fixed gear and lacks recognition of the real-time operation of the range hood, resulting in obvious blindness and lag in air volume regulation and low operating efficiency of the range hood; this solution innovatively combines the precise monitoring of the range hood regulation direction based on neural network with the dynamic generation of regulation strategy based on optimization algorithm, which significantly improves the accuracy, timeliness and reliability of range hood air volume regulation, while greatly improving the energy-saving performance and noise control effect of the range hood, effectively reducing the energy consumption and noise of the equipment, enhancing the user's comfort, and effectively avoiding the blindness and instability of air volume regulation, so as to efficiently realize the adaptive intelligent control and intelligent decision-making of range hood air volume.
[0047] (2) To address the limitations of single optimization criteria and noise optimization strategies in the high-dimensional feature selection technology of traditional range hood airflow intelligent control systems, which easily lead to feature redundancy, noise feature residue, and omission of key features, resulting in low accuracy of the output results of the range hood control and monitoring model, this solution innovatively constructs a dual-objective feature set adaptation function, integrating the two core objectives of monitoring result differentiation accuracy and feature simplification. It also proposes an adaptive beneficial noise mechanism to dynamically adjust the noise type and intensity. This avoids feature redundancy or information loss caused by traditional methods that only pursue a single index, efficiently eliminates redundant and noise features, reduces redundant load in the model training process, accurately retains key features strongly correlated with range hood control and monitoring, improves the accuracy of feature selection, avoids traditional single noise strategies from falling into local optima, improves the reliability of feature selection, and ultimately improves the accuracy of the output results of the range hood control and monitoring model, achieving precise judgment of the range hood airflow control direction.
[0048] (3) In view of the technical problems in the existing range hood control and monitoring models, which are unable to effectively capture the multi-scale time-series features of range hood operation data, have weak time-series adaptive capabilities, and are difficult to make full use of bidirectional time-series information, resulting in low accuracy and delayed response of range hood control and monitoring results, this solution innovatively improves the multi-scale time-series adaptive bidirectional gated recurrent neural network to construct the range hood control and monitoring model. By introducing a multi-scale time-series feature extraction structure and combining the bidirectional time-series information capture capability of the bidirectional gated recurrent neural network, it can accurately capture the multi-scale time-series correlation features for range hood control direction judgment, enhance the model's adaptability to data fluctuations at different time-series scales, improve the model's utilization efficiency of bidirectional time-series information, and significantly improve the accuracy, timeliness and stability of range hood control and monitoring results, ultimately achieving accurate monitoring and efficient judgment of range hood operation status and control direction.
[0049] (4) In view of the technical problem that the existing methods for obtaining control strategies lack dynamic adaptability of the control optimization objective function, resulting in poor adaptability of the obtained control strategies, large fluctuations in control effect and insufficient stability, this solution innovatively sets the weight coefficient of each objective sub-item in the control optimization objective function based on the real-time range hood control monitoring results, so as to achieve accurate adaptation of the objective function with the real-time operation scenario and improve the adaptability and stability of the control strategy. Attached Figure Description
[0050] Figure 1 A schematic diagram of a module for an intelligent control system for range hood airflow based on big data, provided for this invention;
[0051] Figure 2 This is a flowchart illustrating the multi-source feature adaptive filtering module.
[0052] Figure 3 This is a flowchart illustrating the control and monitoring module for a range hood.
[0053] Figure 4 A flowchart illustrating the process of generating intelligent control strategies;
[0054] Figure 5 A flowchart illustrating the process of constructing a range hood control and monitoring model within the range hood control and monitoring module;
[0055] 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
[0056] 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.
[0057] 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 system 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.
[0058] Example 1, see Figure 1The present invention provides an intelligent control system for range hood air volume based on big data, including a data acquisition module, a multi-source feature adaptive screening module, a range hood regulation and monitoring module, an intelligent regulation strategy generation module, and a range hood air volume intelligent control module.
[0059] The data acquisition module is used to collect the raw data required to realize the intelligent control of the range hood's air volume. Specifically, it obtains the optimized data for intelligent control of the range hood's air volume through data acquisition operations and data optimization processing, and sends the data to the multi-source feature adaptive filtering module.
[0060] The multi-source feature adaptive screening module receives data sent by the data acquisition module and is used to perform high-dimensional feature screening on the intelligent control optimization data of the range hood air volume, eliminating redundant and noisy features, and retaining key features strongly correlated with the range hood control monitoring. Specifically, it first constructs an initial set of candidate feature weight vectors, then distinguishes between the dual objectives of accuracy and feature simplification by fusing monitoring results, constructs a feature set fit function, calculates the feature set fit value corresponding to each candidate weight vector, and then obtains the globally optimal feature weight vector through high-quality weight vector replication, dynamic optimization of beneficial noise based on an adaptive beneficial noise mechanism, weight set update, and iterative optimization of feature set fit at iteration termination. Finally, it performs optimal feature screening for control monitoring based on the globally optimal feature weight vector to obtain the optimal feature set for control monitoring, and sends the data to the range hood control monitoring module.
[0061] The range hood control and monitoring module receives data sent by the multi-source feature adaptive filtering module to accurately determine the current control direction of the range hood, clarify whether the current control direction is suitable for the real-time operating status of the range hood, whether it needs to be adjusted, and if so, to adjust the direction. Specifically, it first constructs a range hood control and monitoring model using an improved multi-scale temporal adaptive bidirectional gated recurrent neural network, then trains the model, and finally performs real-time monitoring of the range hood control to obtain the real-time range hood control and monitoring results, and sends the data to the intelligent control strategy generation module and the range hood airflow intelligent control module.
[0062] The intelligent control strategy generation module receives data sent by the range hood control monitoring module. Specifically, it first constructs a set of control parameter optimization boundaries based on the real-time range hood control monitoring results and designs a control optimization objective function. Then, it uses a particle swarm optimization algorithm to iteratively search and optimize the control parameters to obtain the optimal control parameter combination. Finally, it outputs the optimal range hood airflow control strategy based on the optimal control parameter combination and sends the data to the range hood airflow intelligent control module.
[0063] The range hood airflow intelligent control module receives data from the range hood regulation and monitoring module and the intelligent regulation strategy generation module. Specifically, it realizes adaptive control and intelligent decision-making of the range hood airflow based on the real-time range hood regulation and monitoring results and the optimal range hood airflow regulation strategy.
[0064] By performing the above operations, this solution addresses the technical problems of traditional range hood airflow intelligent control systems, which rely on a single oil fume concentration parameter to trigger fixed settings and lack real-time monitoring of the range hood's operation, resulting in significant blindness, delayed response, and low operating efficiency in airflow control. This innovative solution combines precise monitoring of the range hood's control direction based on neural networks with dynamic generation of control strategies based on optimization algorithms. This significantly improves the accuracy, timeliness, and reliability of range hood airflow control, while also greatly enhancing energy efficiency and noise control. It effectively reduces equipment operating energy consumption and noise, improves user comfort, and avoids the blindness and instability of airflow control, enabling efficient adaptive intelligent control and intelligent decision-making for range hood airflow.
[0065] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, the data acquisition module collects data through sensors and the range hood data control platform to obtain raw data for intelligent control of range hood airflow, and optimizes the raw data to obtain optimized data for intelligent control of range hood airflow. The raw data for intelligent control of range hood airflow includes historical range hood operation scenario monitoring data and real-time range hood operation scenario monitoring data.
[0066] The data optimization process specifically includes data cleaning, data standardization, and data encoding.
[0067] Both the historical and real-time range hood operation monitoring data include data from inside the kitchen, range hood operation data, cooking data, user interaction data, and data from outside the kitchen. The data from inside the kitchen includes ambient temperature, humidity, smoke concentration, indoor air pressure, odor residue concentration, kitchen area, ceiling height, exhaust vent status, door and window status, and exhaust duct length. The range hood operation data includes motor power, motor speed, motor temperature, and operating time. The cooking data includes cooktop surface temperature and cooktop power setting. The user interaction data includes the user's noise tolerance threshold, user's preferred delay shutdown duration, user's energy-saving priority, and user's tolerance for cooking fumes; the kitchen external environment data includes outdoor wind pressure, outdoor wind direction, outdoor wind speed, outdoor ambient temperature, and outdoor ambient humidity; the historical range hood operation scenario monitoring data also includes historical range hood control monitoring results; the range hood control monitoring results include no adjustment, low-noise standby mode, normal smoke exhaust mode, rapid smoke exhaust mode with smoke burst, wind pressure compensation anti-backflow mode, and equipment safety speed reduction mode;
[0068] The data cleaning process specifically includes outlier removal and missing value handling. Outlier removal involves statistical distribution analysis of the original data using the interquartile range (IMR) to identify and remove abnormal data points that deviate from the normal distribution range. Missing value handling involves filling in missing values in the original data using the nearest neighbor imputation method. Data standardization is used to unify the processing of data from different dimensions and sources, eliminating differences in data units and dimensions. Specifically, it uses the minimum-max normalization (MPM) method to map the numerical data of the original data to a unified numerical range. Data encoding is used to numerically represent unstructured or categorical data. Specifically, it uses a label encoding method to encode the category fields in the original data, converting them into structured numerical values.
[0069] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The multi-source feature adaptive screening module is used to perform high-dimensional feature screening on the intelligent control optimization data of the range hood air volume, remove redundant and noisy features, and retain key features that are strongly correlated with the range hood control and monitoring, so as to provide accurate feature input for the subsequent training of the range hood control and monitoring model. Specifically, it includes the following steps:
[0070] Candidate weight vector initialization, specifically based on the total feature dimension D of the intelligent control optimization data for range hood airflow, defines the candidate feature weight vector as follows: D-dimensional continuous vectors within the interval, and N candidate feature weight vectors are randomly generated to construct an initial set of candidate feature weight vectors;
[0071] Feature set fit calculation is used to adapt the comprehensive performance of the feature sets corresponding to the quantified candidate weight vectors, serving as the core optimization objective for feature selection. Specifically, it constructs a feature set fit function by fusing the range hood control monitoring results to differentiate between two objectives: accuracy and feature simplification. The function then quantifies and evaluates the feature set corresponding to each candidate weight vector to obtain the feature set fit value. The formula used is as follows:
[0072] ;
[0073] In the formula, This represents the fitness value of the k-th candidate feature weight vector. Represented as a balance parameter, with a range of values. This is used to balance the discrimination accuracy and feature simplification of the monitoring results for range hood control. The error rate of the output result of the range hood control and monitoring model is represented by the feature set of the k-th candidate feature weight vector. This represents the weight of the j-th feature in the k-th candidate feature weight vector. Indicates the feature selection threshold. Indicates an indicator function, when it satisfies Returns 1 if the condition is met, otherwise returns 0.
[0074] Feature set fit iteration optimization is used to continuously improve the overall performance of the feature set in terms of the accuracy of distinguishing range hood control monitoring results and the simplification of features by dynamically adjusting the candidate feature weight vectors in multiple rounds. It gradually eliminates inferior feature sets with low fit, ensuring that the final selected feature set accurately matches the training requirements of the range hood control monitoring model. Simultaneously, it avoids the problems of inefficient model training and decreased generalization ability caused by redundant features. Specifically, based on the initial candidate feature weight vector set, it obtains the globally optimal feature weight vector through high-quality weight vector replication, dynamic optimization with beneficial noise, weight set update, and iteration termination. The process includes the following steps:
[0075] The optimal copying of high-quality weight vectors is used to prioritize the inheritance of high-quality feature sets based on the comprehensive performance ranking of candidate feature weight vectors. Specifically, based on the fitness value of each candidate feature weight vector, a roulette wheel selection mechanism is used to calculate the copying probability of each candidate feature weight vector, and the copying quantity of each candidate feature weight vector is calculated according to the copying probability. Finally, the copying operation is performed on each candidate feature weight vector according to the copying quantity to generate a set of high-quality candidate feature weight copy vectors. The formula used is as follows:
[0076] ;
[0077] ;
[0078] In the formula, This represents the replication probability of the k-th candidate feature weight vector. This represents the fitness value of the m-th candidate feature weight vector. This represents the number of copies of the k-th candidate feature weight vector;
[0079] Beneficial noise dynamic optimization is used to precisely adjust the copy vectors of high-quality candidate feature weights in the set of high-quality candidate feature weight copy vectors. This balances the need for global exploration of potential key features with the need for local fine-tuning of effective features, avoiding iterative optimization from getting stuck in local optima. Specifically, it adaptively adjusts the noise type and intensity through an adaptive beneficial noise mechanism. First, the noise intensity of the high-quality candidate feature weight copy vector is calculated. Then, the noise type is selected differentially based on its fitness value. Finally, the optimized vector of high-quality candidate feature weights is generated. The process includes the following steps:
[0080] The noise intensity is calculated to assign differentiated perturbation amplitudes to each high-quality candidate feature weight copy vector. This avoids excessive disruption to the high-fitness weight structure of high-fitness vectors while providing sufficient perturbation for low-fitness vectors to break through local optima. It also provides a quantitative adjustment benchmark for subsequent noise injection. Specifically, the noise intensity is calculated based on the fitness value of the high-quality candidate feature weight copy vectors in the set of high-quality candidate feature weight copy vectors. The formula used is as follows:
[0081] ;
[0082] In the formula, This represents the noise intensity of the weight copy vector of the k-th high-quality candidate feature. This represents the fit value of the weight copy vector of the k-th high-quality candidate feature;
[0083] Differentiated noise type selection is used to avoid optimization bias caused by a single noise type. Specifically, based on the fit value of the weight replication vector of each high-quality candidate feature, the selection probability of Gaussian noise, Cauchy noise, and difference noise is calculated, and then randomly generated. Uniformly distributed random numbers in an interval The noise type is selected based on the relationship between the random number and the selection probability, and finally a d-dimensional noise vector is generated based on the selected noise type.
[0084] The noise type selection is specifically as follows: If so, choose Gaussian noise. If the condition is met, then Cauchy noise should be selected; otherwise, differential noise should be selected.
[0085] The noise vector generation constraints are as follows: each element of the Gaussian noise vector follows a standard normal distribution; each element of the Cauchy noise vector is generated based on a uniformly distributed random number in [0,1]; and the differential noise vector is obtained by subtracting the k-th high-quality candidate feature weight copy vector from another different vector randomly selected from the set of high-quality candidate feature weight copy vectors.
[0086] The formula used is as follows:
[0087] ;
[0088] ;
[0089] In the formula, This represents the selection probability of Gaussian noise. This represents the selection probability of Cauchy noise. This represents the selection probability of the differential noise;
[0090] The generation of optimized vectors for high-quality candidate feature weights involves injecting differential noise into the high-quality candidate feature weight copy vector to achieve precise weight adjustment while ensuring the legitimacy of the physical meaning of the optimized weights. Specifically, the noise vector, combined with its intensity, is injected into the high-quality candidate feature weight copy vector to generate an initial optimized vector. Then, boundary truncation is used to remove outlier weights, resulting in the final optimized vector for high-quality candidate feature weights. The formula used is as follows:
[0091] ;
[0092] In the formula, This represents the optimized weight vector for the k-th high-quality candidate feature. This represents the weight copy vector of the k-th high-quality candidate feature. This represents the noise vector representing the copy vector of the weights of the k-th high-quality candidate feature. This represents the noise amplitude control parameter, used to fine-tune the overall amplitude of noise disturbance to avoid weight distortion or ineffective optimization.
[0093] The weight set is updated to maintain the stability and diversity of the candidate feature weight vector set. Specifically, the candidate feature weight vector set of the current iteration number is merged with the optimized set of high-quality candidate feature weight vectors to obtain a temporary merged set. All candidate feature weight vectors in the temporary merged set are then sorted in descending order according to their corresponding fitness values, and the top-ranked vectors are selected. The candidate feature weight vectors are used as the set of candidate feature weight vectors for the next iteration. Where t represents the current iteration number;
[0094] The iteration terminates when the set of candidate feature weight vectors for the next iteration is obtained. When at least one candidate feature weight vector has a fitness value higher than the fitness threshold, or when the current iteration number reaches the maximum iteration number, the iteration is terminated and the candidate feature weight vector with the highest fitness value is selected from the candidate feature weight vector set of the next iteration as the global optimal feature weight vector.
[0095] The optimal feature selection for regulation and monitoring is specifically based on the global optimal feature weight vector combined with the feature selection threshold. The threshold judgment is performed on each feature weight of the global optimal feature weight vector, and all features that meet the selection conditions are summarized to form the optimal feature set for regulation and monitoring.
[0096] The threshold determination is specifically as follows: Then retain the j-th feature weight in the globally optimal feature weight vector. The corresponding features, if Then the j-th feature weight in the globally optimal feature weight vector is removed. Corresponding characteristics.
[0097] By performing the above operations, this solution addresses the limitations of traditional high-dimensional feature selection techniques for intelligent range hood airflow control systems. These techniques rely on single optimization criteria and noise optimization strategies, which can lead to feature redundancy, residual noise features, and omission of key features, resulting in low accuracy of the range hood control and monitoring model's output. This solution innovatively constructs a dual-objective feature set adaptation function, integrating two core objectives: the accuracy of monitoring result differentiation and the degree of feature simplification. It also proposes an adaptive beneficial noise mechanism to dynamically adjust the noise type and intensity. This avoids feature redundancy or information loss caused by traditional methods that only pursue a single indicator. It efficiently eliminates redundant and noise features, reduces redundant load during model training, accurately retains key features strongly correlated with range hood control and monitoring, improves the accuracy of feature selection, avoids traditional single noise strategies getting trapped in local optima, and enhances the reliability of feature selection. Ultimately, this improves the accuracy of the range hood control and monitoring model's output, enabling precise judgment of the range hood's airflow control direction.
[0098] Example 4, see Figure 1 , Figure 3 and Figure 5 This embodiment is based on the above embodiment. The range hood control and monitoring module is used to accurately determine the current control direction of the range hood, and to determine whether the current control direction is suitable for the real-time operating status of the range hood, whether adjustment is needed, and if so, the adjustment direction. Specifically, it includes the following steps:
[0099] A range hood control and monitoring model is constructed, specifically by using an improved multi-scale temporal adaptive bidirectional gated recurrent neural network; this includes the following steps:
[0100] Deep extraction of temporal global features is used to accurately extract long-term dependencies of dynamic temporal features of range hoods, while capturing both forward and reverse changes in time-series data, improving adaptability to complex operating scenarios of range hoods. Specifically, the optimal feature set for regulation and monitoring is first input, and then a bidirectional gated recurrent neural network is used to extract high-dimensional global features from the dynamic temporal features from both forward and reverse temporal directions, obtaining high-dimensional forward temporal feature vectors and high-dimensional reverse temporal feature vectors respectively. These are then fused using element-level addition to obtain the high-dimensional temporal global feature vector. The formula used is as follows:
[0101] ;
[0102] In the formula, This represents the high-dimensional temporal global feature vector at time l. This represents the high-dimensional forward temporal feature vector at time l. This represents the high-dimensional reverse temporal eigenvector at time l. This represents an element-level addition fusion operation;
[0103] A multi-scale temporal detail feature layer extraction method is used to supplement the local details of high-dimensional global features, capture the short-to-medium-term fluctuations and instantaneous changes in the operation of range hoods, solve the problem that a single global feature is difficult to distinguish similar operating states, realize the gradient extraction of features, and gradually refine from global to local to improve the comprehensiveness of feature representation. Specifically, a two-layer nested bidirectional gated recurrent neural network is constructed. First, the high-dimensional temporal global feature vector is input into the first layer of the bidirectional gated recurrent neural network to extract the short-to-medium-term dependency features of the temporal data and output the intermediate-scale temporal feature vector. Then, the intermediate-scale temporal feature vector is input into the second layer of the bidirectional gated recurrent neural network to extract local detail features and output the local intermediate feature vector.
[0104] Multi-scale temporal static feature fusion is used to integrate multi-scale and multi-type features, solving the problem of incomplete representation by single-scale features. It ensures that the fused features can accurately adapt to the feature differences in the range hood control monitoring results. At the same time, the addition of static features enhances the rationality of the control direction and the pertinence of the judgment. Specifically, it fuses the high-dimensional temporal global feature vector, intermediate-scale temporal feature vector, and local intermediate feature vector by element-level addition, and concatenates the static features in the optimal feature set of control monitoring to finally obtain the multi-scale fused feature vector.
[0105] The control direction determination is specifically achieved by first inputting a multi-scale fused feature vector, then mapping the feature dimensions through a fully connected layer, and finally normalizing the output result using a softmax activation function to obtain the range hood control monitoring result; the formula used is as follows:
[0106] ;
[0107] In the formula, This represents the monitoring result of the range hood control at time l. and These represent the weight matrix and bias vector of the output layer, respectively.
[0108] The monitoring model training specifically involves using the optimal feature set of historical regulation and monitoring as the training data for the monitoring model to train the model and obtain the trained range hood regulation and monitoring model.
[0109] The optimal feature set for historical regulation and monitoring is specifically the optimal feature set obtained by processing historical range hood operation scenario monitoring data from the intelligent control optimization data of range hood air volume through a multi-source feature adaptive filtering module.
[0110] The model training specifically employs the category cross-entropy loss function and uses the backpropagation algorithm and gradient descent optimization method to iteratively update the weight matrix and bias parameters of the range hood control and monitoring model. The model parameters are continuously optimized through multiple iterations. When the preset maximum number of training iterations is reached or the loss function converges to a set threshold, the iterative training stops.
[0111] Real-time monitoring of range hood control involves inputting the optimal feature set of real-time control monitoring into the trained range hood control monitoring model to obtain the real-time range hood control monitoring results.
[0112] The optimal feature set for real-time control and monitoring is specifically obtained by processing the real-time range hood operation scenario monitoring data in the intelligent control optimization data of range hood air volume through a multi-source feature adaptive filtering module.
[0113] By performing the above operations, this solution addresses the technical problems of existing range hood control and monitoring models, which suffer from ineffective capture of multi-scale temporal features of range hood operating data, weak temporal adaptive capabilities, and difficulty in fully utilizing bidirectional temporal information, resulting in low accuracy and delayed response in range hood control and monitoring results. This solution innovatively improves the multi-scale temporal adaptive bidirectional gated recurrent neural network for constructing the range hood control and monitoring model. By introducing a multi-scale temporal feature extraction structure and combining it with the bidirectional temporal information capture capability of the bidirectional gated recurrent neural network, it can accurately capture the multi-scale temporal correlation features for judging the control direction of the range hood, enhance the model's adaptability to data fluctuations at different temporal scales, improve the model's utilization efficiency of bidirectional temporal information, and significantly improve the accuracy, timeliness, and stability of the range hood control and monitoring results. Ultimately, it achieves accurate monitoring and efficient judgment of the range hood's operating status and control direction.
[0114] Example 5, see Figure 1 and Figure 4This embodiment is based on the above embodiment. The intelligent control strategy generation module is used to output a precise range hood control strategy based on the real-time range hood control monitoring results, specifically including the following steps:
[0115] A set of optimization boundaries for control parameters is constructed to limit the reasonable optimization range for subsequent control parameter optimization algorithms, avoid ineffective optimization and control failure caused by parameter over-limit, improve the efficiency and safety of parameter optimization, and provide boundary constraints for solving the optimal control parameters; specifically, based on the current real-time range hood control monitoring results, the preset control parameter boundary table is retrieved to confirm the upper and lower limits of each control parameter search, forming a set of parameter optimization boundaries;
[0116] The control parameters are the core controllable parameters that the range hood can directly execute, including motor speed, air volume adjustment gradient, wind pressure compensation value, and motor operating power;
[0117] The preset control parameter boundary table is a pre-configured standardized parameter constraint table, which is pre-calibrated based on the factory hardware limits of the range hood, historical operation optimization data, control requirements of different operating states, and the solution range of the objective function.
[0118] A control optimization objective function is designed to quantify the comprehensive adaptability of different combinations of control parameters, clarify the core direction of parameter optimization, balance the three core control requirements of range hood smoke emission compliance, energy saving and consumption reduction, and noise control, and avoid the imbalance of other requirements caused by single-objective optimization. Specifically, the smoke emission compliance deviation, energy consumption optimization, and noise control are taken as three sub-objectives of the control optimization objective function. First, all sub-objectives are normalized to units, and weight coefficients for each sub-objective are set based on real-time range hood control monitoring results. Finally, the weighted sum of each sub-objective is performed to construct the control optimization objective function; the formula used is as follows:
[0119] ;
[0120] ;
[0121] ;
[0122] ;
[0123] In the formula, This indicates the adjustment and optimization of the objective function value. , and These are the weighting coefficients for the smoke emission compliance deviation item, energy consumption optimization item, and noise control item, respectively. , and These are respectively: smoke emission compliance deviation items, energy consumption optimization items, and noise control items. This represents the constant concentration of basic kitchen fumes from the range hood. Indicates the smoke extraction coefficient based on motor speed. Indicates the motor speed. This represents the wind pressure influence coefficient. This represents the outdoor reference wind pressure constant. Indicates the wind pressure compensation value. This indicates the target value for the concentration of cooking fumes. This represents the maximum safe threshold constant for oil fume concentration. Indicates the load correction factor. Indicates the motor's operating power. Indicates the motor's operating power coefficient. This represents the constant of the motor's rated maximum power. This represents the basic noise constant of the range hood. Indicates the current operating noise target threshold. This represents the maximum permissible threshold constant for noise.
[0124] The iterative search optimization of control parameters is used to dynamically iteratively search and optimize the core control parameters within a preset set of control parameter optimization boundaries. Specifically, it employs a particle swarm optimization algorithm to dynamically optimize and search the control parameters, obtaining the optimal combination of control parameters that best suits the current real-time range hood control monitoring results. This includes the following steps:
[0125] Particle swarm initialization specifically involves using the control parameters as the individual particle position vectors in the optimization algorithm, and randomly generating M individual particle position vectors based on the parameter optimization boundary set. This completes the initialization of the particle population;
[0126] Each individual particle represents a candidate combination of control parameters;
[0127] The particle fitness value is calculated to evaluate the overall fitness performance of each candidate combination of control parameters, clarify the superiority or inferiority of each candidate combination, provide core quantitative basis for subsequent particle position updates, and ensure that iterative optimization always moves towards a better overall fitness. Specifically, the particle fitness calculation function is based on the control optimization objective function, thereby calculating the fitness value of each particle in the population.
[0128] Particle update specifically involves first updating the particle velocity, and then updating the particle position based on the updated particle velocity; the formula used is as follows:
[0129] ;
[0130] ;
[0131] In the formula, It indicates that the i-th particle is in the... Speed in the next iteration It indicates that the i-th particle is in the... Speed in the next iteration and express Random numbers within a range It indicates that the i-th particle is in the... Position in the next iteration Indicates the first The optimal individual particle position in the next iteration. This represents the particle self-learning coefficient of the i-th particle. This represents the swarm learning coefficient of the i-th particle. Indicates the current iteration number. Indicates the first Inertia weights in the next iteration;
[0132] The particle search terminates by first updating the optimal particle position, and then terminating the search and obtaining the global optimal particle position when the particle fitness value is higher than the fitness threshold or when the maximum number of iterations is reached; the global optimal particle position specifically refers to the optimal combination of control parameters.
[0133] The particle optimal position update specifically involves re-evaluating the fitness value of all updated particles, and comparing the current particle's fitness value with the current particle's global optimal position. If the current particle's fitness value is better, then the particle's global optimal position is updated.
[0134] The optimal control strategy output is specifically based on the optimal combination of control parameters, generating and outputting the optimal range hood airflow control strategy under the current real-time range hood control monitoring results.
[0135] By performing the above operations, this solution addresses the technical problem in existing methods for obtaining control strategies that lack dynamic adaptability of the control optimization objective function, resulting in poor adaptability of the obtained control strategy, large fluctuations in control effect, and insufficient stability. This solution innovatively sets the weight coefficient of each objective sub-item in the control optimization objective function based on the real-time range hood control monitoring results, thereby achieving precise adaptation of the objective function to the real-time operating scenario and improving the adaptability and stability of the control strategy.
[0136] Example 6, see Figure 1This embodiment is based on the above embodiment. Specifically, the intelligent control module for range hood air volume first generates and judges an intelligent control strategy based on the real-time range hood control monitoring results. If the real-time range hood control monitoring results indicate no control and no adjustment, the intelligent control strategy generation module outputs the corresponding optimal range hood air volume control strategy based on the real-time range hood control monitoring results. Finally, the dynamic control of the range hood air volume is completed according to the strategy. Otherwise, the current range hood operating parameters are maintained, realizing adaptive control and intelligent decision-making of the range hood air volume, ensuring that the range hood can be in the optimal air volume operating state throughout the entire cooking cycle.
[0137] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0138] 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.
[0139] 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. A smart control system for range hood airflow based on big data, characterized in that: It includes a data acquisition module, a multi-source feature adaptive filtering module, a range hood control and monitoring module, an intelligent control strategy generation module, and a range hood airflow intelligent control module; The data acquisition module specifically obtains intelligent control optimization data for the range hood's airflow through data collection and data optimization processing. The multi-source feature adaptive screening module first constructs an initial candidate feature weight vector set, then integrates monitoring results to distinguish between two objectives: accuracy and feature simplification, constructs a feature set fit function, calculates the feature set fit value corresponding to each candidate weight vector, and then obtains the globally optimal feature weight vector through high-quality weight vector replication, dynamic optimization of beneficial noise based on an adaptive beneficial noise mechanism, weight set update, and iterative optimization of feature set fit at iteration termination. Finally, the optimal feature set for regulation and monitoring is screened based on the globally optimal feature weight vector to obtain the optimal feature set for regulation and monitoring. The range hood control and monitoring module specifically employs an improved multi-scale temporal adaptive bidirectional gated recurrent neural network to construct a range hood control and monitoring model, then trains the model, and finally performs real-time monitoring of the range hood control to obtain real-time range hood control and monitoring results. The intelligent control strategy generation module first constructs a set of control parameter optimization boundaries and designs a control optimization objective function based on the real-time range hood control monitoring results, and then uses a particle swarm optimization algorithm to iteratively search and optimize the control parameters to obtain the optimal control parameter combination. Finally, it outputs the optimal range hood airflow control strategy based on the optimal control parameter combination. The intelligent airflow control module for the range hood specifically achieves adaptive control of the range hood's airflow based on real-time monitoring results and the optimal airflow control strategy.
2. The intelligent control system for range hood airflow based on big data as described in claim 1, characterized in that: The multi-source feature adaptive filtering module specifically includes the following steps: Candidate weight vector initialization, specifically based on the total feature dimension D of the intelligent control optimization data for range hood airflow, defines the candidate feature weight vector as follows: D-dimensional continuous vectors within the interval, and N candidate feature weight vectors are randomly generated to construct an initial set of candidate feature weight vectors; The feature set fit calculation specifically involves constructing a feature set fit function by fusing the range hood control monitoring results to differentiate between two objectives: accuracy and feature simplification. The feature set corresponding to each candidate weight vector is then quantitatively evaluated to obtain the feature set fit value. The formula used is as follows: ; In the formula, This represents the fitness value of the k-th candidate feature weight vector. Represented as a balance parameter, with a range of values. This is used to balance the discrimination accuracy and feature simplification of the monitoring results for range hood control. The error rate of the output result of the range hood control and monitoring model is represented by the feature set of the k-th candidate feature weight vector. This represents the weight of the j-th feature in the k-th candidate feature weight vector. Indicates the feature selection threshold. Indicates an indicator function, when it satisfies Returns 1 if the condition is met, otherwise returns 0. The feature set fit iterative optimization is specifically based on the initial candidate feature weight vector set. It obtains the globally optimal feature weight vector through high-quality weight vector copying, beneficial noise dynamic optimization, weight set update and iteration termination. The optimal feature selection for regulation and monitoring is specifically based on the global optimal feature weight vector combined with the feature selection threshold. The threshold judgment is performed on each feature weight of the global optimal feature weight vector, and all features that meet the selection conditions are summarized to form the optimal feature set for regulation and monitoring.
3. The intelligent control system for range hood airflow based on big data according to claim 2, characterized in that: The iterative optimization of the feature set fit specifically includes the following steps: The high-quality weight vector is selectively copied. Specifically, based on the fitness value of each candidate feature weight vector, the copy probability of each candidate feature weight vector is calculated through a roulette wheel selection mechanism. The copy number of each candidate feature weight vector is calculated according to the copy probability. Finally, the copy operation is performed on each candidate feature weight vector according to the copy number to generate a set of high-quality candidate feature weight copy vectors. The beneficial noise dynamic optimization is specifically achieved by adaptively adjusting the noise type and intensity through an adaptive beneficial noise mechanism. First, the noise intensity of the high-quality candidate feature weight copy vector is calculated. Then, the noise type is selected differentially based on its fitness value. Finally, the high-quality candidate feature weight optimization vector is generated. The process includes the following steps: The noise intensity is calculated specifically by calculating the fitness value of the high-quality candidate feature weight copy vector in the set of high-quality candidate feature weight copy vectors. Differentiated noise type selection involves calculating the selection probability of Gaussian noise, Cauchy noise, and difference noise based on the fit value of the weight replication vector for each high-quality candidate feature, and then randomly generating... Random numbers are uniformly distributed across intervals. Noise types are selected based on the relationship between the random numbers and the selection probability. Finally, a d-dimensional noise vector is generated based on the selected noise type. The generation of high-quality candidate feature weight optimization vectors involves injecting the noise vector and noise intensity into the high-quality candidate feature weight copy vector to generate an initial optimization vector. Then, abnormal weights are removed through boundary truncation to obtain the final high-quality candidate feature weight optimization vector. The weight set is updated by merging the candidate feature weight vector set of the current iteration with the optimized high-quality candidate feature weight vector set to obtain a temporary merged set. All candidate feature weight vectors in the temporary merged set are then sorted in descending order according to their corresponding fitness values, and the top-ranked vectors are selected. The candidate feature weight vectors are used as the set of candidate feature weight vectors for the next iteration; The iteration terminates when at least one candidate feature weight vector in the set of candidate feature weight vectors for the next iteration has a fitness value higher than the fitness threshold, or when the current iteration count reaches the maximum iteration count. In this case, the iteration terminates and the candidate feature weight vector with the highest fitness value is selected from the set of candidate feature weight vectors for the next iteration as the globally optimal feature weight vector.
4. The intelligent control system for range hood airflow based on big data as described in claim 1, characterized in that: The range hood control and monitoring module specifically includes the following steps: A range hood control and monitoring model is constructed, specifically by using an improved multi-scale temporal adaptive bidirectional gated recurrent neural network to construct the range hood control and monitoring model. The monitoring model training specifically involves using the optimal feature set of historical regulation and monitoring as the training data for the monitoring model to train the model and obtain the trained range hood regulation and monitoring model. Real-time monitoring of range hood control involves inputting the optimal feature set of real-time control monitoring into the trained range hood control monitoring model to obtain the real-time range hood control monitoring results.
5. The intelligent control system for range hood airflow based on big data according to claim 4, characterized in that: The construction of the range hood control and monitoring model specifically includes the following steps: The deep extraction of global features in time series involves first inputting the optimal feature set of regulation and monitoring, and then using a bidirectional gated recurrent neural network to extract high-dimensional global features from the dynamic time series features from both forward and reverse time series directions. This yields high-dimensional forward time series feature vectors and high-dimensional reverse time series feature vectors, which are then fused using element-level addition to obtain a high-dimensional global time series feature vector. The multi-scale detailed feature extraction of time series is specifically achieved by constructing a two-layer nested bidirectional gated recurrent neural network. First, the high-dimensional time series global feature vector is input into the first layer of the bidirectional gated recurrent neural network to extract the short-to-medium-term dependency features of the time series data and output the intermediate-scale time series feature vector. Then, the intermediate-scale time series feature vector is input into the second layer of the bidirectional gated recurrent neural network to extract local detailed features and output the local intermediate feature vector. Multi-scale temporal static feature fusion specifically involves fusing high-dimensional temporal global feature vectors, intermediate-scale temporal feature vectors, and local intermediate feature vectors through element-level addition, while simultaneously concatenating static features from the optimal feature set of regulation and monitoring to obtain a multi-scale fused feature vector. The control direction determination is specifically achieved by first inputting a multi-scale fusion feature vector, mapping the feature dimensions through a fully connected layer, and then normalizing the output result through an activation function to obtain the range hood control monitoring result.
6. The intelligent control system for range hood airflow based on big data according to claim 1, characterized in that: The intelligent control strategy generation module specifically includes the following steps: Construct a set of optimization boundaries for control parameters. Specifically, based on the current real-time monitoring results of range hood control, retrieve the preset control parameter boundary table to confirm the upper and lower limits of each control parameter search, thus forming a set of parameter optimization boundaries. The design of the regulation and optimization objective function involves taking the smoke exhaust compliance deviation, energy consumption optimization, and noise control as three sub-objectives of the regulation and optimization objective function. First, all sub-objectives are normalized to units, and then the weight coefficients of each sub-objective are set based on the real-time range hood regulation and monitoring results. Finally, the weighted sum of each sub-objective is performed to construct the regulation and optimization objective function. The iterative search and optimization of control parameters specifically involves using a particle swarm optimization algorithm to dynamically optimize and search for the control parameters, obtaining the optimal combination of control parameters that best suits the current real-time monitoring results of the range hood; this includes the following steps: Particle swarm initialization specifically involves using the control parameters as the individual particle position vectors in the optimization algorithm, and randomly generating M individual particle position vectors based on the parameter optimization boundary set, thereby completing the initialization of the particle swarm. The particle fitness value is calculated by using the control optimization objective function as the particle fitness calculation function, thereby calculating the fitness value of each particle in the population. Particle update involves first updating the particle velocity, and then updating the particle position based on the updated particle velocity. The particle search terminates by first updating the optimal particle position, and then terminating the search and obtaining the global optimal particle position when the particle fitness value is higher than the fitness threshold or when the maximum number of iterations is reached; the global optimal particle position specifically refers to the optimal combination of control parameters. The optimal control strategy output is specifically based on the optimal combination of control parameters, generating and outputting the optimal range hood airflow control strategy under the current real-time range hood control monitoring results.
7. The intelligent control system for range hood airflow based on big data according to claim 1, characterized in that: The intelligent control module for range hood airflow first generates and judges an intelligent control strategy based on the real-time range hood control monitoring results. If the real-time range hood control monitoring results indicate no control and no adjustment, the intelligent control strategy generation module outputs the corresponding optimal range hood airflow control strategy based on the real-time range hood control monitoring results. Finally, the dynamic control of the range hood airflow is completed according to the strategy. Otherwise, the current range hood operating parameters are maintained, realizing adaptive control and intelligent decision-making of the range hood airflow.
8. The intelligent control system for range hood airflow based on big data according to claim 1, characterized in that: The data acquisition module specifically acquires raw data for intelligent control of range hood airflow through data collection, and optimizes the raw data to obtain optimized data for intelligent control of range hood airflow. The raw data for intelligent control of range hood airflow includes historical monitoring data of range hood operating scenarios and real-time monitoring data of range hood operating scenarios. The historical and real-time range hood operation scenario monitoring data both include kitchen data, range hood operation data, cooking data, user interaction data, and external kitchen data; the historical range hood operation scenario monitoring data also includes historical range hood control monitoring results; the data optimization processing specifically includes data cleaning, data standardization, and data encoding.
