A Neural Network-Based Method for Predicting Heating Loads in Community Buildings

Abstract

This invention discloses a method for predicting heating load in community buildings based on neural networks, belonging to the field of immune management technology. The method includes: calculating a dynamic thermal inertia value using a neural network based on historical energy consumption and attribute data of buildings collected by sensors at heating stations; further calculating a heating load adjustment coefficient taking into account the heat transfer effect between buildings using this value; generating a preliminary load prediction value for the next 24 hours, taking into account thermal disturbances, by combining meteorological forecast data; achieving online self-optimization by comparing the predicted value with the measured value and adjusting network parameters in real time; and finally outputting a high-precision final heating load prediction result for community buildings to guide heat source plants in on-demand production and dynamic scheduling of the pipeline network. This invention can effectively improve the accuracy of heating load prediction, reduce energy waste, and realize intelligent and adaptive control of the heating system.

Description

Technical Field

[0001] This invention relates to the field of heating system optimization and intelligent control technology, and particularly to a method for predicting heating load in community buildings based on neural networks. Background Technology

[0002] In urban heating systems, accurately predicting the heating load of community buildings is crucial for achieving energy conservation, emission reduction, and precise regulation. Traditional prediction methods mainly rely on historical data statistics or physical modeling, which have significant limitations. Statistical methods such as time series analysis struggle to capture nonlinear factors such as sudden weather changes and shifts in building usage patterns; while physical models based on heat balance equations rely on precise building parameters, resulting in complex modeling and weak generalization ability.

[0003] While existing machine learning methods have shown some improvement, they generally neglect the heat transfer effects of distance and wind direction within building complexes, leading to discrepancies between predicted results and actual loads. Furthermore, most models employ offline training, lacking the ability to adaptively optimize based on real-time operational data, making it difficult to cope with dynamic changes in heating systems. With the widespread adoption of IoT technology in heating stations, the acquisition of multi-source data has become possible, creating an urgent need for an intelligent prediction method that can integrate this data, consider inter-building thermal disturbances, and possess online learning capabilities to improve the overall energy efficiency of heating systems. Summary of the Invention

[0004] A method for predicting heating load in community buildings based on neural networks, comprising: S1. Based on the historical energy consumption and building attribute data collected by the heating station, the dynamic thermal inertia value of each building is calculated and output through an embedded neural network. S2. Using the thermal inertia value, the neural network further calculates the heating load adjustment coefficient that reflects the heat transfer effect between buildings due to distance and wind direction. S3. Combining the heating load adjustment coefficient with real-time forecast data provided by the meteorological department, the neural network generates a preliminary heating load forecast value for the next 24 hours, taking into account the thermal disturbance of the building complex. S4. The system compares the preliminary heating load forecast with the actual load measured at the heating station, and uses the resulting prediction error to adjust the internal parameters of the neural network in real time to achieve online self-optimization of the model. S5. Based on the optimized model, output the final high-precision heating load prediction results for community buildings to guide the heat source plant to produce on demand and the dynamic scheduling of the pipeline network.

[0005] The method for predicting community building heating load based on neural networks, as described above, involves calculating and outputting the dynamic thermal inertia value of each building using historical energy consumption and building attribute data collected from heating stations, through an embedded neural network. The method includes the following sub-steps: Historical building energy consumption data is obtained from the data acquisition system of the heating station, including daily energy consumption records and hourly energy consumption data during the heating season; Extract building structural parameters from the building attribute database, including building area, wall thermal resistance, and window heat transfer coefficient; Using an embedded neural network, historical energy consumption data and building structural parameters are used as inputs to calculate dynamic thermal inertia values. These values ​​reflect the building's heat storage and release characteristics and are then output to subsequent processing modules.

[0006] The community building heating load prediction method based on neural networks, as described above, further calculates a heating load adjustment coefficient that reflects the heat transfer effect between buildings due to distance and wind direction using thermal inertia values. This includes the following sub-steps: Based on the spatial distance between buildings, obtain building location data from the geographic information system and construct a heat transfer network model for the building complex. By combining real-time wind direction and speed data obtained from the meteorological department interface, the heat transfer intensity between buildings is calculated. The wind direction and speed data are transmitted in real time through the meteorological monitoring station. Based on the heat transfer intensity and dynamic thermal inertia value, a heating load adjustment coefficient is calculated through a neural network. This coefficient is used to adjust the heating load of a single building to take into account the thermal disturbance effect of the building complex.

[0007] The method for predicting community building heating load based on neural networks, as described above, involves combining heating load adjustment coefficients with real-time forecast data provided by meteorological departments. The neural network generates a preliminary heating load forecast for the next 24 hours, taking into account thermal disturbances within the building complex. This includes the following sub-steps: The system obtains temperature, humidity, wind speed, and wind direction forecast data for the next 24 hours from the meteorological department's API interface. This data is updated in real time through a meteorological data server. The heating load adjustment coefficient and meteorological forecast data are input into the neural network to generate preliminary heating load prediction values; The predicted values ​​are smoothed, and abnormal fluctuations are eliminated through a moving average algorithm to ensure the continuity and stability of the prediction curve.

[0008] The aforementioned neural network-based method for predicting community building heating load involves comparing preliminary heating load predictions with actual load measurements from a heating station. The resulting prediction errors are used to adjust the internal parameters of the neural network in real time, enabling online self-optimization of the model. This method includes the following sub-steps: The data acquisition and monitoring control system directly connected to the heating station controller collects the measured heating load data of the heating station in real time. Calculate the error between the preliminary prediction and the measured value, and generate an error sequence for evaluating the prediction accuracy; The error sequence is used to adjust the weights and bias parameters of the neural network through the backpropagation algorithm to achieve online learning, and the adjustment step size is limited to prevent overshoot.

[0009] The method for predicting heating load in community buildings based on neural networks, as described above, involves adjusting the weights and bias parameters of the neural network using an error sequence through a backpropagation algorithm to achieve online learning and limiting the adjustment step size to prevent overshoot. The method includes the following sub-steps: Record historical adjustment trajectories, analyze the implicit patterns of parameter adjustment sequences through long short-term memory networks, and generate parameter prediction values ​​to accelerate the convergence process; A sliding window mechanism is adopted, using only the error data of the most recent 24 hours for adjustment, in order to adapt to system changes and seasonal load fluctuations; When the prediction error continues to exceed a preset threshold, the neural network structure is reconstructed, increasing the number of hidden layer neurons to improve the model capacity.

[0010] The community building heating load prediction method based on neural networks, as described above, outputs a final high-precision heating load prediction result for the community buildings based on the optimized model, which guides the heat source plant's on-demand production and the dynamic scheduling of the pipeline network. This includes the following sub-steps: The final heating load forecast results are sent to the production scheduling system of the heat source plant and transmitted via industrial communication protocols. Based on the prediction results, dynamic scheduling instructions for the pipeline network are generated to adjust the valve opening and pump station frequency of each branch line. The system monitors prediction accuracy and system energy consumption in real time. When the deviation exceeds the safe range, it automatically activates the emergency adjustment mode to ensure the stability of community heating.

[0011] A computer storage medium, characterized in that it comprises: at least one memory and at least one processor; Memory, used to store one or more program instructions; A processor for running one or more program instructions to execute a neural network-based method for predicting heating loads in community buildings, as described in any of the preceding embodiments.

[0012] The beneficial effects achieved by this invention are as follows: (1) In terms of data integration and feature extraction, a multi-source data input channel was constructed by integrating historical energy consumption and building attribute data collected by the heating station and real-time forecast information provided by the meteorological department. The dynamic thermal inertia value of the building was accurately calculated by using embedded neural network, which provided a comprehensive and accurate data foundation for load forecasting and significantly improved the reliability of model input.

[0013] (2) In terms of dynamic modeling and disturbance compensation, a heating load adjustment coefficient reflecting the heat transfer effect between buildings was innovatively introduced. By combining neural networks with geographic information and real-time meteorological data, the impact of thermal disturbance of building groups on the load was effectively quantified, making the prediction results more in line with the actual heating needs of complex community environments.

[0014] (3) In terms of model self-optimization and accuracy improvement, by comparing the preliminary predicted value with the measured load of the heating station in real time, and using the prediction error to adjust the neural network parameters, an online self-optimization mechanism was established, which enables the model to continuously adapt to system changes and seasonal fluctuations, and ensure the accuracy and stability of long-term prediction.

[0015] (4) In terms of practical application and system integration, the final high-precision heating load prediction results directly serve the production plan of the heat source plant and the dynamic scheduling of the pipeline network, realizing closed-loop management from prediction to control, effectively improving the energy efficiency ratio and response speed of the community heating system, and providing reliable technical support for the refined operation of the smart heating system. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of a community building heating load prediction method based on neural networks provided in an embodiment of this application. Detailed Implementation

[0018] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1 As shown in the figure, an embodiment of this application provides a method for predicting the heating load of community buildings based on neural networks, including: Step S1: Based on the historical energy consumption and building attribute data collected by the heating station, calculate and output the dynamic thermal inertia value of each building through an embedded neural network. Specifically, historical energy consumption data of the building is acquired through the data acquisition system of the heating station, including daily energy consumption records and hourly energy consumption data during the heating season. Building structural parameters, such as building area, wall thermal resistance, and window heat transfer coefficient, are extracted from the building attribute database. Using an embedded neural network, with historical energy consumption data and building structural parameters as input, a dynamic thermal inertia value is calculated. This value characterizes the building's heat storage and release characteristics and is output to the subsequent processing module, including the following sub-steps: Step S11: Obtain historical building energy consumption data from the data acquisition system of the heating station, including daily energy consumption records and hourly energy consumption data during the heating season; The system utilizes a distributed data acquisition and monitoring control system deployed at the heating stations to establish communication connections with various building subsystems via the OPC UA protocol, automatically collecting daily and hourly energy consumption data throughout the heating season. The system stores the raw data in a time-series database and sets data quality verification rules to mark and repair anomalies such as communication interruptions and out-of-bounds values.

[0020] Step S12: Extract building structural parameters from the building attribute database, including building area, wall thermal resistance, and window heat transfer coefficient; By constructing a digital twin model of the building, the system automatically extracts building geometric and thermal performance parameters from the BIM database. Semantic parsing technology is used to identify building component attributes and convert unstructured data into a standardized parameter matrix. For missing parameters, a machine learning model intelligently completes the data based on building type and age information.

[0021] Step S13: Using an embedded neural network, with historical energy consumption data and building structural parameters as input, calculate the dynamic thermal inertia value. The dynamic thermal inertia value reflects the building's heat storage and release characteristics and is output to the subsequent processing module. An embedded neural network architecture with an attention mechanism is designed, whose input layer receives a feature-engineered energy consumption sequence and a vector of building parameters. The hidden layers employ convolutional layers to extract local features, combined with long short-term memory units to capture long-term dependencies, and a formula is used to calculate the dynamic thermal inertia value: ,in, Represents architecture The dynamic thermal inertia value; This represents the Sigmoid activation function; This represents the output layer weight vector; Represents the time-series energy consumption feature vector; Indicates feature fusion operators; Indicates the total number of time steps; Indicates time step , No. Weight coefficients for each structural feature; Represents the structural feature weight matrix; Represents architecture The structural feature vector; Indicates the output layer bias term; This represents the scaling factor for maximum thermal inertia.

[0022] The process integrates the building's thermal response characteristics with the coupling effect of the external environment. The network training adopts an adaptive moment estimation algorithm, which optimizes the weight parameters through backpropagation, and finally outputs a dynamic thermal inertia index with clear physical meaning, providing a quantitative basis for subsequent heat transfer analysis.

[0023] Step S2: Using the thermal inertia value, the neural network further calculates the heating load adjustment coefficient that reflects the heat transfer effect between buildings due to distance and wind direction; Specifically, based on the spatial distance between buildings, building location data is obtained from a geographic information system to construct a heat transfer network model of the building complex; combined with real-time wind direction and speed data, obtained through a meteorological department interface and transmitted in real time by a meteorological monitoring station, the heat transfer intensity between buildings is calculated; based on the heat transfer intensity and dynamic thermal inertia value, a heating load adjustment coefficient is calculated using a neural network. This coefficient is used to adjust the heating load of individual buildings to comprehensively consider the impact of thermal disturbance effects of the building complex, including the following sub-steps: Step S21: Based on the spatial distance between buildings, obtain building location data from the geographic information system and construct a heat transfer network model of the building complex; Step S22: Combine real-time wind direction and speed data, obtain them from the meteorological department interface, and calculate the heat transfer intensity between buildings. The wind direction and speed data are transmitted in real time through the meteorological monitoring station. The system accesses the meteorological data service platform via a highly secure communication protocol to acquire encrypted gridded wind direction and speed forecast data in real time, and uses digital signature technology to verify data integrity. A spatial interpolation algorithm maps the wind direction and speed data from meteorological observation points to the locations of various buildings, and a formula is used to calculate the wind field correction factor. ,in, Indicates the wind field correction factor; Represents the wind speed vector; This represents the position vector between buildings; The coefficients, representing empirical parameters, are obtained through experimental or numerical simulation fitting and are used to adjust the shape of the first exponential term to characterize the nonlinear relationship between the wind field and the position vector. This represents the empirical parameter, the coefficient of the empirical fit, which controls the power weight of the first factor and affects the magnitude of the wind field correction. This represents the empirical parameters, the coefficients of the empirical fit, the power of the ratio of the control wind speed vector magnitude to the reference wind speed, and characterizes the contribution of wind speed to wind field correction. This represents the empirical parameters, the coefficients of the empirical fit, the rate of adjustment of the exponential decay term, and reflects the decay characteristics of the position vector's correction of the wind field. These represent empirical parameters, empirical coefficients related to the distribution or diffusion characteristics of the location vector, used to refine the influence of the location vector on the wind field; This indicates the reference wind speed.

[0024] This factor is based on the angle between the wind direction vector and the line connecting the buildings, as well as the wind speed.

[0025] The heat transfer intensity calculation integrates the distance attenuation model and the wind field enhancement effect, and uses the formula to calculate the convective heat transfer coefficient: ,in, Represents architecture and The convective heat transfer coefficient between them; Represents the Nusselt number; Indicates the thermal conductivity of air; Indicates the length of architectural features; These represent empirical coefficients, obtained through experimental or numerical simulation fitting, used to adjust the shape of the hyperbolic tangent term and characterize the correction law when the Reynolds number deviates from the critical value. Represents the hyperbolic tangent function; Represents the Reynolds number; Indicates the critical Reynolds number; This represents the wind field correction factor.

[0026] This coefficient comprehensively considers the characteristics of the atmospheric boundary layer, the geometric features of the building facade, and the influence of local topographical obstruction. A real-time feedback mechanism is incorporated into the calculation process to dynamically adjust the heat flux rate parameter, ensuring the accuracy and environmental adaptability of the heat transfer intensity output.

[0027] Step S23: Based on the heat transfer intensity and dynamic thermal inertia value, calculate the heating load adjustment coefficient through a neural network. This coefficient is used to adjust the heating load of a single building to take into account the thermal disturbance effect of the building group. A graph neural network-based regression model is designed, using the heat transfer intensity matrix and dynamic thermal inertia values ​​as input feature vectors. The network architecture employs a multi-head attention mechanism, aggregating the thermal impact information of neighboring buildings through a feature cross-layer, calculating attention weights between nodes, and prioritizing paths with high heat transfer intensity and significant differences in thermal inertia. Historical load data and meteorological sequences are used as supervision signals during training. The network parameters are optimized using a backpropagation algorithm, and the heating load adjustment coefficient is calculated using the following formula: ,in, Represents architecture The heating load adjustment coefficient; This represents the Sigmoid activation function; Indicates the bias term; Represents a learnable weight vector; Indicates the transpose operation; Represents architecture The input feature vector; Indicates learnable parameters; Represents architecture Adjacent buildings with thermal interaction ; Represents architecture and The thermal interaction coefficient between them; Represents architecture and The amount of heat transfer between them; Normalization factor representing heat transfer rate; Represents architecture The dynamic thermal inertia value; Characterization factor representing thermal inertia; This represents a learnable parameter that controls the weights of the spatial gradient term of thermal inertia. Represents the hyperbolic tangent function; Represents architecture The spatial gradient norm of thermal inertia; This represents the maximum norm of the spatial gradient of thermal inertia.

[0028] This coefficient outputs a characteristic scalar value for each building, quantifying the degree to which it is affected by thermal disturbances from surrounding buildings. The model integrates an online learning module that dynamically adjusts network weights based on real-time prediction errors to adapt to seasonal load fluctuations and changes in building usage patterns.

[0029] Step S3: Combining the heating load adjustment coefficient with real-time forecast data provided by the meteorological department, the neural network generates a preliminary heating load forecast value for the next 24 hours, taking into account the thermal disturbance of the building complex. Specifically, the system obtains 24-hour forecast data on temperature, humidity, wind speed, and wind direction through a meteorological department's API interface; this data is updated in real time via a meteorological data server. The heating load adjustment coefficient and the meteorological forecast data are input into a neural network to generate a preliminary heating load prediction. The predicted value is then smoothed using a moving average algorithm to eliminate abnormal fluctuations and ensure the continuity and stability of the prediction curve. This process includes the following sub-steps: Step S31: Obtain the temperature, humidity, wind speed, and wind direction forecast data for the next 24 hours from the meteorological department's API interface. The data is updated in real time through the meteorological data server. Step S32: Input the heating load adjustment coefficient and meteorological forecast data into the neural network to generate a preliminary heating load prediction value; A multi-source heterogeneous data fusion framework is constructed, which deeply integrates the heating load adjustment coefficient, which characterizes the thermal disturbance effect of building clusters, with multi-dimensional meteorological forecast data at the feature level. Feature cross-processing and attention weighting mechanisms are used to enhance the model's ability to extract key features. A hybrid neural network architecture with gated recurrent units and convolutional layers is designed, and feature interaction is performed in the deep network to generate preliminary heating load predictions. A multi-task learning framework is used to simultaneously optimize short-term prediction accuracy and long-term trend consistency. The network output layer uses a linear activation function to generate load values ​​that conform to physical meaning, and combines historical load patterns for rationality verification to ensure that the prediction results reflect both the impact of the external environment and the energy consumption patterns of buildings.

[0030] Step S33: Smooth the predicted values ​​and eliminate abnormal fluctuations by using a moving average algorithm to ensure the continuity and stability of the prediction curve. A post-processing pipeline for prediction is established. To address potential local fluctuations and outliers in the initial prediction sequence, an adaptive smoothing algorithm based on variational mode decomposition is designed. This algorithm decomposes the original prediction sequence into multiple intrinsic mode functions, identifies and separates high-frequency noise components, dynamically determines the smoothing intensity of each data point based on prediction confidence and historical error distribution, and uses multi-scale moving average filtering to maintain the key features of the prediction curve at different time scales while suppressing random fluctuations. A boundary effect correction mechanism is also introduced to reduce the impact of smoothing on the sequence endpoints. The final output is a smooth prediction curve that maintains the true load change characteristics and has good continuity, providing a stable and reliable input for thermal system scheduling decisions.

[0031] Step S4: The system compares the preliminary heating load prediction with the actual load measured at the heating station, and uses the generated prediction error to adjust the internal parameters of the neural network in real time to achieve online self-optimization of the model. Specifically, the system collects real-time measured heating load data of the heating station through a data acquisition and monitoring control system directly connected to the heating station controller. It calculates the error between the preliminary predicted value and the measured value, generates an error sequence for evaluating prediction accuracy, and uses this error sequence to adjust the weights and bias parameters of the neural network through a backpropagation algorithm to achieve online learning. Simultaneously, it limits the adjustment step size to prevent overshoot. The system includes the following sub-steps: Step S41: Collect the measured heating load data of the heating station in real time through the data acquisition and monitoring control system directly connected to the heating station controller; Step S42: Calculate the error between the preliminary predicted value and the measured value, and generate an error sequence for evaluating the prediction accuracy; The system employs a dynamic time warping algorithm to align the predicted and measured sequences, eliminating time-series biases caused by sampling delays. The error calculation module generates an error sequence based on multi-dimensional indicators, including absolute error, relative error, and second-order difference error, and calculates error statistics such as the moving mean and variance. The error sequence undergoes multi-resolution analysis using wavelet transform to extract high-frequency fluctuations and low-frequency trend components, thereby constructing the error energy spectrum and autocorrelation function. The system also integrates an anomaly detection mechanism, identifying systematic biases based on the kurtosis and skewness of the error distribution, and generating an error evolution map for model health assessment.

[0032] Step S43: Use the error sequence to adjust the weights and bias parameters of the neural network through the backpropagation algorithm to achieve online learning, and limit the adjustment step size to prevent overshoot; Step S431: Record the historical adjustment trajectory, analyze the implicit patterns of the parameter adjustment sequence through the Long Short-Term Memory Network, and generate parameter prediction values ​​to accelerate the convergence process; The system constructs a parameter evolution knowledge base, recording the changes in the weight matrix and bias vector in each optimization in a time-series format. A Long Short-Term Memory (LSTM) network accepts the parameter sequence as input, calculates the hidden states and cell states, and captures long-term dependencies through a gating mechanism. The network output layer generates parameter predictions, which are used to initialize the initial parameters for the next round of optimization, thereby reducing the number of convergence iterations. The system also introduces an attention mechanism to enhance focus on key parameter adjustment events and improve prediction accuracy.

[0033] Step S432: Adopt a sliding window mechanism and adjust using only the error data of the most recent 24 hours to adapt to system changes and seasonal load fluctuations; Step S433: When the prediction error continues to exceed the preset threshold, trigger the reconstruction of the neural network structure and increase the number of hidden layer neurons to improve the model capacity. The system sets multiple error thresholds. When the moving average error exceeds the threshold for several consecutive periods, the network reconstruction module is activated. The correlation between model complexity and error is analyzed, and a hidden layer expansion strategy is determined based on gradient importance analysis. The reconstruction process includes adding neurons, reinitializing connection weights while retaining the knowledge transfer from the original network, and optimizing the topology using an evolutionary algorithm. The reconstructed model is compressed and quantized to ensure efficient deployment on resource-constrained devices.

[0034] Step S5: Based on the optimized model, output the final high-precision heating load prediction results for community buildings to guide the heat source plant to produce on demand and the dynamic scheduling of the pipeline network. Specifically, the final heating load forecast results are sent to the production scheduling system of the heat source plant via an industrial communication protocol to guide on-demand production; dynamic scheduling instructions for the pipeline network are generated based on the forecast results to adjust the valve opening and pump station frequency of each branch line, thereby achieving dynamic scheduling of the pipeline network; at the same time, the forecast accuracy and system energy consumption are monitored in real time, and an emergency adjustment mode is automatically activated when the deviation exceeds the safe range to ensure the stability of community heating, including the following sub-steps: Step S51: Send the final heating load forecast results to the production scheduling system of the heat source plant via industrial communication protocol. Step S52: Based on the prediction results, generate dynamic scheduling instructions for the pipeline network and adjust the valve opening and pump station frequency of each branch line; Based on the final load forecast results, a pipeline network optimization scheduling model is established. The system solves the pipeline network operation optimization problem in real time and uses formulas to calculate the optimal flow distribution for each branch: ,in, Indicates branch In time Traffic; Indicates time The load forecast value; Indicates fluid density; Indicates the specific heat capacity of a fluid; Indicates the design temperature difference; Indicates branch The drag coefficient; Indicates the total number of branches; Indicates branch The drag coefficient; This indicates a branch index.

[0035] Based on this, valve opening adjustment commands are generated. Simultaneously, based on load change gradient prediction, the pump station frequency adjustment is analyzed to achieve precise matching between the distribution system and the heat load. After security verification, the dispatch commands are sent to the intelligent actuators of each branch line via industrial Ethernet. The valve controller uses a fuzzy PID algorithm to achieve precise control of the valve opening, and the pump station frequency converter smoothly adjusts the operating frequency based on model predictive control technology, ensuring that the pipeline system is always in optimal operating condition.

[0036] Step S53: Monitor the prediction accuracy and system energy consumption in real time. When the deviation exceeds the safe range, automatically start the emergency adjustment mode to ensure the stability of community heating. A multi-source data fusion monitoring and early warning platform is constructed to collect system operating status parameters in real time. The dynamic deviation between predicted and measured values ​​is analyzed, and the total energy consumption of the system is monitored through energy metering devices to assess system energy efficiency. When the prediction deviation continuously exceeds a safety threshold or the system energy efficiency significantly decreases, the intelligent decision engine automatically triggers an emergency adjustment mode. The system simultaneously activates redundant control loops, ensuring control reliability through multi-controller collaboration, and pushes early warning information to maintenance personnel in real time, forming a complete emergency support system to maximize the stability and safety of the heating system.

[0037] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor; The memory is used to store one or more program instructions; A processor for running one or more program instructions to execute a neural network-based method for predicting heating loads in community buildings; Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide a method for predicting heating loads in community buildings based on neural networks.

[0038] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions that, when executed on a computer, cause the computer to perform the aforementioned neural network-based method for predicting heating loads in community buildings.

[0039] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0040] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0041] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0042] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0043] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0044] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0045] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0046] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting heating load in community buildings based on neural networks, characterized in that, include: S1. Based on the historical energy consumption and building attribute data collected by the heating station, the dynamic thermal inertia value of each building is calculated and output through an embedded neural network. S2. Using the thermal inertia value, the neural network further calculates the heating load adjustment coefficient that reflects the heat transfer effect between buildings due to distance and wind direction. S3. Combining the heating load adjustment coefficient with real-time forecast data provided by the meteorological department, the neural network generates a preliminary heating load forecast value for the next 24 hours, taking into account the thermal disturbance of the building complex. S4. The system compares the preliminary heating load forecast with the actual load measured at the heating station, and uses the resulting prediction error to adjust the internal parameters of the neural network in real time to achieve online self-optimization of the model. S5. Based on the optimized model, output the final high-precision heating load prediction results for community buildings to guide the heat source plant to produce on demand and the dynamic scheduling of the pipeline network.

2. The method for predicting community building heating load based on neural networks according to claim 1, characterized in that, Based on historical energy consumption and building attribute data collected from heating stations, the dynamic thermal inertia value of each building is calculated and output using an embedded neural network, including the following sub-steps: Historical building energy consumption data is obtained from the data acquisition system of the heating station, including daily energy consumption records and hourly energy consumption data during the heating season; Extract building structural parameters from the building attribute database, including building area, wall thermal resistance, and window heat transfer coefficient; Using an embedded neural network, historical energy consumption data and building structural parameters are used as inputs to calculate dynamic thermal inertia values. These values ​​reflect the building's heat storage and release characteristics and are then output to subsequent processing modules.

3. The method for predicting heating load in community buildings based on neural networks according to claim 1, characterized in that, Using the thermal inertia value, the neural network further calculates the heating load adjustment coefficient, which reflects the heat transfer effect between buildings due to distance and wind direction, including the following sub-steps: Based on the spatial distance between buildings, obtain building location data from the geographic information system and construct a heat transfer network model for the building complex. By combining real-time wind direction and speed data obtained from the meteorological department interface, the heat transfer intensity between buildings is calculated. The wind direction and speed data are transmitted in real time through the meteorological monitoring station. Based on the heat transfer intensity and dynamic thermal inertia value, a heating load adjustment coefficient is calculated through a neural network. This coefficient is used to adjust the heating load of a single building to take into account the thermal disturbance effect of the building complex.

4. The method for predicting heating load in community buildings based on neural networks according to claim 1, characterized in that, Combining the heating load adjustment coefficient with real-time forecast data provided by the meteorological department, the neural network generates a preliminary heating load forecast for the next 24 hours, taking into account the thermal disturbance of the building complex. This includes the following sub-steps: The system obtains temperature, humidity, wind speed, and wind direction forecast data for the next 24 hours from the meteorological department's API interface. This data is updated in real time through a meteorological data server. The heating load adjustment coefficient and meteorological forecast data are input into the neural network to generate preliminary heating load prediction values; The predicted values ​​are smoothed, and abnormal fluctuations are eliminated through a moving average algorithm to ensure the continuity and stability of the prediction curve.

5. The method for predicting community building heating load based on neural networks according to claim 1, characterized in that, The system compares the preliminary heating load forecast with the actual load measured at the heating station, and uses the resulting prediction error to adjust the internal parameters of the neural network in real time, achieving online self-optimization of the model. This includes the following sub-steps: The data acquisition and monitoring control system directly connected to the heating station controller collects the measured heating load data of the heating station in real time. Calculate the error between the preliminary prediction and the measured value, and generate an error sequence for evaluating the prediction accuracy; The error sequence is used to adjust the weights and bias parameters of the neural network through the backpropagation algorithm to achieve online learning, and the adjustment step size is limited to prevent overshoot.

6. The method for predicting heating load in community buildings based on neural networks according to claim 5, characterized in that, Using an error sequence to adjust the weights and bias parameters of a neural network via backpropagation enables online learning, while limiting the adjustment step size to prevent overshoot. This includes the following sub-steps: Record historical adjustment trajectories, analyze the implicit patterns of parameter adjustment sequences through long short-term memory networks, and generate parameter prediction values ​​to accelerate the convergence process; A sliding window mechanism is adopted, using only the error data of the most recent 24 hours for adjustment, in order to adapt to system changes and seasonal load fluctuations; When the prediction error continues to exceed a preset threshold, the neural network structure is reconstructed, increasing the number of hidden layer neurons to improve the model capacity.

7. The method for predicting heating load in community buildings based on neural networks according to claim 1, characterized in that, Based on the optimized model, the final high-precision heating load prediction results for community buildings are output to guide the on-demand production of heat source plants and the dynamic scheduling of the pipeline network. This includes the following sub-steps: The final heating load forecast results are sent to the production scheduling system of the heat source plant and transmitted via industrial communication protocols. Based on the prediction results, dynamic scheduling instructions for the pipeline network are generated to adjust the valve opening and pump station frequency of each branch line. The system monitors prediction accuracy and system energy consumption in real time. When the deviation exceeds the safe range, it automatically activates the emergency adjustment mode to ensure the stability of community heating.

8. A computer storage medium, characterized in that, include: At least one memory and at least one processor; Memory, used to store one or more program instructions; A processor for running one or more program instructions to perform a neural network-based method for predicting heating loads in community buildings as described in any one of claims 1-7.

Images