A digital twin monitoring system for building environment

Abstract

The present invention discloses a building environment digital twin monitoring system, including an environment perception physical layer, a communication processing twin layer, an analysis modeling prediction layer, and a building environment monitoring application layer. The present invention belongs to the technical field of building environment monitoring systems, specifically a building environment digital twin monitoring system, the present invention adopts a U-shaped network combined with a convolutional long short-term memory module, and performs data classification and hierarchical storage by a multi-task learning data classification method, thereby improving the overall operation efficiency of the system; adopts a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction, thereby improving the overall availability of the system; adopts a building environment digital twin comprehensive monitoring, obtains a building environment digital twin monitoring comprehensive solution, improves the application layer of the digital twin system, realizes the unification of environmental management, space management, and personnel safety management, and provides a feasible technical reference for further optimization of the building environment digital twin monitoring system.

Description

Technical Field

[0001] The present invention belongs to the technical field of building environment monitoring systems, and specifically relates to a building environment digital twin monitoring system. Background Art

[0002] The digital twin monitoring system for building environment is an intelligent platform based on digital twin technology. It realizes dynamic monitoring and management of the building environment by synchronizing the physical environment data of the building (such as temperature, humidity, and air quality) in real time. The system uses sensors to collect data and data analysis technology for processing. Its main functions are to improve building energy efficiency, optimize space utilization, and enhance the comfort and safety of living and working environments. At the same time, it supports management decisions and realizes intelligent operation and maintenance of the building environment.

[0003] However, in the existing digital twin monitoring systems for building environments, there is a technical problem that the representation of the building environment is complex and difficult, often involving multiple data indicators and prediction tasks, and the reading and writing frequencies of data for different data indicators and tasks are different and the specific content of the required data is also different. The use of a fixed and single data storage method is likely to waste storage resources and reduce system efficiency; in the existing digital twin monitoring systems for building environments, there is a technical problem that when conducting environmental monitoring, traditional methods use discrete and independent methods to predict environmental indicators, environmental occupancy and personnel circulation tasks, which leads to a lack of connectivity between specific tasks of building environment monitoring, ignores potential data connections and thus reduces the availability of predictions; in the existing digital twin monitoring systems for building environments, there is a technical problem that for the prediction of specific building environment indicators, traditional methods cannot form an effective linkage monitoring plan due to independent predictions, thereby reducing the availability and controllability of the digital twin monitoring system when applied to the building environment. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digital twin monitoring system for a building environment. In view of the fact that in the existing digital twin monitoring system for a building environment, the representation of the building environment is complex and difficult, often involving multiple data indicators and prediction tasks, and the reading and writing frequencies of data for different data indicators and tasks are different and the specific contents of the required data are also different, and the use of a fixed single data storage method is prone to waste storage resources and reduce the efficiency of the system. The technical problem, this solution creatively adopts a U-shaped network combined with a convolutional long short-term memory module, and performs a multi-task learning data classification method to perform data classification and hierarchical storage. By performing data recognition based on different task data and storing them at different levels according to the reading and writing frequencies, the overall operating efficiency of the system is improved, and a good data foundation is provided for subsequent prediction tasks; in view of the fact that in the existing digital twin monitoring system for a building environment, the traditional method uses a discrete and independent method to predict the tasks of environmental indicators, environmental occupancy and personnel circulation when performing environmental monitoring, which leads to a lack of connectivity between the specific tasks of building environment monitoring and ignores the potential data. In order to solve the technical problem that the connection between two building environments is poor, the availability of prediction is reduced. This solution creatively adopts a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction. Through data classification storage in the data storage layer and shared feature extraction of the multi-task spatiotemporal convolutional network, the unique features of the tasks and the shared features are connected, thereby obtaining the potential connection between the three tasks, which improves the system prediction performance and monitoring efficiency as a whole, and improves the overall availability of the system. In view of the technical problem that in the existing digital twin monitoring system of the building environment, there are predictions for specific building environment indicators. Traditional methods cannot form an effective linkage monitoring plan due to independent predictions, thereby reducing the availability and controllability of the digital twin monitoring system when applied to the building environment. This solution creatively adopts comprehensive monitoring of digital twins of the building environment to obtain a comprehensive plan for digital twin monitoring of the building environment. By designing the subsequent application of the prediction data of the three tasks, the application layer of the digital twin system is improved, and the unification of environmental management, space management and personnel safety management is realized, providing a feasible technical reference for further optimization of the digital twin monitoring system of the building environment.

[0005] The technical solution adopted by the present invention is as follows: A building environment digital twin monitoring system provided by the present invention includes an environment perception physical layer, a communication processing twin layer, an analysis modeling prediction layer and a building environment monitoring application layer;

[0006] The environment sensing physical layer is used for data collection, and obtains the original building environment monitoring data through data collection, and sends the original building environment monitoring data to the communication processing twin layer;

[0007] The communication processing twin layer is used for data transmission processing and data storage management. Through data transmission processing, building environment monitoring optimization data is obtained, and data intercommunication is achieved by constructing a data transmission unit. Through data storage management, classified and hierarchical data storage is performed to obtain storage data category classification information and data hierarchical index information, thereby achieving data storage optimization.

[0008] The analysis modeling prediction layer is used for building environment monitoring index prediction, obtains building environment monitoring feedback data information through building environment monitoring index prediction, and sends the building environment monitoring feedback data information to the building environment monitoring application layer;

[0009] The building environment monitoring application layer is used for building environment digital twin monitoring. Through the building environment digital twin monitoring, a comprehensive building environment digital twin monitoring solution is obtained.

[0010] Furthermore, the data collection is used to collect basic raw data required for building environment monitoring, specifically by setting up environmental sensors and occupancy detection sensors, and collecting information from security equipment to obtain raw data for building environment monitoring;

[0011] The building environment monitoring raw data specifically includes environmental index data, occupancy index data and personnel circulation index data;

[0012] The environmental index data specifically includes air humidity data, temperature data, light intensity data and air composition data;

[0013] The occupancy index data specifically includes infrared sensor data, acoustic sensor data and numerical representation data;

[0014] The personnel circulation index data specifically includes personnel entry and exit data, personnel stay data and circulation status data.

[0015] Furthermore, the data transmission processing is used to perform processing optimization and preliminary analysis on the original data and provide a data transmission channel, specifically to pre-process and pre-analyze the original building environment monitoring data to obtain optimized building environment monitoring data, and to construct a data transmission unit to perform monitoring system data transmission and communication;

[0016] The data transmission unit specifically includes a communication protocol and a communication device;

[0017] The communication protocol specifically uses a low-power wide area network combined with a standard protocol for data communication;

[0018] The communication equipment specifically uses routers and switches to perform network communication.

[0019] Furthermore, the data storage management is used to efficiently process and manage the optimization data, specifically receiving the building environment monitoring optimization data, adopting a classified and hierarchical data storage management method, optimizing the building environment monitoring data storage management, and constructing a storage management unit;

[0020] The classified and hierarchical data storage management method specifically includes a data classification unit, a data hierarchical unit, a data storage unit and a storage management unit;

[0021] The data classification unit specifically uses a U-shaped network combined with a convolutional long short-term memory module to perform data classification;

[0022] The storage management unit is used to store data for building environment monitoring index prediction and send the data to the monitoring application module through the communication processing module;

[0023] The steps of adopting the classified and hierarchical data storage management method to optimize the building environment monitoring data storage management and constructing a storage management unit include:

[0024] Multi-task learning data classification, specifically constructing a U-shaped network combined with a convolutional long short-term memory module, and performing multi-task learning data classification to obtain storage data category classification information, and using the storage data category classification information as a data index to obtain data hierarchical index information, specifically includes the following steps:

[0025] Constructing a convolutional long short-term memory module includes the following steps:

[0026] Construct an input splicing mechanism to splice the data input and the hidden state of the previous time step in the channel dimension. Specifically, construct a seven-layer splicing dimension and connect seven layers of convolutional layers, long short-term memory layers, and deconvolution layers in sequence. The calculation formula for the input splicing is:

[0027] ;

[0028] In the formula, Z t is the concatenated data tensor, which is used to represent the concatenated value of the data input corresponding to time t and the hidden state vector at the previous time. t is the data input corresponding to time t, is the hidden state vector at the previous moment, where l is the layer index and t is the time index;

[0029] Construct a long short-term memory gating mechanism, specifically construct the forget gate, input gate and output gate in sequence, and construct a basic long short-term memory subnet structure. The calculation formula is:

[0030] ;

[0031] In the formula, F t is the output of the forget gate, I t is the input gate output, O t is the output of the output gate, t is the time index, sig(·) is the S-type activation function, W F is the forget gate weight, W I is the input gate weight, W O is the output gate weight, b F is the forget gate bias term, b I is the input gate bias term, b O is the output gate bias term, Z t is the concatenated data tensor, * is the convolution operator;

[0032] Calculate the long short-term memory cell state update and hidden state update, the calculation formula is:

[0033] ;

[0034] In the formula, C t is the state of the long short-term memory cell at the current time t, F t is the output of the forget gate, C t-1 is the state of the long short-term memory cell at the previous moment t-1, I t is the input gate output, is element-wise multiplication, tanh(·) is the hyperbolic tangent activation function, and W C is the cell state weight, b C is the cell state bias term, Z t is the concatenated data tensor, H t is the hidden state at the current time t, O t is the output gate output;

[0035] Time series data processing, specifically, stacking seven levels of convolutional long short-term memory blocks to process time series data and update cell states and hidden states;

[0036] Building a U-shaped network includes the following steps:

[0037] Constructing the encoding side subnet, specifically, performing seven-level recursive feature extraction through convolution operations and pooling operations to construct the encoding side subnet;

[0038] Construct a decoding side subnet, specifically, restore seven levels of features through deconvolution operations and skip connections to construct a decoding side subnet;

[0039] Subnet U-type integration, specifically, taking the last layer output of the encoding side subnet as the input of the convolutional long short-term memory module, and processing the convolutional long short-term memory module to obtain encoding features, and decoding the encoding features through the decoding side subnet to obtain the U-type network prediction output;

[0040] Multi-task learning data classification model training, through the construction of convolutional long short-term memory module and the construction of U-type network, and the construction of multi-task loss function, multi-task learning model training is performed to obtain the data storage classification model Model DC ;

[0041] The multi-task learning data classification tasks specifically include environmental indicator recognition and classification tasks, occupancy indicator recognition and classification tasks, and personnel flow indicator recognition and classification tasks;

[0042] The multi-task loss function is constructed, specifically refers to a non-deterministic weighted loss function, and the calculation formula is:

[0043] ;

[0044] Where, L S is a multi-task loss function, n is the total number of tasks, the specific value is 3, i is the task index, is a non-deterministic weighting factor, which is obtained by the automatic differentiation effect of the back propagation of the neural network. i is the loss function corresponding to the i-th task, log is the natural logarithm function, is the non-deterministic weighting factor The cumulative multiplication operation of

[0045] Multi-task learning data classification, specifically using the data storage classification model Model DC , classifying the building environment monitoring optimization data to obtain storage data category classification information;

[0046] The stored data category classification information specifically includes building environment index change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data;

[0047] Data stratification, specifically, performing data stratification according to the stored data category classification information to obtain data stratification index information;

[0048] The data layered index information specifically includes a high-frequency read-write layer, a medium-frequency read-write layer, and a low-frequency read-write layer;

[0049] Data storage, specifically, data classification and hierarchical storage according to the data hierarchical index information and the stored data category classification information;

[0050] Storage management, specifically, constructing the storage management unit through the multi-task learning data classification, the data stratification and the data storage, and performing data classification and stratification management on the data in the building environment monitoring optimization data through the storage management unit to obtain data stratification index information, and the data stratification index information is used to query the optimized environment index data, optimized occupancy index data or optimized personnel circulation index data in the storage.

[0051] Furthermore, the building environment monitoring index prediction is used to sequentially predict building environment index changes, building internal occupancy rate prediction and building personnel flow prediction, specifically, based on the data hierarchical index information and reading the building environment index change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data in the storage management unit, and reading the building environment monitoring optimization data, using a multi-task spatiotemporal convolutional network model, to perform three-way building environment monitoring prediction, and obtain building environment monitoring feedback data information;

[0052] The multi-task spatiotemporal convolutional network model specifically includes an input layer, a shared layer and a task branch layer;

[0053] The input layer is used to receive the overall data input;

[0054] The shared layer is used to construct a one-dimensional convolution to extract shared features;

[0055] The task branching layer is used to create dedicated branches for the three prediction tasks and execute the prediction tasks;

[0056] The step of using a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction and obtain building environment monitoring feedback data information includes:

[0057] Constructing an input layer, specifically, reading the building environment index change prediction data, the building internal occupancy prediction data and the building personnel flow prediction data in the storage management unit according to the data hierarchical index information as raw data input samples;

[0058] Constructing a shared layer, specifically constructing a one-dimensional convolution module, and using the one-dimensional convolution module for feature extraction, the one-dimensional convolution module specifically includes causal convolution and dilated convolution, the causal convolution is used to ensure temporal causality, the dilated convolution is used to expand the convolution perception range, the construction calculation formula of the one-dimensional convolution module is:

[0059] ;

[0060] In the formula, is the hidden state of the Lth one-dimensional convolution module at time t, K is the total number of convolution kernels, k is the convolution kernel index, is the weight of the kth convolution kernel corresponding to the Lth one-dimensional convolution module, d k is the dilated convolution coefficient, is the Lth one-dimensional convolution module in td k The hidden state at time b L is the bias term of the Lth one-dimensional convolution module;

[0061] Convolution stacking, specifically, constructing a residual link and activating it with a nonlinear activation function, and constructing a five-layer one-dimensional convolution module stack to obtain a stacked shared convolution layer, and extracting shared features based on the building environment monitoring optimization data through the stacked shared convolution layer to obtain shared feature data;

[0062] The unique feature extraction is specifically to use the stacked shared convolutional layer to extract the unique features of the task according to the building environment index change prediction data, the building internal occupancy rate prediction data and the building personnel flow prediction data, so as to obtain a unique feature data set, wherein the unique feature data set specifically includes building environment index change features, building internal occupancy rate features and building personnel flow features;

[0063] Shared feature integration, specifically, performing feature connection operations on each feature in the shared feature data and the unique feature data set to obtain task-specific shared feature data;

[0064] Construct a task branch layer, specifically construct a three-way task-specific convolution layer, and construct a fully connected layer for each task-specific convolution layer to perform environmental indicator prediction, occupancy prediction, and personnel flow prediction. The specific steps include:

[0065] Construct a task-specific convolutional layer, and the calculation formula is:

[0066] ;

[0067] In the formula, is the hidden state of the Tth task at time t, K is the total number of convolution kernels, k is the convolution kernel index, T is the task index, is the weight value corresponding to the Tth task, The Tth task corresponds to td k The task-specific shared feature data at time d k is the dilated convolution coefficient, b T is the bias term of the Tth task;

[0068] Construct a fully connected layer, the calculation formula is:

[0069] ;

[0070] In the formula, is the predicted output corresponding to the Tth task, is the weight of the fully connected layer corresponding to the Tth task, is the hidden state of the Tth task at time t, is the fully connected layer bias term corresponding to the Tth task;

[0071] Construct a multi-task loss function, specifically using mean square error loss to construct the multi-task loss function;

[0072] The multi-task spatiotemporal convolutional network model training is specifically carried out by constructing the input layer, the shared layer, the convolution stacking, the unique feature extraction, the shared feature integration, the task branch layer and the fully connected layer, according to the multi-task loss function, and the grid search algorithm is used for parameter optimization to train the multi-task spatiotemporal convolutional network model, so as to obtain the building environment monitoring index prediction model Model GP ;

[0073] Building environment monitoring index prediction, specifically using the building environment monitoring index prediction model Model GP , based on the building environment monitoring optimization data, the building environment monitoring index is predicted to obtain the building environment monitoring feedback data information.

[0074] Furthermore, the building environment digital twin monitoring is used to perform three-way monitoring of the building environment, specifically, by obtaining the feedback data information of the building environment monitoring, performing comprehensive monitoring of the building environment digital twin, and obtaining a comprehensive monitoring plan of the building environment digital twin;

[0075] The said comprehensive monitoring of building environment digital twins specifically includes early warning monitoring of environmental changes, monitoring of occupancy changes and monitoring of personnel flow;

[0076] The environmental change early warning monitoring specifically refers to setting environmental indicator thresholds, providing early warning feedback when the predicted value reaches the threshold or falls too far below the threshold, and adjusting the temperature, humidity and air quality through the ventilation system and air conditioning system;

[0077] The occupancy change monitoring specifically refers to optimizing and adjusting specific spaces through occupancy rate prediction and occupancy density prediction;

[0078] The personnel flow monitoring specifically refers to setting a threshold for human flow, issuing a congestion warning and notification when the predicted value reaches the threshold, and optimizing access control and evacuation management through the predicted path for human flow.

[0079] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0080] (1) In the existing digital twin monitoring system of building environment, the representation of building environment is complex and difficult, often involving multiple data indicators and prediction tasks. The reading and writing frequencies of data for different data indicators and tasks are different, and the specific content of the required data is also different. The use of a fixed single data storage method is likely to waste storage resources and reduce system efficiency. This solution creatively adopts a U-shaped network combined with a convolutional long short-term memory module and a multi-task learning data classification method to classify and hierarchically store data. By identifying data based on different task data and storing them at different levels according to the reading and writing frequencies, the overall operation efficiency of the system is improved, and a good data foundation is provided for subsequent prediction tasks.

[0081] (2) In the existing digital twin monitoring system of building environment, the traditional method adopts discrete and independent methods to predict the tasks of environmental indicators, environmental occupancy and personnel circulation when conducting environmental monitoring. This leads to the lack of connectivity between the specific tasks of building environment monitoring, ignores the potential data connection and thus reduces the technical problem of the availability of prediction. This solution creatively adopts a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction. Through data classification storage in the data storage layer and combined with the shared feature extraction of the multi-task spatiotemporal convolutional network, the unique features of the tasks and the shared features are connected, so as to obtain the potential connection between the three tasks, which improves the overall system prediction performance and monitoring efficiency, and improves the overall availability of the system.

[0082] (3) In the existing digital twin monitoring system of building environment, there are predictions for specific building environment indicators. The traditional method cannot form an effective linkage monitoring plan due to independent predictions, which reduces the technical problem of poor availability and controllability of the digital twin monitoring system when applied to the building environment. This solution creatively adopts the comprehensive monitoring of the digital twin of the building environment to obtain a comprehensive monitoring plan for the digital twin of the building environment. By designing the subsequent application of the prediction data of the three tasks, the application layer of the digital twin system is improved, and the unification of environmental management, space management and personnel safety management is realized, which provides a feasible technical reference for the further optimization of the digital twin monitoring system of the building environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A schematic diagram of the structure of a building environment digital twin monitoring system provided by the present invention;

[0084] Figure 2 A schematic flow chart of steps executed by a building environment digital twin monitoring system provided by the present invention;

[0085] Figure 3A flowchart of the steps performed for data transmission processing in the communication processing twin layer;

[0086] Figure 4 A flow chart of the steps performed for data storage management in the communication processing twin layer;

[0087] Figure 5 Flow diagram of the steps performed to model the prediction layer for analytics.

[0088] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0090] Example 1, see Figure 1 and Figure 2 , the invention provides a building environment digital twin monitoring system, including an environment perception physical layer, a communication processing twin layer, an analysis modeling prediction layer and a building environment monitoring application layer;

[0091] The environment sensing physical layer is used for data collection, and obtains the original building environment monitoring data through data collection, and sends the original building environment monitoring data to the communication processing twin layer;

[0092] The communication processing twin layer is used for data transmission processing and data storage management. Through data transmission processing, building environment monitoring optimization data is obtained, and data intercommunication is achieved by constructing a data transmission unit. Through data storage management, classified and hierarchical data storage is performed to obtain storage data category classification information and data hierarchical index information, thereby achieving data storage optimization.

[0093] The analysis modeling prediction layer is used for building environment monitoring index prediction, obtains building environment monitoring feedback data information through building environment monitoring index prediction, and sends the building environment monitoring feedback data information to the building environment monitoring application layer;

[0094] The building environment monitoring application layer is used for building environment digital twin monitoring. Through the building environment digital twin monitoring, a comprehensive building environment digital twin monitoring solution is obtained.

[0095] By performing the above operations, in view of the fact that in the existing digital twin monitoring system of building environment, there are predictions for specific building environment indicators. The traditional method cannot form an effective linkage monitoring plan due to independent predictions, thereby reducing the technical problem of poor availability and controllability of the digital twin monitoring system when applied to the building environment. This solution creatively adopts the comprehensive monitoring of the digital twin of the building environment to obtain a comprehensive solution for the digital twin monitoring of the building environment. By designing the subsequent application of the prediction data of the three tasks, the application layer of the digital twin system is improved, and the unification of environmental management, space management and personnel safety management is realized, which provides a feasible technical reference for the further optimization of the digital twin monitoring system of the building environment.

[0096] Embodiment 2, this embodiment is based on the above embodiment, see Figure 1 and Figure 2 , the data collection is used to collect the basic raw data required for building environment monitoring, specifically by setting up environmental sensors and occupancy detection sensors, and collecting information from security equipment to obtain the raw data for building environment monitoring;

[0097] The building environment monitoring raw data specifically includes environmental index data, occupancy index data and personnel circulation index data;

[0098] The environmental index data specifically includes air humidity data, temperature data, light intensity data and air composition data;

[0099] The occupancy index data specifically includes infrared sensor data, acoustic sensor data and numerical representation data;

[0100] The personnel circulation index data specifically includes personnel entry and exit data, personnel stay data and circulation status data.

[0101] Embodiment 3, this embodiment is based on the above embodiment, see Figure 2 and Figure 3 The data transmission processing is used to perform processing optimization and preliminary analysis on the original data and provide a data transmission channel, specifically to pre-process and preliminarily analyze the original building environment monitoring data to obtain optimized building environment monitoring data, and to construct a data transmission unit to perform data transmission and communication of the monitoring system;

[0102] The data transmission unit specifically includes a communication protocol and a communication device;

[0103] The communication protocol specifically uses a low-power wide area network combined with a standard protocol for data communication;

[0104] The communication equipment specifically uses routers and switches to perform network communication;

[0105] The step of preprocessing and preliminarily analyzing the raw building environment monitoring data to obtain optimized building environment monitoring data includes:

[0106] Environmental indicator data processing, specifically, preprocessing the environmental indicator data in the raw data of building environment monitoring, including data cleaning, sensor data correction and data denoising operations, to obtain optimized environmental indicator data;

[0107] Occupancy index data processing, specifically, performing data preprocessing on the occupancy index data in the raw building environment monitoring data, including sensor data correction and signal denoising operations, to obtain optimized occupancy index data;

[0108] Processing personnel circulation index data, specifically, performing data cleaning operations on the personnel circulation index data in the original building environment monitoring data to obtain optimized personnel circulation index data;

[0109] Aggregating sensor data, specifically aggregating sensor data in the optimized environment index data and the optimized occupancy index data to obtain an aggregated environment sensor data set;

[0110] Preliminary analysis of personnel circulation index data, specifically, basic data analysis of the numerical data in the optimized occupancy index data and the optimized personnel circulation index data, including calculation of the data mean, median, variance, standard deviation and Z score, to construct an environmental attribute analysis data set;

[0111] Data integration and optimization, specifically, by performing overall data optimization processing through the environmental index data processing, occupancy index data processing, personnel circulation index data processing, sensor data aggregation and preliminary analysis of personnel circulation index data, to obtain building environment monitoring optimization data;

[0112] The building environment monitoring optimization data specifically includes optimized environment index data, optimized occupancy index data and optimized personnel circulation index data in terms of content, and specifically includes aggregated environment sensor data set and environment attribute analysis data set in terms of structure;

[0113] Data transmission, specifically, by constructing a data transmission unit, transmits the building environment monitoring optimization data for data storage management and building environment monitoring indicator prediction.

[0114] Embodiment 4, this embodiment is based on the above embodiment, see Figure 2 and Figure 4 The data storage management is used to efficiently process and manage the optimization data, specifically receiving the building environment monitoring optimization data, adopting a classified and hierarchical data storage management method, optimizing the building environment monitoring data storage management, and constructing a storage management unit;

[0115] The classified and hierarchical data storage management method specifically includes a data classification unit, a data hierarchical unit, a data storage unit and a storage management unit;

[0116] The data classification unit specifically uses a U-shaped network combined with a convolutional long short-term memory module to perform data classification;

[0117] The storage management unit is used to store data for building environment monitoring index prediction and send the data to the monitoring application module through the communication processing module;

[0118] The steps of adopting the classified and hierarchical data storage management method to optimize the building environment monitoring data storage management and constructing a storage management unit include:

[0119] Multi-task learning data classification, specifically constructing a U-shaped network combined with a convolutional long short-term memory module, and performing multi-task learning data classification to obtain storage data category classification information, and using the storage data category classification information as a data index to obtain data hierarchical index information, specifically includes the following steps:

[0120] Constructing a convolutional long short-term memory module includes the following steps:

[0121] Construct an input splicing mechanism to splice the data input and the hidden state of the previous time step in the channel dimension. Specifically, construct a seven-layer splicing dimension and connect seven layers of convolutional layers, long short-term memory layers, and deconvolution layers in sequence. The calculation formula for the input splicing is:

[0122] ;

[0123] In the formula, Z t is the concatenated data tensor, which is used to represent the concatenated value of the data input corresponding to time t and the hidden state vector at the previous time. t is the data input corresponding to time t, is the hidden state vector at the previous moment, where l is the layer index and t is the time index;

[0124] Construct a long short-term memory gating mechanism, specifically construct the forget gate, input gate and output gate in sequence, and construct a basic long short-term memory subnet structure. The calculation formula is:

[0125] ;

[0126] In the formula, F t is the output of the forget gate, I t is the input gate output, O t is the output of the output gate, t is the time index, sig(·) is the S-type activation function, W F is the forget gate weight, WI is the input gate weight, W O is the output gate weight, b F is the forget gate bias term, b I is the input gate bias term, b O is the output gate bias term, Z t is the concatenated data tensor, * is the convolution operator;

[0127] Calculate the long short-term memory cell state update and hidden state update, the calculation formula is:

[0128] ;

[0129] In the formula, C t is the state of the long short-term memory cell at the current time t, F t is the output of the forget gate, C t-1 is the state of the long short-term memory cell at the previous moment t-1, I t is the input gate output, is element-wise multiplication, tanh(·) is the hyperbolic tangent activation function, and W C is the cell state weight, b C is the cell state bias term, Z t is the concatenated data tensor, H t is the hidden state at the current time t, O t is the output gate output;

[0130] Time series data processing, specifically, stacking seven levels of convolutional long short-term memory blocks to process time series data and update cell states and hidden states;

[0131] Building a U-shaped network includes the following steps:

[0132] Constructing the encoding side subnet, specifically, performing seven-level recursive feature extraction through convolution operations and pooling operations to construct the encoding side subnet;

[0133] The calculation formula of the convolution operation is:

[0134] Y=sig(W*X+b);

[0135] Where Y is the convolution output, sig(·) is the S-type activation function, W is the convolution kernel weight, X is the convolution input, and b is the convolution kernel bias term;

[0136] The calculation formula of the pooling operation is:

[0137] Y pool =MaxPool(Y);

[0138] Where Y poolis the pooling output, MaxPool(·) is the maximum pooling function, and Y is the convolution output;

[0139] Construct a decoding side subnet, specifically, restore seven levels of features through deconvolution operations and skip connections to construct a decoding side subnet;

[0140] The calculation formula of the deconvolution operation is:

[0141] Y up =TransposeConv(X);

[0142] Where Y up is the deconvolution output, TransposeConv(·) is the transposed convolution operation function, and X is the convolution input;

[0143] The calculation formula of the jump connection is:

[0144] Y cat =[Y up ,Y skip ];

[0145] Where Y cat is the output feature after concatenation, Y up is the deconvolution output, Y skip is the feature of the corresponding layer on the encoding side;

[0146] Subnet U-type integration, specifically, taking the last layer output of the encoding side subnet as the input of the convolutional long short-term memory module, and processing the convolutional long short-term memory module to obtain encoding features, and decoding the encoding features through the decoding side subnet to obtain the U-type network prediction output;

[0147] Multi-task learning data classification model training, through the construction of convolutional long short-term memory module and the construction of U-type network, and the construction of multi-task loss function, multi-task learning model training is performed to obtain the data storage classification model Model DC ;

[0148] The multi-task learning data classification tasks specifically include environmental indicator recognition and classification tasks, occupancy indicator recognition and classification tasks, and personnel flow indicator recognition and classification tasks;

[0149] The multi-task loss function is constructed, specifically refers to a non-deterministic weighted loss function, and the calculation formula is:

[0150] ;

[0151] Where, L S is a multi-task loss function, n is the total number of tasks, the specific value is 3, i is the task index, is a non-deterministic weighting factor, which is obtained by the automatic differentiation effect of the back propagation of the neural network. i is the loss function corresponding to the i-th task, log is the natural logarithm function, is the non-deterministic weighting factor The cumulative multiplication operation of

[0152] Multi-task learning data classification, specifically using the data storage classification model Model DC , classifying the building environment monitoring optimization data to obtain storage data category classification information;

[0153] The stored data category classification information specifically includes building environment index change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data;

[0154] The building environment index change prediction data is specifically composed of optimized environment index data in the aggregated environment sensor data set, and is used for high-frequency real-time building environment index change prediction;

[0155] The building internal occupancy rate prediction data is specifically composed of infrared and acoustic sensor data of optimized occupancy index data in the aggregated environmental sensor data set and numerical data of optimized occupancy index data in the environmental attribute analysis data set, and is used for slow-frequency building internal occupancy rate prediction;

[0156] The building personnel flow prediction data is specifically composed of optimized personnel circulation index data in the environmental attribute analysis data set, and is used for medium-frequency building personnel flow prediction;

[0157] Data stratification, specifically, performing data stratification according to the stored data category classification information to obtain data stratification index information;

[0158] The data layered index information specifically includes a high-frequency read-write layer, a medium-frequency read-write layer, and a low-frequency read-write layer;

[0159] The high-frequency read-write layer is used to store building environment index change prediction data;

[0160] The intermediate frequency read-write layer is used to store building personnel flow prediction data;

[0161] The low-frequency read-write layer is used to store the building internal occupancy prediction data;

[0162] Data storage, specifically, data classification and hierarchical storage according to the data hierarchical index information and the stored data category classification information;

[0163] Storage management, specifically, constructing the storage management unit through the multi-task learning data classification, the data stratification and the data storage, and performing data classification and stratification management on the data in the building environment monitoring optimization data through the storage management unit to obtain data stratification index information, and the data stratification index information is used to query the optimized environment index data, optimized occupancy index data or optimized personnel circulation index data in the storage.

[0164] By performing the above operations, in view of the fact that in the existing digital twin monitoring system of the building environment, the representation of the building environment is complex and difficult, often involving multiple data indicators and prediction tasks, and the reading and writing frequencies of data for different data indicators and tasks are different and the specific content of the required data is also different, the use of a fixed and single data storage method is prone to waste storage resources and reduce the efficiency of the system. The technical problem, this scheme creatively adopts a U-shaped network combined with a convolutional long short-term memory module, and performs a multi-task learning data classification method to classify and hierarchically store data. By performing data recognition based on different task data and storing them at different levels according to the reading and writing frequencies, the overall operation efficiency of the system is improved, and a good data foundation is provided for subsequent prediction tasks.

[0165] Embodiment 5, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 5 The building environment monitoring index prediction is used to predict building environment index changes, building internal occupancy rate prediction and building personnel flow prediction in sequence, specifically, according to the data hierarchical index information and reading the building environment index change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data in the storage management unit, and reading the building environment monitoring optimization data, using a multi-task spatiotemporal convolutional network model, to perform three-way building environment monitoring prediction, and obtain building environment monitoring feedback data information;

[0166] The multi-task spatiotemporal convolutional network model specifically includes an input layer, a shared layer and a task branch layer;

[0167] The input layer is used to receive the overall data input;

[0168] The shared layer is used to construct a one-dimensional convolution to extract shared features;

[0169] The task branching layer is used to create dedicated branches for the three prediction tasks and execute the prediction tasks;

[0170] The step of using a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction and obtain building environment monitoring feedback data information includes:

[0171] Constructing an input layer, specifically, reading the building environment index change prediction data, the building internal occupancy prediction data and the building personnel flow prediction data in the storage management unit according to the data hierarchical index information as raw data input samples;

[0172] Constructing a shared layer, specifically constructing a one-dimensional convolution module, and using the one-dimensional convolution module for feature extraction, the one-dimensional convolution module specifically includes causal convolution and dilated convolution, the causal convolution is used to ensure temporal causality, the dilated convolution is used to expand the convolution perception range, the construction calculation formula of the one-dimensional convolution module is:

[0173] ;

[0174] In the formula, is the hidden state of the Lth one-dimensional convolution module at time t, K is the total number of convolution kernels, k is the convolution kernel index, is the weight of the kth convolution kernel corresponding to the Lth one-dimensional convolution module, d k are the dilated convolution coefficients, is the Lth one-dimensional convolution module in td k The hidden state at time b L is the bias term of the Lth one-dimensional convolution module;

[0175] Convolution stacking, specifically, constructing a residual link and activating it with a nonlinear activation function, and constructing a five-layer one-dimensional convolution module stack to obtain a stacked shared convolution layer, and extracting shared features based on the building environment monitoring optimization data through the stacked shared convolution layer to obtain shared feature data;

[0176] The unique feature extraction is specifically to use the stacked shared convolutional layer to extract the unique features of the task according to the building environment index change prediction data, the building internal occupancy rate prediction data and the building personnel flow prediction data, so as to obtain a unique feature data set, wherein the unique feature data set specifically includes building environment index change features, building internal occupancy rate features and building personnel flow features;

[0177] Shared feature integration, specifically, performing feature connection operations on each feature in the shared feature data and the unique feature data set to obtain task-specific shared feature data;

[0178] Construct a task branch layer, specifically construct a three-way task-specific convolution layer, and construct a fully connected layer for each task-specific convolution layer to perform environmental indicator prediction, occupancy prediction, and personnel flow prediction. The specific steps include:

[0179] Construct a task-specific convolutional layer, and the calculation formula is:

[0180] ;

[0181] In the formula, is the hidden state of the Tth task at time t, K is the total number of convolution kernels, k is the convolution kernel index, T is the task index, is the weight value corresponding to the Tth task, The Tth task corresponds to td k The task-specific shared feature data at time d k is the dilated convolution coefficient, b T is the bias term of the Tth task;

[0182] Construct a fully connected layer, the calculation formula is:

[0183] ;

[0184] In the formula, is the predicted output corresponding to the Tth task, is the weight of the fully connected layer corresponding to the Tth task, is the hidden state of the Tth task at time t, is the fully connected layer bias term corresponding to the Tth task;

[0185] Construct a multi-task loss function, specifically using mean square error loss to construct the multi-task loss function;

[0186] The multi-task spatiotemporal convolutional network model training is specifically carried out by constructing the input layer, the shared layer, the convolution stacking, the unique feature extraction, the shared feature integration, the task branch layer and the fully connected layer, according to the multi-task loss function, and the grid search algorithm is used for parameter optimization to train the multi-task spatiotemporal convolutional network model, so as to obtain the building environment monitoring index prediction model Model GP ;

[0187] Building environment monitoring index prediction, specifically using the building environment monitoring index prediction model Model GP , based on the building environment monitoring optimization data, predict the building environment monitoring indicators and obtain building environment monitoring feedback data information;

[0188] The building environment monitoring feedback data information specifically includes the predicted value of building environment index change, the predicted value of building internal occupancy rate and the predicted value of building personnel flow;

[0189] The predicted values ​​of changes in building environmental indicators specifically include temperature prediction curves, humidity prediction curves and air quality prediction indexes;

[0190] The predicted value of the building internal occupancy rate specifically includes the overall occupancy rate of each area in the building, the predicted value of the peak occupancy period and the occupancy density change curve;

[0191] The building personnel flow prediction value specifically includes the personnel flow prediction amount and the personnel flow prediction path.

[0192] By performing the above operations, in order to solve the technical problem that in the existing digital twin monitoring system of building environment, the traditional method adopts discrete and independent methods to predict the tasks of environmental indicators, environmental occupancy and personnel circulation when conducting environmental monitoring, which leads to the lack of connectivity between the specific tasks of building environment monitoring, ignores the potential data connection and thus reduces the availability of prediction, this scheme creatively adopts a multi-task spatiotemporal convolutional network model to perform three-way building environment monitoring prediction, through the data classification storage of the data storage layer, combined with the shared feature extraction of the multi-task spatiotemporal convolutional network, the connection between the unique features of the task and the shared features is performed, so as to obtain the potential connection between the three tasks, thereby improving the system prediction performance and monitoring efficiency as a whole, and improving the overall availability of the system.

[0193] Embodiment 6, this embodiment is based on the above embodiment, see Figure 1 and Figure 2 The building environment digital twin monitoring is used to perform three-way monitoring of the building environment, specifically, by obtaining the feedback data information of the building environment monitoring, performing comprehensive monitoring of the building environment digital twin, and obtaining a comprehensive monitoring plan for the building environment digital twin;

[0194] The said comprehensive monitoring of building environment digital twins specifically includes early warning monitoring of environmental changes, monitoring of occupancy changes and monitoring of personnel flow;

[0195] The environmental change early warning monitoring specifically refers to setting environmental indicator thresholds, providing early warning feedback when the predicted value reaches the threshold or falls too far below the threshold, and adjusting the temperature, humidity and air quality through the ventilation system and air conditioning system;

[0196] The occupancy change monitoring specifically refers to optimizing and adjusting specific spaces through occupancy rate prediction and occupancy density prediction;

[0197] The personnel flow monitoring specifically refers to setting a threshold for human flow, issuing a congestion warning and notification when the predicted value reaches the threshold, and optimizing access control and evacuation management through the predicted path for human flow.

[0198] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process and method including a series of elements include not only those elements, but also other elements not explicitly listed, or also include elements inherent to such process and method.

[0199] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0200] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A building environment digital twin monitoring system, characterized by: It includes the environment perception physical layer, the communication processing twin layer, the analysis modeling prediction layer and the building environment monitoring application layer; The environment sensing physical layer is used for data collection, and obtains the original building environment monitoring data through data collection, and sends the original building environment monitoring data to the communication processing twin layer; The communication processing twin layer is used for data transmission processing and data storage management. Through data transmission processing, building environment monitoring optimization data is obtained, and data intercommunication is achieved by constructing a data transmission unit. Through data storage management, classified and hierarchical data storage is performed to obtain storage data category classification information and data hierarchical index information, thereby achieving data storage optimization. The classified and hierarchical data storage specifically adopts a classified and hierarchical data storage management method, including a data classification unit, a data stratification unit, a data storage unit and a storage management unit; The data classification unit specifically uses a U-shaped network combined with a convolutional long short-term memory module to perform data classification; the storage management unit is used to predict and store data for building environment monitoring indicators, and send data to the monitoring application module through the communication processing module; The steps of adopting the classified and hierarchical data storage management method to optimize the data storage management of building environment monitoring and constructing a storage management unit include: multi-task learning data classification, data stratification, data storage and storage management; The U-shaped network combined with the convolutional long short-term memory module is used for multi-task learning data classification, which specifically includes the following steps: constructing a convolutional long short-term memory module, constructing a U-shaped network, multi-task learning data classification model training and multi-task learning data classification; The construction of the U-shaped network specifically includes the following steps: Constructing the encoding side subnet, specifically, performing seven-level recursive feature extraction through convolution operations and pooling operations to construct the encoding side subnet; Construct a decoding side subnet, specifically, restore seven levels of features through deconvolution operations and skip connections to construct a decoding side subnet; Subnet U-type integration, specifically, taking the last layer output of the encoding side subnet as the input of the convolutional long short-term memory module, and processing the convolutional long short-term memory module to obtain encoding features, and decoding the encoding features through the decoding side subnet to obtain the U-type network prediction output; The analysis modeling prediction layer is used for building environment monitoring index prediction. Through the building environment monitoring index prediction, building environment monitoring feedback data information is obtained, and the building environment monitoring feedback data information is sent to the building environment monitoring application layer. Specifically, according to the data hierarchical index information, the building environment indicator change prediction data, the building internal occupancy prediction data and the building personnel flow prediction data in the storage management unit are read, and the building environment monitoring optimization data is read. A multi-task spatiotemporal convolutional network model is used to perform three building environment monitoring predictions to obtain the building environment monitoring feedback data information; The multi-task spatiotemporal convolutional network model specifically includes an input layer, a shared layer and a task branch layer; The building environment monitoring application layer is used for building environment digital twin monitoring. Through the building environment digital twin monitoring, a comprehensive building environment digital twin monitoring solution is obtained.

2. A building environment digital twin monitoring system according to claim 1, characterized in that: The data collection is used to collect the basic raw data required for building environment monitoring, specifically by setting up environmental sensors and occupancy detection sensors, and collecting information from security equipment to obtain the raw data for building environment monitoring; The building environment monitoring raw data specifically includes environmental index data, occupancy index data and personnel circulation index data.

3. A building environment digital twin monitoring system according to claim 2, characterized in that: The data transmission processing is used to perform processing optimization and preliminary analysis on the original data and provide a data transmission channel, specifically to pre-process and pre-analyze the original building environment monitoring data to obtain optimized building environment monitoring data, and to construct a data transmission unit to perform data transmission and communication of the monitoring system; The data transmission unit specifically includes a communication protocol and a communication device; The communication protocol specifically uses a low-power wide area network combined with a standard protocol for data communication; The communication equipment specifically uses routers and switches to carry out network communication.

4. A building environment digital twin monitoring system according to claim 3, characterized in that: The data storage management is used to efficiently process and manage the optimization data, specifically receiving the building environment monitoring optimization data, adopting a classified and hierarchical data storage management method, optimizing the building environment monitoring data storage management, and constructing a storage management unit; The classified and hierarchical data storage management method specifically includes a data classification unit, a data hierarchical unit, a data storage unit and a storage management unit; The data classification unit specifically uses a U-shaped network combined with a convolutional long short-term memory module to perform data classification; The storage management unit is used to store data for building environment monitoring index prediction and send the data to the monitoring application module through the communication processing module; The steps of adopting the classified and hierarchical data storage management method to optimize the building environment monitoring data storage management and constructing a storage management unit include: Multi-task learning data classification, specifically, constructing a U-shaped network combined with a convolutional long short-term memory module, and performing multi-task learning data classification to obtain storage data category classification information, and using the storage data category classification information as a data index to obtain data hierarchical index information; Data stratification, specifically, performing data stratification according to the stored data category classification information to obtain data stratification index information; The data layered index information specifically includes a high-frequency read-write layer, a medium-frequency read-write layer, and a low-frequency read-write layer; Data storage, specifically, data classification and hierarchical storage according to the data hierarchical index information and the stored data category classification information; Storage management, specifically, constructing the storage management unit through the multi-task learning data classification, the data stratification and the data storage, and performing data classification and stratification management on the data in the building environment monitoring optimization data through the storage management unit to obtain data stratification index information, and the data stratification index information is used to query the optimized environment index data, optimized occupancy index data or optimized personnel circulation index data in the storage.

5. A building environment digital twin monitoring system according to claim 4, characterized in that: The multi-task learning data classification specifically includes the following steps: constructing a convolutional long short-term memory module, constructing a U-shaped network, multi-task learning data classification model training and multi-task learning data classification; The construction of the convolutional long short-term memory module specifically includes the following steps: Construct an input splicing mechanism to splice the data input and the hidden state of the previous time step in the channel dimension. Specifically, construct a seven-layer splicing dimension and connect seven layers of convolutional layers, long short-term memory layers, and deconvolution layers in sequence. The calculation formula for the input splicing is: In the formula, Z t is the concatenated data tensor, which is used to represent the concatenated value of the data input corresponding to time t and the hidden state vector at the previous time. t is the data input corresponding to time t, is the hidden state vector at the previous moment, where l is the layer index and t is the time index; Construct a long short-term memory gating mechanism, specifically construct the forget gate, input gate and output gate in sequence, and construct a basic long short-term memory subnet structure. The calculation formula is: In the formula, F t is the output of the forget gate, I t is the input gate output, O t is the output of the output gate, t is the time index, sig(·) is the S-type activation function, W F is the forget gate weight, W I is the input gate weight, W O is the output gate weight, b F is the forget gate bias term, b I is the input gate bias term, b O is the output gate bias term, Z t is the concatenated data tensor, * is the convolution operator; Calculate the long short-term memory cell state update and hidden state update, the calculation formula is: In the formula, C t is the state of the long short-term memory cell at the current time t, F t is the output of the forget gate, C t-1 is the state of the long short-term memory cell at the previous moment t-1, I t is the input gate output, ⊙ is the element-wise multiplication, tanh(·) is the hyperbolic tangent activation function, W C is the cell state weight, b C is the cell state bias term, Z t is the concatenated data tensor, H t is the hidden state at the current time t, O t is the output gate output; Time series data processing, specifically, stacking seven levels of convolutional long short-term memory blocks to process time series data and update cell states and hidden states; The multi-task learning data classification model training is performed by constructing the convolutional long short-term memory module and the U-shaped network, and constructing a multi-task loss function to perform multi-task learning model training to obtain a data storage classification model Model DC ; The multi-task learning data classification tasks specifically include environmental indicator recognition and classification tasks, occupancy indicator recognition and classification tasks, and personnel flow indicator recognition and classification tasks; The multi-task loss function is constructed, specifically refers to a non-deterministic weighted loss function, and the calculation formula is: Where, L S is the multi-task loss function, n is the total number of tasks, the specific value is 3, i is the task index, σ i is a non-deterministic weighting factor, which is obtained by the automatic differentiation effect of the back propagation of the neural network. i is the loss function corresponding to the i-th task, log is the natural logarithm function, is the non-deterministic weighting factor σ i The cumulative multiplication operation of The multi-task learning data classification is specifically to use the data storage classification model Model DC , classifying the building environment monitoring optimization data to obtain storage data category classification information; The stored data category classification information specifically includes building environment indicator change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data.

6. A building environment digital twin monitoring system according to claim 5, characterized in that: The building environment monitoring index prediction is used to sequentially predict building environment index changes, building internal occupancy rate prediction and building personnel flow prediction, specifically, according to the data hierarchical index information and reading the building environment index change prediction data, building internal occupancy rate prediction data and building personnel flow prediction data in the storage management unit, and reading the building environment monitoring optimization data, using a multi-task spatiotemporal convolutional network model, to perform three building environment monitoring predictions, and obtain building environment monitoring feedback data information; The multi-task spatiotemporal convolutional network model specifically includes an input layer, a shared layer and a task branch layer; The input layer is used to receive the overall data input; The shared layer is used to construct a one-dimensional convolution to extract shared features; The task branching layer is used to create dedicated branches for the three prediction tasks and execute the prediction tasks.

7. A building environment digital twin monitoring system according to claim 6, characterized in that: The steps of using a multi-task spatiotemporal convolutional network model to perform three building environment monitoring predictions and obtain building environment monitoring feedback data information include: constructing an input layer, constructing a shared layer, convolution stacking, extracting unique features, integrating shared features, constructing a task branch layer, constructing a fully connected layer, constructing a multi-task loss function, training a multi-task spatiotemporal convolutional network model, and predicting building environment monitoring indicators; The constructing of the input layer specifically involves reading the building environment index change prediction data, the building internal occupancy prediction data and the building personnel flow prediction data in the storage management unit according to the data hierarchical index information as the original data input samples; The construction of the shared layer is specifically to construct a one-dimensional convolution module, and use the one-dimensional convolution module to extract features. The one-dimensional convolution module specifically includes causal convolution and dilated convolution. The causal convolution is used to ensure temporal causality, and the dilated convolution is used to expand the convolution perception range. The construction calculation formula of the one-dimensional convolution module is: In the formula, is the hidden state of the Lth one-dimensional convolution module at time t, K is the total number of convolution kernels, k is the convolution kernel index, is the weight of the kth convolution kernel corresponding to the Lth one-dimensional convolution module, d k is the dilated convolution coefficient, is the Lth one-dimensional convolution module in td k The hidden state at time b L is the bias term of the Lth one-dimensional convolution module; The convolution stacking is specifically to construct a residual link and activate it with a nonlinear activation function, and to construct a five-layer one-dimensional convolution module stack to obtain a stacked shared convolution layer, and to extract shared features based on the building environment monitoring optimization data through the stacked shared convolution layer to obtain shared feature data; The unique feature extraction is specifically to use the stacked shared convolutional layer to extract task-specific features based on the building environment index change prediction data, the building internal occupancy rate prediction data and the building personnel flow prediction data, respectively, to obtain a unique feature data set, wherein the unique feature data set specifically includes building environment index change features, building internal occupancy rate features and building personnel flow features; The shared feature integration is specifically to perform feature connection operations on each feature in the shared feature data and the unique feature data set to obtain the shared feature data for a specific task; The construction of the task branch layer specifically constructs three specific task convolution layers, and constructs a fully connected layer for each specific task convolution layer to perform environmental indicator prediction, occupancy prediction, and personnel flow prediction, specifically including the following steps: Construct a task-specific convolutional layer, and the calculation formula is: In the formula, is the hidden state of the Tth task at time t, K is the total number of convolution kernels, k is the convolution kernel index, T is the task index, is the weight value corresponding to the Tth task, The Tth task corresponds to td k The task-specific shared feature data at time d k is the dilated convolution coefficient, b T is the bias term of the Tth task; Construct a fully connected layer, the calculation formula is: In the formula, is the predicted output corresponding to the Tth task, is the weight of the fully connected layer corresponding to the Tth task, is the hidden state of the Tth task at time t, is the fully connected layer bias term corresponding to the Tth task; Construct a multi-task loss function, specifically using mean square error loss to construct the multi-task loss function; The multi-task spatiotemporal convolutional network model training is specifically carried out by constructing the input layer, the shared layer, the convolution stacking, the unique feature extraction, the shared feature integration, the task branch layer and the fully connected layer, according to the multi-task loss function, and using the grid search algorithm to optimize the parameters, and the multi-task spatiotemporal convolutional network model training is carried out to obtain the building environment monitoring index prediction model Model GP ; The building environment monitoring index prediction is specifically to use the building environment monitoring index prediction model Model GP , based on the building environment monitoring optimization data, predict the building environment monitoring indicators and obtain building environment monitoring feedback data information; The building environment monitoring feedback data information specifically includes the predicted value of building environment index change, the predicted value of building internal occupancy rate and the predicted value of building personnel flow.

8. A building environment digital twin monitoring system according to claim 7, characterized in that: The building environment digital twin monitoring is used to perform three monitoring of the building environment, specifically, by obtaining the feedback data information of the building environment monitoring, performing comprehensive monitoring of the building environment digital twin, and obtaining a comprehensive monitoring plan of the building environment digital twin; The said comprehensive monitoring of building environment digital twins specifically includes early warning monitoring of environmental changes, monitoring of occupancy changes and monitoring of personnel flow; The environmental change early warning monitoring specifically refers to setting environmental indicator thresholds, providing early warning feedback when the predicted value reaches the threshold or falls too far below the threshold, and adjusting the temperature, humidity and air quality through the ventilation system and air conditioning system; The occupancy change monitoring specifically refers to optimizing and adjusting specific spaces through occupancy rate prediction and occupancy density prediction; The personnel flow monitoring specifically refers to setting a threshold for human flow, issuing a congestion warning and notification when the predicted value reaches the threshold, and optimizing access control and evacuation management through the predicted path for human flow.

Images