A building energy consumption prediction method, system, device and medium

Abstract

The present application provides a method, system, device and medium for predicting building energy consumption. By performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area, the associated identification features of the energy consumption between the overall building and each functional area are obtained. An energy consumption prediction model is constructed based on all the associated identification features, and the suppression gain coefficient of the corresponding characteristic channel of each functional area in the energy consumption prediction model is determined; the energy consumption fluctuation level during the abnormal energy consumption period is fuzzy evaluated according to the load fluctuation characteristics of each functional area to obtain a fuzzy evaluation value of the energy consumption fluctuation level; when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, the energy consumption feature attention of each functional area in the energy consumption prediction model is cost-constrained according to all the suppression gain coefficients, and then the building energy consumption is predicted based on the cost-constrained energy consumption prediction model. By adopting the solution of the present application, the cost constraint of different feature attentions in the energy consumption prediction model can be achieved.

Description

Technical Field

[0001] The present application relates to the technical field of energy consumption prediction, and more specifically, to a method, system, device and medium for predicting building energy consumption. Background Art

[0002] In building energy management, building energy consumption detection, as a basic link in smart energy regulation, is of great significance for achieving building energy conservation and consumption reduction, load response adjustment and energy utilization optimization. Traditional energy consumption monitoring methods mostly focus on the collection and presentation of total energy consumption data, lack the ability to fine-tune the energy consumption characteristics of each functional area and dynamically respond, and it is difficult to support the early identification and targeted regulation of abnormal energy consumption behavior. With the complexity of building structure and functional layout, refined energy consumption perception and high-timeliness prediction capabilities have gradually become the core requirements of building energy management systems.

[0003] Existing mainstream building energy consumption prediction methods generally have static problems in their model structure design, making it difficult to adapt to the complex and changing energy consumption characteristics of functional areas within a building. In particular, in scenarios with frequent load fluctuations or frequent local energy consumption anomalies, the nonlinear correlation between overall energy consumption data and local energy consumption behavior is difficult to effectively extract, resulting in the model's lack of responsiveness to regional abnormal energy consumption. The fundamental reason for this defect is that traditional methods fail to introduce a feature channel mechanism that can reflect the dynamic correlation between overall and local energy consumption, lack means to suppress the energy consumption characteristics of irrelevant areas, and lack strategies to strengthen the characteristics of key areas. This in turn limits the model's ability to analyze energy consumption anomalies and its predictive stability. Therefore, how to implement cost constraints on the attention of different features in energy consumption prediction models to improve the robustness of building energy consumption prediction has become a difficult problem facing the industry. Summary of the Invention

[0004] The present application provides a building energy consumption prediction method, system, equipment and medium, which can realize the cost constraint of different feature attention in the energy consumption prediction model.

[0005] In a first aspect, the present application provides a building energy consumption prediction method, wherein low-frequency energy consumption monitoring equipment is pre-deployed outside a target building and high-frequency energy consumption monitoring equipment is pre-deployed in various functional areas inside the target building. The building energy consumption prediction method comprises the following steps:

[0006] During the current energy consumption monitoring cycle, the overall energy consumption data of the target building is collected through low-frequency energy consumption monitoring equipment, and the local energy consumption data of each functional area of ​​the target building is collected through high-frequency energy consumption monitoring equipment;

[0007] Performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data to obtain association identification features between the overall energy consumption and the energy consumption of each functional area, and then constructing an energy consumption prediction model based on all the association identification features, and determining the suppression gain coefficient of the feature channel corresponding to each functional area in the energy consumption prediction model;

[0008] Based on the historical energy consumption records of the target building, the energy consumption abnormal period in the next energy consumption monitoring cycle is predicted, and then the energy consumption fluctuation level in the energy consumption abnormal period is fuzzy evaluated according to the load fluctuation characteristics of each functional area to obtain the fuzzy evaluation value of the energy consumption fluctuation level in the energy consumption abnormal period;

[0009] During the energy consumption prediction process of the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, the energy consumption characteristic attention of each functional area in the energy consumption prediction model is cost-constrained based on all the suppression gain coefficients, and then the building energy consumption of the next energy consumption monitoring cycle is predicted based on the cost-constrained energy consumption prediction model.

[0010] Preferably, the overall energy consumption of the target building and the local energy consumption of each functional area are associated with feature mapping based on the overall energy consumption data and all local energy consumption data, and the associated identification features of the energy consumption between the overall building and each functional area are obtained, specifically including:

[0011] Performing time alignment on the overall energy consumption data and the local energy consumption data of each functional area, and extracting correlation mapping features of energy consumption between the overall energy consumption and each functional area within the same time window;

[0012] Determine the dynamic correlation coefficient between the local energy consumption data of each functional area and the overall energy consumption data;

[0013] The correlation identification features of the overall energy consumption and the energy consumption of each functional area are determined through all dynamic correlation coefficients and all correlation mapping features.

[0014] Preferably, building an energy consumption prediction model based on all associated identification features specifically includes:

[0015] Build a prediction model containing multiple feature input channels, where each channel corresponds to a functional area;

[0016] The associated identification features corresponding to each functional area are respectively used as input parameters of the corresponding channel, and the initial parameters of the prediction model are set;

[0017] The prediction model is trained by using a supervised training method in combination with the historical energy consumption data of the target building, and the initial parameters are iteratively optimized to obtain the energy consumption prediction model.

[0018] Preferably, determining the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model specifically includes:

[0019] Extract the deviation distribution characteristics between the energy consumption change rate of each functional area during the historical energy consumption monitoring period and the overall energy consumption change trend;

[0020] Determining the response strength of the characteristic channel corresponding to each functional area in the energy consumption prediction model based on the deviation distribution characteristics;

[0021] The initial inhibition factor of the characteristic channel corresponding to each functional area is set according to all response intensities;

[0022] The initial suppression factor of each characteristic channel is dynamically adjusted through error feedback during the training process of the energy consumption prediction model to obtain the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model.

[0023] Preferably, predicting the abnormal energy consumption period in the next energy consumption monitoring cycle based on the historical energy consumption records of the target building specifically includes:

[0024] Obtain historical overall energy consumption data from the historical energy consumption records of the target building;

[0025] Periodic feature extraction is performed on the historical overall energy consumption data to identify time periods with sudden energy consumption changes and frequent anomalies, and the energy consumption anomaly period within the next energy consumption monitoring cycle is predicted based on the identified time periods.

[0026] Preferably, performing fuzzy evaluation on the energy consumption fluctuation level during the abnormal energy consumption period according to the load fluctuation characteristics of each functional area, and obtaining the fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period specifically includes:

[0027] According to the load fluctuation characteristics of each functional area in the historical period, a fuzzy membership function describing energy consumption fluctuation is constructed, and then the fuzzy membership of each functional area in different fluctuation intervals is determined;

[0028] Perform fuzzy reasoning on the energy consumption fluctuation level of each functional area during the energy consumption abnormality period based on all fuzzy membership degrees to obtain the energy consumption fluctuation credibility of each functional area at different fluctuation levels;

[0029] The fuzzy evaluation value of the energy consumption fluctuation level during the energy consumption abnormality period is determined through all energy consumption fluctuation credibility.

[0030] Preferably, the low-frequency energy consumption monitoring device is a multifunctional power meter.

[0031] In a second aspect, the present application provides a building energy consumption prediction system, comprising:

[0032] The collection module is used to collect the overall energy consumption data of the target building through low-frequency energy consumption monitoring equipment and the local energy consumption data of each functional area of ​​the target building through high-frequency energy consumption monitoring equipment during the current energy consumption monitoring cycle;

[0033] a processing module for performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data, obtaining association identification features between the overall energy consumption and the energy consumption of each functional area, and then constructing an energy consumption prediction model based on all the association identification features, and determining a suppression gain coefficient for a feature channel corresponding to each functional area in the energy consumption prediction model;

[0034] The processing module is further configured to predict an abnormal energy consumption period within the next energy consumption monitoring cycle based on the historical energy consumption records of the target building, and then perform a fuzzy evaluation of the energy consumption fluctuation level during the abnormal energy consumption period according to the load fluctuation characteristics of each functional area to obtain a fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period;

[0035] The execution module is used to, during the energy consumption prediction process of the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, cost-constrain the energy consumption characteristic attention of each functional area in the energy consumption prediction model according to all the suppression gain coefficients, and then predict the building energy consumption of the next energy consumption monitoring cycle based on the cost-constrained energy consumption prediction model.

[0036] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a code, and the processor is configured to obtain the code and execute the above-mentioned building energy consumption prediction method.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which implements the above-mentioned building energy consumption prediction method when executed by a processor.

[0038] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0039] In an embodiment of the present application, during a current energy consumption monitoring cycle, overall energy consumption data of a target building is collected by a low-frequency energy consumption monitoring device, and local energy consumption data of each functional area of ​​the target building is collected by a high-frequency energy consumption monitoring device; feature mapping and association are performed on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data to obtain associated identification features of the energy consumption between the overall building and each functional area, and then an energy consumption prediction model is constructed based on all associated identification features, and a suppression gain coefficient of a feature channel corresponding to each functional area in the energy consumption prediction model is determined; based on the historical energy consumption records of the target building, an abnormal energy consumption period in the next energy consumption monitoring cycle is predicted, and then a fuzzy evaluation is performed on the energy consumption fluctuation level of the abnormal energy consumption period based on the load fluctuation characteristics of each functional area to obtain a fuzzy evaluation value of the energy consumption fluctuation level in the abnormal energy consumption period; during the energy consumption prediction process for the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds a preset energy consumption fluctuation threshold, a cost constraint is applied to the energy consumption feature attention of each functional area in the energy consumption prediction model based on all the suppression gain coefficients, and then the building energy consumption for the next energy consumption monitoring cycle is predicted based on the cost-constrained energy consumption prediction model.

[0040] It can be seen that the present application uses all the suppression gain coefficients to impose cost constraints on the attention of the energy consumption characteristics of each functional area in the energy consumption prediction model, and predicts the energy consumption of the building based on the cost-constrained energy consumption prediction model; first, by performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area, it is possible to reveal the response degree and behavioral characteristics of different functional areas to the overall energy consumption changes under the premise of time consistency, thereby establishing an energy consumption feature expression mechanism with regional resolution capability. This feature mapping process realizes the high-dimensional correlation expression of energy consumption data by extracting the coupling characteristics between the overall and local energy consumption, and constructs an energy consumption prediction model on this basis, which enhances the model's perception of energy consumption differences between regions from the source; secondly, by determining the suppression gain coefficients of the feature channels corresponding to each functional area in the energy consumption prediction model, it is possible to dynamically adjust the expression strength of the feature channels corresponding to each functional area, effectively suppress the interference of abnormal regional features on the overall prediction results, and highlight the main features of key areas. The energy consumption characteristics are guided, and the controllable adjustment of feature attention is realized, providing a clearer semantic basis for the prediction of abnormal periods; then, based on the historical energy consumption data, the abnormal energy consumption period of the next monitoring cycle is predicted, and combined with the load fluctuation characteristics of the functional area, the fuzzy evaluation method is used to quantitatively integrate and evaluate the energy consumption fluctuation level, thereby outputting a fuzzy evaluation value with uncertainty tolerance, which significantly improves the adaptability of the energy consumption prediction model to complex fluctuation behaviors; finally, when the fuzzy evaluation value exceeds the set threshold, the cost constraint mechanism is used to activate the readjustment process of the feature attention in the energy consumption prediction model, so that the energy consumption prediction model can further strengthen the energy consumption feature expression of the key area in the abnormal prediction task, and minimize the misleading effect of local noise features on the judgment results of the energy consumption prediction model. This mechanism constructs an adaptive adjustment path for the energy consumption prediction model in the energy consumption fluctuation scenario through dynamic cost constraints, effectively overcoming the problem of insufficient robustness of traditional prediction methods due to fixed feature expression patterns. In summary, the present application scheme can realize the cost constraint of different feature attentions in the energy consumption prediction model, thereby improving the robustness of building energy consumption prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is an exemplary flow chart of a building energy consumption prediction method according to some embodiments of the present application;

[0042] Figure 2 is a schematic diagram of an application scenario of a building energy consumption prediction system according to some embodiments of the present application;

[0043] Figure 3 is a schematic diagram of a process for determining a fuzzy evaluation value according to some embodiments of the present application;

[0044] Figure 4 is a schematic structural diagram of a building energy consumption prediction system according to some embodiments of the present application;

[0045] Figure 5 It is a structural diagram of a computer device for implementing a building energy consumption prediction method according to some embodiments of the present application. DETAILED DESCRIPTION

[0046] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0047] refer to Figure 1 , which is an exemplary flow chart of a building energy consumption prediction method according to some embodiments of the present application. The building energy consumption prediction method 100 mainly includes the following steps:

[0048] In step 101, during the current energy consumption monitoring cycle, the overall energy consumption data of the target building is collected by low-frequency energy consumption monitoring equipment, and the local energy consumption data of each functional area of ​​the target building is collected by high-frequency energy consumption monitoring equipment.

[0049] It should be noted that in this application, the pre-deployment of low-frequency energy consumption monitoring equipment on the outside of the target building and the deployment of high-frequency energy consumption monitoring equipment in various functional areas inside the target building can be achieved in the following ways, namely: low-frequency energy consumption monitoring equipment can be installed at the main incoming line of the building or the main distribution meter to record the overall energy consumption of the building. The sampling frequency of such equipment is generally set to once every 5 to 15 minutes to meet the needs of overall energy consumption trend monitoring; high-frequency energy consumption monitoring equipment can be installed in various functional areas inside the building (such as office areas, computer rooms, and kitchens). Such equipment is installed in regional branch circuits or local distribution units, and the sampling frequency is usually once every 1 to 5 minutes to capture rapid changes in regional energy consumption.

[0050] In specific implementation, the collection of the overall energy consumption data of the target building through low-frequency energy consumption monitoring equipment can be achieved in the following ways, namely: the overall energy consumption data of the building can be collected through a multi-function power meter, and the sampling frequency is set to once every 5 to 15 minutes, and it is connected to the background data platform or edge computing gateway through a standard communication protocol (such as Modbus), and the data is collected and uploaded regularly; the collection of local energy consumption data of each functional area of ​​the target building through high-frequency energy consumption monitoring equipment can be achieved in the following ways, namely: the energy consumption data of each functional area of ​​the building can be collected through a multi-function power meter, and the sampling frequency is set to once every 1 to 5 minutes, and it is connected to the background data platform or edge computing gateway through a standard communication protocol (such as Modbus), and the data is collected and uploaded regularly; it should be noted that the energy consumption data specifically includes voltage, current, power and total power consumption.

[0051] In some embodiments, reference Figure 2As shown in the figure, this figure is a schematic diagram of the application scenario of the building energy consumption prediction system in some embodiments of the present application. A plurality of monitoring devices 120 are installed in the monitoring scene 110. The energy consumption data of the building is monitored by the monitoring devices, and then the collected energy consumption data is transmitted to the data processor 130. The data processor analyzes and processes the energy consumption data, and predicts the building energy consumption. Finally, the prediction results are transmitted to the data storage 140, and the prediction results are stored in the data storage.

[0052] In step 102, feature mapping association is performed on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data to obtain the associated identification features of the energy consumption between the overall building and each functional area. Then, an energy consumption prediction model is constructed based on all the associated identification features, and the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model is determined.

[0053] In some embodiments, feature mapping is performed on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data to obtain the associated identification features of the energy consumption between the overall building and each functional area. The following steps can be used:

[0054] Time-aligning the overall energy consumption data and the local energy consumption data of each functional area, and extracting correlation mapping features of energy consumption between the overall and each functional area within the same time window;

[0055] Determine the dynamic correlation coefficient between the local energy consumption data of each functional area and the overall energy consumption data;

[0056] The correlation identification features of the overall energy consumption and the energy consumption of each functional area are determined through all dynamic correlation coefficients and all correlation mapping features.

[0057] It should be noted that the association mapping feature in this application is a feature that measures the temporal synchronization between the overall energy consumption and the local energy consumption of each functional area; the dynamic correlation coefficient in this application refers to a statistic used to measure the local linear correlation strength generated by the change of two time series over time in a sliding time window; the association identification feature in this application is a feature used to characterize the association response strength of each functional area in the overall energy consumption structure.

[0058] In specific implementation, first, the overall energy consumption data and the local energy consumption data of each functional area are time-aligned, and the correlation mapping features of the energy consumption between the overall and each functional area in the same time window are extracted. This can be achieved in the following way, namely: the overall energy consumption data and the local energy consumption data of each functional area can be time-aligned, that is, the two are resampled and interpolated according to a unified sampling period (such as every 5 minutes), and an energy consumption data matrix under a unified timestamp is constructed. Then, based on each time window (such as a sliding window length of 60 minutes), the mapping feature pairs between the overall energy consumption and the local energy consumption of each functional area are extracted. For example, the statistical features of normalized difference, sliding mean difference, and energy consumption ratio can be used to form a feature vector, and the feature vector is used as the correlation mapping feature; secondly, the dynamic correlation coefficient of the local energy consumption data of each functional area and the overall energy consumption data can be determined in the following way The method is implemented in the following way, that is, for each functional area, the Pearson correlation coefficient of the sliding window is used to analyze the dynamic correlation between the overall energy consumption and the energy consumption of the functional area over time, and the analysis result is used as the dynamic correlation coefficient between the local energy consumption data of the functional area and the overall energy consumption data, thereby obtaining the dynamic correlation coefficient between the local energy consumption data of each functional area and the overall energy consumption data; then, the association identification feature of the energy consumption between the overall and each functional area is determined through all the dynamic correlation coefficients and all the association mapping features, namely, for each functional area, the vector composed of the dynamic correlation coefficient between the local energy consumption data of the functional area and the overall energy consumption data and the association mapping features of the energy consumption between the overall and the functional area in the same time window is used as the association identification feature of the energy consumption between the overall and the functional area, thereby obtaining the association identification feature of the energy consumption between the overall and each functional area.

[0059] In some embodiments, building an energy consumption prediction model based on all associated identification features can be achieved by using the following steps:

[0060] Build a prediction model containing multiple feature input channels, where each channel corresponds to a functional area;

[0061] The associated identification features corresponding to each functional area are respectively used as input parameters of the corresponding channel, and the initial parameters of the prediction model are set;

[0062] The prediction model is trained by using a supervised training method in combination with the historical energy consumption data of the target building, and the initial parameters are iteratively optimized to obtain the energy consumption prediction model.

[0063] It should be noted that the initial parameters in this application specifically include network structure parameters, activation function type, loss function, optimizer configuration parameters and training control parameters.

[0064] In specific implementation, the following method can be used to construct a prediction model containing multiple feature input channels, namely: first divide the input channels according to the functional area as the dimension, and construct a model structure with multiple input branches, such as using a multi-channel LSTM network or, each channel inputs the associated identification features of a functional area; adopt a supervised training method to train the prediction model in combination with the historical energy consumption data of the target building, and iteratively optimize the initial parameters, and then train to obtain an energy consumption prediction model. The following method can be used, namely: first organize the energy consumption data of the target building in the historical period, and construct a supervised training data set, wherein the input is the associated identification features of each functional area in multiple consecutive time steps, and the label is the overall energy consumption value after the corresponding time step, and then divide the supervised training data set into a training set and a validation set in proportion, and input them into the constructed multi-channel prediction model. During the training process, the mean squared error (MSE) is used. Error (MSE) is used as the loss function, and an optimizer (such as Adam) is used to execute the back propagation algorithm to iteratively optimize the model parameters such as the initially set weights and biases. In each round of iteration, the model calculates the predicted value based on the current parameters, compares it with the true value to generate an error, and the error is back-propagated to update the parameters of each layer to minimize the loss function value. During the training process, a preset iterative stopping strategy can be introduced to prevent overfitting, and the generalization performance of the model is monitored through the validation set. When the model reaches the convergence standard or early termination condition on the validation set, the trained energy consumption prediction model is output, which has the ability to predict the future overall energy consumption based on the energy consumption characteristics of each functional area.

[0065] In some embodiments, determining the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model can be achieved by using the following steps:

[0066] Extract the deviation distribution characteristics between the energy consumption change rate of each functional area during the historical energy consumption monitoring period and the overall energy consumption change trend;

[0067] Determining the response strength of the characteristic channel corresponding to each functional area in the energy consumption prediction model based on the deviation distribution characteristics;

[0068] The initial inhibition factor of the characteristic channel corresponding to each functional area is set according to all response intensities;

[0069] The initial suppression factor of each characteristic channel is dynamically adjusted through error feedback during the training process of the energy consumption prediction model to obtain the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model.

[0070] It should be noted that the deviation distribution characteristics in this application are characteristics that measure the degree of deviation between the energy consumption changes in each functional area and the overall energy consumption change trend; the initial suppression factor in this application refers to a preset weight coefficient used to limit the response intensity of the functional area characteristic channel; the suppression gain coefficient in this application is an indicator that quantifies the degree of suppression of the contribution of the functional area characteristic channel to the energy consumption prediction model.

[0071] In the specific implementation, first, the deviation distribution characteristics between the energy consumption change rate of each functional area in the historical energy consumption monitoring period and the overall energy consumption change trend are extracted in the following way, namely: in multiple historical energy consumption monitoring periods, the energy consumption change rate sequence of each functional area is calculated, and the overall energy consumption change trend of the corresponding period is extracted simultaneously, and the deviation value between the two is calculated point by point using a sliding time window, and the deviation distribution characteristics (such as mean, standard deviation) of each functional area are statistically obtained; secondly, based on the deviation distribution characteristics, the response intensity of the characteristic channel corresponding to each functional area in the energy consumption prediction model is determined in the following way, namely: a mapping function (such as Sigmoid function) is used to map the deviation distribution characteristics to the response value of the characteristic channel corresponding to each functional area. The mapping function can convert the deviation size into a numerical expression of the response intensity, and the calculated response value is used as the characteristic channel corresponding to each functional area. The response strength of the characteristic channel; then, the initial suppression factor of the characteristic channel corresponding to each functional area is set according to all the response intensities, which can be achieved in the following way: according to the response strength of all functional areas, the initial suppression factors of different channels are reasonably set, and a larger deviation is usually assigned a higher suppression factor after normalization to reduce its excessive influence on the model; finally, the initial suppression factor of each characteristic channel is dynamically adjusted through error feedback in the training process of the energy consumption prediction model, and the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model is obtained. The following method is used, that is, in the training process of the energy consumption prediction model, the model error feedback mechanism is combined, and the initial suppression factor is dynamically adjusted by using optimization algorithms such as gradient descent, and the suppression factor is continuously optimized according to error back propagation, and the optimized suppression factor is then used as the suppression gain coefficient of the characteristic channel corresponding to the corresponding functional area.

[0072] In step 103, the abnormal energy consumption period in the next energy consumption monitoring cycle is predicted based on the historical energy consumption records of the target building, and then the energy consumption fluctuation level in the abnormal energy consumption period is fuzzy evaluated according to the load fluctuation characteristics of each functional area to obtain a fuzzy evaluation value of the energy consumption fluctuation level in the abnormal energy consumption period.

[0073] In some embodiments, the following steps may be used to predict the abnormal energy consumption period in the next energy consumption monitoring cycle based on the historical energy consumption records of the target building:

[0074] Obtain historical overall energy consumption data from the historical energy consumption records of the target building;

[0075] Periodic feature extraction is performed on the historical overall energy consumption data to identify time periods with sudden energy consumption changes and frequent anomalies, and the energy consumption anomaly period within the next energy consumption monitoring cycle is predicted based on the identified time periods.

[0076] It should be noted that the historical energy consumption records in this application refer to energy consumption records within multiple historical monitoring cycles; the abnormal energy consumption period in this application refers to the time period when the overall energy consumption of the building fluctuates violently or deviates from the normal pattern within a specific time period.

[0077] In specific implementation, we first extract the daily overall energy consumption data from the historical energy consumption records of the target building, and construct an energy consumption curve arranged in time series. Then, we use periodic analysis methods (such as Fourier transform) to extract the periodic components in the data to identify the energy consumption fluctuation patterns at different time scales. Secondly, we use mutation detection algorithms (such as the cumulative sum algorithm based on moving average difference) to identify mutation points such as sudden increases and decreases in energy consumption in the curve. Further, we statistically analyze the time position and frequency of these mutation points in previous cycles to identify high-frequency abnormal sections, such as morning rush hour and seasonal transition periods. Finally, based on the distribution pattern of these historical high-frequency abnormal time periods, we use time series clustering (such as K-means to cluster the start time of the mutation section) to predict the potential abnormal time intervals in the next energy consumption monitoring cycle.

[0078] In some embodiments, reference Figure 3 As shown in FIG, this figure is a schematic diagram of a process for determining a fuzzy evaluation value in some embodiments of the present application. In this embodiment, a fuzzy evaluation is performed on the energy consumption fluctuation level during the abnormal energy consumption period based on the load fluctuation characteristics of each functional area. The fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period can be obtained by the following steps:

[0079] In step 1031, a fuzzy membership function describing energy consumption fluctuation is constructed based on the load fluctuation characteristics of each functional area in the historical period, and the fuzzy membership of each functional area in different fluctuation intervals is determined;

[0080] In step 1032, fuzzy reasoning is performed on the energy consumption fluctuation level of each functional area during the energy consumption abnormality period based on all fuzzy memberships to obtain the energy consumption fluctuation credibility of each functional area at different fluctuation levels;

[0081] In step 1033, a fuzzy evaluation value of the energy consumption fluctuation level during the energy consumption abnormality period is determined through all energy consumption fluctuation credibility.

[0082] It should be noted that the fuzzy membership in this application is a measurement indicator used to indicate the degree to which an object belongs to a specific fuzzy set; the energy consumption fluctuation credibility in this application refers to the credibility of indicating that the energy consumption fluctuation of a functional area belongs to a certain fluctuation level during the fuzzy evaluation process; the fuzzy evaluation value in this application is an indicator to measure the overall energy consumption fluctuation degree during the abnormal energy consumption period.

[0083] In the specific implementation, first, a fuzzy membership function describing energy consumption fluctuation is constructed based on the load fluctuation characteristics of each functional area in the historical period, and then the fuzzy membership of each functional area in different fluctuation intervals is determined. This can be achieved in the following way, namely: first collect the load power data of each functional area in multiple historical periods, calculate its load fluctuation characteristic indicators, such as standard deviation, variance or coefficient of variation, which reflect the fluctuation amplitude and instability of the load, and divide the load fluctuation amplitude intervals based on statistical results, such as low fluctuation interval, medium fluctuation interval and high fluctuation interval. The intervals are usually set in combination with expert experience or statistical distribution characteristics. The boundary of the interval is determined, and a fuzzy membership function is constructed for each interval. Common function forms include triangular function and trapezoidal function. These functions map the actual load fluctuation value to a membership value in the range of [0, 1], indicating the degree to which the value belongs to the corresponding fluctuation interval. The load fluctuation value of each functional area in the current period is further substituted into the corresponding membership function, and its fuzzy membership in each fluctuation interval is calculated to obtain a fuzzy set reflecting the fluctuation state of the area; based on all the fuzzy memberships, fuzzy reasoning is performed on the energy consumption fluctuation level of each functional area in the energy consumption abnormal period, and the load fluctuation level of each functional area under different fluctuation levels is obtained. The credibility of energy consumption fluctuation can be achieved in the following way: taking the fuzzy membership of each functional area as input, constructing a fuzzy rule base, the rule form usually adopts the "if-then" structure, for example, "if the load fluctuation membership is high, then the energy consumption fluctuation level is high", these rules can be determined based on expert experience or historical data statistics, and using the Mamdani fuzzy inference method, the input fuzzy membership value is fuzzy matched with the rule base, and the activation strength is calculated by the intersection and union operations of fuzzy sets to obtain the output fuzzy set of each rule, and then synthesize the output fuzzy set of all rules to form a comprehensive output membership function. The comprehensive output membership function is defuzzified by the centroid method or the maximum membership method to obtain the energy consumption fluctuation credibility of the functional area at different fluctuation levels; finally, the fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period is determined by all the energy consumption fluctuation credibility. This can be achieved in the following way, namely: all the energy consumption fluctuation credibility can be fuzzy fused to obtain the membership function of the comprehensive energy consumption fluctuation of the entire building during the period, and then converted into a specific fuzzy evaluation value through the defuzzification method, which is used to quantitatively reflect the size and credibility of the energy consumption fluctuation, thereby realizing the overall fuzzy evaluation of the energy consumption fluctuation during the abnormal period.

[0084] It should be noted that this application introduces a fuzzy membership function to quantitatively model the historical characteristics of load fluctuations in each functional area, thereby realizing the conversion from quantitative data to fuzzy levels, effectively improving the inclusiveness and flexibility of the fluctuation range, and secondly, integrating the fuzzy membership information of multiple functional areas through the fuzzy reasoning mechanism to obtain the credibility distribution under each fluctuation level, thereby enhancing the ability to express the coupling relationship of multi-source data; finally, based on the credibility of each area, the fuzzy evaluation value of the energy consumption abnormal period is comprehensively output, providing a path for dynamic, flexible and interpretable characterization of the energy consumption fluctuation level, thereby improving the accuracy of identifying abnormal risks and the forward-looking response in building energy consumption management. This scheme not only improves the accuracy of energy consumption fluctuation level assessment, but also enhances the adaptability of the assessment system to uncertainty and regional differences.

[0085] In step 104, during the energy consumption prediction process of the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, the energy consumption characteristic attention of each functional area in the energy consumption prediction model is cost-constrained according to all the suppression gain coefficients, and then the building energy consumption of the next energy consumption monitoring cycle is predicted based on the cost-constrained energy consumption prediction model.

[0086] It should be noted that the judgment of "when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold" is set in the present application scheme. The main purpose is to realize dynamic regulation of the behavior of the prediction model to cope with potential high-risk energy consumption fluctuations. The specific reasons are as follows: the fuzzy evaluation value is a comprehensive indicator to measure the overall fluctuation level during the abnormal energy consumption period. Its increase means that the building energy consumption is about to enter a highly volatile and unstable state. At this time, if we continue to rely on the conventional prediction model, the abnormal amplification effect of local energy consumption may easily lead to a significant increase in the overall prediction deviation; and the introduction of a preset energy consumption fluctuation threshold as a decision boundary helps to divide the energy consumption prediction task into two categories: normal state and high fluctuation state, so as to realize differentiated processing of the model in different operating scenarios and improve the stability and security of the prediction; only when the evaluation value exceeds the threshold will the "cost constraint mechanism" be activated, and the feature attention of the abnormal dominant area is adjusted by introducing the suppression gain coefficient, thereby avoiding excessive interference of the abnormal area on the overall prediction result, and realizing the enhanced robustness of the energy consumption prediction model.

[0087] It should also be noted that the energy consumption fluctuation threshold in this application refers to the numerical limit used to determine whether there are significant abnormal energy consumption fluctuations within a certain energy consumption monitoring period. Its setting scheme can be determined based on the statistical distribution characteristics of historical energy consumption data. Specifically, the percentile method can be used to set the upper limit value of the 90% or 95% confidence interval as the threshold according to the fluctuation amplitude distribution to ensure that the identified fluctuation is indeed an abnormal fluctuation area.

[0088] In some embodiments, cost constraints on the energy consumption feature attention of each functional area in the energy consumption prediction model based on all suppression gain coefficients can be implemented by the following steps:

[0089] Extracting energy consumption feature attention of each functional area in the energy consumption prediction model;

[0090] A cost function reflecting the deviation between feature attention and its expected contribution is constructed based on the inhibition gain coefficient of the feature channel corresponding to each functional area;

[0091] The cost function is introduced into the parameter update process of the energy consumption prediction model to iteratively constrain the attention of the energy consumption characteristics of each functional area and suppress the functional area features with excessive contribution.

[0092] It should be noted that the energy consumption feature attention in this application is an indicator to measure the contribution of the energy consumption features of the functional area to the overall energy consumption prediction results.

[0093] In the specific implementation, first, the energy consumption feature attention of each functional area in the energy consumption prediction model is extracted, which can be achieved in the following way, namely: the attention allocation value of each functional area in the prediction process can be obtained by traversing the attention allocation layer (such as the channel attention module) of the energy consumption prediction model; secondly, the cost function reflecting the deviation between the feature attention and its expected contribution is constructed according to the inhibition gain coefficient of the feature channel corresponding to each functional area. It can be achieved in the following way, namely: the inhibition gain coefficient is used as a reference value of the expected attention weight of the corresponding functional area. Generally, the functional area with a smaller inhibition gain coefficient should have a lower feature weight. In order to reflect the deviation between the actual attention and the expected contribution, a cost function is constructed, which can be expressed in the common L2 norm method as follows: for the i-th functional area, the cost term is ,in is the current attention of the model, In order to suppress the gain coefficient, the valence function is introduced into the total loss function of the model as a regular term, and participates in the gradient descent optimization process together with the main loss (such as MSE), so as to continuously suppress the attention distribution that does not meet the expectations during model training, and enhance the prediction stability and robustness; the cost function is introduced into the parameter update process of the energy consumption prediction model, and the attention of the energy consumption characteristics of each functional area is iteratively constrained. Suppressing the functional area features with excessive contribution can be achieved in the following way, namely: using a conventional gradient descent algorithm (such as Adam optimizer) during the training of the energy consumption prediction model, backpropagating and updating the weights of the total loss function, and dynamically adjusting the attention value in each iteration so that it tends to be consistent with the suppression gain coefficient, so as to effectively suppress the excessively prominent feature contribution in the functional area, avoid overfitting of the model to some areas, and improve the robustness and generalization ability of the overall prediction.

[0094] It should be noted that the present application scheme effectively avoids the excessive contribution of some functional area features in the model by cost-constraining the attention of energy consumption features in the energy consumption prediction model based on the suppression gain coefficient of all functional areas, thereby improving the generalization ability and prediction stability of the model. Compared with the existing technology in which the feature weights of each area are fixed or adjusted without constraints, this scheme dynamically suppresses abnormally high contribution features, reduces the impact of noise and abnormal fluctuations on the prediction results, thereby improving the overall prediction accuracy and robustness, and meeting the modeling requirements of complex building energy consumption diversity and nonlinear fluctuations.

[0095] It should also be noted that in this application, predicting the building energy consumption in the next energy consumption monitoring cycle based on the energy consumption prediction model after cost constraints means predicting the building energy consumption in the next energy consumption monitoring cycle through the energy consumption prediction model after cost constraints.

[0096] On the other hand, in some embodiments, the present application provides a building energy consumption prediction system, referring to Figure 4 , which is a schematic diagram of the structure of a building energy consumption prediction system according to some embodiments of the present application. The building energy consumption prediction system 400 includes: a collection module 401, a processing module 402 and an execution module 403, which are described as follows:

[0097] The acquisition module 401 in this application is mainly used to collect the overall energy consumption data of the target building through the low-frequency energy consumption monitoring equipment and the local energy consumption data of each functional area of ​​the target building through the high-frequency energy consumption monitoring equipment during the current energy consumption monitoring cycle;

[0098] Processing module 402, in the present application, is used to perform feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data, obtain the associated identification features of the energy consumption between the overall building and each functional area, and then construct an energy consumption prediction model based on all the associated identification features, and determine the suppression gain coefficient of the feature channel corresponding to each functional area in the energy consumption prediction model;

[0099] The processing module 402 in the present application is further configured to predict an abnormal energy consumption period within the next energy consumption monitoring cycle based on the historical energy consumption records of the target building, and then perform a fuzzy evaluation of the energy consumption fluctuation level during the abnormal energy consumption period according to the load fluctuation characteristics of each functional area, to obtain a fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period;

[0100] Execution module 403. In this application, execution module 403 is mainly used to perform cost constraints on the energy consumption characteristics of each functional area in the energy consumption prediction model according to all suppression gain coefficients during the energy consumption prediction process of the next energy consumption monitoring cycle. Then, the energy consumption prediction model after cost constraints is used to predict the building energy consumption of the next energy consumption monitoring cycle.

[0101] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory storing a code, and the processor being configured to obtain the code and execute the above-mentioned building energy consumption prediction method.

[0102] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing a building energy consumption prediction method according to some embodiments of the present application. The building energy consumption prediction method in the above embodiment can be Figure 5 The computer device 500 shown in FIG. 5 is implemented as shown in FIG. 5 . The computer device 500 includes at least one processor 501 , a communication bus 502 , a memory 503 , and at least one communication interface 504 .

[0103] The processor 501 may be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0104] The communication bus 502 may be used to transmit information between the aforementioned components.

[0105] The memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory 503 may be independent and connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0106] Memory 503 is used to store program code for executing the solution of the present application, and is controlled by processor 501 for execution. Processor 501 is used to execute the program code stored in memory 503. The program code may include one or more software modules. The building energy consumption prediction method in the above embodiment can be implemented by processor 501 and one or more software modules in the program code stored in memory 503.

[0107] The communication interface 504 uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0108] In a specific implementation, as an example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0109] The aforementioned computer device can be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of this application do not limit the type of computer device.

[0110] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned building energy consumption prediction method.

[0111] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0112] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for predicting building energy consumption, wherein: Low-frequency energy consumption monitoring equipment is pre-deployed outside the target building and high-frequency energy consumption monitoring equipment is pre-deployed in each functional area inside the target building. The building energy consumption prediction method includes the following steps: During the current energy consumption monitoring cycle, the overall energy consumption data of the target building is collected through low-frequency energy consumption monitoring equipment, and the local energy consumption data of each functional area of ​​the target building is collected through high-frequency energy consumption monitoring equipment; Performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data to obtain association identification features between the overall energy consumption and the energy consumption of each functional area, and then constructing an energy consumption prediction model based on all the association identification features, and determining the suppression gain coefficient of the feature channel corresponding to each functional area in the energy consumption prediction model; Based on the historical energy consumption records of the target building, the energy consumption abnormal period in the next energy consumption monitoring cycle is predicted, and then the energy consumption fluctuation level in the energy consumption abnormal period is fuzzy evaluated according to the load fluctuation characteristics of each functional area to obtain the fuzzy evaluation value of the energy consumption fluctuation level in the energy consumption abnormal period; During the energy consumption forecasting process of the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, the energy consumption characteristic attention of each functional area in the energy consumption forecasting model is cost-constrained according to all the suppression gain coefficients, and then the building energy consumption of the next energy consumption monitoring cycle is predicted based on the cost-constrained energy consumption forecasting model; The cost constraint on the energy consumption feature attention of each functional area in the energy consumption prediction model based on all the suppression gain coefficients specifically includes: Extracting energy consumption feature attention of each functional area in the energy consumption prediction model; A cost function reflecting the deviation between feature attention and its expected contribution is constructed based on the inhibition gain coefficient of the feature channel corresponding to each functional area; The cost function is introduced into the parameter update process of the energy consumption prediction model to iteratively constrain the attention of the energy consumption characteristics of each functional area and suppress the functional area features with excessive contribution.

2. The method according to claim 1, wherein Based on the overall energy consumption data and all local energy consumption data, feature mapping is performed on the overall energy consumption of the target building and the local energy consumption of each functional area to obtain the associated identification features of the energy consumption between the overall building and each functional area, specifically including: Time-aligning the overall energy consumption data and the local energy consumption data of each functional area, and extracting correlation mapping features of energy consumption between the overall and each functional area within the same time window; Determine the dynamic correlation coefficient between the local energy consumption data of each functional area and the overall energy consumption data; The correlation identification features of the overall energy consumption and the energy consumption of each functional area are determined through all dynamic correlation coefficients and all correlation mapping features.

3. The method according to claim 1, wherein The energy consumption prediction model based on all associated identification features includes: Build a prediction model containing multiple feature input channels, where each channel corresponds to a functional area; The associated identification features corresponding to each functional area are respectively used as input parameters of the corresponding channel, and the initial parameters of the prediction model are set; The prediction model is trained by using a supervised training method in combination with the historical energy consumption data of the target building, and the initial parameters are iteratively optimized to obtain the energy consumption prediction model.

4. The method according to claim 1, wherein Determining the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model specifically includes: Extract the deviation distribution characteristics between the energy consumption change rate of each functional area during the historical energy consumption monitoring period and the overall energy consumption change trend; Determining the response strength of the characteristic channel corresponding to each functional area in the energy consumption prediction model based on the deviation distribution characteristics; The initial inhibition factor of the characteristic channel corresponding to each functional area is set according to all response intensities; The initial suppression factor of each characteristic channel is dynamically adjusted through error feedback during the training process of the energy consumption prediction model to obtain the suppression gain coefficient of the characteristic channel corresponding to each functional area in the energy consumption prediction model.

5. The method according to claim 1, wherein Based on the historical energy consumption records of the target building, the prediction of abnormal energy consumption periods in the next energy consumption monitoring cycle specifically includes: Obtain historical overall energy consumption data from the historical energy consumption records of the target building; Periodic feature extraction is performed on the historical overall energy consumption data to identify time periods with sudden energy consumption changes and frequent anomalies, and the energy consumption anomaly period within the next energy consumption monitoring cycle is predicted based on the identified time periods.

6. The method according to claim 1, wherein The energy consumption fluctuation level during the abnormal energy consumption period is fuzzily evaluated according to the load fluctuation characteristics of each functional area, and the fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period is obtained, specifically including: According to the load fluctuation characteristics of each functional area in the historical period, a fuzzy membership function describing energy consumption fluctuation is constructed, and then the fuzzy membership of each functional area in different fluctuation intervals is determined; Perform fuzzy reasoning on the energy consumption fluctuation level of each functional area during the energy consumption abnormality period based on all fuzzy membership degrees to obtain the energy consumption fluctuation credibility of each functional area at different fluctuation levels; The fuzzy evaluation value of the energy consumption fluctuation level during the energy consumption abnormality period is determined through all energy consumption fluctuation credibility.

7. The method according to claim 1, wherein The low-frequency energy consumption monitoring device is specifically a multifunctional power meter.

8. A building energy consumption prediction system, which uses the method according to any one of claims 1 to 7 to predict building energy consumption, characterized in that: The system includes: The collection module is used to collect the overall energy consumption data of the target building through low-frequency energy consumption monitoring equipment and the local energy consumption data of each functional area of ​​the target building through high-frequency energy consumption monitoring equipment during the current energy consumption monitoring cycle; a processing module for performing feature mapping association on the overall energy consumption of the target building and the local energy consumption of each functional area based on the overall energy consumption data and all local energy consumption data, obtaining association identification features between the overall energy consumption and the energy consumption of each functional area, and then constructing an energy consumption prediction model based on all the association identification features, and determining a suppression gain coefficient for a feature channel corresponding to each functional area in the energy consumption prediction model; The processing module is further configured to predict an abnormal energy consumption period within the next energy consumption monitoring cycle based on the historical energy consumption records of the target building, and then perform a fuzzy evaluation of the energy consumption fluctuation level during the abnormal energy consumption period according to the load fluctuation characteristics of each functional area to obtain a fuzzy evaluation value of the energy consumption fluctuation level during the abnormal energy consumption period; The execution module is used to, during the energy consumption prediction process of the next energy consumption monitoring cycle, when the fuzzy evaluation value exceeds the preset energy consumption fluctuation threshold, cost-constrain the energy consumption characteristic attention of each functional area in the energy consumption prediction model according to all the suppression gain coefficients, and then predict the building energy consumption of the next energy consumption monitoring cycle based on the cost-constrained energy consumption prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a code, wherein: The processor is configured to obtain the code and execute the building energy consumption prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the building energy consumption prediction method according to any one of claims 1 to 7 is implemented.

Images