Building cold load prediction method and system based on combined prediction model, equipment and medium

Abstract

The invention discloses a building cold load prediction method and system based on a combined prediction model, equipment and a medium, and belongs to the technical field of building cold load prediction. The method comprises the following steps: collecting historical data of a building, and preprocessing the data to obtain a training set; using a LightGBM algorithm to train the training set, and determining characteristic factors affecting the building cold load; respectively inputting the data of the characteristic factors influencing the building cold load into a plurality of single prediction models for prediction; and determining a weight coefficient of each single prediction model by adopting a minSSE algorithm according to prediction results of the plurality of single prediction models, and constructing a combined prediction model for prediction. The combined prediction model obtained through the method has the good prediction effect and the high generalization ability, the characteristics and rules of data can be effectively captured, the prediction result is close to real data on the whole, and the performance is more stable at the local extreme value position and the high-load time period.

Description

Technical Field

[0001] The invention belongs to the technical field of building cooling load prediction, and in particular relates to a building cooling load prediction method and system, equipment and medium based on a combined prediction model. Background Art

[0002] Building cooling load prediction is a crucial technology that can effectively reduce the uncertainty of cooling load in central air-conditioning systems and provide key support for balancing supply and demand and achieving optimal control of energy management systems in buildings. At present, there is a widespread hysteresis in central air-conditioning systems, which requires researchers to adopt short-term prediction methods to adjust the optimal settings of equipment in advance to achieve on-demand cooling. In practical applications, in order to solve the unnecessary energy waste that may be caused by hysteresis, researchers have proposed a method that combines cooling load prediction of central air-conditioning systems with energy-saving optimization control. In this way, the central air-conditioning system can accurately and quickly achieve the balance of supply and demand of the entire system while meeting the cooling load demand of the building. This method has become one of the important directions for energy-saving optimization of central air-conditioning systems. In addition, the accuracy of building cooling load prediction directly determines the effectiveness of the control strategy, which directly affects the operating energy consumption of the system. Therefore, in order to ensure the stability and safety of building energy systems in operation, improving the accuracy of load prediction has become a hot topic and focus of research in recent years.

[0003] At present, the commonly used models in the field of building cooling load prediction mainly include two categories: physical method prediction and data-driven prediction. The prediction accuracy of the physical method depends largely on the detailed information of the building system. If the assumptions of the physical principles are not met, the model performance may be inconsistent. Based on the physical method, researchers have developed various building energy consumption simulation software, including EnergyPlus, DeST, TRNSYS, etc. At present, the statistical methods included in the data-driven method and the artificial intelligence-based method have achieved relatively ideal results in the field of building cooling load prediction. The statistical method uses the historical load data, schedule data, and weather data of the building to derive the correlation between the load factors and the actual load to calculate the prediction equation. This method has high requirements on the comprehensiveness and accuracy of the historical data, and it is difficult to deal with complex nonlinear problems. Summary of the invention

[0004] For complex building cooling load prediction problems, the data are often diverse and complex, and a single prediction model is often difficult to fully capture the characteristics and laws of the data, resulting in poor prediction results. Therefore, a building cooling load prediction method, system, equipment and medium with a combined prediction model are provided.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] The present invention provides a building cooling load prediction method based on a combined prediction model, comprising the following steps:

[0007] The actual data of characteristic factors affecting the building cooling load are input into the combined prediction model to obtain the building cooling load;

[0008] The construction process of the combined prediction model is as follows:

[0009] Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0010] The training set is input into the LightGBM algorithm for training to determine the characteristic factors that affect the building cooling load;

[0011] The data of characteristic factors affecting the building cooling load are input into several single prediction models for prediction;

[0012] According to the prediction results of several single prediction models, the minSSE algorithm is used to determine the weight coefficient of each single prediction model and construct a combined prediction model.

[0013] The characteristic factors affecting the building cooling load include the building cooling load at time t-1, outdoor dry bulb temperature, outdoor dew point temperature, radiant illumination at time t-1, personnel flow, outdoor relative humidity, heat transfer coefficient of the enclosure structure, use of lighting equipment, radiant illumination at time t-2 and wind speed.

[0014] The several single prediction models include LSTM model, SVR model and CNN model.

[0015] The weight coefficient of each single prediction model is determined by using the minSSE algorithm according to the prediction results of several single prediction models:

[0016] Several single prediction model weighting coefficients meet the following requirements:

[0017]

[0018] Calculate the sum of squares of the combined forecast errors based on the forecast errors of the combined forecast model;

[0019]

[0020] set up Then J=L T EL;

[0021] Where J is the sum of squares of the combined prediction errors, e t is the prediction error of the combined prediction model at time t, t = 1, 2, …, p, p is the time step;

[0022] Let R = [1KK1] T , determine the weight coefficient vector that minimizes the sum of squares of the combined prediction error J;

[0023] minJ=L T EL,

[0024] Among them, E is the prediction error and R is the determination coefficient.

[0025] Calculate the prediction error of the combined prediction model through the prediction error of the single prediction model;

[0026] e it =(x t -x it )

[0027]

[0028] Among them, e t is the prediction error of the combined prediction model at time t, t = 1, 2, …, p, p is the time step; e it is the prediction error of the i-th single prediction model at time t, i = 1, 2, ... n; x t is the actual value of the building cooling load at time t; x it is the building cooling load prediction value of the ith single prediction model at time t; l i is the weighting coefficient of the i-th single prediction model.

[0029] Assuming that the prediction error vector groups of several single prediction models are linearly independent, the optimal solution to the problem and the optimal value of the objective function are as follows:

[0030]

[0031] Among them, E is the prediction error and R is the determination coefficient.

[0032] The construction of the combined prediction model is specifically as follows:

[0033] Construct a combined prediction model through weighted coefficients and the prediction results of a single prediction model;

[0034]

[0035] in, is the combined predicted value of the building cooling load at time t, kW; x it is the building cooling load prediction value of the i-th single prediction model at time t.

[0036] The present invention also provides a building cooling load prediction system based on a combined prediction model, comprising:

[0037] A prediction module, used for inputting actual data of characteristic factors affecting the building cooling load into the combined prediction model to obtain the building cooling load;

[0038] The data acquisition module is used to collect historical data of building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0039] The data screening module is used to input the training set into the LightGBM algorithm for training and determine the characteristic factors that affect the building cooling load;

[0040] The single prediction model prediction module is used to input the data of the main factors affecting the building cooling load into several single prediction models for prediction;

[0041] A combined prediction model module is constructed to determine the weight coefficient of each single prediction model using the minSSE algorithm based on the prediction results of several single prediction models to construct a combined prediction model.

[0042] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned building cooling load prediction method based on the combined prediction model when executing the computer program.

[0043] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned building cooling load prediction method based on the combined prediction model are implemented.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention first pre-processes the historical data to ensure the quality and reliability of the data, and then uses the LightGBM algorithm to screen out the characteristic factors that affect the building cooling load and eliminate the factors with less influence. The LightGBM algorithm can ensure the speed in the process of processing data and the accuracy of the results, and then selects a single prediction model with good prediction performance, uses the advantages of each model to make predictions, uses the minSSE algorithm to determine the weight or weighting coefficient of each single prediction model, and combines the prediction results of the single prediction model to perform weighted averaging or other combination methods to reduce the overall prediction error. The combined prediction model obtained by the method of the present invention has good prediction effect and strong generalization ability, can effectively capture the characteristics and laws of the data, and the prediction results are generally close to the real data, and are more stable at local extreme values ​​and high-load time periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following briefly introduces the drawings required for use in the embodiment:

[0047] Figure 1 It is a flow chart of a building cooling load prediction method based on a combined prediction model of the present invention;

[0048] Figure 2 A step diagram of a building cooling load prediction method based on a combined prediction model according to the present invention;

[0049] Figure 3 This is a schematic diagram of the histogram optimization algorithm;

[0050] Figure 4 It is the Leaf-wise growth strategy diagram;

[0051] Figure 5 The following is a bar chart showing the analysis of the factors affecting the building cooling load by the LightGBM algorithm in the present invention;

[0052] Figure 6 This is a comparison chart of the prediction results of the combined prediction model and the single prediction model in Example 1;

[0053] Figure 7 This is a comparison chart of the relative errors of the prediction results of the combined prediction model and the single prediction model using the data from June 1 to 28, 2023 in Example 2;

[0054] Figure 8 This is a comparison chart of the relative errors of the prediction results of the combined prediction model and the single prediction model using the data from June 1 to July 28, 2022 in Example 2;

[0055] Fig. 9 A diagram of a building cooling load prediction system based on a combined prediction model according to the present invention;

[0056] Fig.10 This is a schematic diagram of the structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0057] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0058] The following is an explanation of the relevant content:

[0059] Since the factors affecting the cooling load of buildings are discrete and difficult to explain, and there is an imbalance between categories, it is particularly important to choose a suitable multi-label classification algorithm. This method selects One Vs RestClassifier (OVR) as the classifier framework to convert the multi-label classification problem into multiple independent binary classification problems for solution. OVR is a commonly used multi-label classification algorithm that converts the multi-label classification problem into multiple independent binary classification problems for solution. This method is widely used to deal with problems such as unbalanced data sets and has good interpretability. The LightGBM algorithm is then selected as the base classifier of OVR.

[0060] Light gradient boosting machine (LightGBM) is an efficient gradient boosting framework for solving supervised learning problems on large-scale and high-dimensional data sets. Different from the traditional depth-wise growth, the LightGBM algorithm adopts the Leaf-wise growth strategy, such as Figure 4 As shown in , each time the node that can minimize the loss function is selected for splitting, the leaf generation strategy shows excellent performance in minimizing the loss during the growth of the same leaf. This improves the training speed while ensuring the accuracy of the model. In addition, Figure 3 As shown, the LightGBM algorithm uses a histogram algorithm to discretize the data, discretizes continuous features into a specified number of buckets, reduces memory usage and computational overhead, and avoids scanning the complete data set when building the histogram, further improving efficiency. In summary, based on the histogram algorithm and leaf-like tree growth, the LightGBM algorithm provides faster training time and significantly reduced memory consumption compared to other gradient boosting frameworks. For example, the LightGBM algorithm can achieve a training speed 10 times faster than some similar algorithms while ensuring accurate results. In addition, by adopting a leaf-like growth strategy with depth restrictions to further optimize the histogram algorithm, the LightGBM algorithm can reduce computational errors.

[0061] The LightGBM algorithm is a tree model composed of multiple decision trees. Before each non-leaf node is split, the LightGBM algorithm calculates the gradient statistics of each feature variable at the current leaf node, such as the second-order derivative of the gradient. Then, the gain of each feature is calculated based on this information. The gain is measured by considering the error reduction before and after the split, and the feature variable with the largest gain is selected for splitting. The gain of the split point is defined as follows:

[0062]

[0063] In the formula, w i 、w j are the weights of the left and right leaf nodes respectively, w k is the weight of the original node before splitting.

[0064] The gradient boosting decision tree in the LightGBM algorithm is generated by multiple rounds of iterations of the training set data. In each iteration, a new tree is refitted using the gradient information to join the previous iteration tree. In the function space, this process can be regarded as a continuously changing linear combination process, and the formula is expressed as follows:

[0065]

[0066] Where χ is the function space of the iterative tree; f q (x i ) represents the predicted value of the i-th sample in the q-th tree. Each layer of the tree uses the best split point as the split node.

[0067] Each single prediction model has its own advantages and expertise. For example, the LSTM model is suitable for processing time series data with long-term dependencies. The LSTM model can solve the gradient vanishing and gradient exploding problems in traditional RNN, so as to better process long sequence data; however, the LSTM model requires a lot of parameter tuning and is prone to overfitting for small data sets. Support Vector Regression (SVR) is a regression method based on Support Vector Machine (SVM). Unlike traditional regression methods, the goal of SVR is not to fit the data directly, but to find a hyperplane in the feature space so that the sample points are as close to the hyperplane as possible, and a certain error is allowed within the tolerance range. The SVR algorithm has the advantages of strong nonlinear ability, strong robustness to outliers, and the ability to process high-dimensional data; however, the SVR algorithm takes a long time to train, is sensitive to noise and outliers, and is sensitive to parameter selection. The convolutional layer of CNN has local perception and can effectively capture the spatial relationship in the data. In building load forecasting, this helps the model understand the differences in energy usage between various areas inside the building, thereby improving the accuracy of the forecast. Moreover, building loads usually change over time, and CNN can effectively process time series data through appropriate architecture and hierarchy. This makes CNN an ideal choice for processing building energy consumption data with time correlation. In addition, CNN has strong generalization ability when trained on large-scale data sets and can adapt to buildings of different types and sizes. This adaptability enables the model to perform load forecasting in different environments and building structures without redesigning or adjusting the model. However, CNN has weak ability in processing time series data, has a large number of parameters, and is prone to overfitting.

[0068] The present invention proposes a combined prediction model to overcome the limitations of a single prediction model. For example, the selection of a single prediction model will be affected by data characteristics, and different data characteristics may require different types of models for prediction. However, for complex building cooling load prediction problems, data often have diversity and complexity, and a single prediction model often finds it difficult to fully capture the characteristics and laws of the data, resulting in poor prediction results. Therefore, a prediction model with relatively good prediction results is selected as a single prediction model, and then the minimum sum of squares of the combined prediction error algorithm is used to calculate the weighted coefficient of each single prediction model, and the building cooling load prediction data is obtained by combining the prediction results of the single prediction models.

[0069] like Figure 1 As shown, the building cooling load prediction method based on the combined prediction model provided by the present invention comprises the following steps:

[0070] The actual data of characteristic factors affecting the building cooling load are input into the combined prediction model to obtain the building cooling load;

[0071] The construction process of the combined prediction model is as follows:

[0072] Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0073] The training set is input into the LightGBM algorithm for training to determine the characteristic factors that affect the building cooling load;

[0074] The data of characteristic factors affecting the building cooling load are input into several single prediction models for prediction;

[0075] According to the prediction results of several single prediction models, the minSSE algorithm is used to determine the weight coefficient of each single prediction model and construct a combined prediction model.

[0076] The present invention first pre-processes the historical data to ensure the quality and reliability of the data, and then uses the LightGBM algorithm to screen out the characteristic factors that affect the building cooling load and eliminate the factors with less influence. The LightGBM algorithm can ensure the speed in the process of processing data and the accuracy of the results, and then selects a single prediction model with good prediction performance, uses the advantages of each model to make predictions, and uses the minSSE algorithm to determine the weight or weighting coefficient of each single prediction model, and combines the prediction results of the single prediction model for weighted averaging or other combinations to reduce the overall prediction error.

[0077] like Figure 2 As shown, the present invention is further explained:

[0078] (1) Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set, specifically:

[0079] First, it is necessary to collect and organize historical data, which includes data on building cooling load and various influencing factors, and organize the influencing factors as characteristic variables and building cooling load as target variables.

[0080] The collected historical data of building cooling load and various influencing factors are cleaned, missing values ​​and outliers are processed, and necessary feature selection, feature conversion and normalization are performed to ensure the quality and availability of the data. The processed data are used to construct a training set.

[0081] (2) The training set is input into the LightGBM algorithm for training to determine the characteristic factors affecting the building cooling load, specifically:

[0082] The training set is input into the LightGBM algorithm for training. After the training is completed, LightGBM will sort each influencing factor according to the sum of gains and the total number of times each influencing factor is used for classification, so as to obtain the importance ranking of each influencing factor. According to the ranking, factors with less impact on the building cooling load are eliminated to determine the characteristic factors affecting the building cooling load, such as Figure 5 shown.

[0083] In order to reduce the dimension of the input parameters of the prediction model, the factors that have a greater impact on the building's cooling load are considered. The present invention eliminates factors that have a smaller impact on the building's cooling load, such as wind direction, sky clarity, and solar scattered radiation value, and determines that the characteristic factors that affect the building's cooling load include the building's cooling load at time t-1, outdoor dry bulb temperature, outdoor dew point temperature, irradiance at time t-1, personnel flow, outdoor relative humidity, heat transfer coefficient of the enclosure structure, use of lighting equipment, irradiance at time t-2, and wind speed. Through comparative experiments, it is verified that after eliminating the input parameters with smaller influencing factors, the prediction results of the same model have a slight improvement in accuracy, which shows the effectiveness of simplifying the model's input parameters.

[0084] (3) Selection of several single prediction models

[0085] There are two main considerations when choosing a single model to build a good combined forecasting model: First, the accuracy of the single forecasting model should be as high as possible. Second, the single forecasting model should have high computational efficiency.

[0086] The present invention selects three classic single prediction models, LSTM, SVR and CNN, to construct the proposed model, and each model has high prediction accuracy and operation efficiency.

[0087] (4) The weight coefficient of each single prediction model is determined by using the minSSE algorithm according to the prediction results of several single prediction models to construct a combined prediction model, specifically:

[0088] Calculate the prediction error of the combined prediction model through the prediction error of the single prediction model;

[0089] e it =(x t -x it )

[0090]

[0091] Among them, e t is the prediction error of the combined prediction model at time t, t = 1, 2, ..., p, where p is the time step; e it is the prediction error of the i-th single prediction model at time t, i = 1, 2, ... n; x t is the actual value of the building cooling load at time t, kW; x it is the building cooling load prediction value of the ith single prediction model at time t, kW; l i is the weight coefficient of the i-th single prediction model;

[0092] The weighting coefficients of several single prediction models should meet the following requirements:

[0093] Calculate the sum of squares of the combined forecast errors based on the forecast errors of the combined forecast model;

[0094]

[0095] set up

[0096] Then we can get J = L T EL

[0097] Where J is the sum of squares of the combined prediction errors;

[0098] Let R = [1KK1] T , determine the weight coefficient vector that minimizes the sum of squares of the combined prediction error J;

[0099] minJ=L T EL,

[0100] Among them, E is the prediction error and R is the determination coefficient.

[0101] Among them, assuming that the prediction error vector groups of several single prediction models are linearly independent, the optimal solution to the problem and the optimal value of the objective function are as follows:

[0102]

[0103] Construct a combined prediction model through weighted coefficients and the prediction results of a single prediction model;

[0104]

[0105] in, is the combined predicted value of the building cooling load at time t, kW; x it is the building cooling load prediction value of the i-th single prediction model at time t.

[0106] Example 1

[0107] In order to more intuitively compare the cooling load prediction effect of the combined prediction model proposed in the present invention in the commercial complex, the data of the sampling points every hour on July 29, 30 and 31 were used for verification analysis. The single prediction models LSTM, SVR and CNN were selected for the prediction model accuracy comparison test. The prediction results are as follows: Figure 6 shown.

[0108] By observing the differences between historical data and the predicted data of each model. In the prediction of the LSTM model, although most of the predicted values ​​are close to the real data, there are deviations at some local extreme values, such as low-load data points. There are also certain differences between the prediction results of the SVR model and the real data, especially in areas with large data fluctuations, such as the high-load period on the second day. The prediction results of the CNN model are generally consistent with the real data, but there are certain deviations at some data points with large fluctuations, such as the data point of 6320.60kW on the third day. The prediction results of the combined prediction model LSTM-SVR-CNN are generally close to the real data, and compared with the prediction results of a single prediction model, it is more stable at local extreme values ​​and high-load periods.

[0109] In order to objectively evaluate the prediction results of each model and analyze the accuracy, stability and fitting degree of the model to the actual data in more depth, Table 1 shows the actual performance of the prediction results of different models in terms of mean absolute percentage error (MAPE), root mean square error (RMSE) and R2. The evaluation data in the table show that the determination coefficients R2 of different models are 0.9749, 0.8395, 0.8301 and 0.8938 respectively, among which the determination coefficient of the combined prediction model is the highest, close to 1, indicating that the predicted value of the model has the best fitting effect with the original data and the best prediction results. In addition, the RMSE and MAPE obtained by the combined prediction model are 1.7436 and 0.9575% respectively, which are 6.5352-19.5234 and 0.64%-5.2631% lower than those of other prediction models, respectively, and its prediction accuracy is higher.

[0110] Table 1 Prediction accuracy evaluation indicators of combined prediction model and single prediction model

[0111]

[0112] Example 2

[0113] In order to verify the generalization ability of the LSTM-SVR-CNN prediction model proposed in the present invention, the present invention partially deletes the original building cooling load data or replaces the building cooling load data of the cooling period in other years, and measures the generalization ability of the model for different data by observing the prediction error of the prediction model.

[0114] The present invention takes two periods of building cooling load data from June 1 to 28, 2023 and June 1 to July 28, 2022 as examples as training data for the LSTM-SVR-CNN prediction model, and uses the 72 data from the last three days as model verification data. The single prediction models LSTM, SVR and CNN are used as comparison models for verification. The verification results are as follows: Figure 7 and Figure 8 shown.

[0115] from Figure 7 and Figure 8 It can be seen that by using different building cooling load data as the training data of the model, compared with a single prediction model, the LSTM-SVR-CNN combined prediction model can achieve better prediction accuracy. Figure 7It can be seen that when only the building cooling load data in June is used as the model training data, the relative error range of the single prediction model is 4.8%-32.6%, while the relative error range of the combined prediction model is 1.62%-13.2%, which is significantly lower than that of the single prediction model. This shows that the combined prediction model has better prediction effect and stronger generalization ability when using different training data.

[0116] like Fig. 9 As shown, another object of the present invention is to propose a building cooling load prediction system based on a combined prediction model, comprising:

[0117] A prediction module, used for inputting actual data of characteristic factors affecting the building cooling load into the combined prediction model to obtain the building cooling load;

[0118] The data acquisition module is used to collect historical data of building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0119] The data screening module is used to input the training set into the LightGBM algorithm for training and determine the characteristic factors that affect the building cooling load;

[0120] The single prediction model prediction module is used to input the data of the main factors affecting the building cooling load into several single prediction models for prediction;

[0121] A combined prediction model module is constructed to determine the weight coefficient of each single prediction model using the minSSE algorithm based on the prediction results of several single prediction models to construct a combined prediction model.

[0122] like Fig.10 As shown, the third object of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the building cooling load prediction method based on the combined prediction model when executing the computer program.

[0123] The building cooling load prediction method based on the combined prediction model comprises the following steps:

[0124] The actual data of characteristic factors affecting the building cooling load are input into the combined prediction model to obtain the building cooling load;

[0125] The construction process of the combined prediction model is as follows:

[0126] Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0127] The training set is input into the LightGBM algorithm for training to determine the characteristic factors that affect the building cooling load;

[0128] The data of characteristic factors affecting the building cooling load are input into several single prediction models for prediction;

[0129] According to the prediction results of several single prediction models, the minSSE algorithm is used to determine the weight coefficient of each single prediction model and construct a combined prediction model.

[0130] A fourth object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the building cooling load prediction method based on the combined prediction model.

[0131] The building cooling load prediction method based on the combined prediction model comprises the following steps:

[0132] The actual data of characteristic factors affecting the building cooling load are input into the combined prediction model to obtain the building cooling load;

[0133] The construction process of the combined prediction model is as follows:

[0134] Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set;

[0135] The training set is input into the LightGBM algorithm for training to determine the characteristic factors that affect the building cooling load;

[0136] The data of characteristic factors affecting the building cooling load are input into several single prediction models for prediction;

[0137] According to the prediction results of several single prediction models, the minSSE algorithm is used to determine the weight coefficient of each single prediction model and construct a combined prediction model.

[0138] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A building cooling load prediction method based on a combined prediction model, characterized in that: The following steps are involved: The actual data of characteristic factors affecting the building cooling load are input into the combined prediction model to obtain the building cooling load; The construction process of the combined prediction model is as follows: Collect historical data on building cooling load and various influencing factors, and pre-process the historical data to obtain a training set; The training set is input into the LightGBM algorithm for training to determine the characteristic factors that affect the building cooling load; The data of characteristic factors affecting the building cooling load are input into several single prediction models for prediction; According to the prediction results of several single prediction models, the minSSE algorithm is used to determine the weight coefficient of each single prediction model and construct a combined prediction model.

2. The building cooling load prediction method based on the combined prediction model according to claim 1 is characterized in that: The characteristic factors affecting the building cooling load include the building cooling load at time t-1, outdoor dry bulb temperature, outdoor dew point temperature, radiant illumination at time t-1, personnel flow, outdoor relative humidity, heat transfer coefficient of the enclosure structure, use of lighting equipment, radiant illumination at time t-2 and wind speed.

3. The building cooling load prediction method based on the combined prediction model according to claim 1 is characterized in that: The several single prediction models include LSTM model, SVR model and CNN model.

4. The building cooling load prediction method based on the combined prediction model according to claim 1 is characterized in that: The weight coefficient of each single prediction model is determined by using the minSSE algorithm according to the prediction results of several single prediction models: Several single prediction model weighting coefficients meet the following requirements: Calculate the sum of squares of the combined forecast errors based on the forecast errors of the combined forecast model; set up Then J=L T EL; Where J is the sum of squares of the combined prediction error, et is the prediction error of the combined prediction model at time t, t = 1, 2, …, p, and p is the time step; Let R = [1KK1] T , determine the weighting coefficient vector that minimizes the sum of squares of the combined prediction error J; Among them, E is the prediction error and R is the determination coefficient.

5. The building cooling load prediction method based on the combined prediction model according to claim 4 is characterized in that: Calculate the prediction error of the combined prediction model through the prediction error of the single prediction model; et(X-Xit Where, et is the prediction error of the combined prediction model at time t, t = 1, 2, ..., p, p is the time step; eit is the prediction error of the i-th single prediction model at time t, i = 1, 2, ...n; x t is the actual value of the building cooling load at time t; t is the building cooling load prediction value of the ith single prediction model at time t; l i is the weighting coefficient of the i-th single prediction model.

6. The building cooling load prediction method based on the combined prediction model according to claim 4 is characterized in that If the prediction error vector groups of several single prediction models are linearly independent, then the optimal solution to the problem and the optimal value of the objective function are as follows: Among them, E is the prediction error and R is the determination coefficient.

7. The building cooling load prediction method based on the combined prediction model according to claim 4 is characterized in that: The construction of the combined prediction model is specifically as follows: Construct a combined prediction model through weighted coefficients and the prediction results of a single prediction model; in, is the combined predicted value of the building cooling load at time t, kW; x it is the building cooling load prediction value of the i-th single prediction model at time t.

8. A building cooling load prediction system based on a combined prediction model, characterized in that: include: A prediction module, used for inputting actual data of characteristic factors affecting the building cooling load into the combined prediction model to obtain the building cooling load; The data acquisition module is used to collect historical data of building cooling load and various influencing factors, and pre-process the historical data to obtain a training set; The data screening module is used to input the training set into the LightGBM algorithm for training and determine the characteristic factors that affect the building cooling load; The single prediction model prediction module is used to input the data of the main factors affecting the building cooling load into several single prediction models for prediction; A combined prediction model module is constructed to determine the weight coefficient of each single prediction model using the minSSE algorithm based on the prediction results of several single prediction models to construct a combined prediction model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the building cooling load prediction method based on the combined prediction model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the building cooling load prediction method based on the combined prediction model according to any one of claims 1 to 7.

Images