Intelligent building personalized space management method based on deep neural network

Abstract

The invention discloses an intelligent building personalized space management method based on a deep neural network, and the method comprises the following steps: S1, collecting environment parameters, user behaviors and preference data, and constructing personalized control input; s2, cleaning, aligning, normalizing and missing filling are performed on the acquired data, and a standardized data set is generated; s3, constructing a deep neural network model, and predicting personalized control parameters; s4, inputting a prediction result into a whale optimization algorithm, and generating a control strategy candidate set meeting comfort and energy consumption targets; s5, inputting the candidate strategy to a water wave optimization algorithm, fusing multi-user preferences, and outputting a final control strategy; s6, a control instruction is issued to the building system, and environment adjustment is executed; and S7, collecting adjustment feedback, calculating an error and updating the neural network model to realize continuous optimization. According to the invention, intelligent regulation and control and personalized optimization of the building space environment are realized, and the comfort level and the energy efficiency management level are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of space management technology, and in particular to a personalized space management method for intelligent buildings based on deep neural networks. Background Art

[0002] With the continuous development of smart buildings, green buildings, and human living environment comfort technologies, building automation systems have gradually transitioned from traditional timing control and sensor-linked control to building environmental regulation systems with a certain level of intelligence. Current mainstream smart building systems rely on preset rule templates or logic programming to automatically control environmental parameters such as lighting, temperature control, ventilation, and shading, which has, to a certain extent, improved building energy efficiency and the living experience. However, existing systems are mostly "device-centric" management, and their control strategies lack a deep understanding of individual user behavior and subjective preferences. They often cannot respond to individual differences and dynamic changes in real time, making it difficult to meet the coordinated needs of environmental comfort and energy conservation in multi-user shared spaces.

[0003] In recent years, some research has attempted to introduce artificial intelligence technology into building management, such as using machine learning to model historical data and achieve simple predictions of user habits. However, most of these methods are based on static features or single-variable modeling, making it difficult to effectively capture the time series characteristics of user behavior and the complex coupling relationships between environmental factors. Furthermore, in multi-user scenarios, systems often use simple average weights or priorities to integrate individual preferences, which is unable to handle preference conflicts or achieve true coordinated optimization. Furthermore, existing models generally lack self-learning and adaptation mechanisms after deployment, resulting in gradual degradation of system performance over time and difficulty in maintaining the effectiveness of control strategies in the long term.

[0004] Traditional optimization algorithms have a certain application foundation in environmental control parameter scheduling, but most optimization methods suffer from problems such as being prone to local optimality and low search efficiency, especially in personalized control scenarios with multiple objectives and constraints. Furthermore, existing processing mechanisms for multimodal sensory data from different sources and frequencies often rely on independent channel processing, lacking a unified fusion modeling mechanism, making it difficult to achieve global synergy.

[0005] In summary, existing technologies still have significant shortcomings in addressing the comprehensive control challenges of personalized, multi-user, dynamic, and energy-constrained smart building environments. There is a lack of a comprehensive methodology that integrates deep user modeling, intelligent optimization search, multi-objective trade-offs, and real-time self-learning and updating. Therefore, there is an urgent need for a personalized smart building space management method that addresses multi-user behavior, multi-source sensory data, and complex control objectives to achieve efficient, intelligent, and human-centered control of building space environments.

[0006] Therefore, how to provide a personalized space management method for intelligent buildings based on deep neural networks is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] One purpose of the present invention is to propose a method for personalized space management in intelligent buildings based on deep neural networks. The present invention makes full use of deep neural network models, whale optimization algorithms, and water wave optimization algorithms, integrates environmental parameter data, user behavior data, and user preference data, and describes in detail the entire process of realizing multi-user personalized space control and adaptive optimization. It has the advantages of fast response speed, high personality matching accuracy, and strong energy consumption control and optimization capabilities.

[0008] According to an embodiment of the present invention, a method for personalized space management of an intelligent building based on a deep neural network includes the following steps:

[0009] S1. Collect environmental parameter data, user behavior data, and user preference data. The environmental parameter data includes temperature, humidity, light intensity, and carbon dioxide concentration. The user behavior data includes location information, activity status information, and device usage records. The user preference data includes historical operation records and environmental feedback information.

[0010] S2. performing data cleaning, time alignment, normalization, and missing value filling on the multi-source data to generate a standardized feature dataset;

[0011] S3. Build a deep neural network prediction model based on the standardized feature data set and output personalized environmental control parameter prediction results;

[0012] S4. Input the prediction results into the whale optimization algorithm module to search for multi-objective control strategies and generate a candidate set of control strategies;

[0013] S5. Input the control strategy candidate set into the water wave optimization algorithm module, perform multi-user preference fusion, and output the final control strategy set;

[0014] S6. Sending the final control strategy set to the building automation control system to perform environmental adjustment operations;

[0015] S7. Collect adjustment results and user feedback information, update the deep neural network prediction model, and achieve model adaptive optimization.

[0016] Optionally, the S1 specifically includes:

[0017] S11. Collecting environmental parameter data through environmental sensing devices installed in the smart building. The environmental sensing devices include temperature sensors, humidity sensors, light sensors, and air quality sensors. The environmental parameter data include real-time values ​​of temperature, humidity, light intensity, and carbon dioxide concentration in each spatial area;

[0018] S12. Collecting user behavior data through a user behavior sensing device deployed in the building space, wherein the user behavior sensing device includes a personnel positioning module, a motion detection module, and a device interaction monitoring module. The user behavior data includes the user's location information in the space, activity status information, and smart device usage records;

[0019] S13. Collecting user preference data through a user terminal system, wherein the user terminal system includes an intelligent control panel, a mobile application, or a voice interaction device, wherein the user preference data includes the user's historical environment adjustment operation records and subjective feedback information on the current space environment;

[0020] S14: Integrate the environmental parameter data, user behavior data, and user preference data to form a multi-source original data set.

[0021] Optionally, the S2 specifically includes:

[0022] S21. Clean the collected multi-source raw data set, delete data records containing missing fields, abnormal formats, or values ​​that significantly deviate from the normal range, and generate a cleaned data set;

[0023] S22. Standardize the timestamp fields of each data item in the cleaned data set in a unified format, and perform synchronous matching processing on environmental parameter data, user behavior data, and user preference data from different sources according to a unified time axis to generate a time-aligned data set;

[0024] S23. Normalize various numerical features in the time-series aligned data set, standardize features of different dimensions and numerical ranges to a unified interval, ensure that the input data has a unified scale, and generate a normalized data set;

[0025] S24. Fill in the missing values ​​for empty value fields or discontinuous records in the normalized data set, using linear interpolation or historical average filling strategy to generate a completely preprocessed data set.

[0026] Optionally, the S3 specifically includes:

[0027] S31, input the data set D as input to the deep neural network prediction model M;

[0028] S32. Construct a prediction model M, which includes an input layer, a time series modeling layer, a feature fusion layer, and an output layer. The time series modeling layer is used to process changes in user behavior, the feature fusion layer is used to integrate environmental parameters, user behavior, and preference features, and the output layer generates prediction results.

[0029] S33, training the model M, optimizing the model weights based on the historical control parameters in the data set D as the target;

[0030] S34. Use the trained model M to predict the real-time data set D and output the personalized environment control parameter result P.

[0031] Optionally, the S4 specifically includes:

[0032] S41, inputting the personalized environmental control parameter prediction result P generated in step S3 into the whale optimization algorithm module W;

[0033] S42. Initializing search population individuals in the whale optimization algorithm module W, where each individual is represented by a control parameter vector including a temperature setting value, a lighting brightness value, and an air circulation rate value;

[0034] S43. Construct a multi-objective fitness function f, which is in the form of:

[0035] f=α×S u -β×E s ;

[0036] Among them, α and β are preset weight coefficients, S u represents the user comfort score, E s Indicates the system energy consumption score;

[0037] S44. According to the position update mechanism of the whale algorithm, iteratively update the individuals in the population and dynamically search for the control parameter vector that satisfies the optimal solution of the fitness function;

[0038] S45. Output a control strategy candidate set C1 that meets the optimal conditions.

[0039] Optionally, the S5 specifically includes:

[0040] S51, inputting the control strategy candidate set C1 output in step S4 into the water wave optimization algorithm module B, wherein the control strategy candidate set C1 includes multiple optimal control parameter vectors for a single user;

[0041] S52. Construct a fusion objective function F based on a multi-user usage scenario, where:

[0042]

[0043] Among them, n represents the number of users in the current space, w i represents the weight coefficient of the i-th user, C i It represents the control strategy assigned to it, P i It represents the prediction result of its personalized control parameters, D(C i ,P i ) represents the preference deviation function;

[0044] S53, through the disturbance propagation mechanism in the water wave optimization algorithm, adjust the individual solutions in the control strategy candidate set C1, and search for the optimal combination strategy under the fusion objective function F;

[0045] S54, using local fluctuation search and global diffusion strategy to alternately execute to avoid falling into local optimum;

[0046] S55. Output the fused and optimized control strategy set C2 for subsequent generation and issuance of environmental control instructions.

[0047] Optionally, the S6 specifically includes:

[0048] S61, converting the fused control strategy set C2 into an executable control instruction set I, wherein the control instruction set I includes a temperature setting instruction, a lighting adjustment instruction, and an air circulation control instruction;

[0049] S62: Send the control instruction set I to the building automation control system S, where the building automation control system S includes a temperature control subsystem, a lighting subsystem, and an air quality control subsystem;

[0050] S63, the control system S controls the execution state of each target device according to the instruction set I, and the setting value change process is recorded as:

[0051] V t =V t-1 +ΔV;

[0052] Among them, V t Indicates the target device parameter value at the current time point, V t-1 Indicates the device parameter value at the previous time point, and ΔV indicates the control change amplitude;

[0053] S64. According to the device status feedback interface, the response status and execution results of each subsystem are collected in real time as the data basis for controlling the execution effect.

[0054] Optionally, the S7 specifically includes:

[0055] S71. Collect environmental change data and user subjective feedback data after each subsystem executes the control instruction in step S6 to form an actual feedback set F. The feedback set F includes the final execution temperature, lighting, air circulation status, and the user's rating of the adjustment result;

[0056] S72: Perform error calculation on the feedback set F and the personalized environmental control parameter prediction result P output by the model M in step S3 to form an error information set E, where the error calculation formula is:

[0057] E = |FP|;

[0058] Where E represents the absolute error vector between the predicted result and the actual execution effect;

[0059] S73. Using the error set E as backpropagation input, update the network weight parameters of the deep neural network prediction model M to enhance the model's ability to respond to user behavior and environmental changes;

[0060] S74: Replace the original model with the updated model M, and participate in the loop execution of steps S3 to S7 to achieve continuous learning and dynamic optimization of the model.

[0061] The beneficial effects of the present invention are:

[0062] (1) The present invention constructs a deep neural network prediction model to integrate environmental parameter data, user behavior data and user preference data, which can accurately predict the comfort needs of individual users in a specific time and space environment, and handle preference conflicts between multiple users through optimization algorithms to achieve coordinated regulation of different user needs in public spaces. Compared with traditional averaging processing methods, the matching accuracy and applicability are significantly improved.

[0063] (2) The present invention introduces the whale optimization algorithm and the water wave optimization algorithm, which are used for global search of the initial control strategy and multi-user conflict fusion processing respectively. By constructing a multi-objective fitness function that considers user comfort scores and system energy consumption scores, the dynamic balance between comfort and energy saving of the environmental control strategy is achieved, which is significantly better than the traditional single-objective or fixed-rule control method.

[0064] (3) The present invention utilizes user feedback results and device response data to construct an error feedback mechanism and perform online optimization on the deep neural network model, so that the model has adaptive learning capabilities and can be dynamically adjusted according to changes in user behavior and environmental fluctuations, effectively overcoming the problem of traditional system performance decaying over time and improving the accuracy and stability of the system in long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0066] Figure 1 This is a flowchart of a method for personalized space management of intelligent buildings based on deep neural networks proposed by the present invention;

[0067] Figure 2 This is a data input and structural diagram of a deep neural network prediction model for a smart building personalized space management method based on a deep neural network proposed in the present invention;

[0068] Figure 3 This is a schematic diagram of the personalized initial control strategy search performed by the whale optimization algorithm of the intelligent building personalized space management method based on deep neural network proposed in the present invention. DETAILED DESCRIPTION

[0069] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0070] refer to Figure 1-3 , a personalized space management method for intelligent buildings based on deep neural networks, comprising the following steps:

[0071] S1. Collect environmental parameter data, user behavior data, and user preference data. The environmental parameter data includes temperature, humidity, light intensity, and carbon dioxide concentration. The user behavior data includes location information, activity status information, and device usage records. The user preference data includes historical operation records and environmental feedback information.

[0072] S2. performing data cleaning, time alignment, normalization, and missing value filling on the multi-source data to generate a standardized feature dataset;

[0073] S3. Build a deep neural network prediction model based on the standardized feature data set and output personalized environmental control parameter prediction results;

[0074] S4. Input the prediction results into the whale optimization algorithm module to search for multi-objective control strategies and generate a candidate set of control strategies;

[0075] S5. Input the control strategy candidate set into the water wave optimization algorithm module, perform multi-user preference fusion, and output the final control strategy set;

[0076] S6. Sending the final control strategy set to the building automation control system to perform environmental adjustment operations;

[0077] S7. Collect adjustment results and user feedback information, update the deep neural network prediction model, and achieve model adaptive optimization.

[0078] In this embodiment, S1 specifically includes:

[0079] S11. Collecting environmental parameter data through environmental sensing devices installed in the smart building. The environmental sensing devices include temperature sensors, humidity sensors, light sensors, and air quality sensors. The environmental parameter data include real-time values ​​of temperature, humidity, light intensity, and carbon dioxide concentration in each spatial area;

[0080] S12. Collecting user behavior data through a user behavior sensing device deployed in the building space, wherein the user behavior sensing device includes a personnel positioning module, a motion detection module, and a device interaction monitoring module. The user behavior data includes the user's location information in the space, activity status information, and smart device usage records;

[0081] S13. Collecting user preference data through a user terminal system, wherein the user terminal system includes an intelligent control panel, a mobile application, or a voice interaction device, wherein the user preference data includes the user's historical environment adjustment operation records and subjective feedback information on the current space environment;

[0082] S14: Integrate the environmental parameter data, user behavior data, and user preference data to form a multi-source original data set.

[0083] In this embodiment, S2 specifically includes:

[0084] S21. Clean the collected multi-source raw data set, delete data records containing missing fields, abnormal formats, or values ​​that significantly deviate from the normal range, and generate a cleaned data set;

[0085] S22. Standardize the timestamp fields of each data item in the cleaned data set in a unified format, and perform synchronous matching processing on environmental parameter data, user behavior data, and user preference data from different sources according to a unified time axis to generate a time-aligned data set;

[0086] S23. Normalize various numerical features in the time-series aligned data set, standardize features of different dimensions and numerical ranges to a unified interval, ensure that the input data has a unified scale, and generate a normalized data set;

[0087] S24. Fill in the missing values ​​for empty value fields or discontinuous records in the normalized data set, using linear interpolation or historical average filling strategy to generate a completely preprocessed data set.

[0088] In this embodiment, S3 specifically includes:

[0089] S31, input the data set D as input to the deep neural network prediction model M;

[0090] S32. Construct a prediction model M, which includes an input layer, a time series modeling layer, a feature fusion layer, and an output layer. The time series modeling layer is used to process changes in user behavior, the feature fusion layer is used to integrate environmental parameters, user behavior, and preference features, and the output layer generates prediction results.

[0091] S33, training the model M, optimizing the model weights based on the historical control parameters in the data set D as the target;

[0092] S34. Use the trained model M to predict the real-time data set D and output the personalized environment control parameter result P.

[0093] In this embodiment, the S4 specifically includes:

[0094] S41, inputting the personalized environmental control parameter prediction result P generated in step S3 into the whale optimization algorithm module W;

[0095] S42. Initializing search population individuals in the whale optimization algorithm module W, where each individual is represented by a control parameter vector including a temperature setting value, a lighting brightness value, and an air circulation rate value;

[0096] S43. Construct a multi-objective fitness function f, which is in the form of:

[0097] f=α×S u -β×E s ;

[0098] Among them, α and β are preset weight coefficients, S u represents the user comfort score, E s Indicates the system energy consumption score;

[0099] S44. According to the position update mechanism of the whale algorithm, iteratively update the individuals in the population and dynamically search for the control parameter vector that satisfies the optimal solution of the fitness function;

[0100] S45. Output a control strategy candidate set C1 that meets the optimal conditions.

[0101] In this embodiment, the S5 specifically includes:

[0102] S51, inputting the control strategy candidate set C1 output in step S4 into the water wave optimization algorithm module B, wherein the control strategy candidate set C1 includes multiple optimal control parameter vectors for a single user;

[0103] S52. Construct a fusion objective function F based on a multi-user usage scenario, where:

[0104]

[0105] Among them, n represents the number of users in the current space, w i represents the weight coefficient of the i-th user, C i It represents the control strategy assigned to it, P i It represents the prediction result of its personalized control parameters, D(C i ,P i ) represents the preference deviation function;

[0106] S53, through the disturbance propagation mechanism in the water wave optimization algorithm, adjust the individual solutions in the control strategy candidate set C1, and search for the optimal combination strategy under the fusion objective function F;

[0107] S54, using local fluctuation search and global diffusion strategy to alternately execute to avoid falling into local optimum;

[0108] S55. Output the fused and optimized control strategy set C2 for subsequent generation and issuance of environmental control instructions.

[0109] In this embodiment, S6 specifically includes:

[0110] S61, converting the fused control strategy set C2 into an executable control instruction set I, wherein the control instruction set I includes a temperature setting instruction, a lighting adjustment instruction, and an air circulation control instruction;

[0111] S62: Send the control instruction set I to the building automation control system S, where the building automation control system S includes a temperature control subsystem, a lighting subsystem, and an air quality control subsystem;

[0112] S63, the control system S controls the execution state of each target device according to the instruction set I, and the setting value change process is recorded as:

[0113] V t =V t-1 +ΔV;

[0114] Among them, V t Indicates the target device parameter value at the current time point, V t-1 Indicates the device parameter value at the previous time point, and ΔV indicates the control change amplitude;

[0115] S64. According to the device status feedback interface, the response status and execution results of each subsystem are collected in real time as the data basis for controlling the execution effect.

[0116] In this embodiment, the S7 specifically includes:

[0117] S71. Collect environmental change data and user subjective feedback data after each subsystem executes the control instruction in step S6 to form an actual feedback set F. The feedback set F includes the final execution temperature, lighting, air circulation status, and the user's rating of the adjustment result;

[0118] S72: Perform error calculation on the feedback set F and the personalized environmental control parameter prediction result P output by the model M in step S3 to form an error information set E, where the error calculation formula is:

[0119] E = |FP|;

[0120] Where E represents the absolute error vector between the predicted result and the actual execution effect;

[0121] S73. Using the error set E as backpropagation input, update the network weight parameters of the deep neural network prediction model M to enhance the model's ability to respond to user behavior and environmental changes;

[0122] S74: Replace the original model with the updated model M, and participate in the loop execution of steps S3 to S7 to achieve continuous learning and dynamic optimization of the model.

[0123] Example 1:

[0124] In order to verify the feasibility of the present invention in implementation, the present invention was applied to a large-scale comprehensive office building intelligent management project in Beijing. The building has a total of 15 floors with a total area of ​​more than 30,000 square meters. The number of office workers exceeds 600 per day. It has a typical multi-user, high-frequency dynamic entry and exit scenario, and is equipped with multiple public spaces, including conference rooms, shared office areas, coffee break areas, etc., which are suitable for evaluating the actual effect of personalized space management systems. The original environmental control system of the building mainly relies on time scheduling strategies and simplified threshold trigger logic. It lacks user behavior recognition, preference modeling and multi-user conflict coordination functions. As a result, in actual use, the air conditioner is often turned on but no one uses it, the lighting does not respond as needed, and the environmental parameters do not match individual differences. These problems not only cause energy waste, but also affect the comfort experience of office staff.

[0125] In this embodiment, the research team completed system deployment in the building based on the technical solution of the present invention. First, environmental sensing modules containing temperature, humidity, light intensity, and carbon dioxide concentration were installed in key areas. Simultaneously, positioning and activity recognition modules were deployed in densely populated areas and variable-use areas to collect user location and behavior data. All sensor data is connected to the system in real time for aggregation and preprocessing, and is then connected to a user preference prediction model built using a deep neural network. Users can actively mark their satisfaction with the current environment through a mobile app, and the system continuously optimizes preference tags based on these tags for continuous learning.

[0126] In practice, when a user enters a shared workspace, the system combines their historical operating habits, recent behavior, and current environmental data to predict the optimal combination of space control parameters. For example, it automatically sets the air conditioning temperature to 24 degrees Celsius, the lighting brightness to 65%, and the fresh air system to 25% power. If multiple users are present in the space, the system uses the Whale Optimization algorithm to preliminarily select individual strategies. This is then combined with the Water Wave Optimization algorithm to resolve conflicting preferences and ultimately output a multi-party equilibrium control strategy. The control results are distributed by the central controller to each subsystem, which continuously records changes in device responses and user satisfaction.

[0127] The entire test lasted four weeks, during which the system automatically controlled operations an average of 320 times a day. By comparing the data before and after the test, it was found that the system saved an average of about 87.5 kWh of energy per week after optimization, and the energy saving rate was about 12.7%. In terms of user satisfaction, the average satisfaction score was 3.9 points (out of 5 points) before the introduction of the system, and it increased to 4.6 points after the introduction, an increase of nearly 18%. At the same time, for the conference room scenario, the system successfully reduced the response time from the original approximately 40 seconds to 11 seconds, significantly improving the real-time response performance. Data shows that after the introduction of the present invention, the efficiency of building space utilization has been significantly improved, especially during peak hours, the environmental adaptability of public spaces has been better, and the equipment idling rate has dropped by more than 15%.

[0128] Table 1: Comparison of key performance indicators before and after system deployment

[0129]

[0130]

[0131] As can be seen from the table, the introduction of this personalized space management system, based on deep neural networks and intelligent optimization algorithms, significantly improved both overall building environmental control efficiency and user experience. First, system response time was reduced from an average of approximately 40.2 seconds in the original system to just 11.3 seconds, a 72% improvement in response speed. This demonstrates that the system can more quickly detect changes and make adjustments to meet users' immediate needs.

[0132] In terms of user satisfaction ratings, through intelligent prediction and multi-user preference integration, the system achieves environmental control that better meets users' actual needs, with the rating increasing from 3.9 to 4.6, an increase of nearly 18%. This indicates that the system has received more positive feedback and recognition from users during actual use.

[0133] In terms of energy savings, the original system's average weekly energy consumption was 688.4 kWh, but after optimization, this system dropped to 600.9 kWh, a 12.7% reduction. This represents a significant improvement in large commercial buildings, particularly during the high-energy consumption periods of summer and winter. Furthermore, by introducing an environmental matching algorithm, the system's spatial matching accuracy increased from 76.5% to 91.2%, a 14.7% improvement, demonstrating that the control strategy better aligns with the actual needs of users and space conditions.

[0134] In terms of system proactiveness, the average daily frequency of control actions increased from 118 times to 320 times, an increase of 171%. This reflects that the system has higher dynamic response capabilities and can continuously optimize and adjust according to real-time changes.

[0135] Furthermore, the system's air conditioning idle rate has been significantly reduced from 21.3% to 5.9%, a reduction of over 15%, effectively addressing the energy waste caused by air conditioning running for extended periods without use. Notably, the introduction of this invention has achieved a 98.7% success rate in resolving conflicts among multiple users' preferences, a problem that was difficult to resolve in the original system. This provides reliable environmental management for shared spaces.

[0136] The above data show that the present invention not only improves the intelligence and humanization level of the system, but also significantly reduces energy consumption through precise prediction and optimized control, improves the efficiency of space resource utilization, and has good practicality, promotion value and economic benefits.

[0137] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for personalized space management of intelligent buildings based on deep neural networks, characterized in that: The steps include: S1. Collect environmental parameter data, user behavior data, and user preference data. The environmental parameter data includes temperature, humidity, light intensity, and carbon dioxide concentration. The user behavior data includes location information, activity status information, and device usage records. The user preference data includes historical operation records and environmental feedback information. S2. performing data cleaning, time alignment, normalization, and missing value filling on the multi-source data to generate a standardized feature data set; S3. Build a deep neural network prediction model based on the standardized feature data set and output personalized environmental control parameter prediction results; S4. Input the prediction results into the whale optimization algorithm module to search for multi-objective control strategies and generate a candidate set of control strategies; S5. Input the control strategy candidate set into the water wave optimization algorithm module, perform multi-user preference fusion, and output the final control strategy set; S6. Sending the final control strategy set to the building automation control system to perform environmental adjustment operations; S7. Collect adjustment results and user feedback information, update the deep neural network prediction model, and achieve model adaptive optimization.

2. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: Said S1 specifically includes: S11. Collecting environmental parameter data through environmental sensing devices installed in the smart building. The environmental sensing devices include temperature sensors, humidity sensors, light sensors, and air quality sensors. The environmental parameter data include real-time values ​​of temperature, humidity, light intensity, and carbon dioxide concentration in each spatial area; S12. Collecting user behavior data through a user behavior sensing device deployed in the building space, wherein the user behavior sensing device includes a personnel positioning module, a motion detection module, and a device interaction monitoring module. The user behavior data includes the user's location information in the space, activity status information, and smart device usage records; S13. Collecting user preference data through a user terminal system, wherein the user terminal system includes an intelligent control panel, a mobile application, or a voice interaction device, wherein the user preference data includes the user's historical environment adjustment operation records and subjective feedback information on the current space environment; S14: Integrate the environmental parameter data, user behavior data, and user preference data to form a multi-source original data set.

3. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S2 specifically includes: S21. Clean the collected multi-source raw data set, delete data records containing missing fields, abnormal formats, or values ​​that significantly deviate from the normal range, and generate a cleaned data set; S22. Standardize the timestamp fields of each data item in the cleaned data set in a unified format, and perform synchronous matching processing on environmental parameter data, user behavior data, and user preference data from different sources according to a unified time axis to generate a time-aligned data set; S23. Normalize various numerical features in the time-series aligned data set, standardize features of different dimensions and numerical ranges to a unified interval, ensure that the input data has a unified scale, and generate a normalized data set; S24. Fill in the missing values ​​for empty value fields or discontinuous records in the normalized data set, using linear interpolation or historical average filling strategy to generate a completely preprocessed data set.

4. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S3 specifically includes: S31, taking the data set D as input and inputting it into the deep neural network prediction model M; S32. Construct a prediction model M, which includes an input layer, a time series modeling layer, a feature fusion layer, and an output layer. The time series modeling layer is used to process changes in user behavior, the feature fusion layer is used to integrate environmental parameters, user behavior, and preference features, and the output layer generates prediction results. S33, training the model M, optimizing the model weights based on the historical control parameters in the data set D as the target; S34. Use the trained model M to predict the real-time data set D and output the personalized environment control parameter result P.

5. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S4 specifically includes: S41, inputting the personalized environmental control parameter prediction result P generated in step S3 into the whale optimization algorithm module W; S42. Initializing search population individuals in the whale optimization algorithm module W, where each individual is represented by a control parameter vector including a temperature setting value, a lighting brightness value, and an air circulation rate value; S43. Construct a multi-objective fitness function f, which is in the form of: f=α×S u -β×E s ; Among them, α and β are preset weight coefficients, S u represents the user comfort score, E s Indicates the system energy consumption score; S44. According to the position update mechanism of the whale algorithm, iteratively update the individuals in the population and dynamically search for the control parameter vector that satisfies the optimal solution of the fitness function; S45. Output a control strategy candidate set C1 that meets the optimal conditions.

6. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S5 specifically includes: S51, inputting the control strategy candidate set C1 output in step S4 into the water wave optimization algorithm module B, wherein the control strategy candidate set C1 includes multiple optimal control parameter vectors for a single user; S52. Construct a fusion objective function F based on a multi-user usage scenario, where: Among them, n represents the number of users in the current space, w i represents the weight coefficient of the i-th user, C i It represents the control strategy assigned to it, P i It represents the prediction result of its personalized control parameters, D(C i ,P i ) represents the preference deviation function; S53, through the disturbance propagation mechanism in the water wave optimization algorithm, adjust the individual solutions in the control strategy candidate set C1, and search for the optimal combination strategy under the fusion objective function F; S54, using local fluctuation search and global diffusion strategy to alternately execute to avoid falling into local optimum; S55. Output the fused and optimized control strategy set C2 for subsequent generation and issuance of environmental control instructions.

7. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S6 specifically includes: S61, converting the fused control strategy set C2 into an executable control instruction set I, wherein the control instruction set I includes a temperature setting instruction, a lighting adjustment instruction, and an air circulation control instruction; S62: Send the control instruction set I to the building automation control system S, where the building automation control system S includes a temperature control subsystem, a lighting subsystem, and an air quality control subsystem; S63, the control system S controls the execution state of each target device according to the instruction set I, and the setting value change process is recorded as: V t =V t-1 +ΔV; Among them, V t Indicates the target device parameter value at the current time point, V t-1 Indicates the device parameter value at the previous time point, and ΔV indicates the control change amplitude; S64. According to the device status feedback interface, the response status and execution results of each subsystem are collected in real time as the data basis for controlling the execution effect.

8. The method for personalized space management of intelligent buildings based on deep neural networks according to claim 1, characterized in that: The S7 specifically includes: S71. Collect environmental change data and user subjective feedback data after each subsystem executes the control instruction in step S6 to form an actual feedback set F. The feedback set F includes the final execution temperature, lighting, air circulation status, and the user's rating of the adjustment result; S72: Perform error calculation on the feedback set F and the personalized environmental control parameter prediction result P output by the model M in step S3 to form an error information set E, where the error calculation formula is: E = |FP|; Where E represents the absolute error vector between the predicted result and the actual execution effect; S73. Using the error set E as backpropagation input, update the network weight parameters of the deep neural network prediction model M to enhance the model's ability to respond to user behavior and environmental changes; S74: Replace the original model with the updated model M, and participate in the loop execution of steps S3 to S7 to achieve continuous learning and dynamic optimization of the model.

Images