Artificial intelligence-based method and device for evaluating energy consumption of a heating system

Abstract

The present application relates to a kind of based on artificial intelligence's heating system energy consumption evaluation method and equipment, belong to heating system energy consumption evaluation technical field.It includes the following steps: obtaining the multi-source data of heating system;Multi-source data is aligned in time and space, obtain the multi-source data set aligned in time and space;According to the data in multi-source data set aligned in time and space, generate basic features and cross features by feature engineering, construct enhanced feature set;Heating system energy consumption prediction model is constructed, the model includes Transform coding layer, LSTM time sequence layer and graph attention network;The data in enhanced feature set is input into heating system energy consumption prediction model and is handled, obtains energy consumption prediction result;According to energy consumption prediction result, calculate unit area energy consumption intensity, if unit area energy consumption intensity is higher than set threshold, then trigger high energy consumption alarm.The present application can improve the accuracy of energy consumption evaluation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of heat supply system energy consumption evaluation, and particularly relates to a heat supply system energy consumption evaluation method and device based on artificial intelligence. BACKGROUND

[0002] As an important part of urban infrastructure, the heat supply system plays an irreplaceable role in ensuring residents' winter heating and industrial heat supply. With the development of computer technology and sensor technology, heat supply system energy consumption evaluation has gradually realized automation and intelligentization. By installing various sensors, the operation parameters of the heat supply system, such as temperature, pressure, flow, etc., can be monitored in real time, providing rich data support for energy consumption evaluation.

[0003] Most current energy consumption evaluation methods are based on static data, i.e., assuming that the operation parameters of the heat supply system remain unchanged over a period of time. However, in reality, the operation of the heat supply system is dynamically changing, influenced by weather, user demand, equipment status, and other factors. Due to limitations in data acquisition and processing capabilities, existing energy consumption evaluation methods often have difficulty in combining multiple data to timely discover and solve energy consumption problems in the heat supply system. SUMMARY

[0004] To solve the above problems, the application provides a heat supply system energy consumption evaluation method and device based on artificial intelligence.

[0005] To achieve the above purpose, the application realizes the following technical solutions:

[0006] The application provides a heat supply system energy consumption evaluation method based on artificial intelligence, comprising the following steps:

[0007] S1. Obtain multi-source data of the heat supply system; the multi-source data includes heat source data, pipe network data, user end data, and meteorological data;

[0008] S2. Time-space align the multi-source data to obtain a time-space aligned multi-source data set;

[0009] S3. According to the data in the time-space aligned multi-source data set, generate basic features and cross features through feature engineering to construct an enhanced feature set;

[0010] S4. Construct a heat supply system energy consumption prediction model, which includes a Transformer encoding layer, an LSTM time series layer, and a graph attention network; input the data in the enhanced feature set into the heat supply system energy consumption prediction model for processing to obtain an energy consumption prediction result;

[0011] S5. Calculate the unit area energy consumption intensity according to the energy consumption prediction result, and if the unit area energy consumption intensity is higher than a set threshold, trigger a high energy consumption alarm.

[0012] Further, the step S1 specifically comprises:

[0013] The heat source data comprises heat source temperature , heat source pressure and fuel flow ; the pipe network data comprises pipe network flow , supply water temperature and return water temperature ; the user end data comprises indoor temperature , set temperature and heat consumption area; and the meteorological data comprises outdoor temperature , wind speed and sunshine intensity .

[0014] Further, the step S2 specifically comprises:

[0015] S21. Time interpolation is performed on the multi-source data, and the granularity of the time stamp is unified, and the time interpolation adopts a cubic spline interpolation formula; and the heat source temperature, the supply water temperature, the return water temperature and the outdoor temperature after the cubic spline interpolation is performed are iteratively optimized through thermodynamic constraint interpolation, and the thermodynamic constraint interpolation adopts a superimposed heat conduction equation constraint, and the formula is expressed as follows:

[0016] ,

[0017] wherein, represents a thermal diffusion coefficient; represents a convection coefficient; represents the outdoor temperature, represents a temperature field to be interpolated, represents an optimized temperature field;

[0018] S22. A pipe network graph structure is constructed , wherein a node represents a heat source, a pipe network branch or a building unit, and an edge represents a connection relationship; a spatial correlation matrix is generated based on the pipe network graph structure , represents the dimension of the matrix; and the calculation process of an element in the spatial correlation matrix is as follows: a breadth-first search is performed on the pipe network graph structure, and the shortest topological path distance between nodes is calculated; the Euclidean distance is calculated according to the GIS coordinates of the nodes; and the shortest topological path distance and the Euclidean distance are fused to obtain the element , the formula is as follows:

[0019] ,

[0020] ,

[0021] wherein, represents a correction distance, represents a fusion coefficient; represents an attenuation coefficient; represents an exponential function with as the base; generating a spatio-temporally aligned dataset based on interpolated multi-source data and a spatial correlation matrix wherein, represents a time step, represents the total number of nodes, represents the number of categories of multi-source data.

[0022] Further, step S3 specifically comprises:

[0023] S31. The basic features include fuel heat value efficiency and equivalent outdoor temperature; the calculation formula of the fuel heat value efficiency is as follows:

[0024] ,

[0025] wherein, represents the specific heat capacity of water; represents the water supply and return temperature difference, ; the formula of the equivalent outdoor temperature is as follows: ;

[0026] S32. The cross features include nonlinear response intensity and regional heat load density field; the nonlinear relationship between fuel flow and equivalent outdoor temperature is modeled by polynomial regression to obtain a heat source-meteorological response surface, and then the nonlinear response intensity is obtained, the formula is as follows:

[0027] ,

[0028] wherein, represents the regression coefficient of the th polynomial, represents the equivalent outdoor temperature of the th polynomial, represents a residual term; the calculation formula of the regional heat load density field is as follows:

[0029] ,

[0030] wherein, represents the regional heat load density field, represents the pipe network flow of the first node, represents the heating area of the first node; represents the pipe network flow of the first node, represents the heating area of the first node; represents the building set associated with the pipe network topology;

[0031] S33. Merge the base features and the cross features and the spatio-temporal aligned dataset in the time-node dimension to obtain an enhanced feature set , wherein, represents the feature category in the feature set.

[0032] Further, step S4 specifically comprises:

[0033] S41. Input the fuel heat value efficiency, equivalent outdoor temperature, heat source temperature, heat source pressure, fuel flow and nonlinear response strength in the enhanced feature set into the Transformer encoding layer for encoding to obtain a global context embedding ; input the pipe network flow, water supply temperature, return water temperature, indoor temperature, set temperature, outdoor temperature, wind speed and solar radiation intensity in the strong feature set into the LSTM time sequence layer for processing to obtain short-term time sequence features , and the LSTM time sequence layer adopts a bidirectional LSTM network structure; input the spatial correlation matrix and regional heat load density field in the strong feature set into the graph attention network to obtain spatial topology constraint features ;

[0034] S42. Fuse the global context embedding , the short-term time sequence features and the spatial topology constraint features by a gating fusion mechanism to obtain fused features , and the formula is as follows:

[0035] ,

[0036] wherein, represents a Sigmoid activation function, represents a gating weight, represents a feature transformation matrix, represents that the global context embedding , the short-term time sequence features and the spatial topology constraint features are spliced, represents a Hadamard product;

[0037] S43. The fused features are processed by a multilayer perceptron to obtain an energy consumption prediction result .

[0038] Furthermore, in step S5, the energy intensity per unit area is obtained. The formula is expressed as: ,in, Indicates the first Predicted energy consumption of each node. .

[0039] Furthermore, in step S2, the convection coefficient is dynamically calculated using wind speed data from meteorological data. .

[0040] Furthermore, the nonlinear response intensity in step S3 The formula is expressed as: ,in, This represents the regression coefficient of the second polynomial. This represents the regression coefficient of the third polynomial. This indicates taking the absolute value.

[0041] This invention also provides an artificial intelligence-based heating system energy consumption assessment device, which performs the above-described artificial intelligence-based heating system energy consumption assessment method, including:

[0042] Data acquisition unit: used to acquire multi-source data from the heating system;

[0043] Spatiotemporal alignment unit: used to perform spatiotemporal alignment on multi-source data to obtain a spatiotemporally aligned multi-source dataset;

[0044] Data augmentation unit: used to process data from spatiotemporally aligned multi-source datasets through feature engineering, generate basic and cross features, and construct an augmented feature set;

[0045] Prediction Unit: Used to construct an energy consumption prediction model for the heating system, the model including a Transformer encoding layer, an LSTM temporal layer and a graph attention network; and inputs the data from the enhanced feature set into the energy consumption prediction model for processing to obtain the energy consumption prediction result;

[0046] Alarm unit: used to calculate the energy consumption intensity per unit area based on the energy consumption prediction results. If the energy consumption intensity per unit area is higher than the set threshold, a high energy consumption alarm will be triggered.

[0047] The advantages of this invention are:

[0048] The present application solves the problem of mismatching of time and space resolution in traditional methods by integrating four-dimensional heterogeneous data of heat source, pipe network, user end and meteorology, adopting spatiotemporal alignment strategy of cubic spline interpolation superimposed heat conduction equation constraint; secondly, the fuel heat value efficiency, equivalent outdoor temperature basic characteristics are innovatively constructed, and the cross characteristics of heat source-meteorological response surface and regional heat load density field are generated combined with polynomial regression, the nonlinear coupling relationship between heat parameters is deeply mined; the global feature association is captured through the Transformer encoding layer, the time sequence dynamics is extracted through the bidirectional LSTM, and the pipe network topology constraint is modeled through the graph attention network, realizing the collaborative learning of multi-scale spatiotemporal features; the wind speed dynamics is introduced into the calculation of the convection coefficient, and the nonlinear response strength index is defined, which significantly improves the prediction robustness under extreme weather; compared with the traditional method, the present application can improve the prediction accuracy, and through the unit area energy consumption intensity threshold alarm mechanism, it provides accurate quantitative basis for energy saving optimization of the heating system. BRIEF DESCRIPTION OF DRAWINGS

[0049] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application together with the embodiments thereof, and explain the present application, but do not constitute a limitation of the present application.

[0050] Figure 1 The present application is a method for the step flowchart. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0052] Embodiment 1

[0053] In this embodiment, as shown in the figure, the present application provides a heating system energy consumption evaluation method based on artificial intelligence, and the specific steps include: Figure 1

[0054] S1. Obtain multi-source data of the heating system; the multi-source data includes heat source data, pipe network data, user end data and meteorological data;

[0055] Specifically, the heat source data includes heat source temperature , heat source pressure and fuel flow ; the pipe network data includes pipe network flow , water supply temperature and return water temperature ​The user-side data includes indoor temperature. Set temperature and the area of ​​heat use; the meteorological data includes outdoor temperature. Wind speed and solar radiation intensity .

[0056] S2. Perform spatiotemporal alignment on the multi-source data to obtain a spatiotemporally aligned multi-source dataset;

[0057] Specifically, S21. Time interpolation is performed on the multi-source data to unify the granularity of timestamps and ensure the consistency of time intervals between the multi-source data. The time interpolation uses a cubic spline interpolation formula. After cubic spline interpolation, the heat source temperature, supply water temperature, return water temperature, and outdoor temperature are iteratively optimized using thermodynamic constraint interpolation. The thermodynamic constraint interpolation uses superimposed heat conduction equations, and the formula is expressed as follows:

[0058] ,

[0059] in, This represents the thermal diffusivity, calculated from the thermal conductivity of the pipe network material; for example, it is taken as 1.25 × 10⁻⁶ for steel pipes. - 5 m 2 / s; The convection coefficient is represented by wind speed data, which is dynamically calculated from meteorological data. ; Indicates the outdoor temperature. This indicates the temperature field that needs to be interpolated. This indicates the optimized temperature field, ensuring that the interpolated temperature change conforms to the thermal inertia of the pipeline network: the rate of temperature change does not exceed the material's heat capacity limit, and environmental heat exchange: the effect of outdoor temperature on the attenuation of pipeline network temperature.

[0060] S22. Constructing the pipeline network structure ,node Indicates heat source, pipeline branch, building unit, edge Representing connection relationships; generating a spatial correlation matrix based on the network diagram structure. , Representing the dimension of a matrix; spatial incidence matrix Obtaining elements The calculation process is as follows: perform a breadth-first search on the network diagram structure and calculate the shortest topological path distance between nodes. Calculate the Euclidean distance based on the GIS coordinates of the nodes. ; Combine shortest topological path distance Euclidean distance The elements were then obtained by modeling using an exponential decay function. , the formula is expressed as follows:

[0061] ,

[0062] ,

[0063] wherein, represents the corrected distance, represents the fusion coefficient, emphasizing the dominance of the pipe network topology, ; represents the attenuation coefficient, ; represents the exponential function with as the base; based on the interpolated multi-source data and the spatial correlation matrix, a spatio-temporally aligned data set is generated wherein, represents the number of time steps, represents the total number of nodes, represents the number of categories of multi-source data.

[0064] Specifically, the heat source node may be a boiler room, a heat exchange station, etc. heat production unit, each heat source corresponds to a unique node; the attribute field is the heat source temperature, the heat source pressure, the fuel flow, the geographic coordinates (x, y); the pipe network branch node may be a pipe intersection, a valve, a pump station, etc. key position of pipe network; the attribute field is flow, water supply temperature, return water temperature, pipe diameter, material (steel / PE), local resistance coefficient; building unit node may be the end user building (such as residential building, commercial body), divided according to independent heating inlet; the attribute field is indoor temperature, set temperature, heat consumption area, building type (residential / office / hospital). The edge represents the physical connection relationship and energy transmission path between nodes, which is divided into two categories: main pipe edge : connection between heat source and pipe network branch, between pipe network branches; the attribute field is pipe length, flow direction, design flow, actual flow; user branch edge : end connection from pipe network branch to building unit; the attribute field is branch length, insulation thickness.

[0065] S3. According to the data in the spatio-temporally aligned multi-source data set, the basic features and cross features are generated through feature engineering, and the enhanced feature set is constructed;

[0066] Specifically, S31. The basic features include fuel heat value efficiency and equivalent outdoor temperature; the calculation formula of the fuel heat value efficiency is expressed as follows:

[0067] ,

[0068] in, This indicates the specific heat capacity of water; Indicates the temperature difference between supply and return water. The formula for the equivalent outdoor temperature is expressed as follows: ;

[0069] S32. The cross-features include nonlinear response intensity and regional heat load density field; by modeling the nonlinear relationship between fuel flow rate and equivalent outdoor temperature through polynomial regression, the heat source-meteorological response surface is obtained, expressed by the following formula:

[0070] ,

[0071] in, Indicates the first The regression coefficients of the polynomial, Indicates the first The equivalent outdoor temperature of a polynomial This represents the residual term; subsequently, the nonlinear response intensity is obtained. The formula is expressed as: ,in, This represents the regression coefficient of the second polynomial. This represents the regression coefficient of the third polynomial. This indicates taking the absolute value; the calculation formula for the regional heat load density field is as follows:

[0072] ,

[0073] in, Represents the regional heat load density field. Indicates the first The network traffic of each node Indicates the first The heat-consuming area of ​​each node; This represents a set of buildings associated with a pipeline network topology.

[0074] S33. Merge the basic features and cross features with the spatiotemporally aligned dataset along the time-node dimension to obtain the enhanced feature set. ,in, Represents the feature categories in the feature set. .

[0075] S4. Construct a heating system energy consumption prediction model, which includes a Transformer encoding layer, an LSTM temporal layer, and a graph attention network; the data in the enhanced feature set is input into the heating system energy consumption prediction model for processing to obtain the energy consumption prediction result;

[0076] Specifically, in S41, the fuel calorific value efficiency, equivalent outdoor temperature, heat source temperature, heat source pressure, fuel flow rate, and nonlinear response intensity from the enhanced feature set are input into the Transformer encoding layer for encoding to obtain the global context embedding. The strong feature set of pipeline flow rate, supply water temperature, return water temperature, indoor temperature, set temperature, outdoor temperature, wind speed, and solar radiation intensity are input into the LSTM time series layer for processing to obtain short-term time series features. The LSTM time-series layer adopts a bidirectional LSTM network structure; the spatial correlation matrix and regional heat load density field of the strong feature set are input into the graph attention network to obtain spatial topological constraint features. ;

[0077] S42. Embed the global context through a gating fusion mechanism. Short-term time series characteristics and spatial topological constraint features By performing fusion, fusion characteristics are obtained. The formula is expressed as follows:

[0078] ,

[0079] in, This represents the Sigmoid activation function. Indicates the gating weight, Represents the characteristic transformation matrix. This indicates embedding the global context. Short-term time series characteristics and spatial topological constraint features To splice, It represents the Hadamardi (or Hadama) stack;

[0080] S43. The fusion feature After processing by a multilayer perceptron, the energy consumption prediction results are obtained. .

[0081] S5. Calculate the energy consumption intensity per unit area based on the energy consumption prediction results. If the energy consumption intensity per unit area exceeds a set threshold, a high energy consumption alarm is triggered. Specifically, the energy consumption intensity per unit area... The formula is expressed as: ,in, Indicates the first Predicted energy consumption of each node. .

[0082] Example 2

[0083] The embodiment provides a heat supply system energy consumption evaluation device based on artificial intelligence, executes the heat supply system energy consumption evaluation method based on artificial intelligence described in embodiment 1, and comprises the following units:

[0084] A data acquisition unit is configured to acquire multi-source data of the heat supply system.

[0085] A space-time alignment unit is configured to perform space-time alignment on the multi-source data to obtain a space-time aligned multi-source data set.

[0086] A data enhancement unit is configured to process data in the space-time aligned multi-source data set through feature engineering to generate basic features and cross features and construct an enhanced feature set.

[0087] A prediction unit is configured to construct a heat supply system energy consumption prediction model, wherein the model comprises a Transformer encoding layer, an LSTM time sequence layer and a graph attention network; and input data in the enhanced feature set into the heat supply system energy consumption prediction model for processing to obtain an energy consumption prediction result.

[0088] An alarm unit is configured to calculate unit area energy consumption intensity according to the energy consumption prediction result, and trigger a high energy consumption alarm if the unit area energy consumption intensity is higher than a set threshold.

[0089] Embodiment 3

[0090] In the embodiment, a heating area of a certain northern city is taken as an example, which comprises one heat source (a coal-fired boiler room), three pipe network branch nodes (including a pump station, a valve group), five building units (three residential buildings, one commercial complex and one hospital), and a time span of all day (00:00-24:00) on January 10, 2025.

[0091] The collected original data (time stamp: 2025-01-10 08:00) are as follows: heat source data: heat source temperature 85℃, heat source pressure 0.8MPa, fuel flow 2.8kg / s, pipe network data: pipe network flow 120m³ / h, water supply temperature 62℃, return water temperature 42℃, user end data: indoor temperature 21.5℃, set temperature 22℃, heat consumption area 6500㎡, and meteorological data: outdoor temperature -9.2℃, wind speed 1.8m / s, and solar radiation intensity 450W / m 2 .

[0092] Space-time alignment processing: the original data time interval is 15 minutes, and is unified to a 10-minute granularity through three times spline interpolation. Taking heat source temperature interpolation as an example, it can be seen from Table 1 that missing values can be well made up through space-time alignment processing.

[0093] Table 1 Comparison of heat source temperature three times spline interpolation alignment

[0094]

[0095] Space correlation matrix calculation: take node 1 (boiler room) and node 4 (commercial complex) as an example, the topological path distance is 2 (boiler room -> pump station -> commercial complex), and the Euclidean distance is 320 m; the corrected distance is 2.24, and the space correlation weight is obtained: .

[0096] The LSTM baseline method and the method of the application are compared, and the energy consumption prediction results of node 4 (commercial complex) are shown in Table 2.

[0097] Table 2 Comparison of LSTM baseline method and the method of the application

[0098]

[0099] As can be seen from the table, the method of the application can better predict energy consumption, and the prediction result is closer to the true value. The nonlinear coupling relationship between heat source and weather is captured by the Transformer, which can improve the accuracy of data prediction compared with the single time sequence modeling of LSTM. The heat conduction equation is introduced in the interpolation stage, so that the temperature prediction conforms to the pipe network thermal inertia law, and the physically unreliable results caused by pure data driving are avoided.

[0100] The alarm trigger threshold is set to 0.0042 GJ / m², and the unit area energy consumption intensity of node 4 (commercial complex) is calculated by the method of the application: At this time, the alarm is triggered, and the actual manual check finds that the commercial complex has a night heat preservation valve leakage; the predicted value calculated by the LSTM baseline method does not trigger the alarm due to low accuracy. Further illustrates the effectiveness of the method of the application, which reduces the energy consumption prediction error by fusing physical laws and deep learning models.

[0101] Finally, it should be noted that: the above only describes the preferred embodiments of the application and is not intended to limit the application. Although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for assessing the energy consumption of a heating system based on artificial intelligence, characterized in that, Includes the following steps: S1. Acquire multi-source data from the heating system; the multi-source data includes heat source data, pipeline network data, user terminal data, and meteorological data; S2. Perform spatiotemporal alignment on the multi-source data to obtain a spatiotemporally aligned multi-source dataset; the specific steps are as follows: S21. Perform time interpolation on multi-source data, unify the granularity of timestamps, and use cubic spline interpolation formula for time interpolation; after cubic spline interpolation, iteratively optimize the heat source temperature, supply water temperature, return water temperature, and outdoor temperature using thermodynamic constraint interpolation. The thermodynamic constraint interpolation uses superimposed heat conduction equations, and the formula is expressed as follows: ， in, Indicates the thermal diffusivity; Indicates the convection coefficient; Indicates the outdoor temperature. This indicates the temperature field that needs to be interpolated. This indicates the optimized temperature field; S22. Constructing the pipeline network structure ,node Indicates heat source, pipeline branch, building unit, edge Representing connection relationships; generating a spatial correlation matrix based on the network diagram structure. , Representing the dimension of a matrix; spatial incidence matrix Obtaining elements The calculation process is as follows: perform a breadth-first search on the network diagram structure and calculate the shortest topological path distance between nodes. Calculate the Euclidean distance based on the GIS coordinates of the nodes. ; Combine shortest topological path distance Euclidean distance The elements were then obtained by modeling using an exponential decay function. The formula is expressed as follows: ， ， in, Indicates the corrected distance. Indicates the fusion coefficient; Indicates the attenuation coefficient; Indicates An exponential function with base 0; generating a spatiotemporally aligned dataset based on interpolated multi-source data and spatial correlation matrix. ,in, Indicates the number of time steps. Indicates the total number of nodes. Indicates the number of categories in multi-source data; S3. Based on the data from the spatiotemporally aligned multi-source dataset, generate basic and cross features through feature engineering to construct an enhanced feature set; the specific steps are as follows: S31. The basic characteristics include fuel calorific value efficiency and equivalent outdoor temperature; S32. The cross-features include nonlinear response intensity and regional heat load density field; by modeling the nonlinear relationship between fuel flow rate and equivalent outdoor temperature through polynomial regression, the heat source-meteorological response surface is obtained, and then the nonlinear response intensity is obtained; S33. Merge the basic features, cross features, and spatiotemporally aligned datasets along the time-node dimension to obtain the enhanced feature set. ,in, Represents the feature categories in the feature set; S4. Construct a heating system energy consumption prediction model, which includes a Transformer encoding layer, an LSTM temporal layer, and a graph attention network; the data in the enhanced feature set is input into the heating system energy consumption prediction model for processing to obtain the energy consumption prediction result; S5. Calculate the energy consumption intensity per unit area based on the energy consumption prediction results. If the energy consumption intensity per unit area is higher than the set threshold, a high energy consumption alarm will be triggered.

2. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 1, characterized in that, Step S1 specifically includes: The heat source data includes the heat source temperature. Heat source pressure and fuel flow The pipeline data includes pipeline flow rate. Water supply temperature and return water temperature The user-side data includes indoor temperature. Set temperature and the area of ​​heat use; the meteorological data includes outdoor temperature. Wind speed and solar radiation intensity .

3. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 2, characterized in that, fuel calorific value efficiency The calculation formula is expressed as follows: ， in, This indicates the specific heat capacity of water; Indicates the temperature difference between supply and return water. The formula for the equivalent outdoor temperature is expressed as follows: ; The formula for calculating the nonlinear response intensity is as follows: ， in, Indicates the first The regression coefficients of the polynomial, Indicates the first The equivalent outdoor temperature of a polynomial The residual term is represented by the formula for calculating the regional heat load density field as follows: ， in, Represents the regional heat load density field. Indicates the first The network traffic of each node Indicates the first The heat-consuming area of ​​each node; A set of buildings representing the topological associations of a pipeline network.

4. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 3, characterized in that, Step S4 specifically includes: S41. The fuel calorific value efficiency, equivalent outdoor temperature, heat source temperature, heat source pressure, fuel flow rate, and nonlinear response intensity from the enhanced feature set are input into the Transformer encoding layer for encoding to obtain the global context embedding. The enhanced feature set, including pipeline flow rate, supply water temperature, return water temperature, indoor temperature, set temperature, outdoor temperature, wind speed, and solar radiation intensity, is input into the LSTM time series layer for processing to obtain short-term time series features. The LSTM temporal layer employs a bidirectional LSTM network structure; the spatial correlation matrix and regional heat load density field from the enhanced feature set are input into the graph attention network to obtain spatial topological constraint features. ; S42. Embed the global context through a gating fusion mechanism. Short-term time series characteristics and spatial topological constraint features By performing fusion, fusion characteristics are obtained. The formula is expressed as follows: ， in, This represents the Sigmoid activation function. Indicates the gating weight, Represents the characteristic transformation matrix. This indicates embedding the global context. Short-term time series characteristics and spatial topological constraint features To splice, It represents the Hadamardi (or Hadama) stack; S43. The fusion feature After processing by a multilayer perceptron, the energy consumption prediction results are obtained. .

5. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 4, characterized in that, In step S5, the energy intensity per unit area is obtained. The formula is expressed as: ,in, Indicates the first Predicted energy consumption of each node. .

6. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 5, characterized in that, In step S2, the convection coefficient is dynamically calculated using wind speed data from meteorological data. .

7. The energy consumption assessment method for heating systems based on artificial intelligence according to claim 6, characterized in that, In step S3, the nonlinear response intensity The formula is expressed as: ,in, This represents the regression coefficient of the second polynomial. This represents the regression coefficient of the third polynomial. This indicates taking the absolute value.

8. An artificial intelligence-based heating system energy consumption assessment device, executing the artificial intelligence-based heating system energy consumption assessment method as described in claim 1, characterized in that, include: Data acquisition unit: used to acquire multi-source data from the heating system; Spatiotemporal alignment unit: used to perform spatiotemporal alignment on multi-source data to obtain a spatiotemporally aligned multi-source dataset; Data augmentation unit: used to process data from spatiotemporally aligned multi-source datasets through feature engineering, generate basic and cross features, and construct an augmented feature set; Prediction Unit: Used to construct an energy consumption prediction model for the heating system, the model including a Transformer encoding layer, an LSTM temporal layer and a graph attention network; and inputs the data from the enhanced feature set into the energy consumption prediction model for processing to obtain the energy consumption prediction result; Alarm unit: used to calculate the energy consumption intensity per unit area based on the energy consumption prediction results. If the energy consumption intensity per unit area is higher than the set threshold, a high energy consumption alarm will be triggered.

Images