Heat supply user room temperature measurement method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of multi-source data fusion, in particular to a heat supply user room temperature measurement method and system based on multi-source data fusion, and the method comprises the following steps: S1, heat source data collection: directly collecting boiler operation data from a heat source plant distributed control system, and uploading the heat source data to an Internet of Things cloud platform; s2, multi-source data are collected, the multi-source data comprise user indoor temperature data, heat supply pipe network data and meteorological data, and the multi-source data are uploaded to an Internet of Things cloud platform; s3, data fusion: fusing the data in the S1 and the S2 in an Internet of Things cloud platform, and dynamically calibrating indoor temperature data; and S4, the system users of different levels pass authority authentication, and the fused and calibrated data are acquired from the Internet of Things cloud platform. According to the invention, a double-layer algorithm is constructed by optimizing the LSTM network and improving the grey wolf algorithm, the room temperature calibration capability in a dynamic environment is improved, and the efficiency precision is improved.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion technology, and more specifically, to a method and system for measuring room temperature of heating users based on multi-source data fusion. Background Technology

[0002] In centralized heating systems, accurate measurement of user room temperature is the core foundation for achieving on-demand heating and balancing energy consumption with user comfort. However, current mainstream heating user room temperature measurement technologies still have many limitations: First, existing room temperature measurements mostly rely on single-point sensors inside the user's room, which are easily affected by factors such as installation location and indoor airflow. Second, the heat source data, which is the basis for room temperature calculation, itself has unstable transmission problems, further amplifying the deviation in room temperature measurement.

[0003] The invention disclosed in CN113888737A is a method for measuring room temperature of heat users based on temperature measurement images captured by a smart device. The method includes: heat users acquiring room temperature measurement values ​​at different locations in real time through a room temperature measuring device during preset time periods; using a smart terminal device to collect images of multiple room temperature measurement values ​​and sending them to a room temperature monitoring host computer via a communication module, and generating incentive results according to preset rules to encourage users to measure room temperature; the room temperature monitoring host computer preprocessing, segmenting, and extracting labels from the real-time acquired room temperature images, and then inputting them into a constructed convolutional neural network model for image recognition, determining the validity of the room temperature measurement location of the heat user in the image and identifying the corresponding room temperature measurement value.

[0004] This invention reduces communication costs associated with existing wireless room temperature data acquisition and ensures the validity of measurement locations, providing reliable room temperature data for heating network scheduling and saving operating costs. However, it is highly dependent on operation, requiring users to photograph the temperature measuring device according to specifications (such as avoiding obstructions and maintaining a horizontal angle). In practical applications, the reading error rate caused by non-standard user operation increases, and data gaps are prone to occur at night or when users are away. Furthermore, it does not solve systemic problems, focusing only on "user-end image recognition" without integrating heat source and pipeline network data. It cannot distinguish whether abnormal room temperature is due to "user measurement operation error" or "system-end supply problem," making it difficult to support the overall control of the heating system. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for measuring room temperature of heating users based on multi-source data fusion, so as to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, this invention provides a method for measuring room temperature of heating users based on multi-source data fusion, comprising the following steps: S1. Heat source data collection: directly collect the boiler operation data from the distributed control system of the heat source plant and upload the heat source data to the Internet of Things cloud platform; S2. Multi-source data collection: the multi-source data includes the indoor temperature data of users, the heat supply network data and the meteorological data, and upload the multi-source data to the Internet of Things cloud platform; S3. Data fusion: in the Internet of Things cloud platform, fuse the data in S1 and S2 and dynamically calibrate the indoor temperature data; S4. System users at different levels pass the permission authentication and obtain the fused and calibrated data from the Internet of Things cloud platform.

[0007] As a further improvement of this technical solution, in S1, the distributed control system collects the heat source operation data through the 485 board of the DCS host. The heat source operation data includes the boiler outlet water temperature, flow rate and pressure, and uses the gateway to transmit it to the Internet of Things platform through the MQTT protocol.

[0008] This setting reduces the intermediate transfer equipment in the data collection link and reduces the risk of data transmission interruption caused by equipment failures.

[0009] As a further improvement of this technical solution, in S2, the indoor temperature data of users is obtained by collecting the real-time temperature of different areas in the user's room through distributed sensors, and is sent to the Internet of Things cloud platform through the communication module supporting the distributed sensors; the heat supply network data includes the water supply temperature, flow rate and pressure data of the heat network, and the data is collected through the temperature, pressure and flow transmitters on the pipeline and sent through the supporting gateway; the meteorological data is obtained from the meteorological information of the National Meteorological Bureau by regularly calling the software API and sent through the Internet.

[0010] This setting constructs a multi-dimensional data collection system from the user side, the pipe network side and the environment side: the distributed sensors can cover different areas in the room, avoiding the measurement deviation caused by the installation position of traditional single-point sensors; the pipe network transmitters directly collect the water supply temperature, flow rate and pressure, and reflect the heat network transmission status in real time; the meteorological data is obtained through the National Meteorological Bureau API to ensure the authority of the environmental factor data such as outdoor temperature and wind force. The three work together to provide comprehensive and reliable multi-source original data for subsequent data fusion and calibration of the room temperature.

[0011] As a further improvement of this technical solution, in S3, for data fusion, in the Internet of Things cloud platform, fuse the data in S1 and S2 and dynamically calibrate the indoor temperature data, including the following steps: S3.1. Receive the heat source data sent in S1 and the multi-source data sent in S2, and then perform preprocessing, where the preprocessing includes outlier cleaning, missing value processing and data standardization; S3.2 Align the preprocessed data in both time and space dimensions; S3.3. Optimize the neural network to dynamically calibrate indoor temperature data.

[0012] As a further improvement to this technical solution, step S3.2 aligns the preprocessed data in both time and space dimensions, including the following steps: A1. For data with different collection frequencies, convert them into time series with the same time interval to ensure that the timestamps of all data are synchronized and mark the original collection frequency of the data. A2. Based on the Geographic Information System (GIS), establish a spatial mapping model to associate the latitude and longitude information of the heat source plant location, pipeline nodes, and user indoor temperature monitoring points, output location association data, divide the grid with the heat station as the center, and bind the user temperature data in each grid with the flow and pressure data of the corresponding pipeline branch.

[0013] As a further improvement to this technical solution, step S3.3, which involves optimizing the neural network to dynamically calibrate indoor temperature data, includes the following steps: B1. The data output by A1 and A2 will be used as the training sample set. The input feature vector includes heat source operation data, user indoor temperature data, heating network data, meteorological data and location-related data. The output label is the reference room temperature benchmark value calculated by a high-precision calibration instrument or based on the energy balance model, which is used to supervise network training. B2. Construct an optimized LSTM network structure and use an improved gray wolf algorithm to optimize the parameters in the LSTM network; B3. Configure the LSTM network with parameters optimized by the improved Grey Wolf algorithm and perform supervised training; B4. Input the real-time collected and preprocessed multi-source data into the trained LSTM network, and calculate and output the current dynamic calibration benchmark value of the user's indoor temperature through forward propagation. B5. Regular model updates and validation: Establish an online model update mechanism and regularly use newly collected data for incremental training and fine-tuning.

[0014] As a further improvement to this technical solution, in B2, an optimized LSTM network structure is constructed and the parameters in the LSTM network are optimized using an improved Grey Wolf algorithm. The optimized LSTM network structure includes an input layer, an LSTM layer, a fully connected layer, and an output layer, wherein: The input layer is used to receive the input feature vector from B1; The LSTM layer processes data through a gating mechanism, introducing a temporal attention mechanism and a temperature influence factor gating mechanism into the LSTM layer; The temporal attention mechanism dynamically calculates the attention weights at each time step, highlighting the impact of key operation points, expanding the input of the LSTM layer, and adding context vectors. This allows the network to take into account information from key historical moments when calculating all gating signals; Temperature influence factor gating mechanism based on outdoor temperature change sequence As an additional input, it specifically captures the delayed decay effect of outdoor temperature changes on indoor temperature and integrates it into cell state updates, directly encoding the delayed effect of external temperature changes into the model's long-term memory. The fully connected layer receives the final output of the LSTM layer and uses the ReLU activation function; The output layer contains a single neuron that uses a linear activation function to output a continuous baseline prediction of indoor temperature.

[0015] As a further improvement to this technical solution, in B2, an improved gray wolf algorithm is used to optimize the parameters in the LSTM network, including the following steps: B2.1 Population Initialization: Set the population size, spatial dimension, and maximum number of iterations. Randomly initialize the position of each individual. The position of an individual ,in, Indicates the number of LSTM layers. This indicates the number of neurons in each LSTM layer. The length of the input sequence; B2.2 Fitness Calculation: The root mean square error of the LSTM network is used as the fitness function. The top three best gray wolves are defined as follows: , and The remaining gray wolves were defined as ; B2.3 Update the position of the individual, and introduce a sinusoidally nonlinearly decreasing convergence factor and a position update that incorporates inertia weights; B2.4. Repeat B2.2 and B2.3 until the maximum number of iterations is reached.

[0016] As a further improvement to this technical solution, in B3, the LSTM network is configured with parameters optimized by the improved Grey Wolf algorithm and subjected to supervised training. The supervised training includes the following steps: B3.1 Training data preprocessing: Extract input feature vectors and corresponding output labels from the training sample set, standardize the input features, and divide the training subset and validation subset according to a preset ratio to ensure consistent data distribution; B3.2 Training Hyperparameter Settings: Set the training batch size, maximum number of iterations, learning rate, and gradient clipping threshold. B3.3 Forward propagation computation: In each round of training, a batch of training subsets are input into the LSTM network. Feature extraction and temporal modeling are performed through the gating mechanism inside the network, and finally the predicted room temperature value of the current batch is output. B3.4 Loss function calculation: The loss value for the current batch is calculated using the predicted room temperature value and the corresponding reference room temperature baseline value as inputs and the mean square error. B3.5 Backpropagation optimization: Based on the loss value, the gradient of each parameter is calculated through the time backpropagation algorithm. The LSTM network parameters are iteratively updated in combination with the Adam optimizer. At the same time, gradient clipping is used to avoid the gradient explosion problem. B3.6 Validation set performance evaluation: After each training cycle, the validation subset is input into the network to calculate the validation loss and prediction accuracy, and to monitor the model's generalization ability in real time. B3.7 Early stopping strategy: When the validation loss remains unchanged for several consecutive rounds, stop training to avoid model overfitting and preserve the current optimal parameter state; B3.8 Training process recording: Record the training loss, validation loss, and key parameter update trajectory information for each round of training, and generate training logs for subsequent model analysis and optimization. B3.9 Final Model Determination: After training, the model with the best performance on the validation set is selected as the final trained LSTM network for subsequent real-time temperature calibration calculations.

[0017] These six settings include standardizing the spatiotemporal alignment of multi-source data to eliminate bias, upgrading the model to a dynamic response mode with real-time output and periodic updates, improving calibration accuracy under complex operating conditions through temporal attention and temperature gating mechanisms, optimizing parameters and improving efficiency with an improved gray wolf algorithm, and ensuring model generalization and traceability through full-process control, providing key technical support for heating energy conservation and user comfort.

[0018] This invention also provides a heating user room temperature measurement system based on multi-source data fusion, used to implement the above-mentioned heating user room temperature measurement method based on multi-source data fusion, including: Heat source data acquisition module: used to directly acquire boiler operation data from the distributed control system of the heat source plant and upload the heat source data to the Internet of Things cloud platform; Multi-source data acquisition module: used for acquiring multi-source data, including user indoor temperature data, heating network data and meteorological data, and uploading the multi-source data to the Internet of Things cloud platform; Data fusion module: used to fuse data from the heat source data acquisition module and the multi-source data acquisition module in the IoT cloud platform, and dynamically calibrate indoor temperature data based on an optimized neural network; Data interaction module: Used to authenticate users at different levels. Authenticated users can retrieve the merged and calibrated data according to their permissions.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The heating user room temperature measurement method and system based on multi-source data fusion collects data from multiple dimensions, including heat source, indoor environment, pipeline network, and meteorology. Compared with traditional single-point collection, it has more comprehensive coverage and avoids the problems of single-point interference and data isolation, laying the foundation for subsequent fusion.

[0020] 2. In this heating user room temperature measurement method and system based on multi-source data fusion, the distributed control system collects heat source operation data through the 485 board of the DCS host. The heat source operation data includes boiler outlet water temperature, flow rate and pressure, and is transmitted to the Internet of Things platform through the MQTT protocol using a gateway, reducing intermediate equipment and facilities, reducing error and failure rate, and ensuring data timeliness.

[0021] 3. In this heating user room temperature measurement method and system based on multi-source data fusion, a two-layer algorithm is constructed by optimizing the LSTM network and improving the Grey Wolf algorithm to enhance the room temperature calibration capability under dynamic environments. The optimized LSTM layer in the LSTM network introduces a temporal attention mechanism and a temperature influence factor gating mechanism to accurately capture environmental changes. The improved Grey Wolf algorithm introduces a sinusoidal nonlinear decreasing convergence factor and a position update of the fusion inertia weight through reasonable population and iteration rules. The LSTM network parameters are optimized to improve calibration efficiency and measurement accuracy, ensuring that the LSTM network outputs an accurate dynamic room temperature calibration benchmark value. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system structure diagram of the present invention; The meanings of the various markings in the diagram are as follows: 100, heat source data acquisition module; 200, multi-source data acquisition module; 300, data fusion module; 400, data interaction module. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] As shown Figure 1 in the figure, this embodiment provides a heating user room temperature measurement method based on multi-source data fusion, including the following steps: S1. Heat source data collection: directly collect boiler operation data from the distributed control system of the heat source plant, and upload the heat source data to the Internet of Things cloud platform; S2. Multi-source data collection: multi-source data includes user indoor temperature data, heating pipe network data, and meteorological data, and upload the multi-source data to the Internet of Things cloud platform; S3. Data fusion: in the Internet of Things cloud platform, fuse the data in S1 and S2, and dynamically calibrate the indoor temperature data; S4. System users at different levels pass the permission authentication and obtain the fused and calibrated data from the Internet of Things cloud platform, which not only ensures data security but also ensures that each role can obtain appropriate and accurate data, providing core technical support for on-demand heating, balancing energy consumption and user comfort.

[0025] In this embodiment, in S1, the distributed control system collects heat source operation data through the 485 board of the DCS host. The heat source operation data includes boiler outlet water temperature, flow rate, and pressure, and uses the gateway to transmit it to the Internet of Things platform through the MQTT protocol, adapting to the real-time transmission requirements of heat source data, ensuring that the data is quickly uploaded to the Internet of Things cloud platform, avoiding the problem of poor data timeliness in traditional transmission methods, providing real-time and reliable heat source basic data for subsequent multi-source data fusion, and ensuring the accuracy of room temperature calibration after fusion.

[0026] In this embodiment, in S2, the user indoor temperature data is obtained by collecting the real-time temperature of different areas in the user's room through distributed sensors, and is sent to the Internet of Things cloud platform through the communication module supporting the distributed sensors, solving the measurement deviation caused by the position limitation of traditional single-point sensors, and the distributed sensors can work continuously, avoiding data断层 in scenarios such as when users go out or at night. For example, the distributed sensors are built with NB-IoT communication modules and directly send the data to the Internet of Things cloud platform through the operator's cellular network (4G / 5G network), or the distributed sensors wirelessly send the data to the LoRa gateway deployed in the community or building, and the LoRa gateway then uploads the aggregated data to the Internet of Things cloud platform through Ethernet or 4G network; the heating pipe network data includes the supply water temperature, flow rate, and pressure data of the heat pipe network, and the data is collected through temperature, pressure, and flow transmitters on the pipeline and sent through the supporting gateway; the meteorological data is obtained from the meteorological information of the National Meteorological Administration by regularly calling the software API and sent through the Internet.

[0027] In this embodiment, in S3, for data fusion, in the Internet of Things cloud platform, fuse the data in S1 and S2, and dynamically calibrate the indoor temperature data, including the following steps: S3.1 Receive the heat source data sent in S1 and the multi-source data sent in S2, and then perform preprocessing. The preprocessing includes outlier cleaning, missing value handling and data standardization, which can filter invalid data, fill in missing information, unify data dimensions, improve data availability, and avoid abnormal data from interfering with the fusion results. S3.2 Align the preprocessed data in both time and space dimensions; S3.3 By optimizing the neural network to dynamically calibrate indoor temperature data, the real-time performance and accuracy of room temperature calibration are improved, and the final output is indoor temperature data that is more in line with reality.

[0028] Specifically, in S3.2, the preprocessed data is aligned in both time and space dimensions, including the following steps: A1. For data with different collection frequencies, convert them into time series with the same time interval to ensure that the timestamps of all data are synchronized and mark the original collection frequency of the data for traceability. For example, heat source data at the second level, pipeline data at the minute level, indoor temperature data at the 15-minute level, and meteorological data at the hour level are converted into time series with a 15-minute interval using linear interpolation. A2. Based on the Geographic Information System (GIS), establish a spatial mapping model to associate the latitude and longitude information of the location of the heat source plant, pipeline nodes (including heating stations and branch pipes), and user indoor temperature monitoring points, output location association data, divide the grid with the heating station as the center, and bind the user temperature data in each grid with the flow and pressure data of the corresponding pipeline branch.

[0029] Furthermore, in S3.3, the indoor temperature data is dynamically calibrated by optimizing the neural network, including the following steps: B1 uses the data output from A1 and A2 as the training sample set. The input feature vector includes heat source operation data, user indoor temperature data, heating network data, meteorological data, and location-related data. The output label is the reference room temperature benchmark value calculated by a high-precision calibrated instrument or based on an energy balance model, which is used to supervise network training. B2. Construct an optimized LSTM network structure and use an improved gray wolf algorithm to optimize the parameters in the LSTM network; An optimized LSTM network structure is constructed, and an improved Grey Wolf algorithm is used to optimize the parameters in the LSTM network. The optimized LSTM network structure includes an input layer, an LSTM layer, a fully connected layer, and an output layer, wherein: The input layer is used to receive the input feature vector from B1; The LSTM layer processes data through a gating mechanism, and the corresponding expression is: Input Gate: ; Forgotten Gate: ; Candidate cell status: ; Update cell status: ; Output gate: ; Output status: ; in, Indicates the current time step. Indicates the previous time step. express The output of the time-input gate controls how much of the "new information at the current time" should be retained. This indicates that no new information will be retained. (This indicates that all new information is retained). This represents the sigmoid activation function (output range 0 to 1), used to map the calculation results to a range of retention or forgetting ratios. This represents the weight matrix of the input gate. express The output of the LSTM layer at each time step (containing short-term temporal information from the previous time step). express The input feature vector at time t, Indicates will LSTM layer output at time t and The input feature vectors at time step are concatenated along the feature dimension (forming a longer vector). Indicates the bias of the input gate; express The output of the Forget Gate The weight matrix represents the forget gate, used to perform a linear transformation on the input features. Indicates the bias of the forget gate; express The candidate cell state at time step (ranging from -1 to 1) contains new information about the input feature vector at the current time step. This represents the hyperbolic tangent activation function, used to map new information to a symmetric numerical range. The weight matrix representing the candidate cell state. The bias representing the candidate cell state; express The cell state at any given time is used to preserve long-term dependency information in time-series data. express Cellular state at any given moment This indicates element-wise multiplication, which means multiplying corresponding elements of two vectors. express The output of the time-of-time output gate (value range is 0 to 1). This represents the weight matrix of the output gate. Indicates the bias of the output gate; express The output of the LSTM layer at time 1; A temporal attention mechanism and a temperature influence factor gating mechanism are introduced into the LSTM layer; The temporal attention mechanism dynamically calculates the attention weights at each time step, highlighting the impact of key operational moments such as boiler start-up, shutdown, and regulation. It also expands the input of the LSTM layer by adding a context vector. Concatenate the input Expand to , Indicates will LSTM layer output at time step Input feature vector and context vector at time step By concatenating features along the feature dimension, the network can incorporate information from key historical moments when calculating all gated signals. Specific expressions include: ; ; ; in, Indicates a historical moment, that is, earlier than The time step of a moment This represents the attention score, which measures... Moments and Historical Moments The stronger the correlation, the higher the score, the more important it is. Represents the rating weight vector. This represents the transpose operation, used to convert a high-dimensional vector into a scalar. Represents the projection matrix of the state at a historical moment. express The output of the LSTM layer at each historical moment. Represents the projection matrix of the target state at that time. Indicates the attention bias parameter; This represents the normalized attention weights. Represents the natural logarithm function. This represents the total number of steps in a historical moment, i.e., the total number of historical moments that need to be reviewed; This represents a context vector containing historical information, retaining only the context vectors. Always-useful key information, filtering out redundant information; Temperature influence factor gating mechanism based on outdoor temperature change sequence (For example, the first-order difference over the past few hours) is used as an additional input to specifically capture the delayed decay effect of outdoor temperature changes on indoor temperature, and the corresponding expression is: Temperature-affected doors: ; Temperature modulation factor: ; in, express Temperature affects the door's output at all times. The weight matrix representing the effect of temperature on the gate. Indicates will LSTM layer output at time step Input feature vector and context vector at time step outdoor temperature change sequence Concatenate along the feature dimension. This indicates how temperature affects the door's bias. Indicates the temperature modulation factor. The weight matrix represents the temperature modulation factor. Indicates the bias of the temperature modulation factor; This is integrated into cell state updates, directly encoding the delayed effects of external temperature changes into the model's long-term memory. The corrected cell state formula is as follows: ; The fully connected layer receives the final output of the LSTM layer and uses the ReLU activation function; The output layer contains a single neuron that uses a linear activation function to output a continuous baseline prediction of indoor temperature. The improved Grey Wolf algorithm is used to optimize the parameters in the LSTM network, including the following steps: B2.1 Population Initialization: Set the population size, spatial dimension, and maximum number of iterations. Randomly initialize the position of each individual. The position of an individual ,in, Indicates the number of LSTM layers. This indicates the number of neurons in each LSTM layer. The length of the input sequence; B2.2 Fitness Calculation: The root mean square error of the LSTM network is used as the fitness function. The top three best gray wolves are defined as follows: , and The remaining gray wolves were defined as ; B2.3 Update the individual's position, and introduce a sinusoidally nonlinearly decreasing convergence factor and a fused inertia weight for position update. The corresponding expression is: ;

[0030] ; ; ; ; ; ; ; ; ; in, Indicates the number of iterations. Indicates the maximum number of iterations. Represents the sine function. Indicates the exploration factor. Indicates the weighting factor. Indicates the first The convergence factor at the next iteration and All are random numbers between [0,1]; Indicates the first In the next iteration, the current gray wolf Location, Indicates the first In the next iteration, Gray Wolf Location, Indicates the first In the next iteration, the current gray wolf and the gray wolf The distance between them; Indicates the first In the next iteration, Gray Wolf Location, Indicates the first In the next iteration, the current gray wolf and the gray wolf The distance between them; Indicates the first In the next iteration, Gray Wolf Location, Indicates the first In the next iteration, the current gray wolf and the gray wolf The distance between them This indicates taking the absolute value. , and All are weighting factors, by It occurs three times, and each of the three occurrences is independent of the others. , and All are exploratory factors, by It occurs three times, and each of the three occurrences is independent of the others. Indicates the current gray wolf Towards the Gray Wolf The step size of movement, Indicates the current gray wolf Towards the Gray Wolf The step size of movement, Indicates the current gray wolf Towards the Gray Wolf The step size of movement; Indicates the first In the next iteration, the current gray wolf Location, Indicates the first Inertia weights in the next iteration This represents the maximum value of the inertia weight. This represents the minimum inertia weight; B2.4, Repeat B2.2 and B2.3 until the maximum number of iterations is reached; B3. Configure the LSTM network using the parameters optimized by the improved Grey Wolf algorithm, and perform supervised training. Supervised training includes the following steps: B3.1 Training data preprocessing: Extract input feature vectors and corresponding output labels from the training sample set, standardize the input features, and divide the training subset and validation subset according to a preset ratio (usually 7:3 or 8:2) to ensure consistent data distribution; B3.2 Training Hyperparameter Settings: Set the training batch size, maximum number of iterations, learning rate, and gradient clipping threshold. B3.3 Forward propagation computation: In each round of training, a batch of training subsets are input into the LSTM network. Feature extraction and temporal modeling are performed through the gating mechanism inside the network, and finally the predicted room temperature value of the current batch is output. B3.4 Loss function calculation: Using the predicted room temperature value and the corresponding reference room temperature baseline value as input, the mean square error is used to calculate the loss value of the current batch, providing a clear direction for reverse optimization; B3.5 Backpropagation optimization: Based on the loss value, the gradient of each parameter is calculated through the time backpropagation algorithm. Combined with the Adam optimizer, the parameters of the LSTM network are iteratively updated. At the same time, gradient clipping is used to avoid the gradient explosion problem, which improves the efficiency of parameter update and avoids the gradient explosion problem, thus ensuring training stability. B3.6 Validation set performance evaluation: After a certain number of training rounds (set as needed), input the validation subset into the network, calculate the validation loss and prediction accuracy, and monitor the model's generalization ability in real time. B3.7 Early stopping strategy: When the validation loss remains unchanged for several consecutive rounds (usually 5 to 10 rounds), stop training to avoid model overfitting and preserve the current optimal parameter state; B3.8 Training process recording: Record the training loss, validation loss, and key parameter update trajectory information for each round of training, and generate training logs for subsequent model analysis and optimization. B3.9 Final Model Determination: After training, the model with the best performance on the validation set is selected as the final trained LSTM network for subsequent real-time temperature calibration calculations. B4. Input the real-time collected and preprocessed multi-source data into the trained LSTM network, and calculate and output the current dynamic calibration benchmark value of the user's indoor temperature through forward propagation. B5. Regular model updates and validation: Establish an online model update mechanism and regularly use newly collected data for incremental training and fine-tuning.

[0031] like Figure 2 As shown, this embodiment also provides a heating user room temperature measurement system based on multi-source data fusion, used to implement the above-mentioned heating user room temperature measurement method based on multi-source data fusion, including: Heat source data acquisition module 100: used to directly acquire boiler operation data from the distributed control system of the heat source plant and upload the heat source data to the Internet of Things cloud platform; Multi-source data acquisition module 200: Used for acquiring multi-source data, including user indoor temperature data, heating network data and meteorological data, and uploading the multi-source data to the Internet of Things cloud platform; Data fusion module 300: Used to fuse data from heat source data acquisition module 100 and multi-source data acquisition module 200 in the Internet of Things cloud platform, and dynamically calibrate indoor temperature data based on optimized neural network; Data interaction module 400: Used to authenticate users at different levels. Authenticated users can retrieve the fused and calibrated data according to their permissions.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for measuring room temperature of heating users based on multi-source data fusion, characterized in that, It includes the following steps: S1. Heat source data collection: directly collect the boiler operation data from the distributed control system of the heat source plant, and upload the heat source data to the Internet of Things cloud platform; S2. Multi-source data collection: the multi-source data includes the indoor temperature data of users, the heat supply network data and the meteorological data, and upload the multi-source data to the Internet of Things cloud platform; S3. Data fusion: in the Internet of Things cloud platform, fuse the data in S1 and S2, and dynamically calibrate the indoor temperature data; S4. System users at different levels pass the permission authentication and obtain the fused and calibrated data from the Internet of Things cloud platform.

2. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 1, characterized in that: In S1, the distributed control system collects the heat source operation data through the 485 board card of the DCS host. The heat source operation data includes the boiler outlet water temperature, flow rate and pressure, and uses the gateway to transmit it to the Internet of Things platform through the MQTT protocol.

3. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 2, characterized in that: In S2, the indoor temperature data of users is obtained by collecting the real-time temperature of different areas in the user's room through distributed sensors, and is sent to the Internet of Things cloud platform through the communication module supporting the distributed sensors; the heat supply network data includes the water supply temperature, flow rate and pressure data of the heat supply network. The data is collected through the temperature, pressure and flow transmitters on the pipeline, and the data is sent through the supporting gateway; the meteorological data is obtained from the meteorological information of the National Meteorological Bureau by regularly calling the software API, and the data is sent through the Internet.

4. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 3, characterized in that, In S3, for data fusion, in the Internet of Things cloud platform, fuse the data in S1 and S2, and dynamically calibrate the indoor temperature data, including the following steps: S3.

1. Receive the heat source data sent in S1 and the multi-source data sent in S2, and then perform preprocessing. The preprocessing includes outlier cleaning, missing value processing and data standardization; S3.

2. Align the preprocessed data in the time and space dimensions; S3.

3. Dynamically calibrate the indoor temperature data through an optimized neural network.

5. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 4, characterized in that, In S3.2, align the preprocessed data in the time and space dimensions, including the following steps: A1. For data with different collection frequencies, uniformly convert them into time series with the same time interval, ensure the synchronization of the timestamps of all data, and mark the original collection frequency of the data; A2. Establish a spatial mapping model based on the geographic information GIS system, associate the longitude and latitude information of the heat source plant location, pipeline network nodes and indoor temperature monitoring points of users, output the location association data, divide the grid centered on the heat substation, and bind the user temperature data in each grid with the flow rate and pressure data of the corresponding pipeline branch.

6. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 5, characterized in that, In S3.3, dynamically calibrate the indoor temperature data through an optimized neural network, including the following steps: B1. Use the data output by A1 and A2 as the training sample set. The input feature vectors include the heat source operation data, the indoor temperature data of users, the heat supply network data, the meteorological data and the location association data. The output label is the reference room temperature benchmark value obtained by a high-precision calibration instrument or calculated based on the energy balance model, which is used to supervise the network training; B2. Construct an optimized LSTM network structure and use an improved grey wolf algorithm to optimize the parameters in the LSTM network; B3. Configure the LSTM network with parameters optimized by the improved Grey Wolf algorithm and perform supervised training; B4. Input the real-time collected and preprocessed multi-source data into the trained LSTM network, and calculate and output the current dynamic calibration benchmark value of the user's indoor temperature through forward propagation. B5. Regular model updates and validation: Establish an online model update mechanism and regularly use newly collected data for incremental training and fine-tuning.

7. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 6, characterized in that, In B2, an optimized LSTM network structure is constructed, and an improved Grey Wolf algorithm is used to optimize the parameters in the LSTM network. The optimized LSTM network structure includes an input layer, an LSTM layer, a fully connected layer, and an output layer, wherein: The input layer is used to receive the input feature vector from B1; The LSTM layer processes data through a gating mechanism, introducing a temporal attention mechanism and a temperature influence factor gating mechanism into the LSTM layer; The temporal attention mechanism dynamically calculates the attention weights at each time step, highlighting the impact of key operation points, expanding the input of the LSTM layer, and adding context vectors. This allows the network to take into account information from key historical moments when calculating all gating signals; Temperature influence factor gating mechanism based on outdoor temperature change sequence As an additional input, it specifically captures the delayed decay effect of outdoor temperature changes on indoor temperature and integrates it into cell state updates, directly encoding the delayed effect of external temperature changes into the model's long-term memory. The fully connected layer receives the final output of the LSTM layer and uses the ReLU activation function; The output layer contains a single neuron that uses a linear activation function to output a continuous baseline prediction of indoor temperature.

8. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 7, characterized in that, In B2, the improved Grey Wolf algorithm is used to optimize the parameters in the LSTM network, including the following steps: B2.1 Population Initialization: Set the population size, spatial dimension, and maximum number of iterations. Randomly initialize the position of each individual. The position of an individual ,in, Indicates the number of LSTM layers. This indicates the number of neurons in each LSTM layer. The length of the input sequence; B2.2 Fitness Calculation: The root mean square error of the LSTM network is used as the fitness function. The top three best gray wolves are defined as follows: , and The remaining gray wolves were defined as ; B2.3 Update the position of the individual, and introduce a sinusoidally nonlinearly decreasing convergence factor and a position update that incorporates inertia weights; B2.

4. Repeat B2.2 and B2.3 until the maximum number of iterations is reached.

9. The method for measuring room temperature of heating users based on multi-source data fusion according to claim 8, characterized in that, In B3, the LSTM network is configured with parameters optimized by the improved Grey Wolf algorithm and subjected to supervised training. The supervised training includes the following steps: B3.1 Training data preprocessing: Extract input feature vectors and corresponding output labels from the training sample set, standardize the input features, and divide the training subset and validation subset according to a preset ratio to ensure consistent data distribution; B3.2 Training Hyperparameter Settings: Set the training batch size, maximum number of iterations, learning rate, and gradient clipping threshold. B3.3 Forward propagation computation: In each round of training, a batch of training subsets are input into the LSTM network. Feature extraction and temporal modeling are performed through the gating mechanism inside the network, and finally the predicted room temperature value of the current batch is output. B3.4 Loss function calculation: The loss value for the current batch is calculated using the predicted room temperature value and the corresponding reference room temperature baseline value as inputs and the mean square error. B3.5 Backpropagation optimization: Based on the loss value, the gradient of each parameter is calculated through the time backpropagation algorithm. The LSTM network parameters are iteratively updated in combination with the Adam optimizer. At the same time, gradient clipping is used to avoid the gradient explosion problem. B3.6 Validation set performance evaluation: After each training cycle, the validation subset is input into the network to calculate the validation loss and prediction accuracy, and to monitor the model's generalization ability in real time. B3.7 Early stopping strategy: When the validation loss remains unchanged for several consecutive rounds, stop training to avoid model overfitting and preserve the current optimal parameter state; B3.8 Training process recording: Record the training loss, validation loss, and key parameter update trajectory information for each round of training, and generate training logs for subsequent model analysis and optimization. B3.9 Final Model Determination: After training, the model with the best performance on the validation set is selected as the final trained LSTM network for subsequent real-time temperature calibration calculations.

10. A heating user room temperature measurement system based on multi-source data fusion, used to implement the steps of the heating user room temperature measurement method based on multi-source data fusion as described in any one of claims 1-9, characterized in that, include: Heat source data acquisition module (100): used to directly acquire boiler operation data from the distributed control system of the heat source plant and upload the heat source data to the Internet of Things cloud platform; Multi-source data acquisition module (200): used for acquiring multi-source data, including user indoor temperature data, heating network data and meteorological data, and uploading the multi-source data to the Internet of Things cloud platform; Data fusion module (300): used to fuse the data from the heat source data acquisition module (100) and the multi-source data acquisition module (200) in the Internet of Things cloud platform, and dynamically calibrate indoor temperature data based on an optimized neural network; Data interaction module (400): Used to authenticate users at different levels. Authenticated users can retrieve the fused and calibrated data according to their permissions.

Images