An energy saving window control system based on artificial intelligence

Abstract

The application discloses an energy-saving window control system based on artificial intelligence, which comprises a data collection module, a data optimization processing module, a dynamic energy-saving strategy judgment module, an energy-saving strategy optimization module and an energy-saving window intelligent control module. The application relates to the technical field of data processing, and particularly discloses an energy-saving window control system based on artificial intelligence. The scheme innovatively proposes a dual-mode switching mechanism combining real-time judgment and switching prediction judgment, significantly reduces indoor energy consumption, and improves indoor comfort. The scheme innovatively designs three types of memory layers, i.e., a steady-state environment memory layer, a multi-factor interaction mechanism memory layer and a steady-state correction memory layer, improves the accuracy of future state prediction of the indoor environment, and improves the intelligent level of energy-saving window control. The improved grey wolf optimization algorithm with the introduction of individual search boundary dynamic adjustment and the probability-based step-by-step adjustment mechanism improves the accuracy of optimal strategy optimization and improves the overall performance of window intelligent control.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an energy-saving window control system based on artificial intelligence. Background Technology

[0002] An energy-saving window control system based on artificial intelligence refers to a system that uses artificial intelligence technology to intelligently control building windows. This system collects multi-source indoor and outdoor environmental data, processes and analyzes the data using artificial intelligence algorithms, generates corresponding window control strategies, and executes control commands through drive devices. This effectively reduces energy consumption and achieves building energy conservation and intelligent operation while ensuring indoor environmental comfort and safety.

[0003] However, traditional energy-saving window control systems generally rely solely on real-time monitoring data for strategy judgment, lacking the ability to predict future indoor environmental trends. This results in the inability to adjust control strategies in advance when facing sudden environmental changes or potential future risks, leading to technical problems such as response lag and increased energy consumption. Existing indoor environment prediction models can only model time series data, lacking the ability to suppress environmental noise and non-stationarity, and are insufficient in modeling the nonlinear interaction relationships between multiple environmental variables, leading to a decrease in the accuracy and stability of indoor environment prediction. Existing window energy-saving strategy optimization methods suffer from poor global optimization capabilities, resulting in insufficient optimization accuracy of window control parameter combinations and unsatisfactory energy-saving effects. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an energy-saving window control system based on artificial intelligence. Traditional energy-saving window control systems generally rely solely on real-time monitoring data for strategy judgment, lacking the ability to predict future indoor environmental changes. This results in a failure to adjust control strategies in advance to address sudden environmental changes or potential future risks, leading to response lag and increased energy consumption. This solution innovatively proposes a dual-mode switching mechanism combining real-time switching judgment and predictive switching judgment. It introduces energy consumption and comfort control decision objective functions as comprehensive criteria, taking into account both the current real-time state and future prediction results. While balancing energy efficiency optimization and comfort assurance, it achieves effective response to abnormal environments. This solution effectively overcomes the shortcomings of traditional methods, such as slow response, excessive energy consumption, and insufficient comfort, by providing rapid response and forward-looking adjustment to future trends. It significantly reduces indoor energy consumption, improves indoor comfort, and enhances the accuracy and stability of energy-saving window control. Addressing the technical problem that existing indoor environment prediction models can only model time series data, lack the ability to suppress environmental noise and non-stationarity, and are insufficient in modeling the nonlinear interactions between multiple environmental variables, leading to decreased accuracy and stability in indoor environment prediction, this solution innovatively designs three types of memory layers in the model: a steady-state environment memory layer, a multi-factor interaction mechanism memory layer, and a steady-state correction memory layer. The steady-state environment memory layer introduces an environmental smoothing factor and environmental baseline... A quasi-state mechanism steady-state environmental memory neural network effectively suppresses the effects of input fluctuations and non-stationarity, maintaining the stability of temporal modeling. The multi-factor interaction mechanism memory layer employs a long short-term memory structure to model the nonlinear interactions between multi-dimensional features, enhancing the ability to express multi-factor coupling relationships. The steady-state correction memory layer utilizes the steady-state environmental memory network to dynamically smooth and correct prior memory states, significantly reducing the impact of noise and accumulated errors. Through these improvements, the accuracy and stability of predicting future indoor environmental states are significantly enhanced, providing more precise environmental trend input for energy-saving window control. This allows the control system to adjust energy-saving window control parameters in advance, achieving effective energy consumption reduction and continuous assurance of indoor comfort. This approach enhances the intelligence level and overall operational efficiency of energy-saving window control. Addressing the technical issue of poor global optimization capabilities in existing window energy-saving strategy optimization methods, which leads to insufficient accuracy and unsatisfactory energy-saving effects in window control parameter combination optimization, this solution innovatively employs an improved Grey Wolf optimization algorithm that incorporates dynamic adjustment of individual search boundaries and a probability-based stepwise adjustment mechanism. This enables the window control parameter combination optimization process to possess both rapid convergence and global search capabilities, significantly improving the accuracy and stability of optimal solution finding and effectively enhancing the global robustness of energy-saving strategy optimization. Consequently, it achieves a significant reduction in indoor energy consumption and continuous assurance of comfort, thereby improving the overall performance and operational efficiency of intelligent energy-saving window control.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an energy-saving window control system based on artificial intelligence, including a data collection module, a data optimization and processing module, a dynamic energy-saving strategy judgment module, an energy-saving strategy optimization module, and an energy-saving window intelligent control module;

[0006] Specifically, the data collection module obtains raw data for energy-saving window control through data acquisition.

[0007] The data optimization processing module specifically performs data cleaning, data normalization, and data time alignment on the original energy-saving window control data to obtain optimized energy-saving window control data.

[0008] The dynamic energy-saving strategy judgment module is used to determine whether the currently executed energy-saving strategy needs to be switched based on a comprehensive consideration of the real-time status and future prediction results. Specifically, it constructs and trains an indoor environment prediction model, an indoor comfort assessment model, and an indoor energy consumption prediction model. Based on the output results of each model, it calculates the real-time decision value and the predicted decision value by combining the real-time control decision objective function and the predictive control decision objective function, and performs real-time and predictive switching judgments in sequence to form a dual-mode switching mechanism for energy-saving strategy switching judgment, and obtains the energy-saving strategy switching result.

[0009] The energy-saving strategy optimization module is used to dynamically optimize window control parameters in order to minimize indoor energy consumption and ensure comfort. Specifically, it adopts an improved gray wolf optimization algorithm and uses the predictive control decision objective function as the fitness function to iteratively optimize and search for the window control parameter combination to obtain the optimal window control parameter combination, thereby obtaining the optimal intelligent control strategy for the window.

[0010] The energy-saving window intelligent control module specifically converts the optimal intelligent control strategy of the window into executable control commands to realize intelligent control of the energy-saving window.

[0011] Furthermore, the data collection module specifically obtains energy-saving window control raw data by collecting the raw data required for energy-saving window control; the energy-saving window control raw data includes historical window control data and real-time window control data; both the historical window control data and the real-time window control data include indoor environmental data, outdoor environmental data, window execution data, indoor user data, and indoor energy consumption equipment operation data.

[0012] Furthermore, the data optimization processing module specifically includes the following steps:

[0013] Data cleaning specifically involves imputing missing values, removing outliers, and deduplicating data from the original data.

[0014] Data normalization specifically involves using linear normalization to scale the numerical range of the cleaned raw data, unifying different data values ​​into the same numerical range.

[0015] Data time alignment specifically involves slicing normalized data into time windows based on a unified timestamp, ensuring consistency of data from different sources and with different sampling frequencies on the same time scale, thus obtaining time-aligned energy-saving window control optimization data.

[0016] Furthermore, the dynamic energy-saving strategy determination module specifically includes the following steps:

[0017] Building and training an indoor environment prediction model includes the following steps:

[0018] The steady-state environment memory layer design specifically involves processing indoor environment data, outdoor environment data, and window execution data through a steady-state environment memory neural network to obtain a steady-state environmental feature vector.

[0019] The steady-state environment memory neural network is specifically an improved neural network that incorporates an environmental smoothing factor and an environmental baseline state; the formula used is as follows:

[0020] ;

[0021] ;

[0022] ;

[0023] ;

[0024] ;

[0025] In the formula, This represents the input gate value at the current moment. This represents the forgetting threshold at the current moment. This represents the output gate value at the current moment. This represents the current state value of the cell unit. This represents the environmental smoothing factor at the current moment. This represents the activation function. This indicates taking the maximum value. This indicates the current input gate calibration value. This represents the current forget gate calibration value. This represents the environmental smoothing factor of the previous time step. This represents the current environmental baseline state value. This represents the environmental baseline state at the previous moment. This represents the steady-state feature vector of the environment at the current moment;

[0026] The multi-factor interaction mechanism memory layer design specifically adopts a unidirectional long short-term memory neural network structure, takes the environmental steady-state feature vector as input, and learns from multi-dimensional environmental features such as temperature, carbon dioxide concentration and illuminance to obtain the interaction memory feature vector;

[0027] The steady-state correction memory layer design specifically involves dynamically smoothing and numerically correcting the interactive memory features in the time dimension through a steady-state environment memory neural network, resulting in a stabilized correction memory feature vector.

[0028] The key temporal feature aggregation layer design specifically involves introducing a temporal attention mechanism to perform weighted calculations on the correction memory feature vector in the time dimension, combining features from different times according to their importance to obtain the key temporal aggregated feature vector;

[0029] The environmental prediction output layer is designed by using a fully connected neural network to linearly transform the input key temporal aggregated feature vector and perform mapping operations using a nonlinear activation function to obtain the output results of the indoor environmental prediction model. The output results of the indoor environmental prediction model include indoor temperature prediction, indoor carbon dioxide concentration prediction, and indoor illuminance prediction.

[0030] Predictive model training specifically involves using indoor environment data, outdoor environment data, and window execution data from historical window control data as training data to train the indoor environment prediction model, resulting in a trained indoor environment prediction model.

[0031] The indoor comfort assessment model is constructed and trained. Specifically, the indoor comfort assessment model is constructed through a multilayer perceptron neural network, and the indoor environment data and user preference parameters in the historical window control data are used as the training data for the indoor comfort assessment model to obtain the trained indoor comfort assessment model.

[0032] The indoor energy consumption prediction model is constructed and trained. Specifically, the indoor energy consumption prediction model is constructed through a multilayer perceptron neural network, and the indoor environmental data, outdoor environmental data and indoor energy consumption equipment operation data in the historical window control data are used as the training data for the indoor energy consumption prediction model. The trained indoor energy consumption prediction model is obtained.

[0033] The energy-saving strategy switching decision-making process specifically employs a dual-mode switching mechanism consisting of real-time switching decision-making and predictive switching decision-making. Logical judgments are performed sequentially, and the decision on whether to switch the energy-saving strategy is based on the judgment results, thus obtaining the energy-saving strategy switching result. This includes the following steps:

[0034] The real-time switching judgment is as follows: First, the indoor environment data and user preference parameters in the real-time window control data are used as input data for the trained indoor comfort assessment model to generate a real-time indoor comfort assessment value. Then, the real-time indoor energy consumption value is calculated based on the indoor energy consumption equipment operation data. Finally, the real-time decision value is calculated based on the real-time control decision objective function. If the real-time decision value exceeds the preset switching decision threshold, the energy-saving strategy switching is triggered and the energy-saving strategy optimization module is entered. Otherwise, the energy-saving strategy switching prediction and judgment stage is carried out.

[0035] The switching prediction judgment process involves first inputting the indoor environment data, outdoor environment data, and window execution data from the real-time window control data into a trained indoor environment prediction model to generate an indoor environment prediction result for a future period. Then, the indoor environment prediction result and user preference parameters are used as input data for a trained indoor comfort assessment model to generate an indoor comfort prediction assessment value. Simultaneously, the indoor environment prediction result, outdoor environment prediction data, and indoor energy consumption equipment operation data are used as input data for a trained indoor energy consumption prediction model to obtain an indoor energy consumption prediction value. Finally, based on the predictive control decision objective function, a predictive decision value is calculated. If the predictive decision value exceeds a preset decision threshold, an energy-saving strategy switch is triggered in advance, and the energy-saving strategy optimization module is entered; otherwise, the current energy-saving window control strategy remains unchanged.

[0036] Furthermore, the energy-saving strategy optimization module specifically includes the following steps:

[0037] Initialize the search population by mapping the window control parameter combination to the position vector of the search individual in the gray wolf optimization algorithm, and generate the position vectors of M search individuals through a random initialization method to obtain the initial search population.

[0038] The window control parameter combination is specifically the control variables in the window execution data;

[0039] The individual fitness value is calculated by calculating the fitness value of the search individuals in the population. The predictive control decision objective function is used as the fitness function, and the generated predictive decision value is used as the fitness value of the search individuals. All search individuals in the population are sorted in descending order according to the fitness value, and the individual with the best fitness and its position, the second best fitness individual and its position, and the third best fitness individual and its position are defined.

[0040] The individual search boundary is dynamically adjusted, specifically based on the number of iterations. If an individual search boundary is not adjusted, then the individual search boundary is maintained; otherwise, the individual search boundary remains unchanged. The formula used is as follows:

[0041] ;

[0042] ;

[0043] In the formula, Indicates the first The lower boundary of the j-th dimension during iteration, Indicates the first The upper boundary of the j-th dimension during iteration, Indicates the first The lower boundary of the j-th dimension is adjusted during iteration. Indicates the first The upper boundary of the j-th dimension during iteration, This indicates a fixed reduction in the neighborhood radius parameter. This represents the position of the fitness-optimal individual in dimension j. This indicates that the threshold for adjusting the iteration boundary is... This represents the floor function. This represents the floor function;

[0044] Individual location-guided updates specifically involve updating the locations of other search individuals based on the core search individual.

[0045] Individual location probability update specifically involves updating the location of the searched individual based on a probability-based stepwise adjustment mechanism; the formula used is as follows:

[0046] ;

[0047] ;

[0048] ;

[0049] In the formula, This represents the deviation between an individual's current position and the target mean position. This represents the probability of an individual updating. This represents a randomness parameter within the range [0,1]. This represents a randomness parameter within the range [0.5, 1]. This represents the probability update of the candidate position for the i-th search individual in the j-th dimension. Indicates that the i-th search individual is in the first position. The position of the j-th dimension during iteration. This indicates the position of the second-best fit individual in dimension j. This indicates the position of the third-best fitter individual in dimension j;

[0050] Search boundary check correction, specifically, involves updating candidate positions obtained through individual location guidance. Candidate positions updated via individual position probabilities Perform boundary checks. If an individual's position exceeds the search space, truncate and correct it to the upper and lower limits of the corresponding variable to obtain the boundary correction position.

[0051] Update the position comparison, specifically by recalculating the search boundary checks and corrections. and The fitness value, and the position of the individual in the current iteration. The fitness values ​​are compared, and the position of the individual with the best fitness value is selected as the initial position of that individual in the next iteration;

[0052] The search for the optimal position of an individual involves evaluating the fitness values ​​of all individuals in the current population after each iteration. If the fitness of an individual's position is better than that of the current global optimal individual, then that individual's global optimal position is updated.

[0053] The search iteration terminates when the fitness value of the search individual is higher than the decision threshold or when the maximum number of iterations is reached, and the search is terminated and the globally optimal individual position is obtained. The globally optimal individual position specifically refers to the optimal combination of window control parameters.

[0054] The optimal window control strategy output specifically involves generating and outputting the optimal intelligent window control strategy based on the optimal combination of window control parameters.

[0055] Furthermore, the energy-saving window intelligent control module specifically generates a corresponding window control command sequence based on the optimal intelligent control strategy for the window, and finally intelligently adjusts the energy-saving window according to the window control command sequence to realize automatic control and energy efficiency optimization of the energy-saving window operation.

[0056] The beneficial effects achieved by the present invention using the above solution are as follows:

[0057] (1) In view of the fact that traditional energy-saving window control systems generally rely solely on real-time monitoring data for strategy judgment, lacking the ability to predict future indoor environmental changes, resulting in the inability to adjust control strategies in advance when facing sudden environmental changes or potential future risks, thus leading to technical problems such as response lag and increased energy consumption, this solution innovatively proposes a dual-mode switching mechanism that combines real-time switching judgment with predictive switching judgment. It also introduces energy consumption and comfort control decision objective functions as comprehensive criteria, taking into account the current real-time state and future prediction results. On the basis of balancing energy efficiency optimization and comfort assurance, it realizes rapid response to abnormal environments and forward-looking adjustment of future trends, effectively overcoming the defects of traditional methods such as response lag, excessive energy consumption and insufficient comfort, significantly reducing indoor energy consumption, improving indoor comfort, and enhancing the accuracy and stability of energy-saving window control.

[0058] (2) To address the technical problems of existing indoor environmental prediction models that can only model time series individually, lack the ability to suppress environmental noise and non-stationarity, and are insufficient in modeling the nonlinear interaction relationships between multiple environmental variables, leading to a decrease in the accuracy and stability of indoor environmental prediction, this solution innovatively designs three types of memory layers in the model: a steady-state environmental memory layer, a multi-factor interaction mechanism memory layer, and a steady-state correction memory layer. The steady-state environmental memory layer effectively suppresses the influence of input fluctuations and non-stationarity by introducing a steady-state environmental memory neural network with an environmental smoothing factor and an environmental baseline state mechanism, thus maintaining the stability of time series modeling. The multi-factor interaction mechanism... The memory layer employs a long short-term memory structure to model the nonlinear interactions between multidimensional features, enhancing the ability to express the coupling relationships of multiple factors. The steady-state correction memory layer utilizes a steady-state environmental memory network to dynamically smooth and correct the preceding memory state, significantly reducing the impact of noise and accumulated errors. Through these improvements, the accuracy and stability of predicting the future state of the indoor environment are significantly enhanced, thus providing more precise environmental trend input for energy-saving window control. This enables the control system to adjust the energy-saving window control parameters in advance, achieving effective reduction of energy consumption and continuous assurance of indoor comfort, while also improving the intelligence level and overall operating efficiency of energy-saving window control.

[0059] (3) In view of the technical problem that the existing window energy-saving strategy optimization methods have poor global optimization capabilities, resulting in insufficient optimization accuracy of window control parameter combination and unsatisfactory energy-saving effect, this solution innovatively adopts an improved gray wolf optimization algorithm that introduces dynamic adjustment of individual search boundaries and a probability-based stepwise adjustment mechanism. This makes the window control parameter combination optimization process have both fast convergence and global search capability, which not only significantly improves the accuracy and stability of finding the optimal solution, but also effectively improves the global robustness of energy-saving strategy optimization. This achieves a significant reduction in indoor energy consumption and continuous assurance of comfort, and improves the overall performance and operating efficiency of intelligent control of energy-saving windows. Attached Figure Description

[0060] Figure 1 A schematic diagram of a module for an energy-saving window control system based on artificial intelligence provided by the present invention;

[0061] Figure 2 A flowchart illustrating the dynamic energy-saving strategy judgment module;

[0062] Figure 3 A flowchart illustrating the energy-saving strategy optimization module;

[0063] Figure 4 A flowchart illustrating the process of building and training an indoor environment prediction model in the dynamic energy-saving strategy judgment module;

[0064] 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

[0065] 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.

[0066] 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.

[0067] Example 1, see Figure 1 The present invention provides an energy-saving window control system based on artificial intelligence, including a data collection module, a data optimization and processing module, a dynamic energy-saving strategy judgment module, an energy-saving strategy optimization module, and an energy-saving window intelligent control module;

[0068] The data collection module is used to collect multi-source raw data required for energy-saving window control. Specifically, it obtains raw data for energy-saving window control through data collection and sends the data to the data optimization and processing module.

[0069] The data optimization processing module receives data sent by the data collection module to improve the accuracy and consistency of the original data. Specifically, it performs data cleaning, data normalization, and data time alignment on the original data of energy-saving window control to obtain optimized data of energy-saving window control, and sends the data to the dynamic energy-saving strategy judgment module and the energy-saving strategy optimization module.

[0070] The dynamic energy-saving strategy judgment module receives data sent by the data optimization processing module and is used to determine whether the currently executed energy-saving strategy needs to be switched based on a comprehensive consideration of the real-time status and future prediction results. Specifically, it constructs and trains an indoor environment prediction model, an indoor comfort assessment model, and an indoor energy consumption prediction model. Based on the output results of each model, it calculates the real-time decision value and the predicted decision value by combining the real-time control decision objective function and the predictive control decision objective function, and performs real-time switching judgment and switching prediction judgment in sequence to form a dual-mode switching mechanism for energy-saving strategy switching judgment, obtains the energy-saving strategy switching result, and sends the data to the dynamic energy-saving strategy judgment module.

[0071] The energy-saving strategy optimization module receives data sent by the dynamic energy-saving strategy judgment module and the data optimization processing module, and uses it to dynamically optimize the window control parameters in order to minimize indoor energy consumption and ensure comfort. Specifically, it adopts an improved gray wolf optimization algorithm and uses the predictive control decision objective function as the fitness function to iteratively optimize and search for the window control parameter combination to obtain the optimal window control parameter combination, thereby obtaining the optimal intelligent control strategy for the window, and sends the data to the energy-saving window intelligent control module.

[0072] The energy-saving window intelligent control module receives data sent by the energy-saving strategy optimization module, specifically converting it into executable control commands based on the window's optimal intelligent control strategy, thereby realizing intelligent control of the energy-saving window.

[0073] Example 2, see Figure 1This embodiment is based on the above embodiment. Specifically, the data collection module collects the raw data required for energy-saving window control from multi-source environmental sensors and user behavior management platforms deployed indoors and outdoors, thus obtaining energy-saving window control raw data. The energy-saving window control raw data includes historical window control data and real-time window control data. Both historical and real-time window control data include indoor environmental data, outdoor environmental data, window execution data, indoor user data, and indoor energy-consuming equipment operation data. The indoor environmental data includes indoor temperature, indoor carbon dioxide concentration, and indoor illuminance. The outdoor environmental data includes outdoor temperature, outdoor humidity, wind speed, wind direction, outdoor illuminance, and outdoor rainfall. The window execution data includes window opening degree, window glass transmittance, window opening degree execution time, and window glass transmittance execution time. The window opening degree controls the air exchange volume, affecting indoor carbon dioxide concentration and temperature changes, and determining the degree to which natural ventilation replaces mechanical ventilation. The window glass transmittance determines the amount of natural light and solar heat entering the room, affecting indoor illuminance and air conditioning energy consumption. The indoor energy consumption equipment operation data includes indoor air conditioning operating power, indoor lighting power, and indoor ventilation equipment power. The indoor user data includes temperature preference settings, illuminance preference settings, ventilation preference settings, number of people in the room, and median age of people in the room. The historical window control data includes user indoor comfort scores.

[0074] Example 3, see Figure 1 This embodiment is based on the above embodiment. The data optimization processing module is used to improve the accuracy and consistency of the original data. Specifically, it performs data cleaning, data normalization, and data time alignment on the original data of energy-saving window control to obtain optimized data for energy-saving window control. This includes the following steps:

[0075] Data cleaning is used to remove outliers and noisy data from the original data of energy-saving window control and to repair missing data. Specifically, it involves filling in missing values, removing outliers, and deduplicating data in the original data.

[0076] The missing value processing specifically involves filling in the missing values ​​in the original data using the nearest neighbor interpolation method;

[0077] The removal of outliers specifically involves detecting and eliminating unreasonable data in the original data using a box plot detection method.

[0078] The deduplication process specifically involves using a data primary key comparison method to detect and delete redundant data records generated due to multiple data collections, communication errors, or duplicate storage.

[0079] Data normalization is used to eliminate different units and numerical ranges; specifically, it involves using linear normalization to scale the numerical range of the cleaned raw data, unifying different data values ​​into the same numerical range.

[0080] Data time alignment is used to solve the problem of data being out of sync in the time dimension. Specifically, it involves slicing normalized data into time windows based on a unified timestamp, so that data from different sources and with different sampling frequencies can maintain consistency on the same time scale, resulting in time-aligned energy-saving window control optimization data.

[0081] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. The dynamic energy-saving strategy judgment module is used to determine whether to switch the currently executed energy-saving strategy based on a comprehensive consideration of the real-time status and future prediction results. This enables a dual-mode switching judgment that can both quickly respond to the current abnormal status and prevent potential future risks, thereby optimizing energy consumption while ensuring indoor comfort. Specifically, it includes the following steps:

[0082] Constructing and training an indoor environment prediction model to build a predictive model that can reflect the changing patterns of the indoor environment, specifically including the following steps:

[0083] The steady-state environment memory layer is designed to achieve numerical stability in time series modeling when the input data is fluctuating, noisy, and non-stationary. Specifically, it uses a steady-state environment memory neural network to process indoor environment data, outdoor environment data, and window execution data to obtain a steady-state feature vector that conforms to the indoor environment change pattern.

[0084] The steady-state environment memory neural network is specifically an improved neural network that incorporates an environmental smoothing factor and an environmental baseline state.

[0085] The formula used is as follows:

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] In the formula, This represents the input gate value at the current moment. This represents the forgetting threshold at the current moment. This represents the output gate value at the current moment. This represents the current state value of the candidate cell. This represents the current state value of the cell unit. This indicates the state of the cell unit at the previous moment. This represents the environmental smoothing factor at the current moment. This represents the activation function. This indicates taking the maximum value. This indicates the current input gate calibration value. This represents the current forget gate calibration value. This represents the environmental smoothing factor of the previous time step. This represents the current environmental baseline state value. This represents the environmental baseline state at the previous moment. This represents the steady-state feature vector of the environment at the current moment;

[0097] This represents the data input at the current moment. , , and These represent the weight matrices from the input data to the input gate, forget gate, output gate, and candidate unit, respectively, used to map the input data to the corresponding gated units. , , and These represent the weight matrices from the hidden state to the input gate, forget gate, output gate, and candidate unit, respectively, used to incorporate historical information into the calculation of the current gated unit. , , and These represent the bias vectors for the input gate, forget gate, output gate, and candidate unit, respectively. This represents the steady-state feature vector of the environment at the previous moment;

[0098] A multi-factor interaction mechanism memory layer is designed to model the nonlinear interaction mechanism between different environmental factors in indoor environmental prediction. Specifically, a unidirectional long short-term memory neural network structure is adopted, taking the environmental steady-state feature vector as input. By learning from multi-dimensional environmental features such as temperature, carbon dioxide concentration, and illuminance, an interaction memory feature vector representing the multi-factor interaction characteristics of the environment is obtained. The formula used is as follows:

[0099] ;

[0100] In the formula, This represents the interactive memory feature vector at the current moment. This represents a unit function in a unidirectional long short-term memory neural network. Represents the interactive memory feature vector of the previous time step;

[0101] A steady-state correction memory layer is designed to suppress and correct accumulated errors and noise in prior memory states during indoor environment prediction. Specifically, it uses a steady-state environmental memory neural network to dynamically smooth and numerically correct interactive memory features over time, resulting in a stabilized correction memory feature vector. The formula used is as follows:

[0102] ;

[0103] In the formula, This represents the corrected memory feature vector at the current moment. This represents the steady-state long short-term memory neural network unit function. This represents the corrected memory feature vector from the previous time step;

[0104] A key temporal feature aggregation layer is designed to aggregate and filter feature information at different time scales during indoor environment prediction, highlighting key temporal features that have a significant impact on future environmental changes. Specifically, by introducing a temporal attention mechanism, the correction memory feature vector is weighted along the time dimension, and features from different times are combined according to their importance to obtain a key temporal aggregated feature vector that can centrally reflect the main environmental evolution trends. The formula used is as follows:

[0105] ;

[0106] ;

[0107] In the formula, Represents the key time-series aggregated feature vector. Represents the total number of time steps. , and These represent the query projection matrix, key projection matrix, and value projection matrix, respectively. Indicates the scaling factor. This indicates the importance of the corrected memory features at time t in the final aggregation. This indicates the transpose operation of the key projection matrix;

[0108] The environmental prediction output layer is designed to map key time-series aggregated feature vectors to future predicted values ​​of the indoor environment. Specifically, it uses a fully connected neural network to perform a linear transformation on the input key time-series aggregated feature vectors and combines this with a nonlinear activation function for mapping operations to obtain the output results of the indoor environment prediction model. The output results of the indoor environment prediction model include predicted values ​​for indoor temperature, indoor carbon dioxide concentration, and indoor illuminance. The formulas used are as follows:

[0109] ;

[0110] In the formula, This indicates the indoor environment prediction results. This represents the weight matrix of the corresponding output. Indicates the bias term parameters of the output layer;

[0111] Predictive model training is used to optimize the parameters of the indoor environment prediction model. Specifically, indoor environment data, outdoor environment data, and window execution data from historical window control data are used as training data to train the indoor environment prediction model, resulting in a trained indoor environment prediction model. This enables high-precision prediction of future indoor temperature, carbon dioxide concentration, and illuminance.

[0112] The model training specifically involves using mean squared error as the loss function to calculate the difference between the predicted result and the true value, and using backpropagation algorithm and gradient descent optimization method to iteratively update the model's weight matrix and bias parameters. During the training process, 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.

[0113] An indoor comfort assessment model is constructed and trained to obtain an assessment model that can reflect the comfort of the indoor environment under the current window control strategy. Specifically, an indoor comfort assessment model is constructed by using a multilayer perceptron neural network, and indoor environmental data and user preference parameters from historical window control data are used as training data for the indoor comfort assessment model to obtain the trained indoor comfort assessment model.

[0114] An indoor energy consumption prediction model is constructed and trained to obtain a prediction model that can reflect the indoor energy consumption level under the current window control strategy. Specifically, an indoor energy consumption prediction model is constructed through a multilayer perceptron neural network, and indoor environmental data, outdoor environmental data and indoor energy consumption equipment operation data in the historical window control data are used as training data for the indoor energy consumption prediction model to obtain the trained indoor energy consumption prediction model.

[0115] The energy-saving strategy switching decision-making process specifically employs a dual-mode switching mechanism consisting of real-time switching decision-making and predictive switching decision-making. Logical judgments are performed sequentially, and the decision on whether to switch the energy-saving strategy is based on the judgment results, thus obtaining the energy-saving strategy switching result. This includes the following steps:

[0116] The real-time switching judgment is used to determine whether the current control strategy needs to be switched based on real-time data. Specifically, it first uses indoor environmental data and user preference parameters from real-time window control data as input data for a trained indoor comfort assessment model to generate a real-time indoor comfort assessment value. Then, it calculates the real-time indoor energy consumption value based on the operating data of indoor energy-consuming equipment and calculates the real-time decision value based on the real-time control decision objective function. If the real-time decision value exceeds the preset switching decision threshold, it triggers the energy-saving strategy switch and enters the energy-saving strategy optimization module; otherwise, it performs the energy-saving strategy switch prediction and judgment step. The real-time control decision objective function is specifically constructed by weighting and summing the real-time indoor comfort assessment value and the real-time indoor energy consumption value according to preset weights.

[0117] The formula used is as follows:

[0118] ;

[0119] ;

[0120] In the formula, Indicates the real-time decision value. This represents the real-time energy consumption prediction value. This represents a real-time assessment value of indoor comfort. Indicates the time of the i-th energy-consuming device Operating power Indicates the sampling time interval. and These are the weighting parameters for the indoor comfort assessment value and the indoor energy consumption value, respectively.

[0121] The switching prediction judgment is used to determine in advance whether an energy-saving strategy switch is needed based on future changes in the indoor environment, thereby achieving a proactive adjustment of the control strategy. Specifically, firstly, indoor environment data, outdoor environment data, and window execution data from real-time window control data are input into a trained indoor environment prediction model to generate an indoor environment prediction result for a future period. Then, the indoor environment prediction result and user preference parameters are used as input data for a trained indoor comfort assessment model to generate an indoor comfort prediction assessment value. Simultaneously, the indoor environment prediction result, outdoor environment prediction data, and indoor energy-consuming equipment operation data are used as input data for a trained indoor energy consumption prediction model to obtain an indoor energy consumption prediction value. Finally, a prediction decision value is calculated based on a predictive control decision objective function. If the prediction decision value exceeds a preset decision threshold, an energy-saving strategy switch is triggered in advance, and the energy-saving strategy optimization module is entered; otherwise, the current energy-saving window control strategy remains unchanged. The predictive control decision objective function is specifically constructed by weighting and summing the indoor comfort prediction assessment value and the indoor energy consumption prediction value according to preset weights.

[0122] The formula used is as follows:

[0123] ;

[0124] In the formula, Indicates the predicted decision value. This represents the predicted indoor energy consumption value. This represents the predicted assessment value for indoor comfort.

[0125] The outdoor environmental forecast data is specifically obtained through a meteorological forecasting system.

[0126] By performing the above operations, this solution addresses the technical problems of traditional energy-saving window control systems, which generally rely solely on real-time monitoring data for strategy judgment and lack the ability to predict future indoor environmental changes. This results in the inability to adjust control strategies in advance when facing sudden environmental changes or potential future risks, leading to response lag and increased energy consumption. This solution innovatively proposes a dual-mode switching mechanism that combines real-time switching judgment with predictive switching judgment. It introduces energy consumption and comfort control decision objective functions as comprehensive criteria, comprehensively considering the current real-time state and future prediction results. While balancing energy efficiency optimization and comfort assurance, it achieves rapid response to abnormal environments and forward-looking adjustment of future trends, effectively overcoming the shortcomings of traditional methods such as response lag, excessive energy consumption, and insufficient comfort. This significantly reduces indoor energy consumption, improves indoor comfort, and enhances the accuracy and stability of energy-saving window control. Furthermore, existing indoor environment prediction models can only model time series data, lacking the ability to suppress environmental noise and non-stationarity, and are inadequate for modeling the nonlinear interaction relationships between multiple environmental variables. To address the technical problem of decreased accuracy and stability in indoor environmental prediction, this solution innovatively incorporates three types of memory layers into the model: a steady-state environmental memory layer, a multi-factor interaction mechanism memory layer, and a steady-state correction memory layer. The steady-state environmental memory layer, through the introduction of an environmental smoothing factor and an environmental baseline state mechanism-based steady-state memory neural network, effectively suppresses the effects of input fluctuations and non-stationarity, maintaining the stability of temporal modeling. The multi-factor interaction mechanism memory layer employs a long short-term memory structure to model the nonlinear interactions between multi-dimensional features, enhancing the ability to express the coupling relationships of multiple factors. The steady-state correction memory layer utilizes the steady-state environmental memory network to dynamically smooth and correct the preceding memory state, significantly reducing the impact of noise and accumulated errors. Through these improvements, the accuracy and stability of predicting future indoor environmental states are significantly enhanced, providing more precise environmental trend input for energy-saving window control. This allows the control system to adjust energy-saving window control parameters in advance, achieving effective energy consumption reduction and continuous assurance of indoor comfort, while also improving the intelligence level and overall operating efficiency of energy-saving window control.

[0127] Example 5, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The energy-saving strategy optimization module is used to optimize window control parameters to minimize indoor energy consumption. Specifically, an improved gray wolf optimization algorithm is used to dynamically optimize and search for the combination of window control parameters to obtain the optimal combination of window control parameters. The steps include:

[0128] Initialize the search population by mapping the window control parameter combination to the position vector of the search individual in the gray wolf optimization algorithm, and generate the position vectors of M search individuals through a random initialization method to obtain the initial search population.

[0129] The window control parameter combination specifically refers to the control variables in the window execution data, including window opening degree, window glass transmittance, window opening execution time, and window glass transmittance execution time.

[0130] The individual fitness value is calculated by calculating the fitness value of the search individuals in the population. The predictive control decision objective function is used as the fitness function, and the generated predictive decision value is used as the fitness value of the search individuals. All search individuals in the population are sorted in descending order according to the fitness value, and the individual with the best fitness and its position, the second best fitness individual and its position, and the third best fitness individual and its position are defined.

[0131] Specifically, the generated prediction decision value involves converting the location vector of the searched individual into a combination of window control parameters, inputting it along with the current indoor and outdoor environmental data into a trained indoor environment prediction model to generate an indoor environment prediction result for a future period. Then, the indoor environment prediction result and user preference parameters are used as input data for a trained indoor comfort assessment model to generate an indoor comfort prediction assessment value. Simultaneously, the indoor environment prediction result, outdoor environment prediction data, and indoor energy consumption equipment operation data are used as input data for a trained indoor energy consumption prediction model to obtain an indoor energy consumption prediction value. Finally, the prediction decision value is calculated based on the prediction control decision objective function.

[0132] The individual search boundary is dynamically adjusted to focus on the optimal individual in the later stages of iteration, accelerating convergence and achieving steady-state search. Specifically, this adjustment is made based on the number of iterations. If necessary, adjust the individual search boundary; otherwise, keep the individual search boundary unchanged.

[0133] The original range of the individual search boundary is determined by the physical constraints of the controlled object and the preset operational requirements.

[0134] The formula used is as follows:

[0135] ;

[0136] ;

[0137] In the formula, Indicates the first The lower boundary of the j-th dimension during iteration, Indicates the first The upper boundary of the j-th dimension during iteration, Indicates the first The lower boundary of the j-th dimension is adjusted during iteration. Indicates the first The upper boundary of the j-th dimension during iteration, This indicates a fixed reduction in the neighborhood radius parameter. This represents the position of the fitness-optimal individual in dimension j. This indicates that the threshold for adjusting the iteration boundary is... This represents the floor function. This represents the floor function;

[0138] Individual position-guided updates; specifically, updating the positions of other search individuals based on the core search individual; the formula used is as follows:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] In the formula, Indicates the i-th search individual's direction The new location Indicates the i-th search individual's direction The new location Indicates the i-th search individual's direction The new location Indicates that the i-th search individual is in the first position. The position of the j-th dimension during iteration. This represents the guided update candidate position of the i-th search individual in the j-th dimension. , and This represents a randomness parameter within the range [0,1]. This represents the individual with the best fitness. This indicates an individual with the second-best fitness. This indicates the individual with the third best fitness. This indicates the position of the second-best fit individual in dimension j. This indicates the position of the third-best fitter individual in dimension j. This represents the convergence factor of the current iteration;

[0144] Individual location probability update specifically involves updating the location of the searched individual based on a probability-based stepwise adjustment mechanism; the formula used is as follows:

[0145] ;

[0146] ;

[0147] ;

[0148] In the formula, This represents the deviation between an individual's current position and the target mean position. This represents the probability of an individual updating. This represents a randomness parameter within the range [0,1]. This represents a randomness parameter within the range [0.5, 1]. This represents the probability update of the candidate position for the i-th search individual in the j-th dimension;

[0149] Search boundary check correction, specifically, involves updating candidate positions obtained through individual location guidance. Candidate positions updated via individual position probabilities Boundary checks are performed. If an individual's position exceeds the search space, it is truncated and corrected to the upper and lower limits of the corresponding variable to obtain the boundary correction position. The formula used is as follows:

[0150] ;

[0151] In the formula, This indicates the guided update correction position of the i-th search individual in the j-th dimension. This indicates the probability update and correction position of the i-th search individual in the j-th dimension;

[0152] Update the position comparison, specifically by recalculating the search boundary checks and corrections. and The fitness value, and the position of the individual in the current iteration. Fitness values ​​are compared, and the position of the individual with the best fitness value is selected as the initial position of that individual in the next iteration; the formula used is as follows:

[0153] ;

[0154] In the formula, Indicates that the i-th search individual is in the first position. The position of the j-th dimension in the iterative population. Represents the fitness function;

[0155] The search for the optimal position of an individual involves evaluating the fitness values ​​of all individuals in the current population after each iteration. If the fitness of an individual's position is better than that of the current global optimal individual, then that individual's global optimal position is updated.

[0156] The search iteration terminates when the fitness value of the search individual is higher than the decision threshold or when the maximum number of iterations is reached, and the search is terminated and the globally optimal individual position is obtained. The globally optimal individual position specifically refers to the optimal combination of window control parameters.

[0157] The optimal window control strategy output is used to transform the parameter combination into an actually executable control scheme. Specifically, it generates and outputs the optimal intelligent control strategy for the window based on the optimal window control parameter combination.

[0158] By performing the above operations, this solution addresses the technical problem of poor global optimization capability in existing window energy-saving strategy optimization methods, which leads to insufficient optimization accuracy and unsatisfactory energy-saving effects in window control parameter combination optimization. It innovatively adopts an improved Grey Wolf optimization algorithm that introduces dynamic adjustment of individual search boundaries and a probability-based stepwise adjustment mechanism. This enables the window control parameter combination optimization process to possess both rapid convergence and global search capability, significantly improving the accuracy and stability of optimal solution finding and effectively enhancing the global robustness of energy-saving strategy optimization. This results in a significant reduction in indoor energy consumption and continuous assurance of comfort, improving the overall performance and operational efficiency of intelligent control of energy-saving windows.

[0159] Example 6, see Figure 1 This embodiment is based on the above embodiment. Specifically, the energy-saving window intelligent control module generates a corresponding window control command sequence according to the optimal intelligent control strategy of the window, and finally intelligently adjusts the energy-saving window according to the window control command sequence to realize automatic control and energy efficiency optimization of the energy-saving window operation.

[0160] 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.

[0161] 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.

[0162] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An energy-saving window control system based on artificial intelligence, characterized in that: It includes a data collection module, a data optimization and processing module, a dynamic energy-saving strategy judgment module, an energy-saving strategy optimization module, and an intelligent control module for energy-saving windows; Specifically, the data collection module obtains raw data for energy-saving window control through data acquisition. The data optimization processing module specifically performs data cleaning, data normalization, and data time alignment on the original energy-saving window control data to obtain optimized energy-saving window control data. The dynamic energy-saving strategy judgment module specifically constructs and trains an indoor environment prediction model, an indoor comfort assessment model, and an indoor energy consumption prediction model. It calculates real-time decision values ​​and predicted decision values ​​by combining the output results of each model with the real-time control decision objective function and the predictive control decision objective function, and performs sequential switching real-time judgment and switching predictive judgment to form a dual-mode switching mechanism for energy-saving strategy switching judgment, thereby obtaining the energy-saving strategy switching result. The construction and training of the indoor environment prediction model includes the design of a steady-state environment memory layer, a multi-factor interaction mechanism memory layer, a steady-state correction memory layer, a key temporal feature aggregation layer, an environment prediction output layer, and prediction model training. The energy-saving strategy optimization module specifically employs an improved gray wolf optimization algorithm that introduces dynamic adjustment of individual search boundaries and a probability-based stepwise adjustment mechanism. It uses the predictive control decision objective function as the fitness function to iteratively optimize and search for the window control parameter combination, thereby obtaining the optimal window control parameter combination and thus the optimal intelligent control strategy for the window. The energy-saving window intelligent control module specifically converts the optimal intelligent control strategy of the window into executable control commands to realize intelligent control of the energy-saving window.

2. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The dynamic energy-saving strategy judgment module specifically... Includes the following steps: Build and train an indoor environment prediction model; The indoor comfort assessment model is constructed and trained. Specifically, the indoor comfort assessment model is constructed through a multilayer perceptron neural network, and the indoor environment data and user preference parameters in the historical window control data are used as the training data for the indoor comfort assessment model to obtain the trained indoor comfort assessment model. The indoor energy consumption prediction model is constructed and trained. Specifically, the indoor energy consumption prediction model is constructed through a multilayer perceptron neural network, and the indoor environmental data, outdoor environmental data and indoor energy consumption equipment operation data in the historical window control data are used as the training data for the indoor energy consumption prediction model. The trained indoor energy consumption prediction model is obtained. The energy-saving strategy switching judgment is specifically composed of a dual-mode switching mechanism consisting of real-time switching judgment and switching prediction judgment. Logical judgments are performed sequentially, and the energy-saving strategy switching is determined based on the judgment results to obtain the energy-saving strategy switching result.

3. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The construction and training of the indoor environment prediction model specifically includes the following steps: The steady-state environment memory layer design specifically involves processing indoor environment data, outdoor environment data, and window execution data through a steady-state environment memory neural network to obtain a steady-state environmental feature vector. The multi-factor interaction mechanism memory layer design specifically adopts a unidirectional long short-term memory neural network structure, takes the environmental steady-state feature vector as input, and learns from multi-dimensional environmental features such as temperature, carbon dioxide concentration and illuminance to obtain the interaction memory feature vector; The steady-state correction memory layer design specifically involves dynamically smoothing and numerically correcting the interactive memory features in the time dimension through a steady-state environment memory neural network, resulting in a stabilized correction memory feature vector. The key temporal feature aggregation layer design specifically involves introducing a temporal attention mechanism to perform weighted calculations on the correction memory feature vector in the time dimension, combining features from different times according to their importance to obtain the key temporal aggregated feature vector; The environmental prediction output layer is designed by using a fully connected neural network to linearly transform the input key temporal aggregated feature vector and perform mapping operations using a nonlinear activation function to obtain the output results of the indoor environmental prediction model. The output results of the indoor environmental prediction model include indoor temperature prediction, indoor carbon dioxide concentration prediction, and indoor illuminance prediction. Predictive model training involves using indoor environment data, outdoor environment data, and window execution data from historical window control data as training data to train the indoor environment prediction model, resulting in a trained indoor environment prediction model.

4. The energy-saving window control system based on artificial intelligence according to claim 3, characterized in that: The steady-state environment memory neural network is specifically an improved neural network that incorporates an environmental smoothing factor and an environmental baseline state; the formula used is as follows: ； ； ； ； ； In the formula, This represents the input gate value at the current moment. This represents the forgetting threshold at the current moment. This represents the output gate value at the current moment. This represents the current state value of the cell unit. This represents the environmental smoothing factor at the current moment. This represents the activation function. This indicates taking the maximum value. This indicates the current input gate calibration value. This represents the current forget gate calibration value. This represents the environmental smoothing factor of the previous time step. This represents the current environmental baseline state value. This represents the environmental baseline state at the previous moment. This represents the steady-state feature vector of the environment at the current moment.

5. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The energy-saving strategy switching judgment specifically includes the following steps: The real-time switching judgment is as follows: First, the indoor environment data and user preference parameters in the real-time window control data are used as input data for the trained indoor comfort assessment model to generate a real-time indoor comfort assessment value. Then, the real-time indoor energy consumption value is calculated based on the indoor energy consumption equipment operation data. Finally, the real-time decision value is calculated based on the real-time control decision objective function. If the real-time decision value exceeds the preset switching decision threshold, the energy-saving strategy switching is triggered and the energy-saving strategy optimization module is entered. Otherwise, the energy-saving strategy switching prediction and judgment stage is carried out. The switching prediction judgment process involves first inputting the indoor environment data, outdoor environment data, and window execution data from the real-time window control data into a trained indoor environment prediction model to generate an indoor environment prediction result for a future period. Then, the indoor environment prediction result and user preference parameters are used as input data for a trained indoor comfort assessment model to generate an indoor comfort prediction assessment value. Simultaneously, the indoor environment prediction result, outdoor environment prediction data, and indoor energy consumption equipment operation data are used as input data for a trained indoor energy consumption prediction model to obtain an indoor energy consumption prediction value. Finally, based on the predictive control decision objective function, a predictive decision value is calculated. If the predictive decision value exceeds a preset decision threshold, an energy-saving strategy switch is triggered in advance, and the energy-saving strategy optimization module is entered; otherwise, the current energy-saving window control strategy remains unchanged.

6. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The energy-saving strategy optimization module specifically includes the following steps: Initialize the search population by mapping the window control parameter combination to the position vector of the search individual in the gray wolf optimization algorithm, and generate the position vectors of M search individuals through a random initialization method to obtain the initial search population. The window control parameter combination is specifically the control variables in the window execution data; The individual fitness value is calculated by calculating the fitness value of the search individuals in the population. The predictive control decision objective function is used as the fitness function, and the generated predictive decision value is used as the fitness value of the search individuals. All search individuals in the population are sorted in descending order according to the fitness value, and the individual with the best fitness and its position, the second best fitness individual and its position, and the third best fitness individual and its position are defined. The individual search boundary is dynamically adjusted, specifically based on the number of iterations. If an individual search boundary is not adjusted, then the individual search boundary is maintained; otherwise, the individual search boundary remains unchanged. The formula used is as follows: ； ； In the formula, Indicates the first The lower boundary of the j-th dimension during iteration, Indicates the first The upper boundary of the j-th dimension during iteration, Indicates the first The lower boundary of the j-th dimension is adjusted during iteration. Indicates the first The upper boundary of the j-th dimension during iteration, This indicates a fixed reduction in the neighborhood radius parameter. This represents the position of the fitness-optimal individual in dimension j. This indicates that the threshold for adjusting the iteration boundary is... This represents the floor function. This represents the floor function; Individual location-guided updates specifically involve updating the locations of other search individuals based on the core search individual. Individual location probability update specifically involves updating the location of the searched individual based on a probability-based stepwise adjustment mechanism; the formula used is as follows: ； ； ； In the formula, This represents the deviation between an individual's current position and the target mean position. This represents the probability of an individual updating. This represents a randomness parameter within the range [0,1]. This represents a randomness parameter within the range [0.5, 1]. This represents the probability update of the candidate position for the i-th search individual in the j-th dimension. Indicates that the i-th search individual is in the first position. The position of the j-th dimension during iteration. This indicates the position of the second-best fit individual in dimension j. This indicates the position of the third-best fitter individual in dimension j; Search boundary check correction, specifically, involves updating candidate positions obtained through individual location guidance. Candidate positions updated via individual position probabilities Perform boundary checks. If an individual's position exceeds the search space, truncate and correct it to the upper and lower limits of the corresponding variable to obtain the boundary correction position. Update the position comparison, specifically by recalculating the search boundary checks and corrections. and The fitness value, and the position of the individual in the current iteration. The fitness values ​​are compared, and the position of the individual with the best fitness value is selected as the initial position of that individual in the next iteration; The search for the optimal position of an individual involves evaluating the fitness values ​​of all individuals in the current population after each iteration. If the fitness of an individual's position is better than that of the current global optimal individual, then that individual's global optimal position is updated. The search iteration terminates when the fitness value of the search individual is higher than the decision threshold or when the maximum number of iterations is reached, and the search is terminated and the globally optimal individual position is obtained. The globally optimal individual position specifically refers to the optimal combination of window control parameters. The optimal window control strategy output specifically involves generating and outputting the optimal intelligent window control strategy based on the optimal combination of window control parameters.

7. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The energy-saving window intelligent control module specifically generates a corresponding window control command sequence based on the optimal intelligent control strategy for the window, and finally intelligently adjusts the energy-saving window according to the window control command sequence to realize automatic control and energy efficiency optimization of the energy-saving window operation.

8. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The data collection module specifically obtains energy-saving window control raw data by collecting the raw data required for energy-saving window control; the energy-saving window control raw data includes historical window control data and real-time window control data; both historical window control data and real-time window control data include indoor environment data, outdoor environment data, window execution data, indoor user data, and indoor energy consumption equipment operation data.

9. The energy-saving window control system based on artificial intelligence according to claim 1, characterized in that: The data optimization processing module specifically includes the following steps: Data cleaning specifically involves imputing missing values, removing outliers, and deduplicating data from the original data. Data normalization specifically involves using linear normalization to scale the numerical range of the cleaned raw data, unifying different data values ​​into the same numerical range. Data time alignment specifically involves slicing normalized data into time windows based on a unified timestamp, ensuring consistency of data from different sources and with different sampling frequencies on the same time scale, thus obtaining time-aligned energy-saving window control optimization data.

Images