Intelligent apartment resource dynamic optimization control method and system

Abstract

The invention discloses a smart apartment resource dynamic optimization control method and system, and the method comprises the steps: 1, collecting multi-dimensional data in a smart apartment through a multi-source data collection module, including energy use data, environment parameters and resident behavior mode data; 2, preprocessing the collected multi-dimensional data by using a data preprocessing module, and converting the preprocessed data into a format suitable for tensor operation; 3, performing optimization processing on the multi-dimensional data obtained in the step 2 by adopting a preset algorithm; and 4, training a machine learning model by using the data processed in the step 3, inputting real-time data into the trained machine learning model to obtain change trend data of the internal environment parameters of the apartment, and dynamically adjusting energy distribution and facility operation plans of the apartment according to the obtained change trend data and user behavior data. According to the method, the life quality of the smart apartment residents can be effectively improved, and the maximum efficiency of resource use and the optimal adjustment of the environment are realized.

Description

Technical Field

[0001] The present invention relates to a control method and system for dynamically optimizing resource allocation and facility operation in a smart apartment, belonging to the technical field of smart building management. Background Art

[0002] With the acceleration of urbanization and the rapid development of science and technology, smart building management systems have become an important tool to improve building efficiency, reduce operating costs and enhance the comfort of residents. In traditional building management systems, resource allocation and facility maintenance often rely on regular manual inspections and fixed scheduling plans, which is not only time-consuming and labor-intensive, but also difficult to respond to emergencies and environmental changes in a timely manner.

[0003] In addition, existing smart building management systems generally lack the ability to effectively process and analyze large amounts of scattered data, and are unable to fully utilize multi-source data collected from the building environment. These data, including energy consumption, environmental parameters, and occupants’ behavior patterns, are key factors in optimizing building management and improving the living experience.

[0004] Although existing technologies have proposed a variety of smart building management solutions, including the use of IoT technology for data collection and simple rule engines for equipment control, these solutions are generally inflexible and cannot effectively respond to complex or changing management needs. For example, they cannot accurately predict fluctuations in energy demand or automatically adjust resource allocation to respond to immediate environmental changes.

[0005] In addition, many systems fail to make full use of effective algorithms, especially in data processing and predictive modeling. Mathematical methods such as tensor decomposition and conjugate gradient method can provide more accurate data analysis and optimization strategies, but they are not applied in existing smart building management systems.

[0006] Therefore, a smart apartment resource management system is needed that uses mathematical models and machine learning methods to achieve efficient and dynamic management of apartment resources and improve the automation and intelligence level of the system. It can automatically analyze and process multi-dimensional data from smart apartments, optimize resource allocation in real time, predict and respond to changes in various environmental and operating conditions, and thus provide a more economical, efficient and comfortable living environment. Summary of the invention

[0007] The technical problem to be solved by the present invention is: how to optimize energy use and living environment through the collected multi-source data, and realize dynamic optimization management of resources in smart apartments.

[0008] In order to solve the above technical problems, the present invention provides a method for dynamically optimizing and controlling resources in a smart apartment, comprising the following steps:

[0009] Step 1. Use the multi-source data acquisition module to collect multi-dimensional data in the smart apartment, including energy usage data, environmental parameters, and resident behavior pattern data;

[0010] Step 2. Use the data preprocessing module to preprocess the collected multidimensional data and convert the preprocessed data into a format suitable for tensor operations;

[0011] Step 3. Use a preset algorithm to optimize the multidimensional data obtained in step 2;

[0012] Step 4. Use the data processed in step 3 to train the machine learning model, and input the real-time data into the trained machine learning model to obtain the change trend data of the internal environmental parameters of the apartment, and dynamically adjust the energy distribution and facility operation plan of the apartment based on the obtained change trend data and user behavior data.

[0013] In step one, the energy usage data includes water, electricity, and natural gas data, and the environmental data includes temperature, humidity, and CO2 concentration.

[0014] In step 2, in the process of removing outliers and noise, Gaussian process is used to detect and correct outliers to improve the system's ability to respond to abnormal events. The formula is as follows:

[0015] ; in is the kernel function, which is used to calculate any two points and The covariance between and is the length scale and amplitude of the Gaussian process, is the noise term, is the Kronecker function.

[0016] The collected data are then normalized to ensure the consistency and standardization of the input data, facilitating subsequent data analysis.

[0017] In step two, data preprocessing includes seasonal adjustment and trend separation of the data.

[0018] In step three, the preset algorithm is: using tensor decomposition method to analyze multidimensional data, extract key information, and use conjugate gradient method to optimize calculation, the formula is as follows: Formula (1);

[0019] Formula (2);

[0020] Formula (3);

[0021] Formula (4);

[0022] Formula (5);

[0023] Indicates The tensor estimate obtained by iteration, in the optimization problem, there is an optimal tensor Make the objective function minimize; Represents the objective function In tensor The gradient at It is The step size factor for the iteration; Indicates The residual vector of the iteration; Indicates The search direction vector for the iteration; It is calculated The coefficient when searching the direction; is a linear operator that maps a tensor to another tensor; is a constant tensor; Represents the inner product operation of a tensor.

[0024] In step 3, the tensor decomposition uses CP decomposition or Tucker decomposition method to decompose the collected multidimensional data to analyze the associations in the data. The decomposition model is as follows: ; Represents the original multi-dimensional data, including high-order tensors consisting of energy usage data, environmental parameters, and resident behavior data collected from smart apartments; is the rank of the tensor decomposition, that is, the number of factors obtained by decomposition; It is The weight coefficient of each component reflects the importance of the component; They are The factor vector of the components in each dimension; Represents the outer product operation of vectors; The symbol indicates that Tensor Add together to get the original tensor .

[0025] In step 4, prediction is performed using a tensor decomposition neural network model, including the following steps:

[0026] 1. Time window division: divide the input time series data into several overlapping time windows, each window corresponds to a sub-tensor, and the size of the initial time window can be determined based on prior knowledge or cross-validation;

[0027] 2) At each prediction step, the window size of the next step is automatically adjusted according to the prediction error of the current time window. If the prediction error is greater than the set value, the time window is reduced to capture the dynamic changes of a shorter scale; if the prediction error is less than the set value, the time window is expanded to utilize the trend information of a longer scale;

[0028] 3) Perform Tucker decomposition or CP decomposition on the sub-tensors corresponding to each time window to obtain a set of factor matrices and core tensors that describe the potential patterns in the corresponding time window;

[0029] 4) Time factor matrix update: Use recurrent neural network to learn the evolution law of the time modal factor matrix decomposed, and predict the factor matrix of the next time window;

[0030] 5) Using the predicted factor matrix, reconstruct the tensor of the next time window to obtain the predicted values ​​of environmental parameters in the future;

[0031] 6) After predicting each time window, slide the window forward by one step and repeat steps 2)-5) to achieve continuous multi-step prediction.

[0032] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0033] A computer-readable storage medium stores a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0034] A computer program product comprises a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0035] Beneficial effects achieved by the present invention: The smart apartment resource dynamic optimization control method of the present invention, through the comprehensive use of tensor decomposition and conjugate gradient method to process the collected multi-source data, uses machine learning models to analyze resident behavior, automatically adjusts energy distribution, and optimizes the HVAC system operation strategy, can effectively improve the management efficiency of smart apartments and the quality of life of residents, achieve maximum efficiency in resource utilization and optimal adjustment of the environment. At the same time, the method of the present invention reduces the parameter space in the algorithm and improves the robustness and accuracy in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of the smart apartment resource dynamic optimization control method of the present invention. DETAILED DESCRIPTION

[0037] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0038] Example 1

[0039] A method for dynamically optimizing and controlling resources of a smart apartment comprises the following steps:

[0040] Step 1. Use the multi-source data acquisition module to collect multi-dimensional data in the smart apartment, including energy usage data, environmental parameters, and resident behavior pattern data;

[0041] In step one, energy usage data, environmental data, and resident behavior data are collected using a variety of sensors installed at various set locations in the apartment;

[0042] The energy usage data includes water, electricity, and natural gas data, and the environmental data includes temperature, humidity, CO2 concentration, etc.;

[0043] Use smart meters to regularly record electricity, water and gas consumption data, and transmit the data to the central database in real time through the embedded system;

[0044] Environmental sensors are placed at multiple points in the apartment to monitor and record data such as temperature, humidity, and light intensity in real time;

[0045] The activity patterns of residents, including time of entry and exit and range of activities, are monitored through cameras and motion sensors, providing a data basis for the apartment's energy needs.

[0046] Step 2. Use the data preprocessing module to preprocess the collected multidimensional data and convert the preprocessed data into a format suitable for tensor operations. The preprocessing includes removing outliers and noise.

[0047] In step 2, data preprocessing includes seasonal adjustment and trend separation of data;

[0048] Seasonal adjustment and trend separation are data preprocessing methods for time series analysis. They separate different components from the original time series data, analyze the inherent laws of the data, and establish a prediction model. Seasonal adjustment and trend separation are important data preprocessing steps for time series modeling. By systematically separating different components, the laws of the data can be more accurately characterized, improving the performance of subsequent prediction or anomaly detection models.

[0049] In step 2, in the process of removing outliers and noise, Gaussian process is used to detect and correct outliers to improve the system's ability to respond to abnormal events. The formula is as follows:

[0050] ; in is the kernel function, which is used to calculate any two points and The covariance between and is the length scale and amplitude of the Gaussian process, is the noise term, is the Kronecker function.

[0051] The collected data are then normalized to ensure the consistency and standardization of the input data, facilitating subsequent data analysis.

[0052] Step 3. Use a preset algorithm to optimize the multidimensional data obtained in step 2 to improve data processing efficiency and analysis accuracy;

[0053] The preset algorithm is: using tensor decomposition method to analyze multidimensional data, extract key information, and use conjugate gradient method to optimize calculation. The formula is as follows: Formula (1);

[0054] Formula (2);

[0055] Formula (3);

[0056] Formula (4);

[0057] Formula (5); Indicates The tensor estimate obtained by iteration, in the optimization problem, there is an optimal tensor Make the objective function minimize;

[0058] Represents the objective function In tensor The gradient at the point where the function value grows fastest. In the optimization algorithm, the variable is updated in the direction of the negative gradient to minimize the objective function.

[0059] It is The step size factor of the iteration is used to control the update amplitude along the negative gradient direction;

[0060] Indicates The residual vector of the iteration is , represents the gap between the current solution and the optimal solution. The conjugate gradient method uses the residual vector to construct the search direction;

[0061] Indicates The search direction vector for the iteration;

[0062] It is calculated The coefficient when searching the direction;

[0063] Is a linear operator that maps a tensor to another tensor. In the optimization problem, Corresponding to the second-order derivative information of the objective function, that is, the Hessian matrix, the conjugate gradient method uses the linear operator Construct orthogonal search directions;

[0064] is a constant tensor representing the "target value" of the optimization problem. hour, is the constant term on the right side of the equation; in the optimization problem, is the negative or zero tensor of the gradient;

[0065] Represents the inner product operation of a tensor, corresponding to the concept of inner product in vector space. The inner product can measure the similarity of two tensors in a certain sense.

[0066] Formula (1)-(5) describe the optimization algorithm combining tensor decomposition and conjugate gradient method. In each iteration, first, the current solution The gradient of the objective function is calculated at , and then follow the negative gradient direction with a step size Update the tensor to get At the same time, using the residual and the previous search direction Calculate new search direction ,in For orthogonalization, the above algorithm utilizes the ability of tensor decomposition to reveal the intrinsic structure of data, and with the help of the convergence of the conjugate gradient method, it can efficiently process high-dimensional data and accelerate the optimization process.

[0067] The tensor decomposition uses CP decomposition or Tucker decomposition method to decompose the collected multidimensional data to analyze the associations in the data. The decomposition model is as follows: ; Represents the original multi-dimensional data, including high-order tensors consisting of energy usage data, environmental parameters, and resident behavior data collected from smart apartments;

[0068] is the rank of the tensor decomposition, that is, the number of factors obtained by decomposition, The value of is determined by balancing the complexity of the data and the required approximate accuracy. The larger it is, the more factors can be decomposed, and the stronger the expression ability is;

[0069] It is The weight coefficient of each component reflects the importance of the component. In CP decomposition, the weight of each component is expressed as a vector is given in the form of;

[0070] They are A factor vector of components in each dimension. For example, if the original tensor is of third order, then It corresponds to feature vectors in three different dimensions (such as time, space, and attributes);

[0071] Represents the outer product operation of vectors, combining multiple vectors into a high-order tensor, for example, Generate a third-order tensor;

[0072] The symbol indicates that Tensor Add together to get the original tensor .

[0073] In the above formula, several weighted low-rank tensors are used to represent high-dimensional complex data. Through tensor decomposition, we can discover the potential patterns and associations in the data and express the original data in a simpler form, thereby greatly reducing storage and computing costs.

[0074] In this embodiment, tensor decomposition is used to analyze the multi-source heterogeneous data of smart apartments to represent the intrinsic connection between energy use, environmental changes and user behavior, providing basic data for subsequent prediction and optimization.

[0075] When processing large-scale data, using tensor decomposition methods to deeply analyze multidimensional data to extract key information can significantly improve computing efficiency.

[0076] Step 4. Use the data processed in step 3 to train the machine learning model, and input the real-time data into the trained machine learning model to obtain the change trend data of the internal environmental parameters of the apartment, and dynamically adjust the energy distribution and facility operation plan of the apartment according to the obtained change trend data and user behavior data to optimize energy efficiency and living comfort;

[0077] In step four, the machine learning model adopts an adaptive time window-based tensor decomposition neural network (ATW-ATDNN) to enhance the flexibility and prediction accuracy of time series data processing.

[0078] The tensor decomposition neural network model is a time series prediction method that can improve the flexibility and prediction accuracy of traditional tensor decomposition models when processing dynamic time series data, and automatically adjust the time scale of tensor decomposition according to the dynamic characteristics of the data, so as to better capture the multi-scale patterns of the time series.

[0079] The prediction using tensor decomposition neural network model includes the following steps:

[0080] 1) Time window division: Divide the input time series data into several overlapping time windows, each window corresponds to a sub-tensor, and the size of the initial time window can be determined based on prior knowledge or cross-validation.

[0081] 2) Adaptive window adjustment: At each prediction step, the window size of the next step is automatically adjusted according to the prediction error of the current time window. If the prediction error is greater than the set value, the time window is reduced to capture the dynamic changes of a shorter scale; if the prediction error is less than the set value, the time window is expanded to utilize the trend information of a longer scale;

[0082] 3) Tensor decomposition: Tucker decomposition or CP decomposition is performed on the sub-tensors corresponding to each time window to obtain a set of factor matrices and core tensors that describe the potential patterns in the corresponding time window;

[0083] 4) Time factor matrix update: Use recurrent neural networks (such as LSTM or GRU) to learn the evolution law of the decomposed time modal factor matrix and predict the factor matrix of the next time window;

[0084] 5) Tensor reconstruction: Use the predicted factor matrix to reconstruct the tensor of the next time window and obtain the predicted values ​​of environmental parameters in the future;

[0085] 6) Sliding window prediction: After predicting each time window, slide the window forward by one step and repeat steps 2)-5) to achieve continuous multi-step prediction.

[0086] The core of the tensor decomposition neural network model is to combine tensor decomposition with deep learning, and to improve the ability of the ATW-ATDNN model to process time series data by introducing adaptive time windows and attention mechanisms.

[0087] The mathematical representation of the tensor decomposition neural network model is as follows:

[0088] Tensor decomposition part: given a high-order tensor ,in Represents the size of each dimension, and decomposes it into a core tensor through Tucker decomposition and several factor matrices The product of:

[0089]

[0090] in Indicated in tensor product in dimensions, is the rank of each dimension.

[0091] Attention mechanism: For the The factor matrix of dimensions , introduce the attention weight matrix , the weighted factor matrix is ​​obtained through the attention mechanism :

[0092]

[0093] Attention weight matrix Through learning, it is possible to adaptively adjust the importance of different dimensions and time steps.

[0094] Time window division: divide the time dimension Divide into The size is time window, each window corresponds to a sub-tensor , the size of the time window Adaptive adjustment through gating mechanism:

[0095]

[0096] in is the hidden state of the previous time window, and are learnable parameters one and two, is the sigmoid activation function;

[0097] Time evolution modeling: For each time window , its factor matrix Enter a Gated Recurrent Unit (GRU) for temporal evolution modeling:

[0098]

[0099]

[0100]

[0101]

[0102] in It is the update gate, is the reset gate, is a candidate hidden state, is the current hidden state, , , are all learnable parameters, Represents element-wise multiplication;

[0103] Prediction output: The hidden state of the last time window Mapped to the predicted value through a fully connected layer :

[0104]

[0105] in is the number of time steps for prediction, and are all learnable parameters;

[0106] When training the model, the predicted value is minimized and the true value The mean square error loss function between them is used to optimize the model parameters:

[0107]

[0108] in is the number of training samples.

[0109] In the prediction stage, historical environmental parameter data (such as temperature, humidity, light intensity, etc.) are constructed into high-order tensors and input into the tensor decomposition neural network model. The model outputs the predicted values ​​of each parameter in the future.

[0110] For example, for temperature prediction:

[0111] [ X temp ] i 1 , i 2 , … , i t - 1 , i t = Temp ( l i 1 , l i 2 , … , t i t )

[0112] in is the temperature data tensor, Indicates different locations, Represents different time steps, Input the trained ATW-ATDNN model:

[0113]

[0114] What you get The future The predicted temperature value for each time step.

[0115] Similarly, data tensors of other parameters such as humidity and light can be constructed and input into the model to obtain the corresponding predicted values.

[0116] Based on the environmental parameter trends predicted by the ATW-ATDNN model, the intelligent control system can adjust the operation strategies of HVAC and other equipment in advance to optimize energy consumption while ensuring comfort.

[0117] Step 5. Generate decision-making information, including energy usage reports and early warning information.

[0118] Provides an intuitive user interface that allows managers to view real-time data on energy consumption and environmental quality, and receive system optimization suggestions and early warning information.

[0119] Users can adjust environmental settings according to personal preferences to enhance user experience.

[0120] The present invention introduces adaptive time windows and attention mechanisms to give tensor decomposition the ability to process time series data, and combines it with deep learning to form an efficient and accurate environmental parameter prediction model to provide support for the dynamic optimization management of smart apartments. In the smart apartment scenario, multidimensional time series such as energy consumption data and environmental parameters can be constructed into a high-order tensor, and then predicted using the ATW-ATDNN model. The output of the model includes the changing trends of environmental parameters such as temperature, humidity, and light intensity in the future.

[0121] Based on the predicted changes in environmental parameters, the operating strategy of the HVAC system can be adjusted in advance. For example, when the temperature is predicted to rise to a certain threshold, the refrigeration system is started in advance to avoid excessively high indoor temperatures that affect living comfort; when the humidity is predicted to exceed the standard, the ventilation volume of the fresh air system is increased to adjust the indoor humidity level in time; according to the predicted changes in light intensity, the curtains and lighting systems are dynamically adjusted to meet lighting needs while reducing unnecessary energy consumption.

[0122] By guiding the intelligent control of the HVAC system through the prediction results of the ATW-ATDNN model, it is possible to achieve energy conservation and efficiency while improving the comfort of residents, giving full play to the advantages of smart apartments. At the same time, the prediction model can also provide support for other intelligent applications, such as early warning of indoor air quality problems, optimizing energy storage and scheduling, etc.

[0123] The machine learning model uses deep reinforcement learning to optimize the apartment’s security monitoring system. It automatically identifies suspicious behaviors by analyzing video surveillance data in real time and issues early warning information in a timely manner.

[0124] Through the above comprehensive technical methods, the present invention can realize dynamic optimization management of resources in smart apartments, improve energy efficiency, enhance the comfort of residents, and provide innovative solutions for future smart building management.

[0125] Example 2

[0126] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0127] Example 3

[0128] A computer-readable storage medium stores a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0129] Example 4

[0130] A computer program product comprises a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0131] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0134] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for dynamic optimization control of smart apartment resources, characterized in that: The following steps are involved: Step 1. Use the multi-source data acquisition module to collect multi-dimensional data in the smart apartment, including energy usage data, environmental parameters, and resident behavior pattern data; Step 2. Use the data preprocessing module to preprocess the collected multidimensional data and convert the preprocessed data into a format suitable for tensor operations; Step 3. Use a preset algorithm to optimize the multidimensional data obtained in step 2; Step 4. Use the data processed in step 3 to train the machine learning model, and input the real-time data into the trained machine learning model to obtain the change trend data of the internal environmental parameters of the apartment, and dynamically adjust the energy distribution and facility operation plan of the apartment based on the obtained change trend data and user behavior data.

2. The method for dynamic optimization control of smart apartment resources according to claim 1, characterized in that: In step one, the energy usage data includes water, electricity, and natural gas data, and the environmental data includes temperature, humidity, and CO2 concentration.

3. The method for dynamic optimization control of smart apartment resources according to claim 1, characterized in that: In step 2, in the process of removing outliers and noise, Gaussian process is used for outlier detection and correction. The formula is as follows: ； in is the kernel function, which is used to calculate any two points and The covariance between and is the length scale and amplitude of the Gaussian process, is the noise term, is the Kronecker function.

4. The method for dynamic optimization control of smart apartment resources according to claim 3 is characterized in that: In step two, data preprocessing includes seasonal adjustment and trend separation of the data.

5. The method for dynamic optimization control of smart apartment resources according to claim 1, characterized in that: In step three, the preset algorithm is: using tensor decomposition method to analyze multidimensional data, extract key information, and use conjugate gradient method to optimize calculation, the formula is as follows: Formula (1); Formula (2); Formula (3); Formula (4); Formula (5); Indicates The tensor estimate obtained by iteration, in the optimization problem, there is an optimal tensor Make the objective function minimize; Represents the objective function In tensor The gradient at It is The step size factor for the iteration; Indicates The residual vector of the iteration; Indicates The search direction vector for the iteration; It is calculated The coefficient when searching the direction; is a linear operator that maps a tensor to another tensor; is a constant tensor; Represents the inner product operation of a tensor.

6. The method for dynamic optimization control of smart apartment resources according to claim 5, characterized in that: In step 3, the tensor decomposition uses CP decomposition or Tucker decomposition method to decompose the collected multidimensional data to analyze the associations in the data. The decomposition model is as follows: ； Represents the original multi-dimensional data, including high-order tensors consisting of energy usage data, environmental parameters, and resident behavior data collected from smart apartments; is the rank of the tensor decomposition, that is, the number of factors obtained by decomposition; It is The weight coefficient of each component reflects the importance of the component; They are The factor vector of the components in each dimension; Represents the outer product operation of vectors; The symbol indicates that Tensor Add together to get the original tensor .

7. The method for dynamic optimization and control of smart apartment resources according to claim 1, characterized in that: In step 4, the tensor decomposition neural network model is used for prediction. The following steps are involved: 1) Divide the input time series data into several overlapping time windows, each window corresponds to a sub-tensor, and the size of the initial time window can be determined based on prior knowledge or cross-validation; 2) At each prediction step, the window size of the next step is automatically adjusted according to the prediction error of the current time window. If the prediction error is greater than the set value, the time window is reduced to capture the dynamic changes of a shorter scale; if the prediction error is less than the set value, the time window is expanded to utilize the trend information of a longer scale; 3) Perform Tucker decomposition or CP decomposition on the sub-tensors corresponding to each time window to obtain a set of factor matrices and core tensors that describe the potential patterns in the corresponding time window; 4) Use recurrent neural network to learn the evolution law of the decomposed time modal factor matrix and predict the factor matrix of the next time window; 5) Using the predicted factor matrix, reconstruct the tensor of the next time window to obtain the predicted values ​​of environmental parameters in the future; 6) After predicting each time window, slide the window forward by one step and repeat steps 2)-5) to achieve continuous multi-step prediction.

8. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images