Method and system for applying DNN and LSTM model in intelligent building control system

Abstract

The invention discloses an application method and system in an intelligent building control system in combination with a DNN and LSTM model, and relates to the technical field of intelligent building control and artificial intelligence fusion, and the method comprises the steps: building a spatial digital twinborn system based on a BIM model, extracting a building physical structure and equipment layout relation, employing multi-dimensional input information, and carrying out the building physical structure and equipment layout relation; constructing a space static parameter code oriented to intelligent control; constructing a multi-source heterogeneous sensing channel driven by the IoT middle station, fusing dynamic environment information and space static parameter coding, and constructing multi-dimensional time sequence input by adopting a fixed time interval and a unified timestamp mechanism; constructing a multilayer neural inference network with heterogeneous fusion of LSTM, Attention and DNN, fusing time sequence modeling and feature mapping capabilities, and constructing a data closed-loop architecture of building information and multi-device linkage; according to the method, fusion modeling of the space structure and the time sequence data is realized, and the multi-device linkage control precision and the self-adaptive capability of the intelligent building are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent building control and artificial intelligence fusion technology, and specifically to an application method and system combining DNN and LSTM models in intelligent building control systems. Background Art

[0002] With the deepening of the concept of intelligent buildings and the continuous improvement of high-quality living standards for "good houses," building environmental control systems based on information technology, sensing systems, control systems, and artificial intelligence are gradually becoming part of the infrastructure of intelligent buildings. Especially in the context of the interconnectedness of multiple subsystems such as building air conditioning, fresh air, lighting, sunshades, and floor heating, how to achieve refined, coordinated, and predictive automatic control has become a key issue in the development of smart buildings. However, current intelligent building control systems still face significant technical shortcomings in structural modeling, data fusion, feature extraction, and dynamic decision-making.

[0003] First, existing technologies generally lack structured representation of building spatial information. While some projects have incorporated BIM models for design assistance and collaborative management during the construction phase, these models often fail to effectively connect with IoT control systems during the building's operational phase, failing to establish a real-time responsive relationship between the building's spatial structure and environmental controls. In particular, static information such as building equipment layout, spatial attributes, and orientation lacks a quantifiable encoding mechanism to support algorithmic model input. This makes it difficult for AI control systems to understand the contextual structure between spatial layout and equipment interaction.

[0004] Secondly, although a large number of sensors have been deployed at the IoT data level, the data upload rhythm is inconsistent, time tags are missing, and device coding is not standardized, making it difficult to effectively integrate multi-source heterogeneous data. Most building IoT platforms still use single-point collection and flat display methods for data integration. They lack system-level time series construction logic and find it difficult to form a high-dimensional time series input structure for intelligent algorithms. At the same time, most of the current mainstream control systems for air conditioning, fresh air, lighting, etc. rely on a "single variable + current state" control strategy, that is, they only adjust according to a current indicator (such as temperature). They fail to consider the changing trends and interactions of indoor temperature, humidity, light and other indicators within a certain time range, and also ignore the cumulative effect and delayed response of comfort during the user's long-term experience.

[0005] Thirdly, at the level of AI algorithm integration, although some building control systems have introduced models such as deep neural networks (DNNs) or long short-term memory networks (LSTMs) for single-point prediction, two core issues remain: limited ability to extract features from multidimensional environmental data, particularly the lack of a scalable input mechanism for fusion encoding of different types of data; and simple sliding window processing of time series data, failing to deeply explore the delayed impact of historical states on current decisions. In this context, relying solely on a single model fails to fully reflect the dynamic scenarios, complex relationships, and multidimensional dependencies required for building control. This limits the level of intelligence and often leads to control strategies experiencing delayed response, improper linkage, and significant differences in user perception. For example, in traditional air conditioning systems, the distributed perception of some users often feels hot while others feel cold, a typical manifestation of the lack of temporal scale judgment and spatial perception encoding.

[0006] Furthermore, judging by existing technical literature and commercial products, the implementation of hybrid neural networks integrating DNN and LSTM structures in intelligent building control scenarios is still in its early stages of exploration, lacking systematic supporting mechanisms such as data input modeling, spatial information structured design, time series consistency establishment, and device instruction mapping. Therefore, establishing a closed-loop AI control process from spatial structure modeling, data perception fusion, to intelligent control decision-making remains a key challenge in current smart building technology research and deployment. Summary of the Invention

[0007] In view of the above-mentioned problems, the present invention is proposed.

[0008] Therefore, the technical problem solved by the present invention is: the existing intelligent building control methods have the problems of difficulty in structured input of spatial structure information, asynchronous dispersion of environmental perception data, weak control model timing modeling capability, and low multi-system linkage coupling, resulting in low control accuracy, untimely response, and coarse comfort control granularity. It also solves the problem of how to realize spatial semantic modeling, multi-source perception fusion, timing feature learning and equipment control closed-loop coordination under a unified framework.

[0009] To solve the above technical problems, the present invention provides the following technical solutions: a method for applying DNN and LSTM models in intelligent building control systems, including constructing a spatial digital twin system based on the BIM model, extracting the relationship between the building's physical structure and equipment layout, and using multi-dimensional input information to construct spatial static parameter coding for intelligent control; constructing a multi-source heterogeneous perception channel driven by the IoT middle platform, integrating dynamic environmental information and spatial static parameter coding, and using a fixed time interval and a unified timestamp mechanism to construct a multi-dimensional time series input; constructing a multi-layer neural inference network that heterogeneously integrates LSTM, Attention, and DNN, integrating time series modeling and feature mapping capabilities, and constructing a data closed-loop architecture for building information and multi-device linkage; using a fixed time interval and a unified timestamp mechanism, including setting a standard data collection cycle during the upload process of IoT perception data, and generating aligned time tags for multiple types of sensors; the data closed-loop architecture includes integrating the spatial dimension vectors provided by the BIM model, the real-time perception data collected by the IoT, and the inference results of the AI ​​module, and feeding them back to the spatial intelligent device through the intelligent control interface.

[0010] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the construction of a spatial digital twin system based on the BIM model includes building a 1:1 restored three-dimensional as-built model through the Revit modeling platform, extracting spatial structure information, and embedding the smart devices installed in the building into the corresponding spatial units in the building model according to their position mapping.

[0011] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the construction of spatial static parameter coding for intelligent control includes field normalization and feature encoding of spatial structural elements extracted from the three-dimensional as-built model as the spatial dimension vector of the neural network input layer.

[0012] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the spatial dimension vector serving as the input layer of the neural network includes establishing a corresponding relationship between spatial static parameter encoding and perception data through a spatial index binding mechanism, and synchronously participating in sequence input construction.

[0013] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the construction of a multi-source heterogeneous perception channel driven by the IoT middle platform includes integrating multiple types of sensor equipment deployed in the spatial area through the smart building control platform, setting a unified transmission cycle, and centrally aggregating and scheduling real-time perception data.

[0014] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the construction of multidimensional time series input includes adding time tags to multiple types of sensor data and generating a standardized time series tensor structure according to the device number and sampling step size combination.

[0015] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the construction of a multi-layer neural inference network with heterogeneous fusion of LSTM, Attention and DNN includes inputting the time series tensor into LSTM to extract the hidden state, generating time series features through multi-head attention weighting, and encoding and splicing the time series features and spatial static parameters and then inputting them into DNN to generate a control vector.

[0016] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the fusion of time series modeling and feature mapping capabilities includes encoding time series data and spatial static parameters in a unified format specification, constructing a splicing input structure, and flowing into LSTM and DNN according to preset channels for joint modeling.

[0017] As a preferred solution of the application method of combining DNN and LSTM models in intelligent building control systems described in the present invention, the data closed-loop architecture for building information and multi-device linkage includes mapping DNN outputs into device control instructions, transmitting them to corresponding devices based on a communication interface, and transmitting status feedback back to the perception layer.

[0018] Another object of the present invention is to provide an application system that combines DNN and LSTM models in intelligent building control systems. By constructing a control flow solution that takes spatial static coding and IoT perception data as input and integrates LSTM, Attention mechanism and DNN reasoning structure, it can solve the current problems in intelligent building control systems, such as the lack of collaborative modeling of spatial structure and sensor data, weak timing response capability, and isolated distribution of equipment control instructions, which have poor coupling and insufficient real-time performance.

[0019] As a preferred solution of the application system combining DNN and LSTM models in intelligent building control described in the present invention, it includes: a spatial modeling and static parameter coding module, a multi-source perception fusion and time series construction module, and a neural reasoning and equipment linkage control module; the spatial modeling and static parameter coding module is used to build a digital twin system through the BIM model, extract building structure and equipment layout information, standardize the spatial attributes and encode them, and generate static input fields adapted to the neural network; the multi-source perception fusion and time series construction module is used to collect multiple types of environmental sensor data, complete multi-source alignment through a unified timestamp mechanism, and fuse static coding to generate standardized multi-dimensional time series input; the neural reasoning and equipment linkage control module is used to input spatial static parameter coding and time series data into the fused neural network structure, complete feature modeling and control instruction generation, and link intelligent devices to dynamically respond.

[0020] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a step of an application method combining a DNN and LSTM model in an intelligent building control system.

[0021] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for applying a DNN and LSTM model in an intelligent building control system.

[0022] Beneficial effects of the present invention: The application method of combining DNN and LSTM models in intelligent building control systems provided by the present invention constructs a spatial twin system through BIM, realizes structured input of equipment and structural data, and enhances the spatial perception ability of the control model; constructs a perception channel through a unified sampling mechanism and time tags to realize temporal alignment and standardized input of heterogeneous data; integrates LSTM, Attention and DNN structures to realize collaborative modeling of spatial and temporal features and improve the accuracy of multi-device linkage control. The present invention has achieved better results in semantic modeling of building spatial structures, standardized processing of multi-source heterogeneous perception data, and spatiotemporal feature fusion reasoning for linkage control. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1This is an overall flow chart of a method for applying a DNN and LSTM model in an intelligent building control system, provided in the first embodiment of the present invention.

[0025] Figure 2 An overall schematic diagram of an application system combining DNN and LSTM models in an intelligent building control system provided by the third embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0027] Example 1, reference Figure 1 , as one embodiment of the present invention, provides an application method combining DNN and LSTM models in an intelligent building control system, comprising: S1: Build a spatial digital twin system based on the BIM model, extract the relationship between the building's physical structure and equipment layout, use multi-dimensional input information, and construct spatial static parameter coding for intelligent control.

[0028] Furthermore, building a spatial digital twin system based on the BIM model includes building a 1:1 restored three-dimensional completion model through the Revit modeling platform, extracting spatial structure information, and embedding the smart devices installed in the building into the corresponding spatial units in the building model according to their location mapping.

[0029] Furthermore, the use of multi-dimensional input information includes integrating IoT information of various sensors and equipment through the smart building management platform. For example, through wired and wireless signal transmission, the total operation volume, deployment quantity and fault quantity of specific equipment types such as regional fire water tanks, air quality, lighting and air conditioning are counted respectively. For example, 514 lighting equipment are in operation and the fault number is 0, indicating that the platform has good monitoring coverage in spatial lighting control. It demonstrates the real-time monitoring capability of the smart building control system on the status of various equipment in the building and the actual embodiment of multi-dimensional data input. The sensor information is transmitted to the platform to realize the storage and recording of multi-dimensional information (temperature, humidity, wind speed, light intensity) and time series data, and is associated with the building space structure and equipment operation data to form a complete time series tensor.

[0030] Furthermore, the room units and smart devices in the Revit model have associated properties, and the smart device layout points have corresponding identifiers in the BIM model. Through mapping through the spatial geometric center, boundary conditions and device coordinates, a "spatial node-device entity" correspondence is established to form a spatial semantic map for the control system.

[0031] It should be noted that the construction of spatial static parameter coding for intelligent control includes field normalization and feature coding of the spatial structural elements extracted from the three-dimensional as-built model as the spatial dimension vector of the neural network input layer.

[0032] It should be noted that field normalization and feature encoding include converting spatial area, orientation, and floor height attributes of different dimensions into a standard form ranging from 0 to 1, constructing a vector matrix with a unified structure, and setting each space as a spatial feature vector with a fixed dimension, so as to facilitate subsequent splicing with dynamic data and input into the neural network.

[0033] It should also be noted that the spatial dimension vector serving as the input layer of the neural network includes spatial static parameter encoding and perception data, which establish a corresponding relationship through a spatial index binding mechanism and synchronously participate in the construction of sequence input.

[0034] S2: Build a multi-source heterogeneous perception channel driven by the IoT middle platform, integrate dynamic environmental information and spatial static parameter encoding, use fixed time intervals and a unified timestamp mechanism to construct multidimensional time series input.

[0035] Furthermore, building a multi-source heterogeneous perception channel driven by the IoT middle platform includes integrating multiple types of sensor devices deployed in the spatial area through the smart building control platform, setting a unified transmission cycle, and centrally aggregating and scheduling real-time perception data.

[0036] Furthermore, setting a unified transmission cycle includes setting the time period length to 1-2 hours for monitored data, and the data interval interval is 10 minutes, that is, 6-12 time series data. For the intelligent temperature control system, the parameter range includes the real-time temperature, real-time humidity, and real-time light intensity monitored for the previous 1-2 hours, as well as the size and orientation of the room where the device intelligent body is located.

[0037] It should be noted that constructing a multi-dimensional time series input includes adding time tags to multiple types of sensor data and generating a standardized time series tensor structure according to the combination of device number and sampling step size.

[0038] It should also be noted that by constructing an IoT middle platform driven multi-source heterogeneous sensing channel, unified access and time synchronization processing of multiple types of environmental sensors are realized, a multi-dimensional tensor structure with timing and spatial identification is generated, a unified input vector is constructed by combining spatial static parameter coding, and a structured spatio-temporal fusion data basis is provided for subsequent neural networks, forming an input closed-loop mechanism supporting linkage control.

[0039] S3: Construct a multi-layer neural inference network with LSTM, Attention and DNN heterogeneous fusion, which integrates time series modeling and feature mapping capabilities, and constructs a data closed-loop architecture for building information and multi-device linkage.

[0040] Further, the multi-layer neural inference network with LSTM, Attention and DNN heterogeneous fusion includes inputting the time series tensor into LSTM to extract hidden states, generating time series features through multi-head attention weighting, and inputting the spliced time series features and spatial static parameter coding into DNN to generate control vectors.

[0041] Further, LSTM represents a long short-term memory neural network that can solve the problem of long-term dependence in long time series through a gate control mechanism, and is an effective way to extract time series element features. Attention represents a multi-head attention mechanism, which is represented as: ; ; wherein, is a query vector, an intelligent device information retrieval request at the current time step; is a linear projection weight matrix of the query vector, representing the intelligent device query demand at the current time step during attention retrieval; is the hidden state of the time step , which is the output of the LSTM network modeling the perception data sequence of the th time slice in the intelligent building, representing the comprehensive semantic representation of the time node in the time series features; is a key vector, representing the identity descriptor of the time step in the attention mechanism; is a linear projection weight matrix of the key vector; is a value vector, representing the intelligent device information carrier transmitted to the final representation at the time step; is a linear projection weight matrix of the value vector; is the attention weight of each time step , representing the importance of the time step to the overall input series; is a normalization function that converts the matching scores of all time steps into a probability distribution; is the transpose matrix of all time step key vectors; is the scaling factor; each time step of the LSTM output is weighted by multi-head attention, which is expressed as: ; in, is the output of the multi-head attention mechanism, which represents the attention result after weighted summation of all heads. represents the number of attention heads, is the splicing operation function, and the weighted calculation is expressed as: ; in, is the temporal feature output of a single attention head, is the time range of the smart device that needs to be weighted, For a certain time step, For each time step The attention weight indicates the importance of this time step to the overall input series. Is a value vector, representing the information carrier of the smart device passed to the final representation at this time step, DNN represents a deep neural network, which is expressed as: ; in, After fusing spatial static parameter encoding and time weighted features, the intermediate state obtained through multi-layer mapping is used to generate the final control instruction. For the The weight matrix of the layer, representing the output from the previous layer The connection strength to the current layer, Output of the previous DNN layer, integrating attention features Concatenated input with spatial static encoding, For the The bias vector of the layer is a constant term that matches the number of neurons in the current layer. is an activation function used to introduce nonlinearity. Through nonlinear calculation, it is expressed as: ; in, is the prediction result of the current intelligent building control status, is the weight matrix of the output layer, mapping the feature vector of the last layer of the neural network to the control instruction space, Output feature vector for the last layer of the neural network, that is, the feature extraction result of the preceding module LSTM→Attention→DNN. The bias term for the output layer is a constant offset corresponding to the control output dimension, providing a mapping from input features to output features.

[0042] It should be noted that the integration of time series modeling and feature mapping capabilities includes encoding time series data and spatial static parameters in a unified format specification, constructing a spliced ​​input structure, and feeding it into LSTM and DNN according to preset channels for joint modeling, which can be expressed as: ; in, is the prediction result of the current intelligent building control status, is the weight matrix of the output layer, mapping the feature vector of the last layer of the neural network to the control instruction space, In order to introduce a multi-head attention mechanism on the hidden state sequence, the time sequence information is dynamically and selectively expressed by weighted focus on key time points. Model time series data, extract long-term dependencies and evolution trends, and generate hidden state sequences for each time step. is the output layer bias term, which is a constant offset corresponding to the control output dimension.

[0043] It should be noted that the model input uses a tensor interface, the LSTM structure processes the temporal feature dimension sequence, and the DNN structure processes the spatial feature vector. During the training process, both networks participate in back propagation simultaneously, and the weights are optimized through a joint loss function, which is expressed as: ; in, Represents the average error of the current batch of samples in predicting the control quantity, represents the number of samples, For the The predicted output of samples is For the The true target value corresponding to each sample forms a fusion learning structure.

[0044] It should also be noted that constructing a data closed-loop architecture for building information and multi-device linkage includes mapping DNN outputs into device control instructions, transmitting them to corresponding devices based on the communication interface, and transmitting status feedback back to the perception layer.

[0045] Example 2, reference Figure 2 , which is an embodiment of the present invention, provides an application system combining DNN and LSTM models in an intelligent building control system, including a spatial modeling and static parameter encoding module 100, a multi-source perception fusion and time series construction module 200, and a neural reasoning and equipment linkage control module 300.

[0046] Among them, X1: the spatial modeling and static parameter encoding module 100 includes a spatial digital twin modeling sub-module 101 and a spatial semantic feature encoding sub-module 102.

[0047] It should be noted that the spatial digital twin modeling submodule 101 is used to construct a three-dimensional BIM model, extract room structure and equipment layout information, establish a correspondence between spatial units and equipment entities, and form a spatial semantic map that can be used for calculation; the spatial semantic feature encoding submodule 102 is used to normalize spatial attributes into fixed-length vectors, bind devices through spatial indexes, realize spatial static parameter vectorization, and construct a neural network spatial input channel.

[0048] It should also be noted that the spatial digital twin modeling submodule 101 provides spatial objects, and the spatial semantic feature encoding submodule 102 completes semantic abstraction, forming a closed-loop process from spatial geometry to network-usable vectors, and transmitting it to the multi-source perception fusion and time series construction module 200.

[0049] X2: The multi-source perception fusion and time series construction module 200 includes a heterogeneous sensing access and time standardization submodule 201 and a spatiotemporal tensor construction and index binding submodule 202.

[0050] It should be noted that the heterogeneous sensor access and time specification submodule 201 is used to uniformly schedule multiple types of sensors, set sampling periods and timestamps, and generate standardized input segments; the spatiotemporal tensor construction and index binding submodule 202 is used to construct the perception data tensor and splice it with the spatial static vector to complete the construction of the multi-dimensional input format and realize spatiotemporal integrated modeling.

[0051] It should also be noted that the heterogeneous sensor access and time specification submodule 201 provides time-aligned dynamic input, and the spatiotemporal tensor construction and index binding submodule 202 realizes spatial injection and structural fusion, completing the input path from perception data to network tensor, and providing a well-structured input tensor for the neural reasoning and device linkage control module 300.

[0052] X3: The neural reasoning and device linkage control module 300 includes an LSTM timing modeling submodule 301, an attention feature aggregation submodule 302, and a DNN decision mapping and control generation submodule 303.

[0053] It should be noted that the LSTM time series modeling submodule 301 is used to model the input time series tensor, extract the dependency between each time step, and form a hidden state sequence; the attention feature aggregation submodule 302 is used to use a multi-head attention mechanism to calculate the time series attention weight, identify key time periods, and improve the model's response ability to emergencies or high-variability intervals; the DNN decision mapping and control generation submodule 303 is used to splice the time series features with static codes into the DNN network, generate a multi-dimensional control vector and send it in real time, and feed the device status back to the perception layer to form a closed loop.

[0054] It should also be noted that the neural reasoning and device linkage control module 300 relies on the spatial structure semantics provided by the spatial modeling and static parameter encoding module 100 and the temporal dynamic features provided by the multi-source perception fusion and time series construction module 200. The three work together to complete the closed-loop task from spatial recognition, environmental perception to control decision-making.

Claims

1. An application method combining DNN and LSTM models in intelligent building control systems, characterized in that: include: Build a spatial digital twin system based on the BIM model, extract the relationship between the building's physical structure and equipment layout, and use multi-dimensional input information to construct spatial static parameter coding for intelligent control; Build a multi-source heterogeneous perception channel driven by the IoT middle platform, integrate dynamic environmental information and spatial static parameter encoding, use fixed time intervals and a unified timestamp mechanism to construct multi-dimensional time series input; Build a multi-layer neural inference network that integrates LSTM, Attention, and DNN, integrating time series modeling and feature mapping capabilities, and construct a data closed-loop architecture for building information and multi-device linkage; Adopting a fixed time interval and unified timestamp mechanism includes setting a standard data collection cycle during IoT sensor data upload and generating aligned time tags for multiple sensor types; The data closed-loop architecture includes integrating the spatial dimension vectors provided by the BIM model, the real-time perception data collected by the Internet of Things, and the inference results of the AI ​​module, and feeding them back to the spatial intelligent devices through the intelligent control interface.

2. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 1, wherein: The construction of a spatial digital twin system based on the BIM model includes: A 1:1 restored 3D completion model is constructed through the Revit modeling platform, spatial structure information is extracted, and the smart devices installed in the building are embedded into the corresponding spatial units in the building model according to their location mapping.

3. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 1 or 2, characterized in that: The construction of spatial static parameter coding for intelligent control includes: The spatial structural elements extracted from the 3D as-built model are subjected to field normalization and feature encoding and used as the spatial dimension vector of the neural network input layer.

4. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 3, wherein: The spatial dimension vector as the input layer of the neural network includes: The spatial static parameter encoding and perception data establish a corresponding relationship through the spatial index binding mechanism and synchronously participate in the sequence input construction.

5. The method for applying the DNN and LSTM models in an intelligent building control system according to any one of claims 1, 2 or 4, characterized in that: The construction of the multi-source heterogeneous perception channel driven by the IoT middle platform includes: Through the smart building control platform, multiple types of sensor devices deployed in the spatial area are integrated, a unified transmission cycle is set, and real-time perception data is centrally aggregated and dispatched.

6. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 5, characterized in that: The constructing of multidimensional time series input includes: Time tags are attached to multi-type sensor data, and a standardized time series tensor structure is generated according to the combination of device number and sampling step.

7. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 6, characterized in that: The multi-layer neural reasoning network constructed by heterogeneous fusion of LSTM, Attention and DNN includes: The time series tensor is input into LSTM to extract the hidden state, and the time series features are generated through multi-head attention weighting. The time series features and spatial static parameter encoding are concatenated and input into DNN to generate the control vector.

8. The method for applying the DNN and LSTM models in an intelligent building control system according to any one of claims 1, 2, and 6, wherein: The fusion of time series modeling and feature mapping capabilities includes, The time series data and spatial static parameters are encoded in a unified format specification, and a spliced ​​input structure is constructed, which flows into LSTM and DNN according to the preset channels for joint modeling.

9. The method for applying the DNN and LSTM models in an intelligent building control system according to claim 8, characterized in that: The data closed-loop architecture for building information and multi-device linkage includes: Map the DNN output into device control instructions, transmit them to the corresponding device based on the communication interface, and transmit the status feedback back to the perception layer.

10. An application system combining DNN and LSTM models in intelligent building control systems, characterized by: It includes a spatial modeling and static parameter encoding module (100), a multi-source perception fusion and time series construction module (200), and a neural reasoning and device linkage control module (300); The spatial modeling and static parameter encoding module (100) is used to construct a digital twin system through the BIM model, extract building structure and equipment layout information, standardize and encode spatial attributes, and generate static input fields that adapt to the neural network; The multi-source perception fusion and time series construction module (200) is used to collect multiple types of environmental sensor data, complete multi-source alignment through a unified timestamp mechanism, and fuse static coding to generate standardized multi-dimensional time series input; The neural reasoning and device linkage control module (300) is used to input spatial static parameter coding and time series data into a fusion neural network structure, complete feature modeling and control instruction generation, and link intelligent devices to dynamic responses.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for applying the DNN and LSTM models in an intelligent building control system according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for applying the DNN and LSTM models in an intelligent building control system according to any one of claims 1 to 9 are implemented.

Images