An AI-based method for predicting building energy consumption

Abstract

This invention belongs to the field of building energy consumption prediction, specifically involving an artificial intelligence-based method for predicting building energy consumption. The method includes collecting historical energy consumption data, indoor and outdoor weather data, and usage data within the building; employing adaptive interval updates to achieve data normalization; extracting key features using mutual information methods; constructing a prediction model by optimizing a multilayer perceptron (MLP) through a genetic algorithm; and using the model to predict energy consumption based on real-time data and displaying the results. This method comprehensively collects data, improves data quality and prediction efficiency, optimizes model accuracy and generalization ability, provides intuitive decision support for building managers, and achieves energy conservation and consumption reduction.

Description

Technical Field

[0001] This invention belongs to the field of building energy consumption prediction and relates to a building energy consumption prediction method based on artificial intelligence. Background Technology

[0002] In today's society, building energy consumption accounts for a significant proportion of global energy consumption, making energy conservation and emission reduction in buildings a top priority. Accurate prediction of building energy consumption is crucial for optimizing building energy management and developing effective energy-saving strategies. However, existing building energy consumption prediction methods often fail to fully consider the combined effects of numerous complex factors within a building, resulting in low prediction accuracy.

[0003] With the rapid development of IoT technology and the widespread application of sensors, we can comprehensively collect historical energy consumption data within buildings, including electricity consumption and water resource consumption. Simultaneously, we can integrate environmental monitoring systems to acquire indoor and outdoor meteorological data, as well as building usage data, such as occupancy, equipment operating status, and building usage time—information spanning multiple dimensions. This rich data provides a foundation for more accurate building energy consumption prediction. Furthermore, in terms of data processing, traditional data normalization methods may not be able to flexibly adapt to the distribution and trends of the data, thus affecting the performance of the prediction model. Effective feature extraction, however, can filter out key features highly correlated with building energy consumption, improving prediction efficiency.

[0004] Therefore, there is an urgent need for an innovative building energy consumption prediction method based on artificial intelligence to make full use of this data, overcome the shortcomings of traditional methods, achieve more accurate energy consumption prediction, provide strong support for building energy management, and thus promote the development of the building industry towards a more energy-efficient and environmentally friendly direction. Summary of the Invention

[0005] This invention addresses the aforementioned technical problems related to building energy consumption by proposing an innovative building energy consumption prediction method based on artificial intelligence.

[0006] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0007] S1. First, use IoT devices and sensors to comprehensively collect historical energy consumption data within the building, and at the same time integrate an environmental monitoring system to collect indoor and outdoor meteorological data, as well as usage data.

[0008] S2. Secondly, adaptive interval update is adopted, which automatically adjusts the normalization interval to achieve data normalization processing based on the distribution and trend of the data.

[0009] S3. Use the mutual information method to extract key features from the preprocessed data. Mutual information measures the correlation between two variables. Calculate the mutual information between each feature and the target variable, and select features with larger mutual information because these features are more correlated with the target variable, which helps to improve prediction efficiency.

[0010] S4. Based on the selected features, a genetic algorithm is used to optimize the multilayer perceptron (MLP) to construct a building energy consumption prediction model. The model is trained through cross-validation to ensure its accuracy and generalization ability.

[0011] S5. Using the trained model, predict energy consumption based on real-time environmental data and building usage, generate energy consumption trends for future time periods, and display the prediction results and influencing factors through visualization tools to provide intuitive decision support for building managers.

[0012] Preferably, step S2 employs adaptive interval updating, which automatically adjusts the normalization interval based on the data distribution and trend to achieve data normalization. The specific implementation process is as follows:

[0013] S21. First, collect initial data samples, denoted as D0 = {X1, X2, ... X}. m}, where m is the initial sample size, calculate the minimum value of the initial data. and maximum value As the initial normalization interval;

[0014] S22. Then, update the data, adding one data point X each time new data arrives. new Calculate the mean of the current dataset. Based on the distribution and trend of the data, an adjustment coefficient α is determined. When the standard deviation is large, it indicates that the data fluctuates greatly, and α can be taken to a larger value to expand the normalization interval; when the standard deviation is small, α is taken to a smaller value to narrow the interval and improve accuracy.

[0015] S23. Next, update the maximum and minimum values ​​of the interval, and the new minimum value. If X new <X min Then update X min =X new The new maximum value If X new >X max Then update X max =X new ;

[0016] S24. Finally, for the new data point X new Normalization is performed using the updated normalization interval.

[0017] Preferably, the mutual information calculation method in step S3 is as follows: Where X and Y are two features, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y, respectively. The six features with the largest mutual information values ​​are selected as key features.

[0018] Preferably, the specific operation of using a genetic algorithm to optimize the multilayer perceptron (MLP) to construct the building energy consumption prediction model in step S4 is as follows:

[0019] S41. First, during population initialization, each individual is represented using a hierarchical encoding mechanism. The encoding of each individual in the population includes the number of hidden layers in the MLP model structure, the number of neurons in each layer, and the hyperparameters of the model. The encoding is as follows: Individual = [L1, L2, ..., L...]. n [,η,activation], where L1,L2,...,L n This represents the number of neurons in each layer, where η is the learning rate and activation is the activation function.

[0020] S42. Next, train the MLP model corresponding to each individual and calculate its fitness based on the building energy consumption prediction task. The fitness function is Fitness = α × RMSE + β × Efficiency, where RMSE is the root mean square error, Efficiency is the improvement in building energy efficiency, which is used to measure the actual improvement effect of the model on building energy consumption management, and α and β are weight coefficients used to balance the weight of accuracy and energy efficiency improvement.

[0021] S43. Next, in the crossover process of the genetic algorithm, multi-scale crossover is adopted, and operations are performed at both the model structure layer and the hyperparameter layer. In the mutation operation, individuals are allowed to borrow some weights or parameters from high-fitness individuals to accelerate convergence. For a certain high-fitness individual, some weights W are randomly selected to replace W. new =W mutated +γ(W best -W mutated ), where γ is the rate of variation, W new This is the latest weight, W best It is the weight of highly fit individuals, W mutated This represents the weight after the mutation;

[0022] S44. Finally, as the genetic algorithm evolves, the crossover rate and mutation rate are dynamically adjusted. The crossover rate is CrossoverRate = Initial Crossover Rate × exp(-k × Generation Number), where Initial Crossover Rate is the initial crossover rate, k is the decay coefficient of the crossover rate, and Generation Number is the current generation. The mutation rate is Mutation Rate = Initial Mutation Rate × exp(-m MR ×Generation Number), where Initial Mutation Rate is the initial mutation rate, m MR It is the decay coefficient of the mutation rate, and the Generation Number is the current generation.

[0023] Preferably, the specific operation of generating and displaying the energy consumption trend in the future time period in step S5 is as follows:

[0024] S51. First, the processed real-time data is input into the previously trained energy consumption prediction model to predict energy consumption for a future time period of h. The model generates the predicted energy consumption value for the future time t+h. in X is the predicted energy consumption at time t+h, W is the trained weights of the model, and X is the energy consumption value at time t+h. t It is the input feature vector at time t;

[0025] S52. Next, trend analysis is performed on the prediction results to generate the energy consumption trend over the future time period. Combined with the real-time input data, sensitivity analysis is used to evaluate the impact of each input feature on energy consumption and calculate the impact factor of each feature. Among them I j This represents the sensitivity of the j-th feature to the energy consumption prediction result;

[0026] S53. Finally, the predicted energy consumption trend and influencing factors are visualized using visualization tools, combining historical energy consumption data and future forecasts, to plot the energy consumption change curve over time.

[0027] Compared with existing technologies, the advantages and positive effects of this invention are that it can comprehensively collect multi-dimensional data, improve data quality through adaptive normalization processing, improve prediction efficiency by extracting key features through mutual information, optimize MLP with genetic algorithm to make the model more accurate and have strong generalization ability, and provide intuitive support for decision-making through visualization, thus effectively achieving energy conservation and consumption reduction in buildings. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram showing the predicted and actual values ​​of shopping mall electricity consumption over six months as predicted by this invention.

[0030] Figure 2 This is a diagram illustrating the predicted and actual values ​​of office building electricity consumption over the next six months as predicted by this invention. Detailed Implementation

[0031] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below with reference to embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0032] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0033] Example 1: With increasing global attention to energy consumption and environmental protection, effective management of building energy consumption has become crucial. In commercial buildings, energy consumption accounts for a significant proportion of costs and also has a substantial impact on the environment. Accurate prediction of building energy consumption can help managers formulate reasonable energy-saving strategies, optimize energy use efficiency, reduce operating costs, and achieve sustainable development goals. This invention proposes an artificial intelligence-based building energy consumption prediction method, which analyzes and predicts building energy consumption from multiple dimensions using artificial intelligence. To comprehensively collect historical energy consumption data and related environmental and usage data within the building, a comprehensive data collection scheme is implemented using IoT devices and sensors. Specifically, this includes installing IoT devices such as smart meters and water meters to acquire electricity consumption data and water resource consumption data; deploying indoor and outdoor temperature and humidity sensors to integrate an environmental monitoring system to collect indoor and outdoor meteorological data; and using personnel counters, equipment monitoring sensors, etc., to acquire personnel flow data, equipment operating status data, and building usage time data. Through the collection of this data, a rich data foundation can be provided for subsequent energy consumption prediction, enabling the prediction results to more accurately reflect the actual energy consumption of the building.

[0034] Secondly, to address the issue that traditional data normalization methods cannot flexibly adapt to data distribution and trends, an adaptive interval update scheme is adopted. First, initial data samples are collected, denoted as D0 = {X1, X2, ... X...}.m}, where m is the initial sample size, calculate the minimum value of the initial data. and maximum value This serves as the initial normalization interval; then, data updates are performed, adding one data point X each time new data arrives. new Calculate the mean of the current dataset. Based on the standard deviation σ, an adjustment coefficient α is determined according to the data distribution and trend. When the standard deviation is large, it indicates that the data fluctuates greatly, and α can be taken to a larger value to expand the normalization interval; when the standard deviation is small, α is taken to a smaller value to narrow the interval and improve accuracy. Then, the maximum and minimum values ​​of the interval are updated, and the new minimum value is... If X new <X min Then update X min =X new The new maximum value If X new >X max Then update X max =X new Finally, for the new data point X new Normalization is performed using the updated normalization interval. This allows the data normalization to better reflect the actual data distribution, improving the stability and accuracy of the model.

[0035] To improve prediction efficiency, a mutual information method is used to extract key features from the preprocessed data. The mutual information calculation method is as follows: Here, X and Y are two features, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y, respectively. The six features with the highest mutual information values ​​are selected as key features. In this way, features with high correlation to building energy consumption can be screened out, reducing the interference of irrelevant features and improving the training efficiency and prediction accuracy of the model.

[0036] To construct a building energy consumption prediction model with high accuracy and generalization ability, a genetic algorithm is used to optimize the Model-Level Processing (MLP). This method can automatically search for the optimal model structure and parameters, improving model accuracy and generalization ability, and adapting to the energy consumption prediction needs of different buildings. First, during population initialization, each individual is represented using a hierarchical encoding mechanism. The encoding of each individual in the population includes the number of hidden layers in the MLP model structure, the number of neurons in each layer, and the model's hyperparameters, as follows: Individual=[L1,L2,...,L... n [,η,activation], where L1,L2,...,L nη represents the number of neurons in each layer, η is the learning rate, and activation is the activation function. Next, the MLP model corresponding to each individual is trained, and its fitness is calculated based on the building energy consumption prediction task. The fitness function is Fitness = α × RMSE + β × Efficiency, where RMSE is the root mean square error, Efficiency is the improvement in building energy efficiency, used to measure the actual improvement effect of the model on building energy consumption management, and α and β are weight coefficients used to balance the weights of accuracy and energy efficiency improvement. Then, in the crossover process of the genetic algorithm, multi-scale crossover is used, operating at both the model structure layer and the hyperparameter layer. In the mutation operation, individuals are allowed to borrow some weights or parameters from high-fit individuals to accelerate convergence. For a high-fit individual, a portion of the weights W is randomly selected and replaced. new =W mutated +γ(W best -W mutated ), where γ is the rate of variation, W new This is the latest weight, W best It is the weight of highly fit individuals, W mutated This represents the weights after mutation; finally, as the genetic algorithm evolves, the crossover rate and mutation rate are dynamically adjusted. The crossover rate is calculated as follows: Crossover Rate = Initial Crossover Rate × exp(-k × GenerationNumber), where Initial Crossover Rate is the initial crossover rate, k is the decay coefficient of the crossover rate, and GenerationNumber is the current generation number; the mutation rate is calculated as follows: Mutation Rate = Initial Mutation Rate × exp(-m MR ×Generation Number), where Initial Mutation Rate is the initial mutation rate, m MR It is the decay coefficient of the mutation rate, and the Generation Number is the current generation.

[0037] Finally, considering the need to provide intuitive decision support for building managers, this invention utilizes a trained model to predict energy consumption and displays the prediction results and influencing factors. First, processed real-time data is input into the previously trained energy consumption prediction model to predict energy consumption over a future time period of h. The model then generates a predicted energy consumption value for the next time point t+h. in X is the predicted energy consumption at time t+h, W is the trained weights of the model, and X is the energy consumption value at time t+h. tFirst, the input feature vector at time t is used. Second, trend analysis is performed on the prediction results to generate the energy consumption trend over the future time period. Combined with the real-time input data, sensitivity analysis is used to evaluate the impact of each input feature on energy consumption, and the influence factor of each feature is calculated. Where I j The j-th feature represents the sensitivity of the energy consumption prediction results. Finally, the predicted energy consumption trend and influencing factors are visualized using visualization tools, combining historical energy consumption data and future predictions to plot the energy consumption change curve over time. This allows building managers to clearly understand the future trends and influencing factors of building energy consumption, thereby making more effective energy-saving decisions.

[0038] To better demonstrate the effectiveness of this invention, energy consumption data from two different types of commercial buildings are selected as examples to verify the applicability and accuracy of this invention in different building types. (The following is a summary of the previous paragraph.) Figure 1 and Figure 2 The image shows line graphs comparing predicted and actual electricity consumption data for office buildings and shopping malls over a six-month period. The graphs demonstrate a high degree of fit between the predicted and actual electricity consumption values ​​for these two different types of commercial buildings. Over the six-month period, the trend of the predicted values ​​largely matches the trend of the actual values, and the values ​​are also quite close. This indicates that the invention has good applicability to different building types and can effectively address the characteristics and energy consumption patterns of various buildings.

[0039] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A building energy consumption prediction method based on artificial intelligence, characterized in that, Includes the following steps: S1. First, use IoT devices and sensors to comprehensively collect historical energy consumption data within the building, and at the same time integrate an environmental monitoring system to collect indoor and outdoor meteorological data, as well as usage data. S2. Secondly, adaptive interval update is adopted, which automatically adjusts the normalization interval to achieve data normalization processing based on the distribution and trend of the data. S3. Use the mutual information method to extract key features from the preprocessed data. Mutual information measures the correlation between two variables. Calculate the mutual information between each feature and the target variable, and select the features with larger mutual information. S4. Based on the selected features, a genetic algorithm is used to optimize the multilayer perceptron (MLP) to construct a building energy consumption prediction model. The model is trained through cross-validation to ensure its accuracy and generalization ability. S5. Using the trained model, predict energy consumption based on real-time environmental data and building usage, generate energy consumption trends for future time periods, and display the prediction results and influencing factors through visualization tools to provide intuitive decision support for building managers. In step S2, adaptive interval update is used. Based on the distribution and trend of the data, the normalization interval is automatically adjusted to achieve data normalization. The specific implementation process is as follows: S21. First, collect initial data samples, denoted as D0 = {X1, X2, ... X}. m }, where m is the initial sample size, calculate the minimum value of the initial data. and maximum value As the initial normalization interval; S22. Then, update the data, adding one data point X each time new data arrives. new Calculate the mean of the current dataset. Based on the distribution and trend of the data, an adjustment coefficient α is determined, along with the standard deviation σ. When the standard deviation is large, α is taken as a larger value to expand the normalization interval; when the standard deviation is small, α is taken as a smaller value to narrow the interval and improve accuracy. S23. Next, update the maximum and minimum values ​​of the interval, and the new minimum value. If X new <X min Then update X min =X new The new maximum value If X new >X max Then update X max =X new ; S24. Finally, for the new data point X new Normalization is performed using the updated normalization interval. The specific operation of using a genetic algorithm to optimize a multilayer perceptron (MLP) to construct a building energy consumption prediction model in step S4 is as follows: S41. First, during population initialization, each individual is represented using a hierarchical encoding mechanism. The encoding of each individual in the population includes the number of hidden layers in the MLP model structure, the number of neurons in each layer, and the model's hyperparameters. The encoding is as follows: Individual = [L1, L2, ..., L...]. n [,η,activation], where L1,L2,...,L n This represents the number of neurons in each layer, where η is the learning rate and activation is the activation function. S42. Next, train the MLP model corresponding to each individual and calculate its fitness based on the building energy consumption prediction task. The fitness function is Fitness = α × RMSE + β × Efficiency, where RMSE is the root mean square error, Efficiency is the improvement in building energy efficiency, which is used to measure the actual improvement effect of the model on building energy consumption management, and α and β are weight coefficients used to balance the weight of accuracy and energy efficiency improvement. S43. Next, in the crossover process of the genetic algorithm, multi-scale crossover is adopted, and operations are performed at both the model structure layer and the hyperparameter layer. In the mutation operation, individuals are allowed to borrow some weights or parameters from high-fitness individuals to accelerate convergence. For a certain high-fitness individual, some weights W are randomly selected to replace W. new =W mutated +γ(W best -W mutated ), where γ is the rate of variation, W new This is the latest weight, W best It is the weight of highly fit individuals, W mutated This represents the weight after the mutation; S44. Finally, as the genetic algorithm evolves, the crossover rate and mutation rate are dynamically adjusted. The crossover rate is CrossoverRate = InitialCrossoverRate × exp(-k × GenerationNumber), where InitialCrossoverRate is the initial crossover rate, k is the decay coefficient of the crossover rate, and GenerationNumber is the current generation number; the mutation rate is MutationRate = InitialMutationRate × exp(-m MR ×GenerationNumber), where InitialMutationRate is the initial mutation rate, m MR It is the decay coefficient of the mutation rate, and GenerationNumber is the current generation.

2. The building energy consumption prediction method based on artificial intelligence according to claim 1, characterized in that, The mutual information calculation method in step S3 is as follows: Where X and Y are two features, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y, respectively. The six features with the largest mutual information values ​​are selected as key features.

3. The building energy consumption prediction method based on artificial intelligence according to claim 1, characterized in that, The specific steps for generating and displaying energy consumption trends over future time periods in step S5 are as follows: S51. First, the processed real-time data is input into the previously trained energy consumption prediction model to predict energy consumption for a future time period of h. The model generates the predicted energy consumption value for the future time t+h. in X is the predicted energy consumption at time t+h, W is the trained weights of the model, and X is the energy consumption value at time t+h. t It is the input feature vector at time t; S52. Next, trend analysis is performed on the prediction results to generate the energy consumption trend over the future time period. Combined with the real-time input data, sensitivity analysis is used to evaluate the impact of each input feature on energy consumption and calculate the impact factor of each feature. Where I j This represents the sensitivity of the j-th feature to the energy consumption prediction result; S53. Finally, the predicted energy consumption trend and influencing factors are visualized using visualization tools, combining historical energy consumption data and future forecasts, to plot the energy consumption change curve over time.

Images