Building energy audit and auxiliary decision-making method and system

Abstract

The application provides a kind of building energy audit and auxiliary decision-making method and system, it is related to the technical field of energy management, the method is by obtaining building static and time sequence operation multi-source data, energy consumption baseline sequence and confidence interval sequence are generated by combining energy consumption prediction model, break through the limitation of traditional energy audit energy consumption baseline static, improve the accuracy of energy efficiency evaluation;By comparing actual energy consumption and confidence interval sequence to identify low energy efficiency working condition period, avoids the subjectivity of traditional artificial qualitative judgment;Quantify the contribution of each input feature to energy efficiency deviation and sort using attribution diagnosis algorithm, realize the quantitative attribution and accurate positioning of energy efficiency anomaly;Based on contribution degree sorting, match the preset energy-saving measures, and calculate energy-saving benefits and economic benefits by combining energy consumption prediction model and energy consumption baseline sequence and output evaluation report, realize the accurate matching and quantitative evaluation of energy-saving measures, provide scientific data support for energy-saving reconstruction investment decision.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and in particular to a method and system for building energy auditing and decision support. Background Technology

[0002] In recent years, with the rapid development of smart park construction, their energy systems have become increasingly complex, and total energy consumption has continued to rise. Conducting systematic energy audits has become a key measure to achieve refined energy management and "dual-carbon" goals in parks. A smart park energy audit is a process of comprehensively monitoring, diagnosing, and evaluating the building energy consumption levels, utilization efficiency, economic benefits, and environmental effects of energy-consuming units within the park. Its core value lies in accurately identifying energy-saving potential and proposing quantifiable and executable differentiated energy-saving optimization solutions, thereby driving the park from passive energy conservation to proactive optimization and achieving a win-win situation for both economic and environmental benefits.

[0003] However, current energy audits in smart parks primarily rely on manual on-site investigations, which suffers from efficiency bottlenecks and data limitations. Diagnostic analysis is mostly based on static threshold comparisons or empirical qualitative judgments, making it difficult to integrate the coupling relationship between building dynamics, environmental changes, and equipment status, resulting in an inability to accurately identify the root causes of energy efficiency anomalies. Furthermore, for proposed energy-saving renovation measures, there is a lack of accurate effect prediction and quantitative assessment capabilities based on historical data and operational characteristics, leading to a lack of scientific basis for investment decisions on energy-saving renovation measures and higher risks. Summary of the Invention

[0004] The purpose of this invention is to provide a building energy audit and auxiliary decision-making method and system to improve the accuracy of energy efficiency assessment, avoid the subjectivity of traditional manual qualitative judgment, realize the quantitative attribution and accurate location of energy efficiency anomalies, and accurately match and quantitatively evaluate energy-saving measures.

[0005] In a first aspect, the present invention provides a building energy audit and auxiliary decision-making method, comprising: acquiring multi-source data of the building to be audited; the multi-source data including: a static building dataset and a time-series operational dataset; generating an energy consumption baseline sequence and its confidence interval sequence of the building to be audited based on the multi-source data and an energy consumption prediction model; the input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time; identifying the low-energy-efficiency operating period of the building to be audited based on the actual energy consumption sequence and confidence interval sequence of the building to be audited; processing the composite vector at all times within the low-energy-efficiency operating period using an attribution diagnosis algorithm to obtain and rank the contribution of each input feature in the composite vector to the energy efficiency deviation; matching preset energy-saving measures based on the contribution ranking results, and calculating energy-saving benefits and economic benefits by combining the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period, and outputting an evaluation report.

[0006] In an optional implementation, the energy consumption prediction model is trained through the following process: acquiring historical multi-source data of efficient operating buildings and cleaning the data to obtain target multi-source data; extracting static and dynamic features of efficient operating buildings from the target multi-source data; concatenating the static features with the dynamic features within a preset sliding window to obtain composite vectors corresponding to multiple consecutive time points; training a deep neural network model with the composite vectors as input and the actual total energy consumption value at the corresponding time point as output to obtain the energy consumption prediction model.

[0007] In an optional implementation, based on multi-source data and an energy consumption prediction model, an energy consumption baseline sequence and its confidence interval sequence for the building to be audited are generated, including: constructing a composite vector corresponding to all times of the building to be audited within a preset audit period based on multi-source data of the building to be audited; traversing the composite vector corresponding to all times within the preset audit period and using the energy consumption prediction model to perform rolling predictions to obtain the energy consumption baseline sequence of the building to be audited within the preset audit period; and calculating the confidence interval sequence corresponding to the energy consumption baseline sequence based on the standard deviation of the prediction error of the energy consumption prediction model and the quantile of the standard normal distribution.

[0008] In an optional implementation, based on the actual energy consumption sequence and confidence interval sequence of the building to be audited, the low-energy-efficiency operating period of the building to be audited is identified, including: if the actual energy consumption value of the building to be audited exceeds the upper limit of the confidence interval sequence for a duration exceeding a preset time threshold, the building to be audited is determined to be in a low-energy-efficiency operating condition; clustering all low-energy-efficiency operating time moments within a preset audit period, and integrating continuous low-energy-efficiency operating time moments into multiple independent low-energy-efficiency operating time periods.

[0009] In an optional implementation, energy efficiency deviation represents the difference between the actual energy consumption value at each moment during the low-energy-efficiency operating period and the corresponding energy consumption baseline value. An attribution diagnosis algorithm is used to process the composite vector at all moments during the low-energy-efficiency operating period to obtain the contribution of each input feature in the composite vector to the energy efficiency deviation. This includes: constructing an abnormal dataset based on the composite vector at all moments during the low-energy-efficiency operating period; extracting samples from the training set of the energy consumption prediction model to construct a background dataset; defining multiple simplified feature vectors and updating the abnormal dataset in combination with the background dataset to obtain a target abnormal dataset; processing the background dataset using the energy consumption prediction model to obtain the model's baseline prediction value; and calculating the average influence strength of each input feature in the composite vector on the energy efficiency deviation based on the energy consumption prediction model, multiple simplified feature vectors, the target abnormal dataset, and the model's baseline prediction value, and using this as the contribution.

[0010] In an optional implementation, preset energy-saving measures are stored in an energy-saving measure library. Each energy-saving measure in the library includes applicable conditions, technical parameters, and investment costs. Matching preset energy-saving measures based on contribution ranking results includes: using a specified number of features ranked first in the contribution ranking results as key input features that cause energy efficiency anomalies; and matching energy-saving measures from the energy-saving measure library that correspond to the applicable conditions and key input features as preset energy-saving measures.

[0011] In an optional implementation, the energy-saving benefits and economic benefits are calculated by combining the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period. This includes: converting the technical parameters corresponding to the preset energy-saving measures into parameter adjustment values ​​of the input features in the composite vector; adjusting the composite vector at all times within the low-energy-efficiency operating period based on the parameter adjustment values ​​to obtain the post-measure composite vector at all times; inputting the post-measure composite vector at all times into the energy consumption prediction model to obtain the simulated energy consumption sequence after the measures; comparing the simulated energy consumption sequence with the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period to calculate the energy saving of the preset energy-saving measures and using it as the energy-saving benefit; and calculating the static investment payback period and net present value of the preset energy-saving measures based on the investment cost, average energy price, discount rate, and lifespan of the measures and using them as the economic benefit.

[0012] Secondly, this invention provides a building energy audit and auxiliary decision-making system, comprising: an acquisition module for acquiring multi-source data of the building to be audited; the multi-source data includes: a static building dataset and a time-series operational dataset; a generation module for generating an energy consumption baseline sequence and its confidence interval sequence of the building to be audited based on the multi-source data and an energy consumption prediction model; the input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time; an identification module for identifying the low-energy-efficiency operating periods of the building to be audited based on the actual energy consumption sequence and confidence interval sequence of the building to be audited; a processing module for processing the composite vector at all times within the low-energy-efficiency operating period using an attribution diagnosis algorithm to obtain and rank the contribution of each input feature in the composite vector to the energy efficiency deviation; and a calculation module for matching preset energy-saving measures based on the contribution ranking results, and calculating energy-saving benefits and economic benefits by combining the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period, and outputting an evaluation report.

[0013] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the building energy audit and auxiliary decision-making method described in any of the foregoing embodiments.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the building energy audit and decision support method described in any of the foregoing embodiments.

[0015] This invention acquires multi-source data on building static and temporal operation, and combines this data with an energy consumption prediction model that uses a composite vector of static and sliding window dynamic features as input to generate energy consumption baseline and confidence interval sequences. This overcomes the limitations of traditional static energy audit baselines, enabling the construction of a dynamic energy efficiency reference system that aligns with actual building operation patterns, thus improving the accuracy of energy efficiency assessments. By comparing actual energy consumption with confidence interval sequences, it identifies periods of low energy efficiency, achieving automated screening of energy efficiency anomalies and avoiding the subjectivity of traditional manual qualitative judgments. Finally, it utilizes an attribution diagnostic algorithm to quantify and rank the contribution of each input feature to energy efficiency deviations. This system solves the problem that traditional auditing cannot trace the root causes of energy efficiency anomalies across systems, enabling quantitative attribution and precise location of energy efficiency anomalies. Based on contribution ranking, it matches preset energy-saving measures and calculates energy-saving and economic benefits by combining energy consumption prediction models and energy consumption baseline sequences, and outputs evaluation reports. This achieves precise matching and quantitative evaluation of energy-saving measures, providing scientific data support for energy-saving renovation investment decisions. It makes up for the lack of precise effect prediction of energy-saving measures in traditional auditing, and improves the automation and precision of energy auditing in smart parks, promoting the transformation of park energy management from passive auditing to proactive optimization. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart of a building energy audit and decision support method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating how to generate an energy consumption baseline sequence and its confidence interval sequence for a building to be audited, based on multi-source data and an energy consumption prediction model, is provided as an embodiment of the present invention. Figure 3 A functional module diagram of a building energy audit and auxiliary decision-making system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0021] Example 1 Figure 1 A flowchart of a building energy audit and decision support method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method specifically includes the following steps: Step S102: Obtain multi-source data of the building to be audited; the multi-source data includes: static building dataset and time-series running dataset.

[0022] At the outset of the implementation of this method, this embodiment of the invention requires the collection of comprehensive data to support energy auditing, namely the aforementioned multi-source data. Data collection can be achieved through API interfaces, IoT gateways, access to BIM (Building Information Modeling) models, and third-party system integration, etc., to access various data within the existing digital platform of the smart park. The multi-source data is divided into two main categories: static building datasets and time-series operational datasets. Static building datasets characterize the inherent physical properties and fundamental system attributes of a building; their data attributes do not change with time or only change slowly over time, serving as a crucial basis for analyzing the basic characteristics of building energy consumption. Time-series operational datasets characterize various monitoring data that dynamically change over time during building operation; they are core data reflecting the actual energy consumption status, operating environment, and equipment working status of the building. Acquiring these two types of datasets lays a complete data foundation for subsequent energy efficiency baseline generation, operating condition identification, and other full-process auditing work.

[0023] Specifically, static building dataset Includes: basic building attributes Thermal parameter matrix of building envelope and equipment ledger database .

[0024] ,in Total building area For air-conditioned area, For the number of floors, Year of construction This refers to the building type.

[0025] ,in , , The heat transfer coefficients of exterior walls, roof, and windows. The solar heat gain coefficient of the exterior window. The ratio of windows to walls.

[0026] The attribute tuple of device h is ,in Rated power, For rated efficiency, For equipment type, This indicates the installation location.

[0027] Time-series running dataset Includes: Total energy consumption time series Energy consumption time series Power generation / heat generation time series Key energy-consuming equipment operating parameter set, indoor and outdoor environmental parameter set, personnel quantity sequence and date types.

[0028] in, This represents the total amount of electricity, gas, and other energy consumed by the building at time t; This represents the energy consumption of the j-th sub-item loop (such as air conditioning system, lighting system, power system, special functional area) at time t; This represents the real-time output power or heating / cooling capacity of the k-th power-producing device (such as a photovoltaic system, a cooling / heating unit, etc.) at time t.

[0029] The set of operating parameters for key energy-consuming equipment includes: The operating status of the chiller at time t: ,in In start / stop state. The performance coefficient characterizing instantaneous energy conversion efficiency, Instantaneous load rate, and The supply and return temperatures for chilled water.

[0030] The water pump's operating status at time t: ,in In start / stop state. For the frequency of the frequency converter, The pressure difference between the inlet and outlet of the water pump. For traffic.

[0031] The fan's operating status at time t: ,in In start / stop state. For the frequency of the frequency converter, For the static pressure of the fan, This refers to air volume.

[0032] Operating status of end-point energy-consuming equipment at time t: It is suitable for lighting circuits, air conditioning units, fan coil units, socket circuits, etc. In start / stop / on / off state, For real-time power consumption, For set values ​​(such as temperature, illuminance).

[0033] The set of indoor and outdoor environmental parameters includes: Indoor environment: ,in The indoor temperature at location p at time t, Let p be the indoor relative humidity at time t. Let p be the indoor carbon dioxide concentration at time t. Let be the indoor illuminance at location p at time t.

[0034] Outdoor environment: ,in Outdoor temperature Outdoor relative humidity, This represents the intensity of solar radiation.

[0035] Personnel Quantity Sequence Indicates the number of people inside the building at time t; date type This indicates the date type corresponding to time t, including season, day of the week, holiday, etc.

[0036] Step S104: Based on multi-source data and energy consumption prediction model, generate the energy consumption baseline sequence and its confidence interval sequence of the building to be audited; the input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time.

[0037] In this embodiment of the invention, the energy consumption prediction model is a pre-trained model for predicting building energy consumption. The input of the model is a composite vector that integrates static features and dynamic features within a preset sliding window. The output is the predicted total energy consumption of the building at the corresponding time. The preset sliding window effectively integrates historical and real-time features of the building, making the model more closely reflect the actual energy consumption patterns of the building. In this step, a feature vector that meets the model input requirements is first constructed based on multi-source data of the building to be audited. This vector is then input into the energy consumption prediction model to predict energy consumption, forming an energy consumption baseline sequence covering the audit period. This sequence represents the energy consumption reference value of the building under ideal conditions. Simultaneously, based on the model's prediction characteristics and statistical regularities, a corresponding confidence interval sequence is matched to this energy consumption baseline sequence. This confidence interval sequence is used to define the fluctuation range of normal building energy consumption, providing a judgment standard for subsequent identification of low-energy-efficiency conditions.

[0038] Step S106: Based on the actual energy consumption sequence and confidence interval sequence of the building to be audited, identify the low-energy-efficiency operating periods of the building to be audited.

[0039] The actual energy consumption sequence of the building to be audited is a statistical sequence of the actual energy consumed by the building at each moment within the audit period, which serves as the basis for reflecting the building's true energy consumption status. The core judgment logic of this step is to compare the energy consumption value at each moment in the actual energy consumption sequence with the upper limit threshold of the corresponding moment in the confidence interval sequence. By setting reasonable judgment rules, the normal energy consumption conditions and low energy efficiency conditions of the building can be distinguished. At the same time, the identified low energy efficiency conditions are integrated and processed to form continuous low energy efficiency period periods, thereby locking in the time range for energy efficiency anomaly diagnosis, making the subsequent attribution analysis more targeted.

[0040] Step S108: Use the attribution diagnosis algorithm to process the composite vector at all times during the low energy efficiency period, obtain the contribution of each input feature in the composite vector to the energy efficiency deviation and sort them.

[0041] In this embodiment of the invention, energy efficiency deviation is the difference between the actual energy consumption of the building to be audited and the energy consumption baseline. It is a core indicator reflecting the degree of abnormality in building energy efficiency. Each input feature in the composite vector represents various factors affecting building energy consumption. The core of this step is to decompose and analyze the feature data during periods of low energy efficiency using an attribution diagnostic algorithm, quantify the impact of each input feature on the energy efficiency deviation, i.e., obtain the contribution of each feature, and then sort all input features according to the magnitude of their contribution. This identifies the key influencing factors leading to abnormal building energy efficiency and provides a direct basis for the accurate matching of subsequent energy-saving measures.

[0042] Step S110: Based on the contribution ranking results, match the preset energy-saving measures, and combine the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low energy efficiency period to calculate the energy-saving benefits and economic benefits, and output the evaluation report.

[0043] Specifically, the preset energy-saving measures are pre-designed energy-saving renovations or optimizations with practical implementation value, targeting various energy efficiency anomalies in buildings. Based on the contribution ranking results, precise matching of energy-saving measures with key influencing factors of energy efficiency anomalies can be achieved, avoiding blind selection of energy-saving measures. After obtaining the corresponding preset energy-saving measures, the building's energy consumption status after the implementation of the energy-saving measures is simulated and predicted using an energy consumption prediction model. Combined with the energy consumption baseline sequence during low-energy-efficiency operating periods, the energy consumption reduction effect that can be achieved after the implementation of the energy-saving measures is quantitatively calculated, i.e., energy-saving benefits. At the same time, the economic benefit indicators of the energy-saving measures are quantitatively calculated by combining factors such as the implementation cost of the energy-saving measures, energy prices, and economic accounting rules. Finally, the audit, diagnosis, and evaluation results of the entire process are integrated to form and output an evaluation report, providing a comprehensive, scientific, and implementable basis for energy optimization decisions in smart park buildings.

[0044] This invention, through acquiring multi-source data on building static and temporal operation, combines this data with an energy consumption prediction model that uses a composite vector of static and sliding window dynamic features as input to generate energy consumption baseline and confidence interval sequences. This overcomes the limitations of traditional static energy audit baselines, enabling the construction of a dynamic energy efficiency reference system that aligns with actual building operation patterns, thus improving the accuracy of energy efficiency assessments. By comparing actual energy consumption with confidence interval sequences, it identifies periods of low-efficiency operation, achieving automated screening of energy efficiency anomalies and avoiding the subjectivity of traditional manual qualitative judgments. Furthermore, it utilizes attribution diagnostic algorithms to quantify the contribution of each input feature to energy efficiency deviations. The ranking system solves the problem that traditional auditing cannot trace the root causes of energy efficiency anomalies across systems, enabling quantitative attribution and precise location of energy efficiency anomalies. Based on contribution ranking, it matches preset energy-saving measures and calculates energy-saving and economic benefits by combining energy consumption prediction models and energy consumption baseline sequences, and outputs evaluation reports. This achieves precise matching and quantitative evaluation of energy-saving measures, providing scientific data support for energy-saving renovation investment decisions. It makes up for the lack of precise effect prediction of energy-saving measures in traditional auditing, and improves the automation and precision of energy auditing in smart parks, promoting the transformation of park energy management from passive auditing to proactive optimization.

[0045] In one alternative implementation, the energy consumption prediction model is trained through the following process: Step S201: Obtain historical multi-source data of the efficiently operating building and perform data cleaning to obtain the target multi-source data.

[0046] In this embodiment of the invention, a high-efficiency operating building refers to a building in the park that has no energy consumption operation problems and meets the energy efficiency standards. Its historical multi-source data includes the building's static data and time-series operation data during its past operation. The data type is consistent with the multi-source data of the building to be audited, which can provide a sample basis that conforms to the actual park operation scenario for model training.

[0047] After acquiring historical multi-source data, data cleaning operations are performed, including identifying and removing outliers generated by sensor acquisition, completing short-term missing values ​​in the data, and uniformly sampling various time-series operational data to the same time granularity and performing spatiotemporal alignment. These cleaning operations eliminate noise and inconsistencies in the data, resulting in target multi-source data whose quality meets the requirements for model training.

[0048] Step S202: Extract the static and dynamic features of the efficient operating building from the target multi-source data.

[0049] Specifically, this embodiment of the invention extracts static features from the static building data of the target multi-source data and dynamic features from the time-series operational data of the target multi-source data. The feature extraction operation specifically involves Z-score standardization of numerical features and one-hot encoding of categorical features. By accurately extracting these two types of features, a comprehensive characterization of the factors influencing the energy consumption of efficiently operating buildings can be achieved.

[0050] Step S203: The static features are concatenated with the dynamic features within the preset sliding window to obtain composite vectors corresponding to multiple consecutive time points.

[0051] For each moment in the historical operation of a high-efficiency building, the extracted static features of the building are fused and concatenated with all dynamic features within a preset sliding window corresponding to that moment, forming a feature vector with unified dimensions, i.e., a composite vector corresponding to that moment. Through the sliding window setting and feature concatenation operation, a single composite vector can simultaneously contain the building's basic attribute features and historical dynamic operation features, enabling the model to learn the building's energy consumption patterns based on more comprehensive feature information. Referring to the above vector construction method, composite vectors corresponding to multiple consecutive historical moments of a high-efficiency building are constructed to form the input feature sample set for model training.

[0052] In this embodiment of the invention, the static dataset of known efficiently running building i is used. Convert to numerical static feature vector And from the known efficient time-series running dataset of building i Extracting key dynamic feature vectors of driving energy consumption For each time point t, a sliding window is used to construct the input feature vector of the model. Includes architectural static features and the past Fusion of dynamic time series features at time steps (e.g., 24 hours). , where ⊕ represents vector concatenation. The corresponding output target (i.e., label) is the actual total energy consumption data at time t. .

[0053] Step S204: Using the composite vector as input and the actual total energy consumption value at the corresponding time as output, train a deep neural network model to obtain an energy consumption prediction model.

[0054] The input feature sample set is divided into training, validation, and test sets according to a preset ratio in chronological order. The numerical features in the input feature sample set have been Z-score standardized, and the categorical features have been one-hot encoded. The standardization parameters are... Calculated only from the training set.

[0055] The training set, validation set, and test set are sequentially input into the deep neural network model for training. During training, a preset loss function is used to measure the deviation between the model's predicted values ​​and the actual total energy consumption. An optimizer is used to continuously adjust the model parameters, and techniques to prevent overfitting and gradient anomalies are introduced. Simultaneously, the model's training effect is monitored in real time using the validation set. After the model converges, the prediction accuracy of the model is evaluated using the test set. When the model's prediction accuracy meets the preset requirements, the training is considered complete, resulting in an energy consumption prediction model that can accurately predict the building's total energy consumption at a corresponding time based on the input composite vector.

[0056] Optionally, the energy consumption prediction model is an LSTM model, where the input layer receives an input tensor of dimension M. LSTM Hidden Layers: 2-3 LSTM layers, each containing 128-256 hidden neurons. Dropout is used between layers to prevent overfitting. Fully Connected Mapping Layers: 1-2 fully connected layers are added after the output of the last LSTM layer to map high-dimensional features to the energy space. L2 regularization is added to the fully connected layers to prevent overfitting. Output Layer: 1 neuron using a linear activation function, outputting the predicted energy value at time t. .

[0057] Model Training: The LSTM training loss function uses mean squared error (MSE) to measure the difference between the predicted and actual values. The Adam optimizer is used with an initial learning rate of 0.001, and gradient clipping is introduced to prevent vanishing or exploding gradients. The loss function is: .

[0058] Model Validation and Evaluation: Evaluate model performance using test set data and calculate root mean square error (RMSE). As an evaluation metric, if the RMSE predicted by the model on the test set is less than 5% of the average total energy consumption, the model is considered successfully trained and can be used in subsequent steps. A successfully trained LSTM model is denoted as . .

[0059] In an optional implementation, step S104 above, as follows: Figure 2 As shown, based on multi-source data and an energy consumption prediction model, the energy consumption baseline sequence and its confidence interval sequence for the building to be audited are generated, specifically including the following steps: Step S1041: Based on the multi-source data of the building to be audited, construct a composite vector corresponding to all times of the building to be audited within the preset audit period.

[0060] Specifically, the preset audit period is the time frame for conducting energy audits on the building to be audited, covering the complete time dimension required for the audit. The composite vector is a feature vector that integrates the building's static features and the dynamic features within the preset sliding window, and is the standard input form for the energy consumption prediction model. In this step, static and dynamic features are first extracted from the multi-source data of the building to be audited, referring to steps S201-S202 above. Then, following the feature fusion rules consistent with those used during the training of the energy consumption prediction model (i.e., step S203), the static features are concatenated with the dynamic features within the preset sliding window corresponding to each time point to form a composite vector for a single time point. Following this rule, feature extraction and fusion operations are performed sequentially for each moment within the preset audit period, ultimately obtaining composite vectors corresponding to all moments within the audit period, ensuring that each moment has feature data that meets the model input requirements.

[0061] Step S1042: Iterate through the composite vectors corresponding to all times within the preset audit period, and use the energy consumption prediction model to perform rolling predictions to obtain the energy consumption baseline sequence of the building to be audited within the preset audit period.

[0062] Specifically, rolling forecasting refers to a method of inputting composite vectors from each moment in chronological order into the model for energy consumption prediction. This approach closely matches the temporal variation characteristics of building energy consumption, enabling continuous forecasting throughout the entire lifecycle. In this embodiment of the invention, the composite vectors corresponding to each moment are input into the energy consumption prediction model sequentially according to the time sequence within a preset audit period. In this process, the model can output the predicted total energy consumption value for each composite vector based on the energy consumption patterns learned during training, expressed as: This predicted value represents the energy consumption baseline of the building under ideal and efficient operation at the corresponding time. Next, the energy consumption baseline values ​​for all times within the preset audit period are integrated chronologically to form a continuous energy consumption data sequence, i.e., the energy consumption baseline sequence of the building under audit. This sequence visually reflects the ideal energy consumption trend of a building when there are no energy consumption anomalies during the audit period.

[0063] Step S1043: Based on the energy consumption baseline sequence, the standard deviation of the prediction error of the energy consumption prediction model, and the standard normal distribution quantile, calculate the confidence interval sequence corresponding to the energy consumption baseline sequence.

[0064] Specifically, the standard deviation of the prediction error of the energy consumption prediction model is a statistical indicator obtained during the testing and verification phase, reflecting the degree of deviation between the model's predicted value and the actual value, and can reflect the prediction accuracy of the model; the standard normal distribution quantile is a fixed value selected according to the preset confidence level, used to determine the upper and lower limits of the confidence interval.

[0065] This invention uses the energy consumption baseline value at each moment in the energy consumption baseline sequence as the center. The product of the prediction error standard deviation and the quantile of the standard normal distribution is added to and subtracted from this baseline value to obtain the upper and lower limits of normal fluctuations in energy consumption at the corresponding moment. These two limits together constitute the energy consumption confidence interval for that moment. Following the above calculation rules, the corresponding confidence intervals can be calculated for each energy consumption baseline value within a preset audit period. These confidence intervals are then integrated in chronological order to form a confidence interval sequence corresponding to the aforementioned energy consumption baseline sequence. This confidence interval sequence provides a clear numerical standard for subsequently determining whether the building's energy consumption at each moment is within the normal range and whether there are low-energy-efficiency operating conditions.

[0066] Optionally, a series of confidence intervals is calculated as the 95% confidence intervals for the dynamic energy efficiency baseline. , , Among them, It is the standard deviation of the prediction error of the energy consumption prediction model on the test set. It is the 97.5th percentile of the standard normal distribution.

[0067] In an optional implementation, step S106 above, based on the actual energy consumption sequence and confidence interval sequence of the building to be audited, identifies the low-energy-efficiency operating periods of the building to be audited, specifically including the following: If the actual energy consumption value of the building to be audited exceeds the upper limit of the confidence interval sequence for a period of time exceeding the preset time threshold, the building to be audited is determined to be in a low energy efficiency condition. All low energy efficiency conditions within the preset audit period are clustered, and the continuous low energy efficiency conditions are integrated into multiple independent low energy efficiency condition periods.

[0068] Specifically, the actual energy consumption sequence of the buildings in the park to be audited will be... Upper bound of the confidence interval sequence of dynamic energy efficiency baseline In contrast, when The duration of consecutive anomalies exceeds a preset time threshold. (For example, 2 hours) is identified as a low-energy-efficiency operating condition. Next, the time points consecutively identified as low-energy-efficiency operating conditions are clustered, thus forming multiple independent low-energy-efficiency operating condition periods. .

[0069] In an optional implementation, the energy efficiency deviation represents the difference between the actual energy consumption value at each moment during the low-energy-efficiency operating period and the corresponding energy consumption baseline value; step S108 above uses an attribution diagnosis algorithm to process the composite vector of all moments during the low-energy-efficiency operating period to obtain the contribution of each input feature in the composite vector to the energy efficiency deviation, specifically including the following steps: Step S1081: Construct an abnormal dataset based on the composite vectors of all times during the low-energy-efficiency operating period, and extract samples from the training set of the energy consumption prediction model to construct a background dataset.

[0070] Specifically, for each low-energy-efficiency operating period identified in step S106 The low-energy-efficiency operating condition dataset is constructed by collecting samples (i.e., composite vectors) from all times within a single low-energy-efficiency operating condition period. In other words, the abnormal dataset. K representative samples are randomly selected from the training set to construct the background dataset. , used to simulate the baseline output of the model.

[0071] Step S1082: Define multiple simplified feature vectors, and update the abnormal dataset in combination with the background dataset to obtain the target abnormal dataset.

[0072] Specifically, in order to attribute the time-series inputs of the energy consumption prediction model, this embodiment of the invention defines a simplified feature vector. , The total dimension of the feature vector is input to the energy consumption prediction model. (Definition) ,in This represents the actual observed value of feature m in the sample during the low-energy-efficiency operating period. 0 indicates that the feature is missing.

[0073] At the same time, define the mapping function. This function is used to simplify the feature vector. To reduce this to the M-dimensional tensor required for the energy consumption prediction model, specifically, for The features of feature m are preserved, and the current observation value of feature m is retained; for The feature with a value of 0 is replaced by the expected value of the corresponding feature in the background dataset. Based on the above feature value rules, the abnormal dataset can be updated to obtain the target abnormal dataset.

[0074] Step S1083: Process the background dataset using the energy consumption prediction model to obtain the model's baseline prediction value.

[0075] Specifically, by using the background dataset matrix Input energy consumption prediction model By averaging these values, we can obtain the baseline prediction value of the model when no feature information is known. ,in, .

[0076] Step S1084: Based on the energy consumption prediction model, multiple simplified feature vectors, target abnormal dataset, and model baseline prediction value, calculate the average influence intensity of each input feature in the composite vector on the energy efficiency deviation, and use it as the contribution.

[0077] Attribution diagnostic algorithms can employ methods such as Kernel SHAP, backpropagation gradient-based attribution algorithms (e.g., DeepSHAP, GradientSHAP), or tree-structure path traversal algorithms (e.g., TreeSHAP). When using the Kernel SHAP method, the contribution of each input feature in the composite vector to the energy efficiency deviation is calculated through the following process.

[0078] For the dataset matrix of low-energy-efficiency operating conditions Each sample in The KernelSHAP interpreter is used to approximate the SHAP value of each input feature m. The specific steps are as follows: Step 1. Random sampling: In the simplified feature vector space D samples were drawn from the middle. Each sample is an M-dimensional vector whose elements are either 0 or 1.

[0079] Step 2. Kernel weight calculation: For each sample Its weight is calculated by the following formula: Based on the above formula, it can be seen that kernel weights tend to give higher weights to feature subsets of moderate size (neither empty nor full). Among them, For the total number of features, For vectors The number of elements that are 1 (i.e. the size of the feature combination). Let be the number of permutations and combinations.

[0080] Step 3. SHAP Value Calculation: Constructing a Linear Function The coefficient vector is solved by weighted least squares. The weights are the kernel weights calculated in step 2. The objective function is solved as follows: The result obtained from the above formula is... That is, feature m in the sample The SHAP value at the corresponding time can be denoted as: .

[0081] For an abnormal time period For all samples within the range, calculate the average SHAP value of each feature. and absolute value average To assess the average direction and intensity of the influence of this feature over the entire time period.

[0082] Among them, the average direction of influence . Indicates feature m during abnormal periods The SHAP value at time t, and the difference between time t and the sample Correspondingly, that is, and Both are expressions of SHAP values.

[0083] This indicates a positive contribution, meaning that the actual value of this characteristic is higher / worse than its "typical value," thereby driving up energy consumption. This indicates a negative contribution, meaning that the feature helps reduce energy consumption, but its effect is insufficient to offset other positive contributing factors.

[0084] Average influence intensity . The larger the value, the greater the influence of the feature.

[0085] In one optional implementation, preset energy-saving measures are stored in an energy-saving measure library, and each energy-saving measure in the library includes applicable conditions, technical parameters, and investment costs.

[0086] Specifically, in this embodiment of the invention, a structured library of energy-saving measures is maintained in advance, with each measure... Includes the following attributes: ID and Description: A unique identifier and a textual description.

[0087] Applicable conditions: The type of problem this measure addresses.

[0088] Technical parameters: Model input features that need to be modified and their value changes for the implementation of the measures.

[0089] Investment cost: includes estimated costs for equipment, installation, labor, etc., denoted as .

[0090] Other attributes include: implementation difficulty, scope of impact, and maintenance requirements.

[0091] In step S110 above, matching preset energy-saving measures based on the contribution ranking results specifically includes the following steps: Step S1101: Select a specified number of features that rank highly in the contribution ranking results as key input features that cause energy efficiency anomalies.

[0092] Specifically, the contribution ranking result is obtained through attribution diagnostic algorithms, ranking the degree of influence of each input feature on energy efficiency deviation. The higher the ranking of a feature, the greater its influence on building energy efficiency anomalies, and the more significant it is as a major contributing factor to low building energy efficiency. Therefore, according to All features are sorted in descending order of their contribution, and the N features with the highest contribution are identified as the main factors causing energy efficiency anomalies, i.e., the key input features. The value of N can be adaptively set according to the actual energy audit requirements and the level of refinement of energy management in the park.

[0093] Step S1102: Match energy-saving measures corresponding to applicable conditions and key input features from the energy-saving measures library, and use them as preset energy-saving measures.

[0094] As described above, the energy-saving measures library is a pre-built structured knowledge base that stores various energy-saving measures that can be implemented. Each energy-saving measure includes core attributes such as applicable conditions, technical parameters, and investment costs. Among them, the applicable conditions clearly define the energy efficiency anomalies and energy consumption influencing factors targeted by each energy-saving measure, which is the core basis for achieving measure matching.

[0095] After identifying the key input features, they are matched one by one with the applicable conditions of each measure in the energy-saving measure library. Energy-saving measures that match the applicable conditions with the key input features are selected. These measures can specifically solve the energy efficiency anomalies caused by the key input features and have practical implementation value and energy-saving effect. They are then identified as the preset energy-saving measures for the energy efficiency anomalies of the building to be audited.

[0096] In an optional implementation, step S110 above, which combines the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period, calculates the energy-saving benefits and economic benefits, specifically includes the following steps: Step S1103: Convert the technical parameters corresponding to the preset energy-saving measures into parameter adjustment values ​​of the input features in the composite vector.

[0097] Specifically, based on the technical requirements of the preset energy-saving measures, the input features in the composite vector that need to be adjusted after the measures are implemented can be obtained. At the same time, the technical parameters are converted into specific adjustment values ​​for the corresponding input features, i.e., parameter adjustment values, so as to achieve a precise match between the technical requirements of the energy-saving measures and the model input features, so that the subsequent model simulation can accurately simulate the building energy consumption status after the measures are implemented.

[0098] Step S1104: Adjust the composite vector at all times during the low-energy-efficiency operating period based on the parameter adjustment value to obtain the composite vector after the measures at all times.

[0099] For each moment within the low-energy-efficiency operating period, the corresponding input feature values ​​in the composite vector at that moment are modified according to the parameter adjustment values ​​obtained in step S1103 above. Input feature values ​​not involved in the adjustment remain unchanged. Through this targeted adjustment, the composite vector at each moment accurately reflects the building's characteristic state after the implementation of energy-saving measures. Finally, the composite vectors after the measures are obtained for all moments within the low-energy-efficiency operating period, providing input data consistent with the implementation scenario for subsequent energy consumption simulations. (For preset energy-saving measures...) After defining the corresponding measures, the input feature vector is: .

[0100] Step S1105: Input the composite vector of measures taken at all times into the energy consumption prediction model to obtain the simulated energy consumption sequence after the measures.

[0101] Specifically, following the temporal sequence of low-energy-efficiency operating conditions, the composite vectors following the measures at each time point are input into the energy consumption prediction model one by one. This outputs the simulated energy consumption values ​​for the corresponding time points, thus obtaining the simulated energy consumption sequence after the implementation of the measures. This sequence reflects the expected energy consumption trend and energy consumption level of a building after implementing preset energy-saving measures during periods of low energy efficiency.

[0102] Step S1106: Compare the simulated energy consumption sequence with the energy consumption baseline sequence corresponding to the low energy efficiency period, calculate the energy saving of the preset energy saving measures, and use it as the energy saving benefit.

[0103] The known energy consumption baseline sequence is the energy consumption reference value of the building under audit in an ideal high-efficiency operating state during a period of low energy efficiency, representing the energy consumption level of the building when there are no energy consumption anomalies during that period; the simulated energy consumption sequence is the expected energy consumption value for that period after implementing energy-saving measures, and the difference between the two is the amount of energy consumption reduction that can be achieved after the implementation of energy-saving measures. The following formula is used to calculate the energy savings in this embodiment of the invention: This indicates that under the same external conditions, due to measures The resulting reduction in energy consumption. This energy saving, as an energy-saving benefit, directly demonstrates the actual effect of the pre-designed energy-saving measures in reducing building energy consumption and improving energy utilization efficiency.

[0104] Step S1107: Based on the investment cost, average energy price, discount rate, and lifespan of the preset energy-saving measures, calculate the static payback period and net present value of the preset energy-saving measures, and use them as economic benefits.

[0105] Specifically, the investment cost of the pre-set energy-saving measures includes all costs such as equipment procurement, installation and construction, and labor required for implementation. The average energy price is the average unit price of various types of energy consumed by the building. The discount rate is an accounting indicator that takes into account the time value of money. The lifespan of the measures is the effective usage time during which the pre-set energy-saving measures can achieve their energy-saving effects.

[0106] Specifically, the embodiments of the present invention use the following formulas to calculate the measures respectively. static investment payback period and net present value The static investment payback period and net present value are used together as economic benefit indicators of the pre-set energy-saving measures. The economic feasibility of the energy-saving measures is evaluated from two dimensions: short-term recovery efficiency and long-term benefit level.

[0107] .

[0108] ,in measures Average annual energy savings For average energy price, The discount rate is... The lifespan of the measure.

[0109] In summary, the building energy audit and decision support method provided by the embodiments of the present invention has the following advantages.

[0110] 1. Improved dynamism and accuracy of energy audit baselines: Traditional energy audits often use fixed values ​​or simple regression as energy efficiency baselines, which are difficult to adapt to fluctuations in building performance under different weather conditions, personnel, and operating conditions. This invention, by accessing multi-source heterogeneous datasets and constructing a deep learning model, can effectively capture the time-lag characteristics of building thermal inertia and energy consumption. Compared to traditional static baselines, the dynamic energy efficiency baseline established by this invention better matches actual operating patterns, significantly reduces errors in energy efficiency assessment, and achieves greater precision in auditing.

[0111] 2. Achieved deep quantitative attribution and precise localization of energy efficiency anomalies: This invention overcomes the limitations of traditional energy audits, which can only detect anomalies but struggle to explain their causes. Through attribution diagnostic algorithms, the complex neural network "black box" model output is decomposed into the marginal contribution of each physical characteristic (such as set temperature, environmental parameters, and equipment status). This processing step quantifies the direction and intensity of each factor's influence on energy consumption deviations, helping managers quickly locate the root cause of low energy efficiency among complex, multi-dimensional influencing factors, thus improving the automated diagnostic capabilities of energy audits.

[0112] 3. A decision verification closed loop based on "digital twins" is constructed: This embodiment of the invention not only provides diagnostic suggestions but also realizes a visualized pre-simulation based on a pre-trained energy consumption prediction model. Before implementing improvement measures, the expected energy savings and return on investment of various energy-saving measures can be quantitatively evaluated by modifying the model input parameters. This model-driven simulation method avoids the risks of blindly modifying existing systems and ensures the scientific nature and economic benefits of optimization decisions.

[0113] 4. It promotes the transformation of energy management from "passive auditing" to "proactive optimization": Based on the embodiments of this invention, a closed-loop system covering the entire process of "data perception - intelligent diagnosis - optimization decision-making - continuous verification" can be constructed. Relying on the deep integration of IoT sensing and artificial intelligence algorithms, the system can continuously monitor and identify low-energy-efficiency operating conditions and automatically match energy-saving measures from the library. This automated closed-loop management model reduces reliance on human experience and significantly improves the intelligence level and continuous operational efficiency of energy management in the park.

[0114] Example 2 This invention also provides a building energy audit and decision support system. This system is mainly used to execute the building energy audit and decision support method provided in Embodiment 1 above. The system provided in this invention will be described in detail below.

[0115] Figure 3 A functional block diagram of a building energy audit and auxiliary decision-making system provided in an embodiment of the present invention is shown below. Figure 3 As shown, the system mainly includes: an acquisition module 10, a generation module 20, a recognition module 30, a processing module 40, and a calculation module 50, wherein: The acquisition module 10 is used to acquire multi-source data of the building to be audited; the multi-source data includes: static building dataset and time-series running dataset.

[0116] The generation module 20 is used to generate the energy consumption baseline sequence and its confidence interval sequence of the building to be audited based on multi-source data and energy consumption prediction model. The input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time.

[0117] The identification module 30 is used to identify the low-energy-efficiency operating periods of the building under audit based on the actual energy consumption sequence and confidence interval sequence of the building under audit.

[0118] The processing module 40 is used to process the composite vector at all times during the low-energy-efficiency operating period using the attribution diagnosis algorithm, and obtain and sort the contribution of each input feature in the composite vector to the energy efficiency deviation.

[0119] The calculation module 50 is used to match preset energy-saving measures based on the contribution ranking results, and combine the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low energy efficiency period to calculate the energy-saving benefits and economic benefits, and output an evaluation report.

[0120] This invention, through acquiring multi-source data on building static and temporal operation, combines this data with an energy consumption prediction model that uses a composite vector of static and sliding window dynamic features as input to generate energy consumption baseline and confidence interval sequences. This overcomes the limitations of traditional static energy audit baselines, enabling the construction of a dynamic energy efficiency reference system that aligns with actual building operation patterns, thus improving the accuracy of energy efficiency assessments. By comparing actual energy consumption with confidence interval sequences, it identifies periods of low-efficiency operation, achieving automated screening of energy efficiency anomalies and avoiding the subjectivity of traditional manual qualitative judgments. Furthermore, it utilizes attribution diagnostic algorithms to quantify the contribution of each input feature to energy efficiency deviations. The ranking system solves the problem that traditional auditing cannot trace the root causes of energy efficiency anomalies across systems, enabling quantitative attribution and precise location of energy efficiency anomalies. Based on contribution ranking, it matches preset energy-saving measures and calculates energy-saving and economic benefits by combining energy consumption prediction models and energy consumption baseline sequences, and outputs evaluation reports. This achieves precise matching and quantitative evaluation of energy-saving measures, providing scientific data support for energy-saving renovation investment decisions. It makes up for the lack of precise effect prediction of energy-saving measures in traditional auditing, and improves the automation and precision of energy auditing in smart parks, promoting the transformation of park energy management from passive auditing to proactive optimization.

[0121] Optionally, the system is also used for: Acquire historical multi-source data of buildings that are operating efficiently, and perform data cleaning to obtain target multi-source data.

[0122] Extract static and dynamic features of efficiently operating buildings from multi-source target data.

[0123] Static features are concatenated with dynamic features within a preset sliding window to obtain composite vectors corresponding to multiple consecutive time points.

[0124] Using a composite vector as input and the actual total energy consumption at the corresponding time point as output, a deep neural network model is trained to obtain an energy consumption prediction model.

[0125] Optionally, the generation module 20 is specifically used for: Based on the multi-source data of the building to be audited, a composite vector is constructed corresponding to the building at all times within the preset audit period.

[0126] The composite vectors corresponding to all times within the preset audit period are traversed, and the energy consumption prediction model is used to make rolling predictions to obtain the energy consumption baseline sequence of the building to be audited within the preset audit period.

[0127] Based on the energy consumption baseline sequence, the standard deviation of the prediction error of the energy consumption prediction model, and the quantile of the standard normal distribution, the confidence interval sequence corresponding to the energy consumption baseline sequence is calculated.

[0128] Optionally, the identification module 30 is specifically used for: If the actual energy consumption value of the building to be audited exceeds the upper limit of the confidence interval sequence for a period of time exceeding a preset time threshold, the building to be audited is determined to be in a low energy efficiency condition.

[0129] Cluster all low-energy-efficiency operating conditions within the preset audit period, and integrate consecutive low-energy-efficiency operating conditions into multiple independent low-energy-efficiency operating condition periods.

[0130] Optionally, the energy efficiency deviation represents the difference between the actual energy consumption value at each moment during the low-energy-efficiency operating period and the corresponding energy consumption baseline value; the processing module 40 is specifically used for: An abnormal dataset is constructed based on composite vectors from all times during periods of low energy efficiency, and a background dataset is constructed by extracting samples from the training set of the energy consumption prediction model.

[0131] Define multiple simplified feature vectors, and update the anomaly dataset in combination with the background dataset to obtain the target anomaly dataset.

[0132] The background dataset was processed using an energy consumption prediction model to obtain the model's baseline prediction values.

[0133] Based on the energy consumption prediction model, multiple simplified feature vectors, target abnormal dataset, and model baseline prediction values, the average influence intensity of each input feature in the composite vector on the energy efficiency deviation is calculated and used as the contribution.

[0134] Optionally, preset energy-saving measures are stored in an energy-saving measure library, and each energy-saving measure in the library includes applicable conditions, technical parameters, and investment costs; the calculation module 50 includes: The determination unit is used to select a specified number of features that rank highly in the contribution ranking results as key input features that lead to energy efficiency anomalies.

[0135] The matching unit is used to match energy-saving measures corresponding to applicable conditions and key input features from the energy-saving measures library, and use them as preset energy-saving measures.

[0136] Optionally, the computing module 50 also includes: The conversion unit is used to convert the technical parameters corresponding to the preset energy-saving measures into parameter adjustment values ​​of the input features in the composite vector.

[0137] The adjustment unit is used to adjust the composite vector at all times during the low-energy-efficiency operating period based on the parameter adjustment value, so as to obtain the post-measure composite vector at all times.

[0138] The processing unit is used to input the post-measure composite vector of all time points into the energy consumption prediction model to obtain the simulated energy consumption sequence after the measures.

[0139] The first calculation unit is used to compare the simulated energy consumption sequence with the energy consumption baseline sequence corresponding to the low energy efficiency period, calculate the energy saving of the preset energy saving measures, and use it as the energy saving benefit.

[0140] The second calculation unit is used to calculate the static payback period and net present value of the preset energy-saving measures based on the investment cost, average energy price, discount rate, and lifespan of the measures, and to use these as economic benefits.

[0141] Example 3 See Figure 4 This invention provides an electronic device, which includes a processor 60, a memory 61, a bus 62, and a communication interface 63. The processor 60, the communication interface 63, and the memory 61 are connected via the bus 62. The processor 60 is used to execute executable modules, such as computer programs, stored in the memory 61.

[0142] The memory 61 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 63 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0143] Bus 62 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0144] The memory 61 is used to store programs. After receiving an execution instruction, the processor 60 executes the program. The method executed by the apparatus defined by the process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 60 or implemented by the processor 60.

[0145] Processor 60 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 60 or by instructions in software form. Processor 60 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 61. Processor 60 reads the information in memory 61 and, in conjunction with its hardware, completes the steps of the above method.

[0146] The computer program product of the building energy audit and auxiliary decision-making method and system provided in the embodiments of the present invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0147] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0148] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0149] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0150] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," "third," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0151] Furthermore, terms such as "horizontal," "vertical," and "sag" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0152] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for building energy auditing and decision support, characterized in that, include: Obtain multi-source data for the building to be audited; The multi-source data includes: a static building dataset and a time-series runtime dataset; Based on the multi-source data and energy consumption prediction model, the energy consumption baseline sequence and its confidence interval sequence of the building to be audited are generated; the input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time. Based on the actual energy consumption sequence of the building to be audited and the confidence interval sequence, identify the low-energy-efficiency operating periods of the building to be audited; The attribution diagnosis algorithm is used to process the composite vector at all times during the low energy efficiency period to obtain and sort the contribution of each input feature in the composite vector to the energy efficiency deviation. Based on the contribution ranking results, preset energy-saving measures are matched, and combined with the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low energy efficiency period, the energy-saving benefits and economic benefits are calculated, and an evaluation report is output.

2. The building energy audit and decision support method according to claim 1, characterized in that, The energy consumption prediction model is trained through the following process: Acquire historical multi-source data of efficiently operating buildings and perform data cleaning to obtain target multi-source data; Extract the static and dynamic features of the efficient operating building from the target multi-source data; The static features are concatenated with the dynamic features within the preset sliding window to obtain composite vectors corresponding to multiple consecutive time points. Using the composite vector as input and the actual total energy consumption value at the corresponding time as output, a deep neural network model is trained to obtain the energy consumption prediction model.

3. The building energy audit and decision support method according to claim 1, characterized in that, Based on the multi-source data and energy consumption prediction model, an energy consumption baseline sequence and its confidence interval sequence for the building to be audited are generated, including: Based on the multi-source data of the building to be audited, a composite vector corresponding to the building at all times within the preset audit period is constructed. The composite vector corresponding to all times within the preset audit period is traversed, and the energy consumption prediction model is used to perform rolling predictions to obtain the energy consumption baseline sequence of the building to be audited within the preset audit period. Based on the energy consumption baseline sequence, the standard deviation of the prediction error of the energy consumption prediction model, and the quantile of the standard normal distribution, the confidence interval sequence corresponding to the energy consumption baseline sequence is calculated.

4. The building energy audit and decision support method according to claim 1, characterized in that, Based on the actual energy consumption sequence of the building to be audited and the confidence interval sequence, identify the low-energy-efficiency operating periods of the building to be audited, including: If the actual energy consumption value of the building to be audited exceeds the upper bound of the confidence interval sequence for a period of time exceeding a preset time threshold, the building to be audited is determined to be in a low energy efficiency condition. Cluster all low-energy-efficiency operating conditions within the preset audit period, and integrate consecutive low-energy-efficiency operating conditions into multiple independent low-energy-efficiency operating condition periods.

5. The building energy audit and decision support method according to claim 4, characterized in that, The energy efficiency deviation represents the difference between the actual energy consumption value at each moment during the low energy efficiency operating period and the corresponding energy consumption baseline value. The attribution diagnostic algorithm is used to process the composite vector at all times during the low-energy-efficiency operating period to obtain the contribution of each input feature in the composite vector to the energy efficiency deviation, including: An abnormal dataset is constructed based on the composite vectors of all times during the low-energy-efficiency operating period, and a background dataset is constructed by extracting samples from the training set of the energy consumption prediction model. Define multiple simplified feature vectors, and update the anomaly dataset in combination with the background dataset to obtain the target anomaly dataset; The background dataset is processed using the energy consumption prediction model to obtain the model's baseline prediction value; Based on the energy consumption prediction model, the multiple simplified feature vectors, the target abnormal dataset, and the model baseline prediction value, the average influence intensity of each input feature in the composite vector on the energy efficiency deviation is calculated and used as the contribution.

6. The building energy audit and decision support method according to claim 1, characterized in that, The preset energy-saving measures are stored in an energy-saving measure library, and each energy-saving measure in the library includes applicable conditions, technical parameters, and investment costs. Pre-set energy-saving measures are matched based on the contribution ranking results, including: A specified number of features ranked at the top of the contribution ranking results will be used as key input features that lead to energy efficiency anomalies. The energy-saving measures corresponding to the key input features and applicable conditions are matched from the energy-saving measures library and used as the preset energy-saving measures.

7. The building energy audit and decision support method according to claim 6, characterized in that, By combining the energy consumption prediction model and the energy consumption baseline sequence corresponding to low-energy-efficiency operating periods, energy-saving benefits and economic benefits are calculated, including: The technical parameters corresponding to the preset energy-saving measures are converted into parameter adjustment values ​​of the input features in the composite vector; Based on the parameter adjustment value, the composite vector at all times during the low-energy-efficiency operating period is adjusted to obtain the composite vector after the measures at all times. Input the composite vector of measures taken at all times into the energy consumption prediction model to obtain the simulated energy consumption sequence after the measures are taken; By comparing the simulated energy consumption sequence with the energy consumption baseline sequence corresponding to the low-energy-efficiency operating period, the energy saving of the preset energy-saving measures is calculated and used as the energy saving benefit; Based on the investment cost, average energy price, discount rate, and lifespan of the preset energy-saving measures, the static payback period and net present value of the preset energy-saving measures are calculated and used as economic benefits.

8. A building energy audit and decision support system, characterized in that, include: The acquisition module is used to acquire multi-source data of the building to be audited; The multi-source data includes: a static building dataset and a time-series runtime dataset; The generation module is used to generate the energy consumption baseline sequence and its confidence interval sequence of the building to be audited based on the multi-source data and the energy consumption prediction model; the input of the energy consumption prediction model is a composite vector that integrates static features and dynamic features within a preset sliding window, and the output is the total energy consumption prediction value at the corresponding time. The identification module is used to identify the low-energy-efficiency operating periods of the building under audit based on the actual energy consumption sequence of the building under audit and the confidence interval sequence; The processing module is used to process the composite vector at all times during the low-energy-efficiency operating period using the attribution diagnosis algorithm, and obtain and sort the contribution of each input feature in the composite vector to the energy efficiency deviation. The calculation module is used to match preset energy-saving measures based on the contribution ranking results, and combine the energy consumption prediction model and the energy consumption baseline sequence corresponding to the low energy efficiency period to calculate the energy-saving benefits and economic benefits, and output an evaluation report.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the building energy audit and auxiliary decision-making method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the building energy audit and decision support method as described in any one of claims 1 to 7.

Images