A prediction method, training method and prediction device for building energy consumption load

Abstract

The present application provides a prediction method, training method and prediction device for building energy consumption load. The prediction method includes obtaining first difference data and second difference data, using the first difference data as input data and the second difference data as output data to train a prediction model for predicting building energy consumption load; when the difference between the predicted value and the actual value is less than a threshold, the training is completed and an optimized prediction model is obtained; based on the second data set, third difference data is obtained so that the optimized prediction model uses each third difference data as input data for prediction to obtain a predicted building energy consumption load; and reverse differential processing is performed on the predicted building energy consumption load to obtain a target building energy consumption load. In this way, the accuracy of predicting building energy consumption load can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of building energy consumption prediction, and in particular to a prediction method, a training method and a prediction device for building energy consumption load. Background Art

[0002] At present, the energy consumption of air conditioning systems (HVAC) usually accounts for more than 30%-50% of the operating energy consumption of public buildings. Since the air conditioning system is designed according to the maximum demand, it is in a partial load state most of the time during actual use, which will lead to low efficiency and energy waste of the air conditioning system under specific working conditions. Building energy load prediction can help optimize the control strategy of the air conditioning system. By predicting the building energy load of the air conditioning system at future moments, the optimal operating conditions and set points can be determined, and a reasonable air conditioning operation strategy can be formulated.

[0003] There are currently many methods for predicting building energy consumption load. Traditional statistical regression analysis methods, such as the differential autoregressive moving average model, are widely used and have simple principles. However, this method has high requirements on data and has low accuracy in predicting building energy consumption load, which cannot meet the needs of predicting building energy consumption load. Summary of the invention

[0004] In view of the above technical problems existing in the prior art, this application is proposed. This application aims to provide a prediction method, training method and prediction device for building energy consumption load, which can adjust the parameters of the prediction model based on the real collected building energy consumption load data for training, can obtain an optimized prediction model and can improve the accuracy of the prediction of building energy consumption load based on the optimized prediction model.

[0005] According to the first scheme of the present application, a method for predicting building energy consumption load is provided, the prediction method comprising: based on a first data set, obtaining a plurality of first difference data of the building energy consumption load in a first time period before a reference time, the first difference data being the difference between the building energy consumption load of two adjacent representative times in the first time period; obtaining first data of the sum of the building energy consumption load in a second time period after the reference time, relative to second data of the sum of the building energy consumption load in a third time period before the reference time; using each first difference data as input data and each second difference data as output data, to predict the energy consumption load of a building in a first time period before the reference time. The method comprises the following steps: training a prediction model for predicting building energy consumption load; obtaining a prediction value during the training process, and completing the training and obtaining an optimized prediction model when the difference between the prediction value and the actual value is less than a threshold value; obtaining a plurality of third difference data of the building energy consumption load in a first time period before a target time based on a second data set, wherein the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period, so that the optimized prediction model uses each third difference data as input data for prediction to obtain a predicted building energy consumption load; and performing a reverse differential processing on the predicted building energy consumption load to obtain a target building energy consumption load.

[0006] According to the second scheme of the present application, a training method for a neural network model for predicting building energy consumption load is provided, the training method comprising: using each first difference data as input data of the neural network model, wherein the first difference data is the difference between the building energy consumption load of two adjacent representative times in a first time period before a reference time determined based on a training set; using second difference data as output data of the neural network model, wherein the second difference data is the deviation of first data of the sum of the building energy consumption load in a second time period after the reference time determined based on the training set relative to second data of the sum of the building energy consumption load in a third time period before the reference time; training the neural network model based on the first difference data and the second difference data.

[0007] According to a third scheme of the present application, a prediction device for building energy consumption load is provided, the prediction device comprising a processor, the processor being configured to: based on a first data set, obtain a plurality of first difference data of the building energy consumption load in a first time period before a reference time, the first difference data being the difference between the building energy consumption load of two adjacent representative times in the first time period; obtain first data of the sum of the building energy consumption load in a second time period after the reference time, relative to second data of the sum of the building energy consumption load in a third time period before the reference time; use each first difference data as input data, and each second difference data as output data; Output data to train a prediction model for building energy consumption load prediction; obtain the predicted value in the training process, and when the difference between the predicted value and the actual value is less than a threshold, complete the training and obtain an optimized prediction model; based on the second data set, obtain multiple third difference data of the building energy consumption load in a first time period before the target time, the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period, so that the optimized prediction model uses each third difference data as input data for prediction to obtain the predicted building energy consumption load; perform reverse differential processing on the predicted building energy consumption load to obtain the target building energy consumption load.

[0008] According to the fourth scheme of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the processor executes the method for predicting building energy consumption load according to the embodiment of the present application or the method for training a neural network model for predicting building energy consumption load according to the embodiment of the present application.

[0009] Compared with the prior art, the beneficial effects of the embodiments of the present application are:

[0010] The method for predicting building energy consumption load provided in the embodiment of the present application performs differential processing on the collected building energy consumption load, uses multiple first difference data of the building energy consumption load in the first time period as input data, and uses each second difference data as output data to train the prediction model, and optimizes and adjusts each parameter in the prediction model according to the difference between the predicted value and the actual value, until the difference between the predicted value and the actual value is less than a threshold value, the training is completed and the optimized prediction model is obtained. The optimized prediction model is trained based on the data of the actually collected building energy consumption load, and is trained with the second difference data as output data, which can improve the prediction accuracy of the optimized prediction model for the building energy consumption load.

[0011] The third difference data in the second data set is used as input data, and based on the optimization prediction model, the building energy consumption load after the target time is predicted, and the predicted building energy consumption load is reversely differentially processed to obtain the target building energy consumption load. In this way, the prediction accuracy of the building energy consumption load can be improved, which is conducive to the reasonable planning of the operation status of the air conditioning system based on the predicted building energy consumption load, thereby improving the working efficiency of the air conditioning system and reducing resource waste.

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above description and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. Similar reference numerals with letter suffixes or different letter suffixes may represent different examples of similar components. The accompanying drawings generally illustrate various embodiments by way of example and not by way of limitation, and together with the description and claims, are used to illustrate the disclosed embodiments. Such embodiments are illustrative and exemplary and are not intended to be exhaustive or exclusive embodiments of the present method, apparatus, system, or non-transitory computer-readable medium having instructions for implementing the method.

[0014] Figure 1 A flow chart showing a method for predicting building energy load according to an embodiment of the present application is shown.

[0015] Figure 2 A schematic diagram showing the comparison between the predicted value and the actual value of the building energy consumption load within the next hour over a period of ten days according to an embodiment of the present application is shown.

[0016] Figure 3 A flow chart showing a method for training a neural network model for predicting building energy load according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solution of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific implementation examples. The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific implementation examples, but are not intended to limit the present application.

[0018] The words "first", "second" and similar words used in this application do not indicate any order, quantity or importance, but are only used to distinguish. The words "include" or "comprises" and similar words used in this application mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of covering other elements. In this application, the arrows shown in the figures of each step are only examples of the execution order, not limitations. The technical solution of this application is not limited to the execution order described in the embodiments. The various steps in the execution order can be combined, decomposed, or the order can be changed, as long as the logical relationship of the execution content is not affected.

[0019] All terms (including technical terms or scientific terms) used in this application have the same meaning as those understood by ordinary technicians in the field to which this application belongs, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined here. Techniques and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the techniques and equipment should be considered as part of the specification.

[0020] Figure 1 A flow chart of a method for predicting building energy consumption load according to an embodiment of the present application is shown. In step S101, based on a first data set, a plurality of first difference data of building energy consumption load in a first time period before a reference time is obtained, wherein the first difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period.

[0021] The building energy load can be understood as the energy that a building needs to consume within a specific period of time, and is usually used to describe the energy consumption of a building, such as energy requirements for heating, cooling, lighting, ventilation, etc. For example, sensors and other equipment can be used to collect operating data of a building's air-conditioning system, boiler, and other related equipment, and the collected data can be analyzed and processed and stored as the building energy load. There is no limitation on the analysis and processing method, and it can be implemented using commonly used analysis and processing methods, such as using building energy model software or existing mathematical models to calculate based on the collected operating data of the air-conditioning system, such as temperature, humidity, air volume, cooling or heating power, etc. This is only used as an exemplary explanation and does not constitute a specific limitation on the building energy load.

[0022] The first data set can be a training set or a data set of real building energy consumption load collected by other users, and there is no limitation on this. The user can set the reference time and the first time period by himself. For example, the reference time is set to 12:00 noon and the first time period is 11:00-12:00 as an example. Between the first time period 11:00-12:00, multiple representative times can be set, such as 11:10, 11:20, 11:30, 11:40, and 11:50. The difference processing method can be used to obtain the difference in building energy consumption load between two adjacent representative times. For example, it is obtained by the difference between the current building energy consumption load and the building energy consumption load of the previous moment. Specifically, it can be the difference between the current building energy consumption load at the representative time of 12:00 and the building energy consumption load at the previous moment of 11:50, the difference between the current building energy consumption load at the representative time of 11:50 and the building energy consumption load at the previous moment of 11:40, the difference between the current building energy consumption load at the representative time of 11:40 and the building energy consumption load at the previous moment of 11:30, and so on. Six first difference data can be obtained in the first time period 11:00-12:00. This is only used as an example, and the difference calculation between the representative times will not be given as an example.

[0023] That is to say, within the first time period, each first difference data can be determined by using a differential processing method based on the set representative time. The setting of the representative time can be set by the user or determined according to the update time of the sensor collecting the building energy consumption related data. For example, if the update time of the sensor collecting the building energy consumption related data is 5 minutes, 12 first difference data can be obtained within 1 hour, and there is no limit to this.

[0024] In step S102, first data of the total building energy load in a second time period after a reference time and second difference data of the total building energy load in a third time period before the reference time are obtained. Specifically, the first data set includes collected real building energy load data, for example, the building energy load of a certain building within one year may be collected. When training the prediction model, part of the data in the first data set may be used as training data.

[0025] For example, as an exemplary explanation: taking the reference time as 12:00 and the third time period as 11:00-12:00 as an example, several representative time points can be determined within the third time period based on the sensor collection interval. For example, if the sensor collection interval is 10 minutes, the representative times are 11:10, 11:20, 11:30, 11:40, 11:50, and 12:00. Then, the building energy consumption load of each representative time within the third time period can be obtained, and the building energy consumption load of each representative time can be added together to obtain the second data.

[0026] The third time period may be the same as or different from the first time period. For example, assuming the reference time is 12:00 and the first time period is 11:00-12:00, the third time period may also be 11:00-12:00, or 10:00-12:00, or other time periods, without limitation.

[0027] The method for acquiring the first data is similar to the method for acquiring the second data, that is, the first data can be obtained by adding up the building energy consumption loads at each representative time collected in the second time period.

[0028] This is only an example and does not constitute a limitation on the specific solution.

[0029] In step S103, the prediction model for building energy load prediction is trained with each first difference data as input data and each second difference data as output data. In step S104, the prediction value in the training process is obtained, and when the difference between the prediction value and the actual value is less than a threshold value, the training is completed and the optimized prediction model is obtained. Specifically, the prediction model can be a neural network prediction model, and in the process of training the prediction model, various parameters of the prediction model are optimized and adjusted.

[0030] For example, the predicted value of the prediction model during the training process is obtained, and the predicted value is compared with the actual value, and the relevant parameters in the prediction model are adjusted and optimized according to the comparison result, so that the difference between the predicted value output based on the prediction model and the actual value is less than the threshold value, and the training is completed and the optimized prediction model is obtained. The actual value can be obtained based on the collected building energy consumption load provided by the first data set. Alternatively, Bayesian optimization can be performed during the training of the prediction model to obtain the average absolute percentage error, which can be used to explain the error between the predicted value and the actual value.

[0031] In step S105, based on the second data set, a plurality of third difference data of the building energy consumption load in the first time period before the target time is obtained, and the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period, so that the optimization prediction model uses each third difference data as input data for prediction to obtain the predicted building energy consumption load. Specifically, the second data set can be a test set, or a data set of building energy consumption load collected before the target time. The target time may be the same as or different from the reference time. For example, the reference time is 12:00, and based on the optimization prediction model, if you want to predict the building energy consumption load after 13:00, the target time is set to 13:00. This is only used as an example and does not constitute a limitation on the reference time and the target time.

[0032] In the process of training the prediction model, the second difference data is used as the output data for training, which will further improve the prediction accuracy of the optimization prediction model, and the optimization prediction model has the best optimization parameters. Based on the second data set, the third difference data can be determined, and the third difference data can be used as the input data of the optimization prediction model for prediction to obtain the predicted building energy consumption load.

[0033] In step S106, the predicted building energy consumption load is subjected to reverse differential processing to obtain a target building energy consumption load, which is the data of the building energy consumption load after the predicted target time. After verification, the target building energy consumption load is almost consistent with the actual building energy consumption load, indicating that the prediction method provided by this embodiment has a high prediction accuracy.

[0034] In this embodiment, the arrows shown in the figures for each step are merely examples of the execution order, rather than limitations. The technical solution of the present application is not limited to the execution order described in the embodiment, and the steps in the execution order can be combined, decomposed, or the order can be changed, as long as the logical relationship of the execution content is not affected.

[0035] For comparison, for example, the building energy load in each time period can be obtained, for example, the building energy load from 0:00 in the morning of the first day to 0:00 in the evening of the tenth day can be directly obtained and stored. The building energy load in the next hour is predicted based on the time point where the collection interval is 5 minutes, and the predicted value of the building energy load is obtained, which is then compared with the actual value of the building energy load in the next hour.

[0036] Figure 2A schematic diagram showing the comparison between the predicted value and the actual value of the building energy load in the next hour over a period of ten days is shown, wherein the actual value is obtained based on the overall shift of the existing building energy load value back by one hour. It can be seen that the predicted value of the building energy load predicted by the prediction method provided in the embodiment of the present application is highly consistent with the actual value, which further illustrates the reliability of the building energy load prediction method of the present application, and the predicted value of the building energy load predicted by the prediction method of the present application is highly accurate, which can provide more reliable reference data for debugging the operating conditions of equipment such as air conditioning systems in buildings.

[0037] Therefore, the embodiments of the present application can better cope with the time-sharing prediction problem of complex building energy consumption loads, and can adapt to more scenarios, improve the efficiency and reliability of energy management, and provide strong support for energy-saving gains.

[0038] In some embodiments of the present application, the input data also includes relevant parameter data related to the building energy consumption load, and the relevant parameter data includes at least weather temperature data, humidity data, wind speed, secondary pump operation frequency, and secondary water supply temperature. In this way, the relevant parameter data and multiple first difference data in the first time period are used as input data to perform the prediction operation, so as to comprehensively consider the relevant factors affecting the building energy consumption load, which is conducive to further improving the prediction accuracy of the optimized prediction model. Of course, in the process of prediction based on the optimized prediction model, using the third difference data and the relevant parameter data as input data at the same time can further improve the prediction accuracy of the predicted building energy consumption load.

[0039] In some embodiments of the present application, obtaining the difference between the building energy consumption load of two adjacent representative times in the first time period specifically includes: obtaining the collection time interval of the building energy consumption load in the first time period; determining each representative time in the first time period based on the collection time interval; subtracting the building energy consumption load of the two adjacent representative times to obtain the difference. The collection time interval can be set by the user, for example, the collection time interval can be determined according to the update time of the relevant sensor used to calculate the building energy consumption load. For example, if the sensor updates and collects data every five minutes, the collection time interval can be set to five minutes.

[0040] Specifically, still taking the reference time as 12:00, the first time period as 11:00-12:00, and the collection time interval as 5 minutes as an example, the representative times in the first time period 11:00-12:00 can be 11:05, 11:10, 11:15, 11:20, 11:25, 11:30, 11:35, 11:40, 11:45, 11:50, 11:55 and 12:00, and the building energy consumption load at 11:05, 11:10, 11:15, 11:20, 11:25, 11:30, 11:35, 11:40, 11:45, 11:50, 11:55 and 12:00 are obtained respectively. Afterwards, the building energy consumption load at 12:00 can be subtracted from the building energy consumption load at 11:55, and the building energy consumption load at 11:55 can be subtracted from the building energy consumption load at 11:50, and so on, and a total of 12 sets of difference data can be obtained. This is only used as an exemplary description and does not constitute a limitation on the specific solution.

[0041] In some embodiments of the present application, the specific method for obtaining the first data and the second data includes: obtaining the collection time interval of the building energy consumption load in the third time period before the reference time and in the second time period after the reference time; based on the collection time interval and the time length of the third time period and the second time period, respectively obtaining the number of times the building energy consumption load is collected in the third time period and the second time period; adding the building energy consumption loads obtained each time in the third time period and the second time period to obtain the second data and the first data. As described above, the third time period and the first time period may be the same or different, and there is no limitation on this. Specifically, taking the collection time interval of 5 minutes as an example, assuming that the time length of the third time period is 1 hour, 12 times of collection can be performed in the third time period, and the building energy consumption loads obtained in these 12 times can be added to obtain the second data. Similarly, assuming that the time length of the second time period is 1 hour, the 12 times of building energy consumption loads collected in the second time period can be added to obtain the first data.

[0042] In some embodiments of the present application, the second time period is a prediction time period or a partial time period in the prediction time period. Specifically, the reference time can still be taken as 12:00 as an example, assuming that the building energy consumption load during the period of 12:00-13:00 is predicted, then the second time period is the prediction time period (12:00-13:00). Assuming that the building energy consumption load of the four time-sharing time periods of 12:00-13:00, 13:00-14:00, 14:00-15:00, and 15:00-16:00 needs to be predicted, the prediction time period is 12:00-16:00, and the second time periods are 12:00-13:00, 13:00-14:00, 14:00-15:00, and 15:00-16:00 respectively. Therefore, the prediction of the building energy consumption load of the future time-sharing time period can be achieved, and the prediction result has a high accuracy.

[0043] In some embodiments of the present application, the predicted building energy consumption load is subjected to reverse differential processing to obtain the target building energy consumption load, specifically comprising: adding the predicted building energy consumption load to the sum of the building energy consumption load in the third time period before the target time, so as to obtain the target building energy consumption load. For example, the target time is 13:00, and it is desired to predict the target building energy consumption load in the period of 13:00-14:00. It is necessary to obtain the sum of the building energy consumption load in the time period of 12:00-13:00 before the target time 13:00, and this part of the data is the known data actually collected. In the process of prediction, the third difference data and related parameter data are used as input data, and prediction is performed based on the optimized prediction model to generate the predicted building energy consumption load, and then the predicted building energy consumption load is added to the sum of the building energy consumption load in the time period of 12:00-13:00, so as to obtain the target building energy consumption load in the period of 13:00-14:00.

[0044] This is only an exemplary description and does not constitute a limitation on the specific time period and specific reverse differential processing.

[0045] In some embodiments of the present application, a training method for a neural network model for predicting building energy consumption load is provided, the training method comprising: using each first difference data as input data of the neural network model, wherein the first difference data is the difference between the building energy consumption load of two adjacent representative times in a first time period before a reference time determined based on a training set; using second difference data as output data of the neural network model, wherein the second difference data is the deviation of first data of the total building energy consumption load in a second time period after the reference time determined based on the training set relative to second data of the total building energy consumption load in a third time period before the reference time; training the neural network model based on the first difference data and the second difference data.

[0046] Specifically, Figure 3 As shown, in step S301, the building energy consumption load of the target building within a preset time period is obtained, and in step S302, the relevant parameter data that affects the building energy consumption load data is obtained. For example, the temperature, air volume, cooling or heating, and other data related to the calculation of the building energy consumption load of the air-conditioning system of the target building for more than two months or three months can be collected using sensors and other related equipment, and the building energy consumption load of the target building can be calculated using the building energy consumption model software. At the same time, it is also necessary to collect the environmental parameters of the target building, such as weather temperature, relative humidity, wind speed, operating frequency of the secondary pump running in the target building, water supply temperature, etc. These relevant parameter data may have an impact on the building energy consumption load.

[0047] The acquired building energy load and related parameter data of the target building are cleaned and preprocessed. For example, in step S303, data cleaning is performed using methods such as missing value processing and outlier processing. Specifically, missing value processing is mainly for data with an abnormal value of 0 in the cleaned data. In the case of discontinuous missing values, interpolation can be used to complete the missing values. For example, the missing values ​​can be inferred based on the relationship between the data points adjacent to the missing values, and the inferred missing values ​​can be used to replace the data with an abnormal value of 0 to complete the missing values; if there are data with continuous missing values, the building energy load and related parameter data at the time or time period corresponding to the data with continuous missing values ​​are removed from the overall data set. In this way, the interference of outliers on the subsequent training of the neural network model can be avoided.

[0048] In general, the building energy load changes at adjacent moments are relatively smooth. On the same type of days, the building energy load change trends on adjacent days are similar. Therefore, the step change of the building energy load from hour to hour or the sudden change of the building energy load characteristics on adjacent days of the same type are outliers. The method for handling outliers is to regard the data that exceeds 3 times the sample standard deviation in a day as outliers and replace them with the linear average. This is only used as an example, and other methods of data processing are not excluded.

[0049] In step S304, a data set is obtained after preprocessing. For example, the acquired building energy load data and related parameter data are cleaned and then preprocessed. Specifically, the preprocessing operation may be differential processing, for example, obtaining the difference between the current building energy load and the building energy load at the previous moment, that is, obtaining the building energy load change value at each moment. At the same time, all building energy loads in the second time period in the future based on the reference moment are added together.

[0050] When only predicting the building energy consumption load in a single period after the reference time, for example, the reference time is 12:00 noon and the collection time interval is 5 minutes, the building energy consumption load within the next hour from 12:00 to 13:00 is predicted by adding the building energy consumption loads at 12:05, 12:10, 12:15, 12:20, 12:25, 12:30, 12:35, 12:40, 12:45, 12:50 and 13:00.

[0051] When predicting the building energy consumption load in multiple time periods after the reference time, for example, the reference time is 12:00 noon, the collection time interval is 5 minutes, and the building energy consumption load in the next 1, 2, and 3 hours is predicted respectively, the building energy consumption load at each moment in the three time periods of 12:00-13:00, 13:00-14:00, and 14:00-15:00 is added together.

[0052] By performing data cleaning and preprocessing operations on the data, a data set for training the neural network model can be obtained. In step S305, part of the data in the data set is used as a training set, and in step S306, another part of the data in the data set is used as a test set, and then step S307 is performed respectively to normalize the data in the training set, and step S308 is performed to normalize the data in the test set. Normalization refers to converting all data into numbers between [0, 1] or [-1, 1]. The purpose is to cancel the order of magnitude difference between the data of each dimension, avoid excessive network prediction errors due to large order of magnitude differences in input and output data, and play a role in unifying dimensions and preventing small data from being swallowed up.

[0053] After normalization, the input data and output data of the training set and the test set are respectively constructed by a sliding window method. In step S309, the first difference data of the training set is used as input data, and the first difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period before the reference time. In step S310, the second difference data of the training set is used as output data, and the second difference data is the deviation of the first data of the sum of the building energy consumption load in the second time period after the reference time relative to the second data of the sum of the building energy consumption load in the third time period before the reference time. In addition, in some other embodiments, the input data also includes relevant parameter data related to the building energy consumption load, and the relevant parameter data includes at least weather temperature data, humidity data, wind speed, secondary pump operation frequency, and secondary water supply temperature.

[0054] Specifically, when predicting the building energy consumption loads of multiple time-sharing periods after the reference time, the second time period is a partial time period in the prediction time period. For example, the reference time is 12:00 noon, the collection time interval is 5 minutes, and the building energy consumption loads in the first, second, and third hours in the future are predicted respectively. Then the output data can be three sets of deviations between the first data of the sum of the building energy consumption loads in the first, second, and third hours and the second data of the sum of the building energy consumption loads in the past hour, namely, the deviation after the first data of the sum of the building energy consumption loads from 12:00 to 13:00 minus the second data of the sum of the building energy consumption loads from 11:00 to 12:00, the deviation after the first data of the sum of the building energy consumption loads from 13:00 to 14:00 minus the second data of the sum of the building energy consumption loads from 11:00 to 12:00, and the deviation after the first data of the sum of the building energy consumption loads from 14:00 to 15:00 minus the second data of the sum of the building energy consumption loads from 11:00 to 12:00. This is only used as an exemplary description and does not constitute a limitation on the specific scheme.

[0055] Based on the first difference data and the second difference data as the input data and output data of the neural network model, step S313 is executed to train the neural network model, so that the neural network model can better learn the characteristics and laws related to the building energy consumption load. Further, in step S314, the Bayesian optimization algorithm is used to optimize the model parameters during the training process. The Bayesian optimization algorithm is used to train the neural network model for the data of the training set, which can effectively search the model parameter space of the neural network model and find the optimal solution. In the Bayesian optimization process, MAPE (Mean absolute percentage error) can be selected as the evaluation index of the performance of the neural network model, that is, the objective function. MAPE is the mean absolute percentage error. This evaluation index is sensitive to relative errors and will not change due to the global scaling of the target variable. It is suitable for problems with large dimensional differences in the target variable. In the prediction of building energy consumption load, the building energy consumption load value itself may have different scales and amplitudes depending on the size of the target building and the type of building energy consumption load, so the percentage expression of MAPE can better reflect these relative errors. The MAPE formula is as follows:

[0056]

[0057] in, is the predicted value of building energy load, y i is the actual value of the building energy load. When the MAPE value is larger, the error between the predicted value and the actual value is larger, and the prediction effect of the neural network model is worse; when the MAPE value is smaller, the error between the predicted value and the actual value is smaller, and the prediction effect of the neural network model is better.

[0058] In the Bayesian optimization process, when the mean absolute percentage error is within a threshold range, the model parameters of the neural network model are considered to be optimal. In some other embodiments of the present application, the neural network model is an LSTM model, and in the process of training the LSTM model, the model parameters of the LSTM model are optimized using a Bayesian optimization algorithm so that the mean absolute percentage error in the optimization process is within a threshold error range, thereby obtaining an LSTM model with higher accuracy in predicting building energy load.

[0059] Based on step S311, the first difference data of the test set is used as input data, and step S315 is continued to be executed to perform a prediction operation based on the optimized neural network model and the input data. That is to say, based on the trained neural network model, the first difference data in the test set is used as input data to perform a prediction operation, and output data is obtained based on the prediction operation (step S316). Then step S317 is continued to be executed to perform inverse normalization and inverse difference processing on the output data to obtain the predicted value of the building energy consumption load in the second time period. In addition, in step S312, the actual value of the building energy consumption load in the second time period is determined based on the test set, which is the actual value of the building energy consumption load in the second time period calculated based on the relevant data of the building energy consumption load stored in advance. Then, step S318 is executed to perform error detection based on the predicted value of the building energy consumption load and the actual value of the building energy consumption load, and the result is output.

[0060] In this way, the prediction accuracy of the neural network model for the building energy consumption load can be improved. The neural network model trained by this embodiment has the optimal model parameters, and the error between the predicted value and the actual value of the building energy consumption load is small. The training method provided by this embodiment can process, analyze and mine the data of the real building energy consumption load (rather than the simulation data), optimize the hyperparameters of the neural network model, and comprehensively learn the laws and characteristics related to the building energy consumption load, so that the neural network model obtained based on the training method can have a higher prediction accuracy. In addition, the neural network model trained by the training method can predict the building energy load in the time-sharing period after the reference time, for example, it can predict the building energy consumption load in the future 1, 2, 3, and 4 hour periods, further improving the advantages and flexibility of the neural network model obtained by the training method in predicting the building energy consumption load.

[0061] The training method of the neural network model for predicting building energy consumption load provided by this application uses differential processing to better capture the changing trend and periodicity of data. The Bayesian optimization algorithm is also used to tune the hyperparameters in the training process of the neural network model. The traditional grid search or random search method has the problem of wasting computing resources and low efficiency when trying various hyperparameter combinations, especially when the hyperparameter space is large. In contrast, Bayesian optimization can explore the hyperparameter space more intelligently, which helps to accelerate the model tuning process and reduce training time.

[0062] In some embodiments of the present application, a prediction device for building energy consumption load is provided, the prediction device comprising a processor, the processor being configured to: based on a first data set, obtain a plurality of first difference data of building energy consumption load in a first time period before a reference time, the first difference data being the difference of building energy consumption load between two adjacent representative times in the first time period; obtain first data of the sum of building energy consumption load in a second time period after the reference time, relative to second data of the sum of building energy consumption load in a third time period before the reference time; use each first difference data as input data, and use each second difference data as output data. Output data to train the prediction model for building energy consumption load prediction; obtain the prediction value in the training process, and when the difference between the prediction value and the actual value is less than the threshold, complete the training and obtain the optimized prediction model; based on the second data set, obtain multiple third difference data of the building energy consumption load in the first time period before the target time, the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period, so that the optimized prediction model uses each third difference data as input data for prediction to obtain the predicted building energy consumption load; perform reverse differential processing on the predicted building energy consumption load to obtain the target building energy consumption load. In this way, the accuracy of the building energy consumption load prediction can be improved.

[0063] The processor may be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), etc.

[0064] In addition, the method for predicting the building energy consumption load or the method for training the neural network model for predicting the building energy consumption load can also be stored in a computer-readable storage medium in the form of computer program instructions, and when the computer program instructions are executed by the processor, the processor executes the method for predicting the building energy consumption load or the method for training the neural network model for predicting the building energy consumption load according to the various embodiments of the present application to perform the various steps of the various embodiments of the present application. The storage medium may include a read-only memory (ROM), a flash memory, a random access memory (RAM), a dynamic random access memory (DRAM) such as a synchronous DRAM (SDRAM) or a Rambus DRAM, a static memory (e.g., a flash memory, a static random access memory), etc., on which computer executable instructions can be stored in any format.

[0065] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present application with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of the application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the claims and the full scope of their equivalents.

[0066] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the application. This should not be interpreted as an intention that a disclosed feature that does not require protection is necessary for any claim. On the contrary, the subject matter of the present application may be less than all the features of a specific disclosed embodiment. Thus, the claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present application should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

[0067] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A method for predicting building energy consumption load, characterized in that: The prediction method comprises: Based on the first data set, a plurality of first difference data of the building energy consumption load in a first time period before a reference time is obtained, wherein the first difference data is a difference between the building energy consumption loads of two adjacent representative times in the first time period before the reference time; Obtaining a collection time interval of the building energy consumption load in a third time period before a reference time and in a second time period after the reference time; Based on the collection time interval and the lengths of the third time period and the second time period, respectively obtaining the number of times the building energy consumption load is collected in the third time period and the second time period; The second data and the first data can be obtained by adding the building energy consumption loads obtained each time in the third time period and the second time period respectively; Acquire second difference data, where the second difference data is a difference between first data of the total energy consumption load of the building in a second time period after a reference time and second data of the total energy consumption load of the building in a third time period before the reference time; Using each first difference data as input data and each second difference data as output data, a prediction model for predicting building energy consumption load is trained; Obtain the predicted value during the training process, and when the difference between the predicted value and the actual value is less than a threshold, the training is completed and an optimized prediction model is obtained; Based on the second data set, a plurality of third difference data of the building energy consumption load in the first time period before the target time is obtained, wherein the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period before the target time, so that the optimization prediction model uses each third difference data as input data for prediction to obtain the predicted building energy consumption load; The predicted building energy consumption load is subjected to inverse differential processing to obtain the target building energy consumption load.

2. The prediction method according to claim 1, characterized in that: The input data also includes relevant parameter data related to the building energy consumption load, and the relevant parameter data at least includes weather temperature data, humidity data, wind speed, secondary pump operation frequency, and secondary water supply temperature.

3. The prediction method according to claim 1, characterized in that: Obtaining the difference between the building energy consumption loads of two adjacent representative times in the first time period, specifically including: Obtaining a collection time interval of a building energy consumption load within a first time period; Determine each representative time within the first time period based on the acquisition time interval; The difference can be obtained by subtracting the building energy consumption loads of two adjacent representative times.

4. The prediction method according to claim 1, characterized in that: The second time period is a predicted time period or a partial time period in the predicted time period.

5. The prediction method according to claim 1, characterized in that: The predicted building energy consumption load is subjected to reverse differential processing to obtain the target building energy consumption load, specifically including: The predicted building energy consumption load is added to the sum of the building energy consumption loads in the third time period before the target time to obtain the target building energy consumption load.

6. A training method for a neural network model for predicting building energy consumption load, characterized in that: The training method comprises: Using each first difference data as input data of the neural network model, wherein the first difference data is the difference between the building energy consumption loads of two adjacent representative times in a first time period before a reference time determined based on a training set; The second difference data is used as the output data of the neural network model, wherein the second difference data is the difference between the first data of the total energy consumption load of the building in the second time period after the reference time determined based on the training set and the second data of the total energy consumption load of the building in the third time period before the reference time; Training the neural network model based on the first difference data and the second difference data; The specific method of obtaining the first data and the second data includes: Obtaining a collection time interval of the building energy consumption load in a third time period before a reference time and in a second time period after the reference time; Based on the collection time interval and the lengths of the third time period and the second time period, respectively obtaining the number of times the building energy consumption load is collected in the third time period and the second time period; The second data and the first data can be obtained by adding up the building energy consumption loads obtained each time in the third time period and the second time period.

7. The training method according to claim 6, characterized in that: The neural network model is an LSTM model, and in the process of training the LSTM model, a Bayesian optimization algorithm is used to optimize the model parameters of the LSTM model so that the average absolute percentage error in the optimization process is within a threshold error range.

8. The training method according to claim 6, characterized in that: The input data also includes relevant parameter data related to the building energy consumption load, and the relevant parameter data at least includes weather temperature data, humidity data, wind speed, secondary pump operation frequency, and secondary water supply temperature.

9. A device for predicting building energy consumption load, characterized in that: The prediction device comprises a processor, wherein the processor is configured to: Based on the first data set, a plurality of first difference data of the building energy consumption load in a first time period before a reference time is obtained, wherein the first difference data is a difference between the building energy consumption loads of two adjacent representative times in the first time period before the reference time; Obtaining a collection time interval of the building energy consumption load in a third time period before a reference time and in a second time period after the reference time; Based on the collection time interval and the lengths of the third time period and the second time period, respectively obtaining the number of times the building energy consumption load is collected in the third time period and the second time period; The second data and the first data can be obtained by adding the building energy consumption loads obtained each time in the third time period and the second time period respectively; Acquire second difference data, where the second difference data is a difference between first data of the total energy consumption load of the building in a second time period after a reference time and second data of the total energy consumption load of the building in a third time period before the reference time; Using each first difference data as input data and each second difference data as output data, a prediction model for predicting building energy consumption load is trained; Obtain the predicted value during the training process, and when the difference between the predicted value and the actual value is less than a threshold, the training is completed and an optimized prediction model is obtained; Based on the second data set, a plurality of third difference data of the building energy consumption load in the first time period before the target time is obtained, wherein the third difference data is the difference between the building energy consumption load of two adjacent representative times in the first time period before the target time, so that the optimization prediction model uses each third difference data as input data for prediction to obtain the predicted building energy consumption load; The predicted building energy consumption load is subjected to inverse differential processing to obtain the target building energy consumption load.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to execute the method for predicting building energy consumption load according to any one of claims 1 to 5 or the method for training a neural network model for predicting building energy consumption load according to claim 7.

Images