A Machine Learning-Based Smart Heating Load Prediction Method and System

By collecting and processing multi-source data from heating units, dynamically reconstructing spatial influence relationships, and using machine learning models for heating load prediction, the problems of insufficient representation of the implicit state of heating unit operation behavior and insufficient adaptability of spatial influence relationships are solved, achieving more accurate and stable load prediction.

CN123117445APending Publication Date: 2026-11-17DALIAN WORMAN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611313717.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing machine learning-based heating load forecasting methods are unable to effectively characterize the implicit state of heating unit operation behavior, and their dynamic adaptability to spatial influence relationships is insufficient, resulting in large prediction errors.

Method used

By collecting multi-source basic data from heating units and standardizing the data to form a raw time-series data set, the spatial influence relationship of heating units is dynamically reconstructed based on the implicit operating status. A machine learning model is used for load prediction, and the model parameters are updated through prediction error information to achieve continuous optimization of heating load.

Benefits of technology

It improves the accuracy and stability of heating load forecasting, and can dynamically adapt to changes in the operating status of heating units, reducing forecasting errors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117445A_ABST
    Figure CN123117445A_ABST
Patent Text Reader

Abstract

This invention discloses a machine learning-based intelligent heating load prediction method and system, belonging to the field of heating load prediction technology. The method includes collecting multi-source basic data on the operation of heating units and standardizing it to form an original time-series data set; performing sequential learning on the operating behavior of the heating units based on the original time-series data set to generate implicit operating states of the heating unit operating modes; dynamically reconstructing the spatial influence relationships of the heating units based on the implicit operating states to generate a dynamic spatial relationship structure; and predicting and calculating the heating load using a machine learning load prediction model, adjusting the influence propagation intensity in the dynamic spatial relationship structure according to the implicit operating states to generate the heating load prediction result. This invention updates the parameters of the implicit operating state construction model and the heating unit spatial influence relationship model based on prediction error information, achieving continuous optimization of the prediction process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of heating load prediction technology, and in particular to a smart heating load prediction method and system based on machine learning. Background Technology

[0002] With the development of smart energy and artificial intelligence technologies, heating load forecasting is increasingly being applied to heating operation optimization and refined management. Traditional methods mainly rely on historical load data, environmental parameters, and equipment operating parameters to build predictive models. In recent years, machine learning methods have improved heating load forecasting capabilities by exploring the nonlinear relationships between operating data and load changes. Simultaneously, the application of multi-source information such as valve control data, dispatch command data, and heating response data provides a richer data foundation for heating load forecasting.

[0003] Existing machine learning-based heating load prediction methods mainly rely on historical loads and environmental factors, lacking effective representation of the implicit states formed by the operating behavior of heating units. At the same time, they usually use fixed spatial relationships to describe the influence propagation process between heating units, which is difficult to adapt to changes in spatial influence relationships caused by changes in operating status. Existing methods lack a mechanism to continuously optimize operating status and spatial relationships based on prediction error information. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a machine learning-based intelligent heating load prediction method that solves the problems of insufficient representation of the implicit state of heating operation behavior and insufficient dynamic adaptability of spatial influence relationships.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a smart heating load prediction method based on machine learning, which includes:

[0008] Collect multi-source basic data on the operation of heating units and perform standardized processing to form a raw time-series data set;

[0009] Based on the original time-series data set, the operating behavior of the heating unit is sequentially learned to generate the hidden operating state of the heating unit's operating mode.

[0010] Based on the hidden operating state, the spatial influence relationship of the heating unit is dynamically reconstructed to generate a dynamic spatial relationship structure.

[0011] Heating load is predicted and calculated using a machine learning load prediction model. The intensity of the influence propagation in the dynamic spatial relationship structure is adjusted according to the implicit operating state to generate heating load prediction results.

[0012] The heating load forecast results are smoothed by time continuity to obtain the corrected heating load forecast results;

[0013] The revised heating load prediction results are compared and analyzed with the actual heating load data to generate prediction error information, and the parameters of the hidden operation state construction model and the heating unit spatial influence relationship model are updated.

[0014] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, the original time-series data set includes,

[0015] The data acquisition equipment obtains equipment operation status data, environmental status data, heating load data, and control and regulation data during the operation of the heating unit, and generates multi-source basic data of the heating unit operation according to the acquisition time.

[0016] The multi-source basic data of the heating unit operation are standardized by time alignment, data format unification and data scale normalization to form a data set with a unified time identifier, thus obtaining the original time series data set.

[0017] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, wherein: the hidden operating state of the heating unit operation mode includes,

[0018] Based on the original time series data set, the data corresponding to each time point are sorted according to the timestamp, and the sorted data is grouped by sliding in a continuous time window. The changes in the operating status of the heating unit, the changes in control behavior, and the changes in load response within each time window are spliced ​​and encoded in the order of occurrence to form a sequence of heating unit operating behavior that reflects the continuous operation process.

[0019] A hidden operating state construction model is constructed based on the temporal continuity of the heating unit's operating behavior sequence. The hidden operating state construction model includes a sequence encoding layer and a hidden state mapping layer.

[0020] Among them, the sequence coding layer is used to encode the continuous state sequence of the heating unit's operating behavior, and the hidden state mapping layer is used to map the continuous state coding to a unified feature space to form the hidden state of the heating unit's operating behavior.

[0021] The heating unit's operating behavior sequence is input into the hidden operating state to construct a model. The sequence coding layer calculates the behavioral change amplitude between adjacent time nodes and the behavioral change pattern between consecutive time windows. The hidden state mapping layer fuses and encodes the behavioral change amplitude and behavioral change pattern to extract the temporal change characteristics of the heating unit's operating behavior.

[0022] Based on the temporal variation characteristics of the heating unit's operating behavior, the feature correlation between different operating behaviors is calculated, and the temporal variation characteristics of the heating unit's operating behavior are aggregated and mapped according to the feature correlation to generate the hidden operating state of the heating unit's operating mode.

[0023] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, the dynamic spatial relationship structure includes,

[0024] Calculate the spatial influence weights between different heating units based on the implicit operating state of the heating unit's operating mode;

[0025] The spatial influence weight change is calculated based on the characteristic distance between the hidden operating states of the heating unit operation modes, and the spatial influence weight between different heating units is iteratively updated using the spatial influence weight change to form the spatial influence relationship of heating units.

[0026] Based on the spatial influence relationship of heating units, the updated spatial influence weights between each heating unit are read. The connection relationship between heating units is adjusted according to the change in spatial influence weights. The topological association between heating units is reconstructed based on the adjusted connection relationship to generate a dynamic spatial relationship structure.

[0027] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, the heating load prediction results include:

[0028] The connection relationship of heating units in the dynamic spatial relationship structure is used as the spatial association input, and a mapping relationship between the spatial association information between heating units and the heating load is established to form a machine learning load prediction model.

[0029] The implicit operating state of the heating unit operation mode is input into the machine learning load prediction model, and the influence propagation correction coefficient corresponding to each connection relationship in the dynamic spatial relationship structure is calculated based on the implicit operating state of the heating unit operation mode.

[0030] Based on the influence propagation correction coefficient, the influence propagation path in the dynamic spatial relationship structure is adaptively adjusted to obtain the adjusted dynamic spatial relationship structure.

[0031] The adjusted dynamic spatial relationship structure is input into the machine learning load prediction model. Based on the adjusted dynamic spatial relationship structure, the spatial influence propagation results between heating units within the target time period are determined. The heating load is then predicted and analyzed in combination with the spatial influence propagation results to generate the heating load prediction results.

[0032] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, the corrected heating load prediction result includes:

[0033] Based on the heating load forecast results, establish the continuous change relationship between adjacent forecast times in chronological order, identify the abnormal load change intervals in the heating load forecast results, and form time continuity correction information;

[0034] Based on the time continuity correction information, the continuity constraint adjustment is applied to the abnormal load change interval in the heating load prediction results to obtain the corrected heating load prediction results.

[0035] As a preferred embodiment of the machine learning-based intelligent heating load prediction method of the present invention, the step of updating the parameters of the hidden operating state construction model and the heating unit spatial influence relationship model includes:

[0036] The revised heating load forecast results are matched with the actual heating load data according to the same time node. The prediction deviation and change characteristics between the revised heating load forecast results and the actual heating load data are analyzed to generate prediction error information.

[0037] The prediction error information is input into the hidden operating state construction model. Based on the prediction error information, the deviation distribution between the hidden operating state of the heating unit operation mode and the actual heating load data is identified, and the parameters in the hidden operating state construction model are adjusted according to the deviation distribution.

[0038] The prediction error information is input into the spatial influence relationship modeling model of the heating unit. Based on the prediction error information, the propagation deviation between the spatial influence relationship of the heating unit and the actual heating load data is identified, and the parameters in the spatial influence relationship modeling model of the heating unit are adjusted according to the propagation deviation.

[0039] Secondly, this invention provides a machine learning-based intelligent heating load prediction system, comprising:

[0040] The processing module collects multi-source basic data on the operation of the heating unit and performs standardized processing to form a raw time-series data set;

[0041] The learning module performs serialized learning on the operating behavior of the heating unit based on the original time-series data set, and generates the hidden operating state of the heating unit's operating mode.

[0042] The reconstruction module dynamically reconstructs the spatial influence relationship of the heating unit based on the hidden operating state, and generates a dynamic spatial relationship structure.

[0043] The adjustment module uses a machine learning load prediction model to predict and calculate the heating load, and adjusts the intensity of the influence propagation in the dynamic spatial relationship structure according to the hidden operating state to generate the heating load prediction result.

[0044] The correction module performs time continuity smoothing correction on the heating load forecast results to obtain the corrected heating load forecast results;

[0045] The update module compares and analyzes the error between the corrected heating load prediction results and the actual heating load data, generates prediction error information, and updates the parameters of the hidden operation state construction model and the heating unit spatial influence relationship modeling model.

[0046] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the machine learning-based smart heating load prediction method described in the first aspect of the present invention.

[0047] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the machine learning-based smart heating load prediction method as described in the first aspect of the present invention.

[0048] The beneficial effects of this invention are as follows: By collecting multi-source basic data during the operation of heating units and performing standardized processing, a unified original time-series data set is constructed; based on the original time-series data set, the operating behavior of heating units is sequentially learned, the temporal change characteristics of operating behavior are extracted, and the implicit operating state of the heating unit operating mode is generated, thereby achieving effective characterization of the potential operating mode of the heating unit; based on the implicit operating state, the spatial influence weights between different heating units are dynamically adjusted, the spatial influence relationship of heating units is reconstructed, and a dynamic spatial relationship structure is generated, thereby achieving dynamic adaptation of the spatial influence relationship of heating units with changes in operating state; a machine learning load prediction model is constructed using the dynamic spatial relationship structure and the implicit operating state, and the utilization capability of spatial correlation information in the heating load prediction process is improved by adjusting the influence propagation path, and the stability of the prediction results is improved by combining time continuity correction; based on the prediction error information, the parameters of the implicit operating state construction model and the heating unit spatial influence relationship model are updated, thereby achieving continuous optimization of the prediction process. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of a machine learning-based smart heating load prediction method.

[0051] Figure 2 This is a schematic diagram of a machine learning-based smart heating load prediction system. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a smart heating load prediction method based on machine learning, including the following steps:

[0056] S1. Collect multi-source basic data on the operation of the heating unit and perform standardized processing to form a raw time-series data set.

[0057] S1.1 The data acquisition equipment obtains equipment operation status data, environmental status data, heating load data and control adjustment data during the operation of the heating unit, and generates multi-source basic data of the heating unit operation according to the acquisition time.

[0058] Furthermore, equipment operation status data reflects changes in the status of equipment within the heating unit during operation; environmental status data reflects changes in the environmental conditions surrounding the heating unit; heating load data reflects changes in the actual heat demand of the heating unit; and control and regulation data reflects changes in the regulatory behavior of the heating unit during operation. The data acquisition equipment timestamps data from different sources according to the acquisition time, and correlates and arranges equipment operation status data, environmental status data, heating load data, and control and regulation data corresponding to different time points, ensuring that data from different sources correspond to the same operating time process. For example, during the operation of a heating unit, when the ambient temperature changes, environmental status data synchronously records the environmental change process, equipment operation status data records the equipment response process, control and regulation data records the regulation behavior change process, and heating load data records the load change process. By correspondingly associating them according to the collection time, the state change chain during the operation of the heating unit is completely preserved. When there is a situation where the equipment operation status of the heating unit changes but the heating load does not change immediately, by simultaneously collecting equipment operation status data and heating load data, the time relationship between equipment status change and load response is preserved. This avoids the problem of not being able to identify the equipment regulation lag effect when making predictions based solely on load changes, and generates multi-source basic data for the operation of the heating unit.

[0059] S1.2 Standardize the multi-source basic data of the heating unit operation. Through time alignment, data format unification and data scale normalization, a data set with a unified time identifier is formed, and the original time series data set is obtained.

[0060] Furthermore, based on the acquisition time during the operation of the heating unit, data from different sources are time-aligned, and data with time discrepancies are matched according to their corresponding acquisition times, ensuring that equipment operation status data, environmental status data, heating load data, and control and adjustment data correspond to the same time identifier. Secondly, the time-matched data undergoes unified data format processing, converting data generated by different acquisition devices into a unified data format, ensuring that data from different sources are stored and processed using a consistent data structure. After format unification, data scale normalization is performed, scaling the data according to the variation range of different data types, allowing data with different dimensions to be jointly analyzed within the same feature range. For example, when there are differences between the variation range of operating parameters in equipment operation status data and the variation range of environmental parameters in environmental status data, scale normalization reduces the dimensional influence between different data types, enabling the subsequent extraction of the implicit operating status of the heating unit's operating mode to simultaneously utilize information from different data types, resulting in the original time-series data set.

[0061] S2. Based on the original time-series data set, perform serialization learning on the operating behavior of the heating unit to generate the hidden operating state of the heating unit's operating mode.

[0062] S2.1 Based on the original time series data set, sort the data corresponding to each time point according to the timestamp, and then slide group the sorted data into continuous time windows. The changes in the operating status of the heating unit, the changes in control behavior, and the changes in load response within each time window are spliced ​​and encoded in the order of occurrence to form a sequence of heating unit operating behavior that reflects the continuous operation process.

[0063] Furthermore, the data corresponding to each time point is sorted according to the timestamp, so that the changes in equipment operating status, environmental status, control behavior, and heating load are arranged in the actual order of occurrence, forming a data arrangement result with a time sequence. The sorted data is then grouped into sliding groups using continuous time windows, which gradually cover the operating data of different time stages, so that each continuous time window contains the status change information of the heating unit during operation. For example, when changes in control and regulation behavior occur during the operation of a heating unit, these changes may not immediately cause changes in the heating load. By using continuous time windows to simultaneously retain changes in equipment operating status and load response before and after the change in control behavior, the problem of missing information in the operation process caused by analyzing only a single time point is avoided. The data in each time window is arranged according to the order of occurrence of changes in heating unit operating status, control behavior, and load response. Different types of change information are spliced ​​and encoded to form a continuous expression of different change processes within the same time window. For example, when changes in environmental conditions cause the heating unit to adjust its operating status, the changes in heating unit operating status, control behavior, and load response are connected sequentially according to time relationship to form a sequence of heating unit operating behavior reflecting the continuous operation process.

[0064] S2.2 Construct a hidden operating state construction model based on the temporal continuity of the heating unit's operating behavior sequence. The hidden operating state construction model includes a sequence encoding layer and a hidden state mapping layer.

[0065] The sequence coding layer is used to encode the continuous state sequence of the heating unit's operating behavior, and the hidden state mapping layer is used to map the continuous state code to a unified feature space to form the hidden state of the heating unit's operating behavior.

[0066] Furthermore, continuous time window data from the heating unit's operational behavior sequence is acquired. The changes in the heating unit's operational state, control behavior, and load response corresponding to each time window are used as sequence input information, forming a continuous input sequence according to the time window arrangement. A sequence coding layer is constructed based on the state change relationships between adjacent time windows in the heating unit's operational behavior sequence. This sequence coding layer performs correlation analysis on the changes in the heating unit's operational state, control behavior, and load response within adjacent time windows, extracting state transition relationships between different time stages and encoding the change trends within the continuous time windows. This transforms the heating unit's operational behavior sequence from discrete change information into a continuous state code with time correlation. For example, when the equipment's operational state remains stable but the control behavior is adjusted during heating unit operation, the sequence coding layer generates a corresponding state change code based on the state differences before and after the control behavior change within the continuous time window; when the heating load generates a continuous response due to environmental changes, the sequence coding layer forms a continuous state change code based on the load response change process.

[0067] Specifically, the continuous state codes output from the sequence coding layer are input into the hidden state mapping layer. This hidden state mapping layer performs feature mapping on the continuous state codes, fusing features from continuous state codes corresponding to different time stages. This transforms operational behavior information from different sources and with different change patterns into a unified feature space, forming a hidden state expression representing the operational patterns of the heating unit. For example, when multiple heating unit operational behavior sequences exhibit similar change processes, the hidden state mapping layer maps the corresponding continuous state codes to similar feature regions. When heating units exhibit different control behaviors or load response processes, the hidden state mapping layer forms a distinctive hidden state expression. Connecting the sequence coding layer and the hidden state mapping layer, and based on the data transfer relationship between the heating unit operational behavior sequence input, the continuous state code output, and the hidden state mapping output, a hidden operational state construction model is formed. This model completes the transformation from the heating unit operational behavior sequence to the hidden state of the heating unit's operational behavior, thus forming the hidden state of the heating unit's operational behavior.

[0068] Specifically, the implicit operating state construction model is divided into a sequence coding layer and an implicit state mapping layer. This layered processing approach transforms the operational behavior process into implicit operating states. The purpose of the sequence coding layer is to preserve the temporal evolution of the heating unit's operation, ensuring that changes in the heating unit's operating state, control behavior, and load response are no longer independent data features, but rather form a continuous state code with a sequential correlation. For example, after an adjustment in the heating unit's control behavior, the heating load may respond after a certain process; the sequence coding layer preserves the correlation between the control change and the load response. The purpose of the implicit state mapping layer is to reduce the differences in data representation between different types of operational behaviors, enabling changes in equipment state, control behavior, and load response to be expressed in a unified feature space. For example, different heating units may have different equipment parameters and operating modes, but when they share the same operational trend, the implicit state mapping layer forms similar implicit state representations. By extracting the time variation pattern through the sequence coding layer and uniformly expressing the continuous state encoding through the hidden state mapping layer, the hidden operating state construction model can extract the hidden states of the heating unit operating behavior that reflect the changing patterns of the heating unit operating mode from the sequence of heating unit operating behavior.

[0069] S2.3 Input the sequence of heating unit operation behavior into the hidden operation state to construct the model. The sequence coding layer calculates the behavior change amplitude between adjacent time nodes and the behavior change pattern between continuous time windows. The hidden state mapping layer fuses and encodes the behavior change amplitude and behavior change pattern to extract the temporal change characteristics of the heating unit operation behavior.

[0070] Furthermore, based on the differences between the operating behavior vectors of heating units corresponding to adjacent time nodes, the magnitude of behavioral changes between adjacent time nodes is calculated. By comparing the vector differences corresponding to changes in the operating state, control behavior, and load response of heating units within adjacent time nodes, the degree of change in the operating behavior of heating units in the time direction is obtained. The sequence of operating behavior of heating units is divided according to the length of the continuous time window. The operating behavior vectors of heating units corresponding to multiple time nodes contained in the continuous time window are combined to form the operating behavior sequence of the heating unit corresponding to the i-th continuous time window. The average value of the magnitude of behavioral changes within the continuous time window is calculated to characterize the overall trend of the operating behavior of heating units within the continuous time window. For example, when the heating unit is in a stable operating state, the magnitude of behavioral changes between adjacent time nodes is small, and the average value of the magnitude of behavioral changes within the continuous time window remains stable. When the heating unit is affected by environmental changes or control adjustments, the magnitude of behavioral changes between adjacent time nodes changes, and the average value of the magnitude of behavioral changes within the continuous time window reflects the process of change in operating state.

[0071] Specifically, based on the differences between the average values ​​of behavioral change amplitudes within adjacent consecutive time windows, the behavioral change pattern between consecutive time windows is calculated to determine whether the trend of heating unit operation behavior changes continuously. For example, when the behavioral change amplitude between consecutive time windows gradually increases, the behavioral change pattern between consecutive time windows reflects an increasing trend in the operating state of the heating unit; when the behavioral change amplitude between consecutive time windows remains close, the behavioral change pattern between consecutive time windows reflects a stable trend in the operating state of the heating unit. The behavioral change amplitudes between adjacent time nodes and the behavioral change patterns between consecutive time windows are input into the hidden state mapping layer, which fuses and encodes the behavioral change amplitudes and behavioral change patterns. By combining short-term time scale state change information and continuous time scale change trend information, the temporal feature set of the heating unit operation behavior is extracted, forming the temporal feature set of the hidden operation state construction model.

[0072] The magnitude of behavioral change between adjacent time points is defined as:

[0073] ;

[0074] in, For the first The magnitude of behavioral changes at each time point For the first The operational behavior vector of the heating unit corresponding to each time point.

[0075] Let the length of the continuous time window be... Then the first A continuous time window is represented as:

[0076] ;

[0077] in, For the first A sequence of heating unit operation behaviors corresponding to a continuous time window. For the first The operational behavior vector of the heating unit corresponding to each time point. For the first The operational behavior vector of the heating unit corresponding to each time point. For the first The operational behavior vector of the heating unit corresponding to each time point.

[0078] Calculate the average value of the behavioral variation within a continuous time window:

[0079] ;

[0080] in, For the first The average value of the variation in the operating behavior of the heating unit within a continuous time window. The length of the continuous time window. For the first The magnitude of changes in the operating behavior of the heating unit at each time point This refers to the time node number. For the first The starting time node for calculating the magnitude of behavioral changes within a continuous time window.

[0081] Define the behavior change pattern between consecutive time windows as follows:

[0082] ;

[0083] in, For the first The consecutive time window and the first The behavioral change pattern between consecutive time windows For the first The average value of the variation in the operating behavior of the heating unit within a continuous time window. For the first The average value of the variation in the operating behavior of the heating unit within a continuous time window.

[0084] ;

[0085] in, The set of temporal features for constructing a model of hidden running states.

[0086] S2.4 Calculate the feature correlation between different operating behaviors based on the temporal change characteristics of the heating unit's operating behavior, and perform aggregation mapping on the temporal change characteristics of the heating unit's operating behavior according to the feature correlation to generate the hidden operating state of the heating unit's operating mode.

[0087] Furthermore, the time-series change feature vector corresponding to the affected heating unit number is compared with the time-series change feature vector corresponding to the heating unit number that caused the impact. The feature correlation is determined based on the degree of feature matching between the change processes of the two heating units' operating behavior. For example, when the two heating units show similar change trends during environmental changes, equipment adjustments, and load responses, the feature correlation between the two heating units is high; when the change processes of the two heating units' operating behavior differ significantly, the feature correlation between the two heating units is low.

[0088] Specifically, the temporal variation characteristics of heating unit operation behavior are aggregated and mapped based on feature correlation relationships. Correlated temporal variation characteristics are merged, transforming the operational behavior influence relationships between different heating units into a unified state expression. The degree of feature correlation between different heating units is adjusted based on feature correlation relationship adjustment parameters. This ensures that the temporal variation characteristics of heating units with stronger influence relationships have a higher correlation effect during the aggregation process, while the correlation effect of the temporal variation characteristics of heating units with weaker influence relationships is reduced. For example, when the operational state change of an influencing heating unit continuously affects the load response of an affected heating unit, the correlation expression between the temporal variation characteristics of the two heating units is enhanced through feature correlation relationships. When there is no obvious influence relationship between the operation processes of two heating units, the correlation degree between the temporal variation characteristics of the two heating units is reduced through feature correlation relationships. Based on the aggregated and mapped temporal variation characteristics, a latent operational state of the heating unit operation mode is generated, allowing the latent operational state of the heating unit operation mode to simultaneously reflect the operational variation characteristics of an individual heating unit as well as the operational behavior correlation characteristics between different heating units.

[0089] The hidden running state expression is:

[0090] ;

[0091] in, for The hidden operating status of the affected heating unit numbers and the heating unit numbers that caused the impact. The time-series change feature vector of the affected heating unit number. The time-series variation feature vector of the heating unit numbers that have an impact. The parameter is used to adjust the feature association relationship. Number the affected heating units. Number the heating unit that is affected.

[0092] S3. Based on the hidden operating state, the spatial influence relationship of the heating unit is dynamically reconstructed to generate a dynamic spatial relationship structure.

[0093] S3.1 Calculate the spatial influence weights between different heating units based on the hidden operating state of the heating unit operation mode.

[0094] Furthermore, the implicit operating state vectors corresponding to different heating units are obtained. The implicit operating state vectors corresponding to the affected heating unit number and the heating unit number that generates the influence are correlated and calculated. The spatial influence weight is determined by the feature matching relationship between the two implicit operating state vectors. The implicit operating state vector of the affected heating unit number is used as the target state expression in the spatial influence relationship, and the implicit operating state vector of the heating unit number that generates the influence is used as the influence state expression in the spatial influence relationship. The spatial influence weight is calculated based on the correlation between the implicit operating state vectors. When the implicit operating state vectors of the two heating unit operating modes have high similarity in terms of state change trends, control behavior changes, and load response changes, the spatial influence weight between the two heating units increases, indicating a strong operational correlation between the two heating units. When the implicit operating state vectors of the two heating unit operating modes differ significantly, the spatial influence weight between the two heating units decreases, indicating a weak operational correlation between the two heating units. For example, when the load change process of one heating unit is synchronous with the equipment adjustment process of another heating unit, the spatial influence weight between the two heating units is increased by calculating the correlation between the implicit operating state vectors; when the operating state changes of the two heating units are independent of each other, the spatial influence weight between the two heating units is reduced by the difference in the implicit operating state vectors, thus obtaining the spatial influence weight between different heating units.

[0095] Specifically, the dynamic state changes of heating units during operation serve as the basis for judging spatial relationships, enabling spatial influence relationships to reflect the correlation changes during actual operation. For example, two heating units that are physically far apart may be correlated due to the same control strategy or similar operating states, and the above operational correlations can be identified through implicit operating states; two heating units that are physically close may have a weaker influence due to different operating modes, and the spatial influence weight can be reduced through differences in implicit operating states.

[0096] The expression for spatial influence weight is:

[0097] ;

[0098] in, for Spatial influence weights of the affected heating unit numbers and the heating unit numbers that generated the influence. for The transpose of the implicit operating state vector of the affected heating unit number. The implicit operating state vector for the heating unit number that has an impact. for The hidden operating state vector of the affected heating unit number.

[0099] S3.2 Calculate the change in spatial influence weight based on the characteristic distance between the hidden operating states of the heating unit operation modes, and use the change in spatial influence weight to iteratively update the spatial influence weight between different heating units to form the spatial influence relationship of heating units.

[0100] Furthermore, the implicit operating state vectors corresponding to the affected heating unit number and the heating unit number that generated the influence are obtained. The characteristic distance between the two implicit operating state vectors is calculated based on the degree of difference between them, representing the difference in the current operating modes of the two heating units. The spatial influence weight change is calculated based on the characteristic distance, and the degree of change corresponding to the characteristic distance is adjusted using a spatial influence weight change adjustment coefficient, so that the spatial influence weight change adapts to the degree of change in the operating mode. When the characteristic distance between the implicit operating states of the two heating units is small, it indicates that the current operating states of the two heating units are similar, the operational relationship between the two heating units has not changed significantly, and the spatial influence weight change is close to a small change state, keeping the spatial influence weight stable. For example, when the two heating units are continuously in similar operating states, the spatial influence weight change will not be frequently adjusted, avoiding invalid changes in the spatial influence relationship. When the characteristic distance between the implicit operating states of two heating unit operation modes gradually increases, it indicates that the operating modes of the two heating units have changed differently, and the spatial influence weight needs to be adjusted. The change in spatial influence weight increases with the increase in characteristic distance, so that the spatial influence relationship responds to the change in operating mode. The range of change in spatial influence weight is limited by the spatial influence weight change adjustment coefficient to avoid excessive change in spatial influence weight when the operating mode of the heating unit changes abruptly. The change in spatial influence weight is applied to the spatial influence weight between different heating units. Through multiple time update processes, the spatial influence weight is adjusted so that the spatial influence weight gradually matches the correlation degree after the change in the operating mode of the heating unit, thus forming the spatial influence relationship of the heating unit.

[0101] The expression for the ratio of changes in spatial influence weight is:

[0102] ;

[0103] in, for The change in spatial influence weight between the affected heating unit number and the heating unit number that generated the influence. This is the adjustment coefficient for the spatial influence weight change. The characteristic distance between the hidden operating status of the affected heating unit number and the heating unit number that caused the impact.

[0104] When the characteristic distance between the hidden operating states of the heating unit's operating mode is small.

[0105] →0;

[0106] The influence of spatial changes on weights is relatively small.

[0107] When the characteristic distance between the hidden operating states of the heating unit operation mode gradually increases.

[0108] → ;

[0109] This indicates that the change in the spatial influence weight gradually increases, but will not grow indefinitely, thus avoiding drastic fluctuations in the spatial influence weight due to sudden changes in the operating mode of the heating unit.

[0110] S3.3. Based on the spatial influence relationship of heating units, read the updated spatial influence weights between each heating unit, adjust the connection relationship between heating units according to the change in spatial influence weights, and reconstruct the topological association between heating units based on the adjusted connection relationship to generate a dynamic spatial relationship structure.

[0111] Furthermore, the spatial influence weight of the current time stage is compared with that of the previous time stage to obtain the changes in spatial influence weight. When the change in spatial influence weight reaches the condition for adjusting the connection relationship, the connection relationship between the corresponding heating units is adjusted. When the change in spatial influence weight indicates that the influence relationship between two heating units is strengthened, the connection strength between the two heating units is increased, and the connection relationship between the two heating units is established or strengthened, so that the heating units with stronger operational changes remain associated in the spatial relationship. When the change in spatial influence weight indicates that the influence relationship between two heating units is weakened, the connection strength between the two heating units is reduced, and the connection relationship between the two heating units is reduced or deleted, so that the spatial influence of the heating units with weaker operational association is reduced. For example, when the operating mode of one heating unit changes, the load response of another heating unit changes synchronously and intensifies. The change in spatial influence weight indicates that the correlation between the two heating units has strengthened, and the connection between the two heating units is adjusted accordingly. When the two heating units do not have a synchronous change relationship for a long period of time, the change in spatial influence weight indicates that the correlation between the two heating units has weakened, and the connection between the two heating units is adjusted accordingly. The topological relationship between the heating units is reconstructed based on the adjusted connection relationship, and the updated spatial influence weight is used as the expression of the influence intensity in the connection relationship. This allows the connection status and influence degree between the heating units to change synchronously, generating a dynamic spatial relationship structure.

[0112] S4. The heating load is predicted and calculated using a machine learning load prediction model. The intensity of the influence propagation in the dynamic spatial relationship structure is adjusted according to the implicit operating state to generate the heating load prediction result.

[0113] S4.1. The connection relationship of heating units in the dynamic spatial relationship structure is used as the spatial association input, and a mapping relationship between the spatial association information between heating units and the heating load is established to form a machine learning load prediction model.

[0114] Furthermore, the connection relationships of heating units in the dynamic spatial relationship structure are read, and the connection status, spatial influence weight, and connection direction between different heating units are organized as spatial association information, enabling the machine learning load prediction model to obtain the association relationships between heating units. For example, when the operating status change of one heating unit continuously affects the load change of another heating unit, the connection relationship in the dynamic spatial relationship structure reflects the influence path between the two heating units. When there is no operating association between two heating units, the connection relationship in the dynamic spatial relationship structure reduces the degree of information association between the two heating units. The spatial association information between heating units is matched with the heating load change process, and the machine learning load prediction model learns the heating load change patterns under different spatial association states, enabling the machine learning load prediction model to establish a mapping relationship between the spatial association information between heating units and the heating load. For example, under the same environmental conditions, different heating units may produce different load change trends due to different spatial association relationships. The machine learning load prediction model identifies the impact of spatial association changes on the heating load, using the heating unit connection relationships in the dynamic spatial relationship structure as spatial association input to form the machine learning load prediction model.

[0115] S4.2 Input the implicit operating status of the heating unit operation mode into the machine learning load prediction model, and calculate the influence propagation correction coefficients corresponding to each connection relationship in the dynamic spatial relationship structure based on the implicit operating status of the heating unit operation mode.

[0116] Furthermore, the connection relationship between the affected heating unit number and the heating unit number that generates the impact is used as the calculation object. The implicit operating states of the heating unit operation modes corresponding to the two heating units are obtained. Based on the correlation between the implicit operating states of the heating unit operation modes, the degree of impact propagation corresponding to different connection relationships is calculated, so that the impact propagation correction coefficient reflects the strength of spatial impact propagation under the current operating mode. For example, when the change in the operating mode of the heating unit that generates the impact is strongly correlated with the change in the load of the affected heating unit, the impact propagation correction coefficient of the corresponding connection relationship increases, strengthening the impact propagation path in the dynamic spatial relationship structure. When the correlation between the changes in the operating modes of the two heating units is weak, the impact propagation correction coefficient of the corresponding connection relationship decreases, weakening the impact propagation path in the dynamic spatial relationship structure. The impact propagation correction coefficient is correlated with the connection relationships in the dynamic spatial relationship structure, so that the connection relationship between each heating unit has an impact propagation correction coefficient that matches the current operating mode.

[0117] The expression for the influence propagation correction coefficient is:

[0118] ;

[0119] in, for The influence propagation correction coefficient between the affected heating unit number and the heating unit number that caused the influence. for The connection identifier between the affected heating unit number and the heating unit number that caused the impact.

[0120] S4.3. Based on the influence propagation correction coefficient, the influence propagation path in the dynamic spatial relationship structure is adaptively adjusted to obtain the adjusted dynamic spatial relationship structure.

[0121] Furthermore, the influence propagation correction coefficient is used to determine the changes in the influence propagation intensity corresponding to different connection relationships. When the influence propagation correction coefficient indicates that the influence of the current connection relationship is enhanced, the influence propagation intensity corresponding to the connection relationship is increased, so that the operational changes of the heating unit that generates the influence are transmitted more strongly to the affected heating unit along the corresponding influence propagation path. When the influence propagation correction coefficient indicates that the influence of the current connection relationship is weakened, the influence propagation intensity corresponding to the connection relationship is decreased, so that the influence propagation effect of the heating unit with a reduced correlation is reduced. For example, when the operating mode of a heating unit changes, the load response correlation between it and another heating unit is enhanced. By increasing the influence propagation intensity of the corresponding connection relationship, the influence propagation path between the two heating units is strengthened. When the operating mode changes of the two heating units are not synchronized, by decreasing the influence propagation intensity of the corresponding connection relationship, the influence propagation path between the two heating units is weakened. The connection state in the dynamic spatial relationship structure is updated again according to the adjusted connection relationship intensity, so that the influence propagation path in the dynamic spatial relationship structure matches the current changes in the operating mode of the heating unit, resulting in an adjusted dynamic spatial relationship structure.

[0122] S4.4 Input the adjusted dynamic spatial relationship structure into the machine learning load prediction model, determine the spatial influence propagation results between heating units within the target time period based on the adjusted dynamic spatial relationship structure, and perform prediction analysis on the heating load based on the spatial influence propagation results to generate heating load prediction results.

[0123] Furthermore, the spatial impact propagation results within the target time period are determined based on the propagation paths between different heating units. For example, when a change in the operating mode of a heating unit propagates to multiple heating units through enhanced connectivity, the machine learning load prediction model determines the spatial impact on the relevant heating units based on the propagation paths. When the connectivity between heating units weakens, the machine learning load prediction model reduces the corresponding spatial impact propagation results. The spatial impact propagation results are then fused and analyzed with the implicit operating states of the heating unit operating modes and the spatial correlation information in the dynamic spatial relationship structure. The machine learning load prediction model learns the heating load change patterns under different spatial impact propagation states. For example, when multiple heating units are simultaneously affected by the same operating mode change, the spatial impact propagation results reflect the load change correlation between multiple heating units; when a change in the operation of a single heating unit only affects local heating units, the spatial impact propagation results reflect the local load change relationship. Based on the spatial impact propagation results, the heating load within the target time period is predicted and analyzed to obtain the heating load prediction results for each heating unit.

[0124] Specifically, the adjusted dynamic spatial relationship structure is used as the spatial influence input for the machine learning load forecasting model. This allows the heating load forecasting process to comprehensively consider changes in the operating modes of heating units and the spatial influence propagation process. The dynamic spatial relationship structure determines the influence propagation results, so that heating load forecasting is no longer limited to the historical changes of individual heating units, but rather combines the influence propagation relationships between heating units for prediction. For example, when the heating demand in a certain area changes, the dynamic spatial relationship structure reflects the direction and intensity of influence propagation, enabling the machine learning load forecasting model to predict the load changes of related heating units; when the correlation between heating units weakens, the machine learning load forecasting model reduces the influence of irrelevant heating units.

[0125] S5. Perform time continuity smoothing correction on the heating load forecast results to obtain the corrected heating load forecast results.

[0126] S5.1 Based on the heating load forecast results, establish the continuous change relationship between adjacent forecast times in chronological order, identify the abnormal load change intervals in the heating load forecast results, and form time continuity correction information.

[0127] Furthermore, based on the changes in heating load forecast results between adjacent forecast times, it is determined whether there are abnormal change intervals in the heating load forecast results that do not conform to the continuous operation pattern. For example, when the heating load forecast results maintain a stable change between adjacent forecast times, but a sudden increase or decrease occurs at a certain forecast time, the corresponding time range is identified as an abnormal load change interval. When the heating load forecast results show a continuous change trend with changes in the environment or operating status, the corresponding change process is identified as a normal load change interval. Time continuity correction information is generated based on the time position, direction of change, and degree of change corresponding to the abnormal load change interval. The time continuity correction information is used to record the positions and directions of continuous adjustments required in the heating load forecast results. For example, when local changes in the heating load forecast results do not match the change trend between adjacent forecast times, the time continuity correction information records the corresponding abnormal interval. When the heating load forecast results maintain a continuous change overall, the time continuity correction information records lower adjustment needs to reduce the impact on normal forecast results. The time change relationship in the heating load forecast results is analyzed to form time continuity correction information.

[0128] S5.2. Based on the time continuity correction information, the abnormal load change interval in the heating load prediction results is adjusted by continuity constraint to obtain the corrected heating load prediction results.

[0129] Furthermore, based on the heating load forecast results corresponding to adjacent forecast times before and after the abnormal load change interval, the continuous change trend that the abnormal load change interval needs to meet is determined. The heating load forecast results within the abnormal load change interval are compared with the forecast times before and after to determine whether the abnormal change disrupts the overall change trend of the heating load forecast results. For example, when the heating load forecast result within the abnormal load change interval shows a rapid increase compared to the previous forecast time, and then returns to the original change trend at the next forecast time, it is determined that there is a sudden change in the abnormal load change interval, and the degree of abnormal change needs to be reduced. When the heating load forecast result within the abnormal load change interval maintains a continuous change relationship with the forecast times before and after, no adjustment is made to the corresponding forecast results. For the abnormal load change interval that needs adjustment, the direction of the continuity constraint is determined based on the time continuity correction information. By adjusting the change relationship between the forecast value within the abnormal interval and the adjacent forecast times, the heating load forecast result within the abnormal interval gradually approaches the continuous change trend formed by the forecast times before and after. When the abnormal load change interval manifests as a localized sudden increase, the variation amplitude of the heating load forecast results within the abnormal interval is reduced to maintain a smooth connection between the load change within the abnormal interval and the preceding and following time points. When the abnormal load change interval manifests as a localized sudden decrease, the continuity of the heating load forecast results within the abnormal interval is increased to restore the load change within the abnormal interval to a continuous trend. During the adjustment process, the overall direction of change in the heating load forecast results is preserved, and only the abnormal load change interval marked by the time continuity correction information is locally adjusted to avoid the continuity constraint adjustment affecting the normal heating load change process. This corrects the abnormal load fluctuations in the heating load forecast results, resulting in the corrected heating load forecast results.

[0130] S6. Compare and analyze the error between the corrected heating load prediction results and the actual heating load data to generate prediction error information, and update the parameters of the hidden operation state construction model and the heating unit spatial influence relationship modeling model.

[0131] S6.1 Match the corrected heating load forecast results with the actual heating load data according to the same time node, analyze the prediction deviation and change characteristics between the corrected heating load forecast results and the actual heating load data, and generate prediction error information.

[0132] Furthermore, the predicted results at the same time point are correlated with the actual heating load data to form a time-consistent comparison relationship between the predicted and actual results. The prediction deviation between the corrected heating load prediction and the actual heating load data is analyzed based on the differences between corresponding time points, obtaining the variation of the prediction deviation from the actual heating load data at different time points. For example, when the corrected heating load prediction is consistently higher than the actual heating load data, it is judged that the prediction result has an overall upward trend; when the corrected heating load prediction shows increased deviation at some time points while remaining close at other time points, it is judged that the prediction deviation is concentrated in a specific operating phase. The change characteristics of the prediction deviation corresponding to different time points are analyzed to identify the trend relationship of the prediction deviation over time, including continuous increase in prediction deviation, periodic change in prediction deviation, and local abrupt change in prediction deviation. Based on the prediction deviation and its change characteristics, prediction error information is generated, reflecting the difference between the corrected heating load prediction and the actual heating load data.

[0133] S6.2 Input the prediction error information into the hidden operation state construction model, identify the deviation distribution between the hidden operation state of the heating unit operation mode and the actual heating load data based on the prediction error information, and adjust the parameters in the hidden operation state construction model according to the deviation distribution.

[0134] Furthermore, by analyzing the prediction error information, the deviation relationship between the implicit operating state of the heating unit operation mode and the actual heating load data is analyzed. Based on the prediction error information, the prediction deviation changes at different time points are obtained, and the prediction deviation changes are matched with the implicit operating state of the heating unit operation mode to analyze the distribution of prediction errors under different implicit operating states. For example, when a large prediction deviation continuously occurs within the time range corresponding to the implicit operating state of a certain type of heating unit operation mode, it is determined that the current implicit operating state deviates from the expression of the heating operation change pattern. When the prediction deviation remains relatively small within the time range corresponding to the implicit operating state of a certain type of heating unit operation mode, it is determined that the current implicit operating state reflects the heating operation state well. Based on the prediction error information, the deviation distribution relationship corresponding to different implicit operating states is identified, the location of the deviation and the trend of deviation change are determined, and the parameters in the implicit operating state construction model are adjusted according to the deviation distribution. When the prediction error information indicates that some changes in operating behavior are not effectively expressed, the corresponding parameters of the sequence coding layer in the implicit operating state construction model are adjusted to improve the ability of the sequence coding layer to extract the process of changes in the operating behavior of the heating unit. When the prediction error information indicates that there is insufficient differentiation between different operating modes, the corresponding parameters of the implicit state mapping layer in the implicit operating state construction model are adjusted to improve the ability to distinguish the implicit state expression corresponding to different operating modes, so that the implicit operating state construction model continuously optimizes the implicit operating state of the heating unit operation mode based on the prediction error information.

[0135] S6.3 Input the prediction error information into the spatial influence relationship modeling model of the heating unit, identify the propagation deviation between the spatial influence relationship of the heating unit and the actual heating load data based on the prediction error information, and adjust the parameters in the spatial influence relationship modeling model of the heating unit according to the propagation deviation.

[0136] Furthermore, by analyzing the propagation deviation between the spatial influence relationship of heating units and the actual heating load data through prediction error information, the changes in prediction deviation at different time points in the prediction error information are obtained. Combined with the connection relationship of heating units in the dynamic spatial relationship structure, spatial correlation analysis is performed on the prediction deviation generation process to determine whether the prediction deviation is related to the spatial influence propagation process between heating units. For example, when the load change prediction of a certain heating unit is relatively accurate, while the prediction results of other affected heating units continuously deviate from the actual heating load data, it is determined that there is a deviation in the spatial influence propagation intensity between heating units. When multiple related heating units show prediction deviations simultaneously, it is determined that there is an overall propagation deviation in the spatial influence relationship between heating units. Based on the prediction error information, the propagation deviation between the spatial influence relationship of heating units and the actual heating load data is identified, and the deviations corresponding to the spatial influence weight, connection relationship, and influence propagation intensity are determined. The parameters in the modeling model of the spatial influence relationship of heating units are adjusted according to the propagation deviation. When the prediction error information indicates that the actual influence intensity between two heating units is higher than the calculated result, the corresponding spatial influence relationship parameters in the modeling model of the spatial influence relationship of heating units are adjusted to improve the expression of influence propagation between the two heating units. When the prediction error information indicates that the actual influence between two heating units is weak but the calculated result is strong, the corresponding parameters are adjusted to reduce the expression of influence propagation between the two heating units. This allows the modeling model of the spatial influence relationship of heating units to optimize the spatial influence relationship between heating units according to the actual heating load changes, forming a spatial influence relationship that is more consistent with the actual operation process.

[0137] This embodiment also provides a machine learning-based intelligent heating load prediction system, including: a processing module that collects multi-source basic data on the operation of heating units and performs standardized processing to form a raw time-series data set;

[0138] The learning module performs serialized learning on the operating behavior of the heating unit based on the original time-series data set, and generates the hidden operating state of the heating unit's operating mode.

[0139] The reconstruction module dynamically reconstructs the spatial influence relationship of the heating unit based on the hidden operating state, and generates a dynamic spatial relationship structure.

[0140] The adjustment module uses a machine learning load prediction model to predict and calculate the heating load, and adjusts the intensity of the influence propagation in the dynamic spatial relationship structure according to the hidden operating state to generate the heating load prediction result.

[0141] The correction module performs time continuity smoothing correction on the heating load forecast results to obtain the corrected heating load forecast results;

[0142] The update module compares and analyzes the error between the corrected heating load prediction results and the actual heating load data, generates prediction error information, and updates the parameters of the hidden operation state construction model and the heating unit spatial influence relationship modeling model.

[0143] This embodiment also provides a computer device applicable to the intelligent heating load prediction method based on machine learning, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the intelligent heating load prediction method based on machine learning as proposed in the above embodiment.

[0144] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0145] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the intelligent heating load prediction method based on machine learning as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0146] In summary, this invention constructs a unified raw time-series data set by collecting and standardizing multi-source basic data during the operation of heating units; it then performs serial learning on the operating behavior of heating units based on this raw time-series data set, extracts the temporal variation features of the operating behavior, and generates the latent operating states of the heating units' operating modes, thus effectively representing the potential operating modes of the heating units; based on the latent operating states, it dynamically adjusts the spatial influence weights between different heating units, reconstructs the spatial influence relationships of heating units, and generates a dynamic spatial relationship structure, achieving dynamic adaptation of the spatial influence relationships of heating units to changes in operating states; it utilizes the dynamic spatial relationship structure and the latent operating states to construct a machine learning load prediction model, improves the utilization of spatial correlation information during the heating load prediction process by adjusting the influence propagation path, and enhances the stability of the prediction results by combining time continuity correction; and it updates the parameters of the latent operating state construction model and the heating unit spatial influence relationship model based on prediction error information, achieving continuous optimization of the prediction process.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart heating load prediction method based on machine learning, characterized in that: include: Collect multi-source basic data on the operation of heating units and perform standardized processing to form a raw time-series data set; Based on the original time-series data set, the operating behavior of the heating unit is sequentially learned to generate the hidden operating state of the heating unit's operating mode. Based on the hidden operating state, the spatial influence relationship of the heating unit is dynamically reconstructed to generate a dynamic spatial relationship structure. Heating load is predicted and calculated using a machine learning load prediction model. The intensity of the influence propagation in the dynamic spatial relationship structure is adjusted according to the implicit operating state to generate heating load prediction results. The heating load forecast results are smoothed by time continuity to obtain the corrected heating load forecast results; The revised heating load prediction results are compared and analyzed with the actual heating load data to generate prediction error information, and the parameters of the hidden operation state construction model and the heating unit spatial influence relationship model are updated.

2. The intelligent heating load prediction method based on machine learning as described in claim 1, characterized in that: The original time-series data set includes, The data acquisition equipment obtains equipment operation status data, environmental status data, heating load data, and control and regulation data during the operation of the heating unit, and generates multi-source basic data of the heating unit operation according to the acquisition time. The multi-source basic data of the heating unit operation are standardized by time alignment, data format unification and data scale normalization to form a data set with a unified time identifier, thus obtaining the original time series data set.

3. The intelligent heating load prediction method based on machine learning as described in claim 2, characterized in that: The hidden operating state of the heating unit's operating mode includes, Based on the original time series data set, the data corresponding to each time point are sorted according to the timestamp, and the sorted data is grouped by sliding in a continuous time window. The changes in the operating status of the heating unit, the changes in control behavior, and the changes in load response within each time window are spliced ​​and encoded in the order of occurrence to form a sequence of heating unit operating behavior that reflects the continuous operation process. A hidden operating state construction model is constructed based on the temporal continuity of the heating unit's operating behavior sequence. The hidden operating state construction model includes a sequence encoding layer and a hidden state mapping layer. Among them, the sequence coding layer is used to encode the continuous state sequence of the heating unit's operating behavior, and the hidden state mapping layer is used to map the continuous state coding to a unified feature space to form the hidden state of the heating unit's operating behavior. The heating unit's operating behavior sequence is input into the hidden operating state to construct a model. The sequence coding layer calculates the behavioral change amplitude between adjacent time nodes and the behavioral change pattern between consecutive time windows. The hidden state mapping layer fuses and encodes the behavioral change amplitude and behavioral change pattern to extract the temporal change characteristics of the heating unit's operating behavior. Based on the temporal variation characteristics of the heating unit's operating behavior, the feature correlation between different operating behaviors is calculated, and the temporal variation characteristics of the heating unit's operating behavior are aggregated and mapped according to the feature correlation to generate the hidden operating state of the heating unit's operating mode.

4. The intelligent heating load prediction method based on machine learning as described in claim 3, characterized in that: The dynamic spatial relationship structure includes, Calculate the spatial influence weights between different heating units based on the implicit operating state of the heating unit's operating mode; The spatial influence weight change is calculated based on the characteristic distance between the hidden operating states of the heating unit operation modes, and the spatial influence weight between different heating units is iteratively updated using the spatial influence weight change to form the spatial influence relationship of heating units. Based on the spatial influence relationship of heating units, the updated spatial influence weights between each heating unit are read. The connection relationship between heating units is adjusted according to the change in spatial influence weights. The topological association between heating units is reconstructed based on the adjusted connection relationship to generate a dynamic spatial relationship structure.

5. The intelligent heating load prediction method based on machine learning as described in claim 4, characterized in that: The heating load forecast results include, The connection relationship of heating units in the dynamic spatial relationship structure is used as the spatial association input, and a mapping relationship between the spatial association information between heating units and the heating load is established to form a machine learning load prediction model. The implicit operating state of the heating unit operation mode is input into the machine learning load prediction model, and the influence propagation correction coefficient corresponding to each connection relationship in the dynamic spatial relationship structure is calculated based on the implicit operating state of the heating unit operation mode. Based on the influence propagation correction coefficient, the influence propagation path in the dynamic spatial relationship structure is adaptively adjusted to obtain the adjusted dynamic spatial relationship structure. The adjusted dynamic spatial relationship structure is input into the machine learning load prediction model. Based on the adjusted dynamic spatial relationship structure, the spatial influence propagation results between heating units within the target time period are determined. The heating load is then predicted and analyzed in combination with the spatial influence propagation results to generate the heating load prediction results.

6. The intelligent heating load prediction method based on machine learning as described in claim 5, characterized in that: The revised heating load forecast results include, Based on the heating load forecast results, establish the continuous change relationship between adjacent forecast times in chronological order, identify the abnormal load change intervals in the heating load forecast results, and form time continuity correction information; Based on the time continuity correction information, the abnormal load change intervals in the heating load forecast results are adjusted by continuity constraints to obtain the corrected heating load forecast results.

7. The intelligent heating load prediction method based on machine learning as described in claim 6, characterized in that: The parameter update of the hidden operating state construction model and the heating unit spatial influence relationship model includes... The revised heating load forecast results are matched with the actual heating load data according to the same time node. The prediction deviation and change characteristics between the revised heating load forecast results and the actual heating load data are analyzed to generate prediction error information. The prediction error information is input into the hidden operating state construction model. Based on the prediction error information, the deviation distribution between the hidden operating state of the heating unit operation mode and the actual heating load data is identified, and the parameters in the hidden operating state construction model are adjusted according to the deviation distribution. The prediction error information is input into the spatial influence relationship modeling model of the heating unit. Based on the prediction error information, the propagation deviation between the spatial influence relationship of the heating unit and the actual heating load data is identified, and the parameters in the spatial influence relationship modeling model of the heating unit are adjusted according to the propagation deviation.

8. A machine learning-based intelligent heating load prediction system, based on the machine learning-based intelligent heating load prediction method according to any one of claims 1 to 7, characterized in that: include: The processing module collects multi-source basic data on the operation of the heating unit and performs standardized processing to form a raw time-series data set; The learning module performs sequential learning on the operating behavior of the heating unit based on the original time-series data set, and generates the hidden operating state of the heating unit's operating mode. The reconstruction module dynamically reconstructs the spatial influence relationship of the heating unit based on the hidden operating state, generating a dynamic spatial relationship structure. The adjustment module uses a machine learning load prediction model to predict and calculate the heating load, and adjusts the intensity of the influence propagation in the dynamic spatial relationship structure according to the hidden operating state to generate the heating load prediction result. The correction module performs time continuity smoothing correction on the heating load forecast results to obtain the corrected heating load forecast results; The update module compares and analyzes the error between the corrected heating load prediction results and the actual heating load data, generates prediction error information, and updates the parameters of the hidden operation state construction model and the heating unit spatial influence relationship modeling model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the machine learning-based smart heating load prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the machine learning-based smart heating load prediction method according to any one of claims 1 to 7.