Deep learning method, system and device based on building group space energy consumption model and computer program product

Abstract

The invention relates to the technical field of building group energy consumption intelligent analysis, and discloses a deep learning method, system and device based on a building group space energy consumption model, and a computer program product, and the method comprises the steps: obtaining a building group space energy consumption model; constructing a multi-scale building group feature set; building a building group space energy consumption relation graph; learning space energy consumption characteristic representation of the building nodes by adopting an information propagation and aggregation mechanism of graph convolution; constructing an energy consumption prediction model based on a deep learning network; executing energy consumption convergence operation; and outputting a total energy consumption predicted value of the building group. In the prior art, a building group energy consumption prediction method mainly depends on single building independent modeling or simple statistical summary, and especially under the application condition that spatial relevance exists in a building group, the technical problem that the energy consumption coupling relation between buildings is difficult to describe accurately is solved. According to the method, a prediction mechanism combining building group space energy consumption relation modeling and deep learning is introduced, so that the accuracy of a building group energy consumption prediction result is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent energy consumption analysis technology for building complexes, and in particular to a deep learning method, system, device, and computer program product based on a building complex spatial energy consumption model. Background Technology

[0002] Currently, with the continuous advancement of smart and low-carbon campus construction, the scale of buildings within university campuses is constantly expanding, exhibiting a diverse range of building types, including teaching buildings, laboratories, libraries, dormitories, canteens, and supporting public facilities. These different building types differ significantly in their functions, population density, operating hours, and energy consumption patterns, resulting in high spatial heterogeneity and temporal fluctuations in the overall energy consumption of university campuses. Accurate prediction of energy consumption across university building complexes is a crucial prerequisite for achieving campus energy dispatch, load assessment, and energy efficiency management.

[0003] Existing technologies primarily focus on the individual building level, typically using linear regression, time series analysis, or traditional machine learning models for modeling and prediction based on historical energy consumption data, meteorological parameters, and equipment operating status. While these methods are applicable to single-building scenarios, their core assumption is often that the energy consumption behavior of each building is independent, failing to reflect the energy consumption relationships within a building complex caused by spatial layout, functional similarities, and overlapping usage behaviors. In building complex-scale energy consumption prediction scenarios, significant spatial coupling often exists between buildings. For example, adjacent buildings may exhibit similar trends in personnel flow, energy usage periods, and heating / cooling loads. Simply relying on independent predictions of individual buildings and then summarizing the results easily overlooks the inter-building correlations, leading to significant deviations in building complex-level energy consumption predictions. Furthermore, energy consumption in university campuses exhibits distinct temporal evolution characteristics, such as periodic or sudden fluctuations due to changes in semester schedules, course rhythms, holidays, and research activities. Traditional methods based on fixed features or shallow models struggle to effectively characterize such complex temporal dependencies.

[0004] Therefore, there is an urgent need for an energy consumption prediction method for university campus building clusters. This method should be able to effectively characterize the spatial relationships and temporal evolution of energy consumption among buildings, especially when the building clusters are large in scale, have complex spatial relationships, and exhibit significant energy consumption evolution characteristics. This would improve the accuracy, stability, and engineering applicability of the energy consumption prediction results for building clusters. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the present invention aims to propose a deep learning method, system, device, and computer program product based on a building cluster spatial energy consumption model. This aims to solve the technical problem that existing building cluster energy consumption prediction methods, which mainly rely on independent modeling of individual buildings or simple statistical summarization, are unable to accurately depict the energy consumption coupling relationship between buildings, especially in applications where there is spatial correlation within the building cluster.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a deep learning method based on a building cluster spatial energy consumption model.

[0007] The deep learning method based on the building cluster space energy consumption model includes:

[0008] Step S10: Obtain the spatial energy consumption model of the building cluster in the university campus. Based on the spatial energy consumption model of the building cluster, use a feature extraction method based on spatial hierarchy perception to extract a multi-scale building cluster feature set D. The multi-scale building cluster feature set includes building spatial location features, building energy consumption time series evolution features, building spatial neighborhood coupling features, and overall building cluster energy consumption trend features.

[0009] Step S20: Based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, a graph construction mechanism with joint constraints of spatial distance and functional similarity is used to perform spatial energy consumption relationship modeling, and output the spatial energy consumption relationship diagram of the building complex. ;

[0010] Step S30: Based on the building space neighborhood coupling characteristics, the overall energy consumption trend characteristics of the building complex, and the energy consumption relationship diagram of the building complex space. A graph convolution-based information propagation and aggregation mechanism is used to perform the task of learning the spatial energy consumption feature representation of building nodes, and outputs the spatial energy consumption embedding sequence H of building clusters.

[0011] Step S40: Preset the building group energy consumption prediction model, input the building group spatial energy consumption embedding sequence H into the building group energy consumption prediction model, and output the building-level predicted energy consumption results;

[0012] Step S50: Perform energy consumption aggregation calculation based on the building-level predicted energy consumption results, and output the total predicted energy consumption value of the building group.

[0013] Preferably, step S10, which involves obtaining a spatial energy consumption model of the building complex within the university campus, and extracting a multi-scale feature set D of the building complex using a feature extraction method based on spatial hierarchy perception, specifically includes:

[0014] Step S101: Obtain the spatial energy consumption model of the building complex within the university campus. Based on the spatial energy consumption model, construct a spatial hierarchy reference structure for the building complex, including individual building hierarchy, local building neighborhood hierarchy, and overall building complex hierarchy. Based on this spatial hierarchy reference structure, generate corresponding building spatial location features for each building. ;

[0015] Step S102: Obtain the first The building is within a preset historical time window Building energy consumption time series At the scale of individual building levels, the time series of building energy consumption. Perform time-series statistical analysis to extract the time-series evolution characteristics of building energy consumption that reflect the building's own energy consumption variation. ;

[0016] Step S103: At the scale of the local neighborhood hierarchy of the building, obtain the first... The set of neighboring buildings within a predefined spatial neighborhood of a building is used to aggregate the energy consumption time series of buildings within this set using a spatial distance-weighted aggregation method. The output is a building spatial neighborhood coupling feature that reflects the spatial energy consumption correlation between buildings. ;

[0017] Step S104: At the scale of the overall building complex, obtain all buildings within a preset historical time window. The energy consumption time series of buildings within the university campus is used to construct an overall energy consumption time series for the building complex. Based on this overall energy consumption time series, a trend analysis method using time window sliding fitting is employed to extract overall energy consumption trend features reflecting the overall operational status of the university campus. .

[0018] Preferably, in step S10, the building cluster spatial energy consumption model is used to represent a spatialized model data established by collecting energy consumption data of the building cluster; the individual building level is used to characterize the independent spatial location of a single building within the university campus; the local neighborhood level is used to characterize the spatial relationship between a building and other buildings within its spatial neighborhood; the overall building cluster level is used to characterize the relative position of the buildings in the overall spatial structure of the university campus; and the building energy consumption time-series evolution characteristics are also included. This includes characteristics of energy consumption change trends, energy consumption fluctuation amplitude, and energy consumption change rate.

[0019] Preferably, in step S20, based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, a graph construction mechanism with joint constraints of spatial distance and functional similarity is used to perform spatial energy consumption relationship modeling, and outputs a spatial energy consumption relationship graph of the building complex. The steps specifically include:

[0020] Step S201: Based on the spatial location characteristics of the buildings, calculate the adjacent spatial distance between any two adjacent buildings, m-th and n-th, using the Euclidean distance calculation method based on spatial coordinates. ;

[0021] Step S202: Based on the temporal evolution characteristics of building energy consumption, calculate the functional similarity factor between the m-th building and the n-th building using a temporal feature similarity analysis method based on cosine similarity. ;

[0022] Step S203: Based on adjacent spatial distance and functional similarity factor An exponentially decaying spatial mapping function is used to map the spatial distance between adjacent buildings, and the distance is then weighted and fused with a functional similarity factor to output a spatial adjacency matrix of building clusters. Based on the spatial adjacency matrix of building clusters Construct a diagram of the spatial energy consumption relationship of a building complex. , ,in Let A represent the set of building nodes within the university campus, and let A represent the spatial energy consumption correlation weight matrix between buildings.

[0023] Preferably, in step S30, the energy consumption is based on the building space neighborhood coupling characteristics, the overall energy consumption trend characteristics of the building complex, and the energy consumption relationship diagram of the building complex space. The steps for performing the spatial energy consumption feature representation learning task of building nodes using a graph convolution-based information propagation and aggregation mechanism, and outputting the spatial energy consumption embedding sequence H of the building complex, specifically include:

[0024] Step S301: Based on the spatial coupling characteristics of building spaces and the overall energy consumption trend characteristics of the building complex, create a spatial energy consumption relationship diagram of the building complex. The spatial energy consumption correlation weight matrix is ​​fused element-wise using the Hadamard product method to output the corrected spatial energy consumption correlation weight matrix. ;

[0025] Step S302: Analyze the energy consumption relationship diagram of the building complex space. The set of building nodes in the data is based on the modified spatial energy consumption correlation weight matrix. The node-level spatial energy consumption feature propagation and aggregation process is performed using a graph convolution-based information propagation and aggregation mechanism, and the node-level spatial energy consumption embedding vector is output.

[0026] Step S303: Arrange the node-level spatial energy consumption embedding vectors in a temporal sequence according to a preset historical time window, and generate and output the building group spatial energy consumption embedding sequence H.

[0027] Preferably, step S40, which involves pre-setting a building complex energy consumption prediction model, inputting the building complex spatial energy consumption embedding sequence H into the building complex energy consumption prediction model, and outputting building-level predicted energy consumption results, specifically includes:

[0028] Step S401: Construct a building cluster energy consumption prediction model based on a deep learning model; the building cluster energy consumption prediction model includes a temporal embedding input layer for receiving the building cluster spatial energy consumption embedding sequence H; a multi-layer recurrent neural network layer for deep modeling of the temporal dependencies in the building cluster spatial energy consumption embedding sequence using a Long Short-Term Memory (LSTM) network; a temporal attention layer for weighting the hidden states of each time step output by the multi-layer recurrent neural network layer within the historical time window using an additive attention method based on Softmax normalization; and a fully connected regression layer for mapping the temporal features weighted by the temporal attention layer to building cluster-level predicted energy consumption values, forming the output of building-level predicted energy consumption results;

[0029] Step S402: Obtain the spatial energy consumption embedding sequence of historical building complexes and the corresponding template set of building-level predicted energy consumption results; use the spatial energy consumption embedding sequence of historical building complexes as the input of the building complex energy consumption prediction model, use the template set of building-level predicted energy consumption results as the output of the building complex energy consumption prediction model, and perform a pre-training process on the building complex energy consumption prediction model in combination with the mean square error loss function.

[0030] Step S403: Input the building cluster spatial energy consumption embedding sequence H into the pre-trained building cluster energy consumption prediction model and output the building-level predicted energy consumption results.

[0031] Preferably, step S50, which involves performing energy consumption aggregation calculations based on building-level predicted energy consumption results and outputting the predicted total energy consumption value for the building complex, specifically includes:

[0032] Step S501: Obtain the building energy consumption benchmark value corresponding to each building, calculate the absolute deviation value based on the building-level predicted energy consumption result and the building energy consumption benchmark value, remove data in the building-level predicted energy consumption result whose absolute deviation value is greater than the preset deviation threshold, and output the optimized building-level predicted energy consumption result.

[0033] Step S502: Finally, based on the optimized building-level predicted energy consumption results, the total predicted energy consumption of the building complex is calculated and output using a simple summation method.

[0034] This invention also provides a deep learning system based on a building complex spatial energy consumption model, comprising:

[0035] The multi-scale spatial energy consumption feature extraction module is used to obtain the spatial energy consumption model of building clusters in the university campus. Based on the spatial energy consumption model of the building clusters, a feature extraction method based on spatial hierarchy perception is used to extract the multi-scale building cluster feature set D. The multi-scale building cluster feature set includes building spatial location features, building energy consumption temporal evolution features, building spatial neighborhood coupling features, and overall building cluster energy consumption trend features.

[0036] The spatial energy consumption relationship graph construction module is used to perform spatial energy consumption relationship modeling based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, and adopts a graph construction mechanism with joint constraints of spatial distance and functional similarity, outputting a spatial energy consumption relationship graph of the building complex. ;

[0037] The spatial energy consumption map feature learning module is used to learn energy consumption based on the spatial neighborhood coupling features of buildings, the overall energy consumption trend features of building groups, and the spatial energy consumption relationship map of building groups. A graph convolution-based information propagation and aggregation mechanism is used to perform the task of learning the spatial energy consumption feature representation of building nodes, and outputs the spatial energy consumption embedding sequence H of building clusters.

[0038] The building cluster energy consumption prediction module is used to preset the building cluster energy consumption prediction model, input the building cluster spatial energy consumption embedding sequence H into the building cluster energy consumption prediction model, and output the building-level predicted energy consumption results.

[0039] The building cluster energy consumption aggregation module is used to perform energy consumption aggregation calculations based on building-level predicted energy consumption results and output the total predicted energy consumption value of the building cluster.

[0040] The present invention also provides a deep learning device based on a building cluster spatial energy consumption model, comprising: a memory, a processor, and a deep learning program based on the building cluster spatial energy consumption model stored in the memory and executable on the processor. When the deep learning program based on the building cluster spatial energy consumption model is executed by the processor, a deep learning method based on the building cluster spatial energy consumption model is implemented.

[0041] The present invention also provides a computer program product, including a deep learning program based on a building cluster spatial energy consumption model, wherein the deep learning program based on the building cluster spatial energy consumption model implements the deep learning method based on the building cluster spatial energy consumption model when executed by a processor.

[0042] The beneficial effects of this invention are as follows: By constructing a spatial energy consumption relationship graph of a building complex and combining it with a graph convolutional network to perform information propagation and feature aggregation on the spatial energy consumption correlation between building nodes, this invention achieves effective modeling of the spatial correlation and energy consumption coupling relationship within a building complex. This avoids the problem of missing spatial information caused by treating buildings as independent objects in the prior art, thereby improving the overall accuracy of building complex energy consumption prediction.

[0043] This invention introduces the embedded sequence of building cluster spatial energy consumption obtained based on graph convolution into a deep learning time-series prediction model to deeply model the temporal evolution of building cluster energy consumption. In the prediction result aggregation stage, it combines the energy consumption benchmark value for robustness processing, thereby improving the stability and reliability of the total energy consumption prediction results of building clusters under complex energy consumption scenarios. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the first embodiment of a deep learning method based on a building cluster spatial energy consumption model according to the present invention.

[0046] Figure 2 This is a schematic diagram illustrating the multi-scale building cluster energy consumption characteristics of spatial hierarchy perception, representing a first embodiment of a deep learning method based on a building cluster spatial energy consumption model according to the present invention.

[0047] Figure 3 This is a schematic diagram of a mapping method based on a spatial energy consumption model of a building complex according to the first embodiment of the present invention, which uses only spatial distance constraints.

[0048] Figure 4 This is a schematic diagram of the mapping based on the joint constraints of distance and functional similarity in the first embodiment of a deep learning method for a building cluster spatial energy consumption model according to the present invention.

[0049] Figure 5 This is a schematic diagram of a device for a deep learning method based on a building cluster spatial energy consumption model according to the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1: As Figure 1The diagram shown is a flowchart of the first embodiment of the deep learning method based on the building cluster space energy consumption model of the present invention, and the first embodiment of the deep learning method based on the building cluster space energy consumption model of the present invention is proposed.

[0052] In the first embodiment, the deep learning method based on the building cluster space energy consumption model includes:

[0053] Step S10: Obtain the spatial energy consumption model of the building cluster in the university campus. Based on the spatial energy consumption model of the building cluster, use a feature extraction method based on spatial hierarchy perception to extract a multi-scale building cluster feature set D. The multi-scale building cluster feature set includes building spatial location features, building energy consumption time series evolution features, building spatial neighborhood coupling features, and overall building cluster energy consumption trend features.

[0054] It should be noted that the "feature extraction method based on spatial hierarchy perception" in this step refers to the following: when constructing a feature set of a multi-scale building complex, not only are features extracted from the spatial location and historical energy consumption changes of a single building, but also the spatial relative relationships between buildings and the energy consumption evolution characteristics at the overall level of the building complex are introduced, thereby forming a feature expression system with a hierarchical structure.

[0055] Understandably, by introducing a spatially hierarchical feature extraction method in this step, the multi-scale building cluster feature set can provide a more complete and structurally informative input for subsequent energy consumption prediction. This allows the prediction model to no longer rely solely on the historical energy consumption data of a single building during the modeling process, but to comprehensively consider the spatial distribution of buildings within the park, the surrounding environment, and the overall energy consumption change background. This enhances the ability to characterize the energy consumption evolution of the building cluster and improves the accuracy of the prediction results in reflecting the actual operational status of the park. For example, ... Figure 2 As shown, by processing historical energy consumption data of multiple buildings within a university campus, the temporal evolution characteristics of energy consumption for each building can be obtained. Based on this, the overall energy consumption trend of the building complex can be calculated. Although the energy consumption changes of different buildings vary, they exhibit a consistent evolutionary trend at the overall level. This phenomenon indicates that the energy consumption of the building complex has obvious spatial hierarchical characteristics. This step introduces a spatial hierarchy-aware feature extraction method to unify the energy consumption characteristics of the different levels into a multi-scale building complex feature set, providing structured input for subsequent building complex energy consumption prediction.

[0056] Step S20: Based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, a graph construction mechanism with joint constraints of spatial distance and functional similarity is used to perform spatial energy consumption relationship modeling, and output the spatial energy consumption relationship diagram of the building complex. ;

[0057] It should be noted that the "graph construction mechanism with joint constraints of spatial distance and functional similarity" in this step refers to the following: when constructing the spatial energy consumption relationship graph of a building complex, the connection relationship is not established solely based on the geometric distance between buildings. Instead, the consistency or similarity of the functional attributes of the buildings is further introduced as a constraint condition, so that the existence and weight of edges in the graph are simultaneously affected by both "spatial proximity" and "functional similarity". Specifically, spatial distance is used to characterize the objective physical proximity of buildings in the park, while functional similarity is used to characterize the potential consistency of different buildings in terms of usage time, personnel activity patterns, and equipment load types. This avoids mistakenly treating spatially close but functionally significantly different buildings as strongly coupled objects, or mistakenly treating spatially relatively far but functionally similar buildings with highly synchronized energy consumption patterns as weakly coupled objects.

[0058] It should be understood that compared to the two common graph construction methods in traditional techniques, this step has more specific improvements: First, for schemes that construct spatial adjacency graphs based solely on distance thresholds or K-nearest neighbor rules, traditional schemes are prone to generating invalid edges when "functional differences lead to inconsistent energy consumption patterns," introducing noise into subsequent feature propagation. This step reduces such invalid edges by introducing functional similarity constraints. Second, for schemes that construct association graphs based solely on energy consumption correlation coefficients or historical curve similarity, traditional schemes often ignore spatial structure, leading to overly dense graph connections or erroneous strong connections across regions. This step suppresses long-distance non-physical associations through spatial distance constraints, making the graph structure sparser and more physically interpretable. Thus, this step simultaneously improves the rationality of the spatial energy consumption relationship graph at both the "edge selection" and "edge weight definition" levels, reducing the risk of structural bias in the subsequent learning process.

[0059] For example, such as Figure 3 and Figure 4 As shown, observing the overall graph structure reveals that the joint constraint graphing method transforms the spatial energy consumption relationship graph from a simple "geometric proximity graph" into an "energy consumption coupling relationship graph" that simultaneously reflects the consistency between spatial structure and energy consumption behavior. Specifically, as... Figure 3As shown, under the same K-nearest neighbor rule constraint, when graphing is based solely on spatial distance, whether connections are established between building nodes depends primarily on their geometric proximity within the campus plan. While this approach can reflect the physical spatial proximity between buildings, in a typical multi-functional mixed-use scenario like a university campus, it's easy for spatially adjacent building nodes with significantly different functions to be directly connected, such as strong connections between teaching buildings and laboratory buildings, or dormitories and canteens. Such connections manifest as dense cross-functional edges in the graph structure, causing energy consumption information from different energy use modes to be indiscriminately propagated and superimposed during subsequent graph-based feature propagation, thus introducing noise associations inconsistent with actual energy consumption behavior into the spatial energy consumption feature representation. In contrast, as... Figure 4 As shown, when a mapping mechanism combining spatial distance and functional similarity constraints is adopted, the edge weights in the graph are not only affected by the attenuation of physical distance between buildings, but also further modulated by the similarity of the temporal evolution characteristics of building energy consumption. For building pairs that are spatially close but have significantly different temporal energy consumption characteristics, their corresponding edge weights are significantly weakened, or even suppressed in cases of insufficient similarity. Conversely, for building pairs with reasonable spatial distances and highly consistent temporal energy consumption trends, their connections are preserved and strengthened in the graph. Therefore, the spatial energy consumption relationship graph after joint constraint mapping exhibits a clearer structural distribution characterized by "enhanced connections within similar buildings and suppression of weak cross-functional connections."

[0060] Step S30: Based on the building space neighborhood coupling characteristics, the overall energy consumption trend characteristics of the building complex, and the energy consumption relationship diagram of the building complex space. A graph convolution-based information propagation and aggregation mechanism is used to perform the task of learning the spatial energy consumption feature representation of building nodes, and outputs the spatial energy consumption embedding sequence H of building clusters.

[0061] It should be noted that the "information propagation and aggregation mechanism based on graph convolution" in this step refers to: using the energy consumption relationship graph of the building cluster as a topological constraint, performing multiple rounds of neighborhood information propagation and weighted aggregation on the energy consumption-related features of building nodes on this graph structure. This ensures that the feature representation of each building node at a specific time not only includes its own energy consumption evolution information but also incorporates the energy consumption impact information of other buildings in its spatial neighborhood. Specifically, the building spatial neighborhood coupling features are used to characterize the local energy consumption correlation between buildings, the overall energy consumption trend features of the building cluster are used to characterize the macroscopic energy consumption change background at the park level, and the building cluster spatial energy consumption relationship graph is used to constrain the information propagation path and propagation intensity, thus forming a node feature learning process with spatial structural constraints.

[0062] Understandably, by introducing a graph convolution-based information propagation and aggregation mechanism in this step, the spatial energy consumption characteristics of building nodes are no longer generated in isolation based on the historical data of individual buildings. Instead, under the constraints of the graph structure, they can selectively absorb energy consumption characteristic information from adjacent building nodes and modulate it in conjunction with the overall energy consumption trend of the building complex. This process enables the generated spatial energy consumption embedding sequence of the building complex to simultaneously reflect the building's own energy consumption behavior, the impact of local spatial coupling, and the overall operational status of the park, providing high-quality input features with spatial contextual information for subsequent deep learning-based energy consumption prediction models.

[0063] For example, taking the energy consumption data of a university campus over a continuous one-month operating cycle, without using graph convolution mechanisms and only modeling features based on the historical energy consumption of individual buildings, the predictive models for multiple teaching buildings located on the periphery of the campus but with similar functions often exhibit significant lag in their response to energy consumption changes, making it difficult to reflect the synchronous energy consumption changes brought about by concentrated class periods or exam weeks. After introducing an information propagation and aggregation mechanism based on graph convolution, these teaching building nodes can perceive the energy consumption change trends from adjacent teaching buildings and buildings with similar functions in advance in the spatial energy consumption embedding sequence, and the consistency of their embedding features in the time dimension is significantly enhanced. At the same time, during periods when the overall energy consumption level decreases significantly, such as winter and summer vacations, the overall energy consumption trend features of the building group can play a background modulation role in the node-level embedding, enabling the predictive model to have a more stable response characteristic when facing overall changes in the campus's operating status.

[0064] Step S40: Preset the building group energy consumption prediction model, input the building group spatial energy consumption embedding sequence H into the building group energy consumption prediction model, and output the building-level predicted energy consumption results;

[0065] It should be noted that the "preset building cluster energy consumption prediction model" in this step refers to a deep learning prediction model that takes the building cluster spatial energy consumption embedding sequence as input and outputs the building-level energy consumption prediction value. This model focuses on modeling the nonlinear characteristics and temporal dependencies of building energy consumption over time. The building cluster spatial energy consumption embedding sequence H is not the original energy consumption data, but a high-dimensional feature representation learned in the previous step based on the spatial energy consumption relationship graph and graph convolution mechanism. It has explicitly integrated the building's own energy consumption evolution characteristics, the influence of spatial neighborhood coupling, and the overall energy consumption trend information of the building cluster, thus providing input data with spatial contextual semantics for the energy consumption prediction model.

[0066] It should be understood that, on the one hand, traditional time series models that rely solely on the historical energy consumption sequences of individual buildings for prediction typically assume that changes in building energy consumption are primarily determined by their own history, making it difficult to reflect the spatial coupling effects within building clusters. On the other hand, traditional methods that directly stitch together energy consumption data from multiple buildings and input it into the prediction model are prone to low model learning efficiency due to excessively high input dimensionality and implicit representations of spatial relationships. This step introduces a spatial energy consumption embedding sequence of building clusters as an intermediate representation, pre-encoding complex spatial relationships. This allows the prediction model to fully utilize spatial information at the building cluster level while maintaining a relatively simple structure, thereby improving the rationality and consistency of building-level energy consumption prediction.

[0067] Step S50: Perform energy consumption aggregation calculation based on the building-level predicted energy consumption results, and output the total predicted energy consumption value of the building group.

[0068] It should be noted that step S50, which involves performing energy consumption aggregation calculations based on building-level predicted energy consumption results to output the total predicted energy consumption value of the building group, specifically includes: Step S501: Obtaining the building energy consumption benchmark value corresponding to each building, calculating the absolute deviation value based on the building-level predicted energy consumption result and the building energy consumption benchmark value, removing data in the building-level predicted energy consumption result whose absolute deviation value is greater than a preset deviation threshold, and outputting the optimized building-level predicted energy consumption result; Step S502: Finally, calculating and outputting the total predicted energy consumption value of the building group using a simple summation method based on the optimized building-level predicted energy consumption result. In step S501, the "building energy consumption benchmark value" refers to the reference energy consumption level calculated for each building based on its historical operating data within a preset statistical period, used to characterize the typical energy consumption range of the building under normal operating conditions. The building energy consumption benchmark value can be determined based on the building's historical energy consumption mean, quantile, or long-term trend value after smoothing, with the aim of providing a stable benchmark for the building-level predicted energy consumption results. By comparing the predicted energy consumption at the building level with the corresponding building energy consumption benchmark and calculating the absolute deviation between the two, abnormal deviations that may occur during the prediction phase can be effectively identified.

[0069] Understandably, during the building cluster energy consumption prediction process, although the preceding steps have reduced errors through spatial energy consumption relationship modeling and deep learning feature representation learning, abnormal deviations in building-level predicted energy consumption results may still occur at individual time points or specific building nodes due to short-term data disturbances, model uncertainties, or extreme operating conditions. By introducing an absolute deviation screening mechanism based on the building energy consumption benchmark value in step S501, prediction results that significantly deviate from the normal energy consumption range can be eliminated or filtered before the total energy consumption of the building cluster is aggregated, thereby avoiding a disproportionate impact of individual building prediction anomalies on the total energy consumption prediction results of the building cluster.

[0070] Example 2: Furthermore, the present invention provides a deep learning system based on a building cluster spatial energy consumption model, employing a deep learning method based on a building cluster spatial energy consumption model as described in the above embodiments, which can solve a technical problem related to deep learning based on a building cluster spatial energy consumption model. The beneficial effects of the deep learning system based on a building cluster spatial energy consumption model provided by the present invention are the same as those of the deep learning method based on a building cluster spatial energy consumption model provided in the above embodiments, and other technical features of the deep learning system based on a building cluster spatial energy consumption model are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0071] Example 3: This invention provides a deep learning device based on a building cluster spatial energy consumption model. Please refer to... Figure 5A deep learning device based on a building cluster spatial energy consumption model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the deep learning method based on a building cluster spatial energy consumption model as described in Embodiment 1 above. The deep learning device based on a building cluster spatial energy consumption model in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This deep learning device based on a building cluster spatial energy consumption model is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A deep learning device based on a building cluster spatial energy consumption model may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the deep learning device based on the building cluster spatial energy consumption model. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows a deep learning device based on a building cluster spatial energy consumption model to wirelessly or wiredly communicate with other devices to exchange data. While the figure illustrates a deep learning device based on a building cluster spatial energy consumption model with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0072] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the deep learning method based on a building cluster spatial energy consumption model as described above. The computer program product provided by this invention can solve a technical problem related to deep learning based on a building cluster spatial energy consumption model. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the deep learning method based on a building cluster spatial energy consumption model provided in the above embodiments, and will not be repeated here.

[0073] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0074] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0075] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A deep learning method based on a building cluster spatial energy consumption model, characterized in that, The methods include: Step S10: Obtain the spatial energy consumption model of the building cluster in the university campus. Based on the spatial energy consumption model of the building cluster, use a feature extraction method based on spatial hierarchy perception to extract a multi-scale building cluster feature set D. The multi-scale building cluster feature set includes building spatial location features, building energy consumption time series evolution features, building spatial neighborhood coupling features, and overall building cluster energy consumption trend features. Step S20: Based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, a graph construction mechanism with joint constraints of spatial distance and functional similarity is used to perform spatial energy consumption relationship modeling, and output the spatial energy consumption relationship diagram of the building complex. ; Step S30: Based on the building space neighborhood coupling characteristics, the overall energy consumption trend characteristics of the building complex, and the energy consumption relationship diagram of the building complex space. A graph convolution-based information propagation and aggregation mechanism is used to perform the task of learning the spatial energy consumption feature representation of building nodes, and outputs the spatial energy consumption embedding sequence H of building clusters. Step S40: Preset the building group energy consumption prediction model, input the building group spatial energy consumption embedding sequence H into the building group energy consumption prediction model, and output the building-level predicted energy consumption results; Step S50: Perform energy consumption aggregation calculation based on the building-level predicted energy consumption results, and output the total predicted energy consumption value of the building group.

2. The deep learning method based on a building cluster spatial energy consumption model as described in claim 1, characterized in that, Step S10 involves obtaining a spatial energy consumption model of the building complex within the university campus, and then extracting a multi-scale feature set D of the building complex using a feature extraction method based on spatial hierarchy perception. This step specifically includes: Step S101: Obtain the spatial energy consumption model of the building complex within the university campus. Based on the spatial energy consumption model, construct a spatial hierarchy reference structure for the building complex, including individual building hierarchy, local building neighborhood hierarchy, and overall building complex hierarchy. Based on this spatial hierarchy reference structure, generate corresponding building spatial location features for each building. ; Step S102: Obtain the first The building is within a preset historical time window Building energy consumption time series At the scale of individual building levels, the time series of building energy consumption. Perform time-series statistical analysis to extract the time-series evolution characteristics of building energy consumption that reflect the building's own energy consumption variation. ; Step S103: At the scale of the local neighborhood hierarchy of the building, obtain the first... The set of neighboring buildings within a predefined spatial neighborhood of a building is used to aggregate the energy consumption time series of buildings within this set using a spatial distance-weighted aggregation method. The output is a building spatial neighborhood coupling feature that reflects the spatial energy consumption correlation between buildings. ; Step S104: At the scale of the overall building complex, obtain all buildings within a preset historical time window. The energy consumption time series of buildings within the university campus is used to construct an overall energy consumption time series for the building complex. Based on this overall energy consumption time series, a trend analysis method using time window sliding fitting is employed to extract overall energy consumption trend features reflecting the overall operational status of the university campus. .

3. The deep learning method based on a building cluster spatial energy consumption model as described in claim 1, characterized in that, In step S10, the building cluster spatial energy consumption model is used to represent a spatialized model data established by collecting energy consumption data of the building cluster. The building unit level is used to represent the independent spatial location of a single building within the university campus; The local neighborhood hierarchy of a building is used to characterize the spatial relationships between a building and other buildings within its spatial neighborhood; the overall hierarchy of the building complex is used to characterize the relative position of a building within the overall spatial structure of the university campus; and the temporal evolution characteristics of building energy consumption are also considered. This includes characteristics of energy consumption change trends, energy consumption fluctuation amplitude, and energy consumption change rate.

4. The deep learning method based on a building cluster spatial energy consumption model as described in claim 1, characterized in that, In step S20, based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, a graph construction mechanism with joint constraints of spatial distance and functional similarity is used to perform spatial energy consumption relationship modeling, and outputs a spatial energy consumption relationship graph of the building complex. The steps specifically include: Step S201: Based on the spatial location characteristics of the buildings, calculate the adjacent spatial distance between any two adjacent buildings, m-th and n-th, using the Euclidean distance calculation method based on spatial coordinates. ; Step S202: Based on the temporal evolution characteristics of building energy consumption, calculate the functional similarity factor between the m-th building and the n-th building using a temporal feature similarity analysis method based on cosine similarity. ; Step S203: Based on adjacent spatial distance and functional similarity factor An exponentially decaying spatial mapping function is used to map the spatial distance between adjacent buildings, and the distance is then weighted and fused with a functional similarity factor to output a spatial adjacency matrix of building clusters. Based on the spatial adjacency matrix of building clusters Construct a diagram of the spatial energy consumption relationship of a building complex. , ,in Let A represent the set of building nodes within the university campus, and let A represent the spatial energy consumption correlation weight matrix between buildings.

5. The deep learning method based on a building cluster spatial energy consumption model as described in claim 4, characterized in that, In step S30, based on the building space neighborhood coupling characteristics, the overall energy consumption trend characteristics of the building complex, and the building complex spatial energy consumption relationship diagram... The steps for performing the spatial energy consumption feature representation learning task of building nodes using a graph convolution-based information propagation and aggregation mechanism, and outputting the spatial energy consumption embedding sequence H of the building complex, specifically include: Step S301: Based on the spatial coupling characteristics of building spaces and the overall energy consumption trend characteristics of the building complex, create a spatial energy consumption relationship diagram of the building complex. The spatial energy consumption correlation weight matrix is ​​fused element-wise using the Hadamard product method to output the corrected spatial energy consumption correlation weight matrix. ; Step S302: Analyze the energy consumption relationship diagram of the building complex space. The set of building nodes in the data is based on the modified spatial energy consumption correlation weight matrix. The node-level spatial energy consumption feature propagation and aggregation process is performed using a graph convolution-based information propagation and aggregation mechanism, and the node-level spatial energy consumption embedding vector is output. Step S303: Arrange the node-level spatial energy consumption embedding vectors in a temporal sequence according to a preset historical time window, and generate and output the building group spatial energy consumption embedding sequence H.

6. The deep learning method based on a building cluster spatial energy consumption model as described in claim 1, characterized in that, Step S40, which involves pre-setting a building complex energy consumption prediction model, inputting the building complex spatial energy consumption embedding sequence H into the building complex energy consumption prediction model, and outputting building-level predicted energy consumption results, specifically includes: Step S401: Construct a building cluster energy consumption prediction model based on a deep learning model; the building cluster energy consumption prediction model includes a temporal embedding input layer for receiving the building cluster spatial energy consumption embedding sequence H; a multi-layer recurrent neural network layer for deep modeling of the temporal dependencies in the building cluster spatial energy consumption embedding sequence using a Long Short-Term Memory (LSTM) network; a temporal attention layer for weighting the hidden states of each time step output by the multi-layer recurrent neural network layer within the historical time window using an additive attention method based on Softmax normalization; and a fully connected regression layer for mapping the temporal features weighted by the temporal attention layer to building cluster-level predicted energy consumption values, forming the output of building-level predicted energy consumption results; Step S402: Obtain the spatial energy consumption embedding sequence of historical building complexes and the corresponding template set of building-level predicted energy consumption results; use the spatial energy consumption embedding sequence of historical building complexes as the input of the building complex energy consumption prediction model, use the template set of building-level predicted energy consumption results as the output of the building complex energy consumption prediction model, and perform a pre-training process on the building complex energy consumption prediction model in combination with the mean square error loss function. Step S403: Input the building cluster spatial energy consumption embedding sequence H into the pre-trained building cluster energy consumption prediction model and output the building-level predicted energy consumption results.

7. The deep learning method based on a building cluster spatial energy consumption model as described in claim 1, characterized in that, Step S50, which involves performing energy consumption aggregation calculations based on building-level predicted energy consumption results and outputting the total predicted energy consumption value for the building complex, specifically includes: Step S501: Obtain the building energy consumption benchmark value corresponding to each building, calculate the absolute deviation value based on the building-level predicted energy consumption result and the building energy consumption benchmark value, remove data in the building-level predicted energy consumption result whose absolute deviation value is greater than the preset deviation threshold, and output the optimized building-level predicted energy consumption result. Step S502: Finally, based on the optimized building-level predicted energy consumption results, the total predicted energy consumption of the building complex is calculated and output using a simple summation method.

8. A deep learning system based on a building cluster spatial energy consumption model, applied to the deep learning method based on a building cluster spatial energy consumption model as described in any one of claims 1 to 7, characterized in that, The deep learning system based on the building cluster space energy consumption model includes: The multi-scale spatial energy consumption feature extraction module is used to obtain the spatial energy consumption model of building clusters in the university campus. Based on the spatial energy consumption model of the building clusters, a feature extraction method based on spatial hierarchy perception is used to extract the multi-scale building cluster feature set D. The multi-scale building cluster feature set includes building spatial location features, building energy consumption temporal evolution features, building spatial neighborhood coupling features, and overall building cluster energy consumption trend features. The spatial energy consumption relationship graph construction module is used to perform spatial energy consumption relationship modeling based on the spatial location characteristics and temporal evolution characteristics of building energy consumption, and adopts a graph construction mechanism with joint constraints of spatial distance and functional similarity, outputting a spatial energy consumption relationship graph of the building complex. ; The spatial energy consumption map feature learning module is used to learn energy consumption based on the spatial neighborhood coupling features of buildings, the overall energy consumption trend features of building groups, and the spatial energy consumption relationship map of building groups. A graph convolution-based information propagation and aggregation mechanism is used to perform the task of learning the spatial energy consumption feature representation of building nodes, and outputs the spatial energy consumption embedding sequence H of building clusters. The building cluster energy consumption prediction module is used to preset the building cluster energy consumption prediction model, input the building cluster spatial energy consumption embedding sequence H into the building cluster energy consumption prediction model, and output the building-level predicted energy consumption results. The building cluster energy consumption aggregation module is used to perform energy consumption aggregation calculations based on building-level predicted energy consumption results and output the total predicted energy consumption value of the building cluster.

9. A deep learning device based on a building complex spatial energy consumption model, characterized in that, The deep learning device based on the building cluster spatial energy consumption model includes: a memory, a processor, and a deep learning program based on the building cluster spatial energy consumption model stored in the memory and executable on the processor. When the deep learning program based on the building cluster spatial energy consumption model is executed by the processor, it implements a deep learning method based on the building cluster spatial energy consumption model according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a deep learning program based on a building cluster spatial energy consumption model. When the deep learning program based on the building cluster spatial energy consumption model is executed by a processor, it implements a deep learning method based on a building cluster spatial energy consumption model as described in any one of claims 1 to 7.

Images