Building cooling and heating load prediction method based on multi-modal environment model

Abstract

The invention belongs to the technical field of building energy conservation, and discloses a building cooling and heating load prediction method based on a multi-modal environment model, which comprises the following steps: collecting multi-modal data, cleaning and preprocessing the data, constructing meteorological data, building structure data and historical load data sub-models, and predicting the cooling and heating load of a building. And fusing the prediction results of the three sub-models into an integrated model by adopting a weighted average model, and finally obtaining an optimal prediction result by optimizing the weight of the integrated model. The integrated model is deployed in a real-time system, prediction can be carried out in each time step, cooling and heating load requirements in several hours in the future can be output, equipment such as an air conditioner, a heater and a ventilation system can be scheduled based on predicted cooling and heating loads, energy consumption and comfort can be optimized, prediction results are integrated by adopting multi-model fusion, and the prediction efficiency is improved. Different from traditional linear weighting and stacking, the method can train a complex model weight close to reality, and has high prediction precision and system stability.

Description

Technical Field

[0001] The present application relates to the technical field of building energy conservation, and more specifically, to a method for predicting building cooling and heating loads based on a multimodal environmental model. Background Art

[0002] At present, with the acceleration of urbanization, the energy consumption of public buildings is getting higher and higher, among which the air conditioning system is the largest energy consumer in public buildings. Therefore, predicting the cooling and heating loads of public buildings is crucial to achieve energy conservation and emission reduction and improve building energy efficiency.

[0003] The main methods for predicting the cooling and heating loads of public buildings are prediction methods based on empirical formulas and prediction methods based on simulation. The prediction method based on empirical formulas is simple and easy to implement, but has low accuracy; the prediction method based on simulation has high accuracy, but due to the many factors that affect the changes in the cooling and heating loads of buildings, the data formats vary greatly, the data volume is large, and some data is stored in a distributed manner. Single models, linearly weighted multi-model integration, and stacked multi-model integration basically use the full amount of data to train and apply models on a single node, which cannot solve the current problem of large amounts of data being distributed and privately stored, resulting in the accuracy of the prediction not being guaranteed due to incomplete data. Summary of the invention

[0004] In order to solve the above problems, the present application provides a method for predicting building cooling and heating loads based on a multimodal environmental model.

[0005] The present application provides a method for predicting building cooling and heating loads based on a multimodal environmental model using the following technical solutions:

[0006] A method for predicting building cooling and heating loads based on a multimodal environmental model comprises the following steps:

[0007] S1. Collect multimodal data, wherein the multimodal data includes meteorological data, building structure data, and historical load data;

[0008] S2, data cleaning and preprocessing, including missing value processing, data standardization / normalization, timestamp alignment, and time series cutting;

[0009] S3. Construct a meteorological data sub-model to process meteorological data, using LSTM network to capture the temporal relationship in meteorological data; construct a building structure data sub-model to process the static structural information of the building, using fully connected neural network DNN deep learning to learn the factors affecting the cooling and heating load of the building; construct a historical load data sub-model to process the historical load data of the building, using LSTM to capture the dynamically changing temporal dependencies;

[0010] S6. The prediction results of the three sub-models are integrated into an integrated model using a weighted average model. The optimal prediction result is finally obtained by optimizing the weight of the integrated model.

[0011] S7, using mean square error, mean absolute error and R 2 The coefficient of determination is used to measure and evaluate the model, and the model is cross-validated to ensure the generalization ability of the model.

[0012] Furthermore, the collecting of multimodal data in S1 also includes adding annotation information to the multimodal local data to generate multimodal label data, and transmitting the local original data and label data to the central node database at regular intervals.

[0013] Furthermore, the missing value processing in S2 mainly includes the following steps:

[0014] S201, determining data points that need to be filled, wherein data points containing missing values ​​are filled;

[0015] S202, select n different data points and construct an n-fold missing value processing function;

[0016] S203. Find the basis functions of the known n data point pairs (x1, y1), (x2, y2)…(xn, yn)

[0017]

[0018] S204. Find the interpolation polynomial of the known n data point pairs (x1, y1), (x2, y2)…(xn, yn)

[0019]

[0020] S205. Substitute the points corresponding to the missing function values ​​into the interpolation polynomial to obtain the missing approximate values ​​L(x).

[0021] Furthermore, the construction of the meteorological data sub-model in S3 also includes time series processing of meteorological data, constructing a time window, and using the data of the previous n hours to predict the cooling and heating loads in the future.

[0022] Furthermore, constructing the building structure data sub-model in S3 also includes standardizing and / or one-hot encoding the static structure information of the building.

[0023] Furthermore, optimizing the weight of the integrated model in S4 includes optimizing the weight of each sub-model in the integration through cross-validation and / or grid search to improve the accuracy of the integrated model.

[0024] In summary, the present application includes at least one of the following beneficial technical effects:

[0025] (1) The integrated model of the present invention is deployed into a real-time system to receive real-time meteorological data, building sensor data, etc. The model can make predictions at each time step and output the cooling and heating load requirements in the next few hours. Based on the predicted cooling and heating loads, the air conditioning, heating, ventilation system and other equipment are dispatched to optimize energy consumption and comfort.

[0026] (2) The present invention uses multi-model fusion to integrate the prediction results. Different from the traditional linear weighting and stacking, this method can train more complex and practical model weights with higher prediction accuracy and system stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the method flow of this application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described clearly and completely below; obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments, and all other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application without making creative work are within the scope of protection of the present application.

[0029] Embodiment 1:

[0030] The present application is further described in detail below in conjunction with the accompanying drawings.

[0031] The present application embodiment discloses a method for predicting building cooling and heating loads based on a multimodal environmental model, comprising the following steps:

[0032] S1. Collect multimodal data, wherein the multimodal data includes meteorological data, building structure data, and historical load data;

[0033] S2, data cleaning and preprocessing, including missing value processing, data standardization / normalization, timestamp alignment, and time series cutting;

[0034] S3. Build a meteorological data sub-model to process meteorological data. Input meteorological data such as temperature, humidity, wind speed, and solar radiation. Process the time series of meteorological data, build a time window, use the data of the previous n hours to predict the cooling and heating loads in the future, and use the LSTM network to capture the time series relationship in meteorological data. LSTM can handle long-term dependency problems, so it is particularly suitable for predicting future load changes.

[0035] Construct a building structure data sub-model to process the static structure information of the building. Input: static structure information of the building, such as window area, wall material, air conditioning system type, etc. These data are usually numerical or categorical data, which can be directly standardized (numerical) or one-hot encoded (categorical). The fully connected neural network DNN deep learning is used to identify the factors that affect the cooling and heating load of the building;

[0036] Construct a historical load data sub-model to process the historical load data of the building (such as temperature records or load records in the past few days). These data are usually continuous numerical data in units of hours or days. Use time series cutting technology to divide the data into windows for training the LSTM model;

[0037] S4, using a weighted average model to fuse the prediction results of the three sub-models into an integrated model, and finally obtaining the optimal prediction result by optimizing the weight of the integrated model;

[0038] S5. The mean square error is used to calculate the difference between the predicted result and the actual result. The smaller the MSE, the higher the prediction accuracy. The mean absolute error is used to measure the average absolute value of the prediction error. 2 The coefficient of determination is used to measure the model's ability to explain the variability of the data. The closer the value is to 1, the better the model is. The model is cross-validated to ensure the generalization ability of the model.

[0039] Visualization tools can be used to compare actual load and predicted load, and analyze the prediction performance of the model in different time periods.

[0040] Preferably, collecting multimodal data in S1 also includes adding annotation information to the multimodal local data to generate multimodal label data, and transmitting the local original data and label data to the central node database at regular intervals.

[0041] Preferably, the missing value processing in S2 mainly includes the following steps:

[0042] S201, determining data points that need to be filled, wherein data points containing missing values ​​are filled;

[0043] S202, select n different data points and construct an n-fold missing value processing function;

[0044] S203. Find the basis functions of the known n data point pairs (x1, y1), (x2, y2)…(xn, yn)

[0045]

[0046] S204. Find the interpolation polynomial of the known n data point pairs (x1, y1), (x2, y2)…(xn, yn)

[0047]

[0048] S205. Substitute the points corresponding to the missing function values ​​into the interpolation polynomial to obtain the missing approximate values ​​L(x).

[0049] Preferably, constructing the meteorological data sub-model in S3 also includes time series processing of meteorological data, constructing a time window, and using the data of the previous n hours to predict the cooling and heating loads in the future.

[0050] Preferably, constructing the building structure data sub-model in S3 also includes standardizing and / or one-hot encoding the static structure information of the building.

[0051] Preferably, optimizing the weight of the integrated model in S4 includes optimizing the weight of each sub-model in the integration through cross-validation and / or grid search to improve the accuracy of the integrated model.

[0052] The integrated model is deployed into a real-time system to receive real-time meteorological data, building sensor data, etc. The model can make predictions at each time step and output the demand for cooling and heating loads in the next few hours. Based on the predicted cooling and heating loads, air conditioning, heating, ventilation systems and other equipment are dispatched to optimize energy consumption and comfort. The present invention uses multi-model fusion to integrate the prediction results. Different from the traditional linear weighting and stacking, this method can train a more complex and practical model weight, with higher prediction accuracy and system stability.

[0053] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting building cooling and heating loads based on a multimodal environmental model, characterized in that: The following steps are involved: S1. Collect multimodal data, wherein the multimodal data includes meteorological data, building structure data, and historical load data; S2, data cleaning and preprocessing, including missing value processing, data standardization / normalization, timestamp alignment, and time series cutting; S3. Construct a meteorological data sub-model to process meteorological data, using LSTM network to capture the temporal relationship in meteorological data; construct a building structure data sub-model to process the static structural information of the building, using fully connected neural network DNN deep learning to learn the factors affecting the cooling and heating load of the building; construct a historical load data sub-model to process the historical load data of the building, using LSTM to capture the dynamically changing temporal dependencies; S4, using a weighted average model to fuse the prediction results of the three sub-models into an integrated model, and finally obtaining the optimal prediction result by optimizing the weight of the integrated model; S5, using mean square error, mean absolute error and R 2 The coefficient of determination is used to measure and evaluate the model, and the model is cross-validated to ensure the generalization ability of the model.

2. The method for predicting building cooling and heating loads based on a multimodal environmental model according to claim 1, characterized in that: The collecting of multimodal data in S1 also includes adding annotation information to the multimodal local data to generate multimodal label data, and transmitting the local original data and label data to the central node database at regular intervals.

3. The method for predicting building cooling and heating loads based on a multimodal environmental model according to claim 1, characterized in that: The missing value processing in S2 mainly includes the following steps: S201, determining data points that need to be filled, wherein data points containing missing values ​​are filled; S202, select n different data points and construct an n-fold missing value processing function; S203, find the known n data point pairs (x1, y1), (x2, y2)…(x n ,y n ) S204, find the n known data point pairs (x1, y1), (x2, y2)…(x n ,y n ) S205. Substitute the points corresponding to the missing function values ​​into the interpolation polynomial to obtain the missing approximate values ​​L(x).

4. The method for predicting building cooling and heating loads based on a multimodal environmental model according to claim 1, characterized in that: The meteorological data sub-model constructed in S3 also includes time series processing of meteorological data, constructing a time window, and using the data of the previous n hours to predict the cooling and heating loads in the future.

5. The method for predicting building cooling and heating loads based on a multimodal environmental model according to claim 1, characterized in that: Constructing the building structure data sub-model in S3 also includes standardizing and / or one-hot encoding the static structure information of the building.

6. The method for predicting building cooling and heating loads based on a multimodal environmental model according to claim 1, characterized in that: Optimizing the weight of the integrated model in S4 includes optimizing the weight of each sub-model in the integration through cross-validation and / or grid search to improve the accuracy of the integrated model.

Images