Building air conditioner load prediction method based on LTSM and ARIMA

Abstract

The invention provides a building air conditioner load prediction method based on LTSM and ARIMA. The building air conditioner load prediction method comprises the following steps of S1, performing data acquisition; s2, a step of carrying out data preprocessing; s3, a step of carrying out modal decomposition; s4, a step of carrying out ARIMA model result analysis; and S5, carrying out result verification. According to the method, the problem of effectiveness of public building air conditioner load prediction is solved, that is, load prediction of a current public building air conditioner system is mainly based on traditional building simulation software or a single-module artificial intelligence algorithm, and the load prediction has certain limitations more or less. And a composite model in which a primary prediction model is established by adopting a long-short-term memory network and then a primary prediction result is corrected through a differential autoregression moving flat residual prediction model can effectively solve related problems in air conditioner load prediction.

Description

Technical Field

[0001] The present invention relates to the field of public building air conditioning load prediction, and in particular to a building air conditioning load prediction method based on LTSM and ARIMA. Background Art

[0002] With the rapid development of economic construction and the gradual improvement of people's requirements for quality of life, the proportion of building energy consumption in the overall energy consumption of society has increased rapidly. By 2022, the completed building area will be about 4 billion square meters, of which public buildings will be about 800 million square meters, accounting for about 20% of the total building area. The total energy consumption of buildings accounts for about 21% of the total energy consumption in the country, and the energy consumption of public buildings accounts for about 38% of the total energy consumption of buildings, of which air conditioning energy consumption accounts for about 50%. Therefore, energy saving and carbon reduction of air conditioning systems is the inevitable way to achieve green and low-carbon buildings.

[0003] The instability and volatility of air-conditioning loads during use cause the power system to face greater peak load pressure during the air-conditioning season, especially in hot weather. In order to further strengthen power regulation and control measures and improve the construction of virtual power plants in power grids, the country is vigorously promoting the construction of virtual power plants. As an adjustable load, air-conditioning load is an important part of the virtual power plant. Effective prediction and scheduling of air-conditioning loads can effectively achieve peak load shaving and valley filling. The commonly used air-conditioning load prediction method currently uses traditional building simulation software to predict cold and hot loads, which requires professional technicians to accurately model the building, which is time-consuming and labor-intensive, and the accuracy may not be guaranteed.

[0004] With the development of artificial intelligence technology, a lot of research has been done on air conditioning load forecasting methods at home and abroad. Commonly used methods include time series analysis, gray theory, artificial neural network, etc. Different air conditioning load forecasting methods have their own advantages, but the application of their models also has certain limitations. Long short-term memory network is an improved recurrent neural network (RNN) that can be used to solve the long-term dependency problem in time series data; differential autoregressive moving average (ARIMA) has a good effect on modeling and predicting the trend and seasonal components in time series data, and can be used to predict future values.

[0005] Based on the above background, the LSTM and ARIMA composite model can be used to effectively predict the air-conditioning load of public buildings. Summary of the invention

[0006] The purpose of the present invention is to provide a building air conditioning load forecasting method based on LTSM and ARIMA in view of the deficiencies in the prior art. The building air conditioning load forecasting method based on LTSM and ARIMA can well solve the above problems.

[0007] In order to achieve the above requirements, the technical solution adopted by the present invention is: to provide a building air conditioning load prediction method based on LTSM and ARIMA, and the building air conditioning load prediction method based on LTSM and ARIMA includes the following steps:

[0008] S1: Data acquisition step, collecting and obtaining factor data related to load prediction of air conditioning systems in public buildings through existing or newly built automation control systems in public buildings, and obtaining weather forecast data at the same time;

[0009] S2: Data preprocessing step, the collected data are grouped into one group for historical data and one group for synchronous data in the forecast period, the data are input into the LSTM model as features, the data are preprocessed, the missing data and abnormal data are filled with the previous unit data, and the data are normalized;

[0010] S3: The step of performing modal decomposition, using ICEEMDAN to decompose the original time series to obtain a series of intrinsic mode functions and a residual sequence;

[0011] S4: Steps for analyzing the results of the ARIMA model;

[0012] S5: Steps for result verification.

[0013] Preferably, the factor data related to the load prediction of the air conditioning system of the public building in step S1 includes historical external environmental factors, historical internal environmental factors, historical air conditioning system factors, historical time factors, weather forecast factors and forecast obtainable information;

[0014] Among them, historical external environmental factors include external temperature, external humidity, light intensity, and wind speed;

[0015] Historical internal environmental factors include indoor temperature, indoor humidity, and human traffic;

[0016] Historical air conditioning system factors include air conditioning loss load;

[0017] Historical time factors include time stamps;

[0018] Weather forecast factors include future maximum and minimum temperatures, future humidity, and future meteorological conditions;

[0019] The forecasted available information includes the expected time period time stamp.

[0020] Preferably, step S3 is specifically as follows:

[0021] S31: Decompose the input signal into several eigenmode functions and a residual component:

[0022]

[0023] Where I(n) is the input signal, IMF k (n) is the eigenmode function of Kth, Res K (n) is the residual;

[0024] S32: Extract IMF for screening by the following steps:

[0025] S321: mark local extreme points;

[0026] S322: connecting the maximum value points to form an upper envelope, and connecting the minimum value points to form a lower envelope;

[0027] S323: Find the mean signal line of the upper and lower lines;

[0028] S324: subtract the upper and lower envelope mean signals from the input signal to obtain an intermediate signal;

[0029] X(t)-m1=h1

[0030] S325: judging whether the intermediate signal satisfies the IMF stop screening condition, if so, the signal is an IMF component; if not, re-analyzing STEP 1 to STEP 4 based on the signal;

[0031] S33: Stop the screening condition. When the SDk obtained in the screening process is less than a given threshold or the number of consecutive screenings is equal to the selected number, the screening process stops;

[0032] SDK standard deviation calculation formula:

[0033]

[0034] Preferably, step S4 is specifically as follows:

[0035] S41: Establish an ARIMA model to conduct forecast analysis on the IMF calculated above;

[0036] S42: Generate a nonlinear residual sequence Rn;

[0037] S43: reorder to generate a new sample sequence, and then predict each part of the sample sequence;

[0038] S44: Finally, the residual Res of the linear prediction result and the ICEEMDAN model prediction result is calculated. K (n) are added together to obtain the final combined residual prediction result.

[0039] Preferably, step S5 is specifically as follows:

[0040] The trained model is used to predict the building air conditioning load for a certain period in the past and compared with the true value. The prediction model is tested using the mean absolute percentage error indicator, requiring MAPE < 20%;

[0041] The mean absolute mutual division error is sensitive to relative error and will not change due to the global scaling of the target variable. The prediction obtained by the prediction model output is subtracted from the true value and then divided by the true value to obtain the relative deviation of a single sample, and then summed and averaged. This can better reflect the average consistency between the predicted value and the true value. The MAPE calculation formula is as follows:

[0042]

[0043] in: is the predicted value of the prediction model, y i is the true value, and n is the number of samples.

[0044] The advantages of the building air conditioning load forecasting method based on LTSM and ARIMA are as follows:

[0045] This method solves the problem of the effectiveness of air conditioning load forecasting in public buildings, that is, the current load forecasting of air conditioning systems in public buildings is mainly based on traditional building simulation software or single-module artificial intelligence algorithms, and load forecasting has certain limitations to a certain extent. The composite model that uses a long short-term memory network (LSTM) to establish a primary forecasting model and then corrects the initial forecast results through a differential autoregressive moving average (ARIMA) residual forecasting model can effectively solve the related problems faced in air conditioning load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The same reference numerals are used in these drawings to represent the same or similar parts. The exemplary embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0047] Figure 1 The flowchart of a building air conditioning load forecasting method based on LTSM and ARIMA according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present application more clear, the present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0049] In the following description, references to "one embodiment", "an embodiment", "an example", "an example", etc. indicate that the embodiment or example described in this way may include specific features, structures, characteristics, properties, elements or limitations, but not every embodiment or example necessarily includes the specific features, structures, characteristics, properties, elements or limitations. In addition, repeated use of the phrase "according to one embodiment of the present application" may refer to the same embodiment, but does not necessarily refer to the same embodiment.

[0050] For the sake of simplicity, certain technical features well known to those skilled in the art are omitted in the following description.

[0051] According to one embodiment of the present application, Figure 1 As shown in the figure, the building air conditioning load forecasting method based on LTSM and ARIMA is specifically described as follows:

[0052] 1) Data acquisition:

[0053] Through the existing or newly built automation control system of public buildings, collect and obtain the factor data related to the load prediction of the air conditioning system of public buildings, and at the same time obtain the weather forecast data. The relevant information data is shown in the following table:

[0054]

[0055] 2) Data preprocessing:

[0056] The collected data are grouped into two groups: one group is historical data, and the other group is synchronous data for the forecast period. The data are input into the LSTM model as features, and the data are preprocessed. The missing data and abnormal data are filled with the previous unit data, and the data are normalized.

[0057] 3) Modal decomposition

[0058] ICEEMDAN is used to decompose the original time series to obtain a series of intrinsic mode functions (IMFs) and a residual sequence. The operation is as follows:

[0059] a) Decompose the input signal into several eigenmode functions and a residual component:

[0060]

[0061] Where I(n) is the input signal, IMF k (n) is the eigenmode function of Kth, Res K (n) is the residual.

[0062] b) Extract IMF for screening through the following steps:

[0063] STEP1: Mark the local extreme points;

[0064] STEP2: Connect the maximum points to form the upper envelope, and connect the minimum points to form the lower envelope;

[0065] STEP3: Find the average signal line of the upper and lower lines;

[0066] STEP4: Subtract the upper and lower envelope mean signals from the input signal to obtain the intermediate signal.

[0067] X(t)-m1=h1

[0068] STEP5: Determine whether the intermediate signal meets the IMF stop screening condition. If so, the signal is an IMF component; if not, repeat the analysis of STEP1 to 4 based on the signal.

[0069] c) Stop filtering conditions

[0070] When the SDk obtained by the screening process is less than a given threshold or the number of consecutive screenings is equal to the selected number, the screening process stops.

[0071] SDK standard deviation calculation formula:

[0072]

[0073] 4) ARIMA model results analysis

[0074] STEP1: Establish an ARIMA model to predict and analyze the IMF calculated above.

[0075] SETP2: Generate nonlinear residual sequence Rn,

[0076] STEP3: Reorder to generate a new sample sequence, and then predict each part of the sample sequence.

[0077] STEP4: Finally, the residual Res of the linear prediction result and the ICEEMDAN model prediction result is calculated. K (n) are added together to obtain the final combined residual prediction result.

[0078] 5) Result verification

[0079] The trained model is used to predict the building air conditioning load for a certain period in the past and compared with the true value. The prediction model is tested using the mean absolute percentage error indicator, requiring MAPE < 20%.

[0080] The mean absolute mutual error (MAPE) is sensitive to relative error and will not change due to the global scaling of the target variable. The prediction obtained by the prediction model output is subtracted from the true value and then divided by the true value to obtain the relative deviation of a single sample, and then summed and averaged, which can better reflect the average consistency between the predicted value and the true value. The MAPE calculation formula is as follows:

[0081]

[0082] in: is the predicted value of the prediction model, y i is the true value, n is the number of samples

[0083] The above-mentioned embodiments only represent several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the claims.

Claims

1. A building air conditioning load forecasting method based on LTSM and ARIMA, characterized in that: The steps include: S1: Data acquisition step, collecting and obtaining factor data related to load prediction of air conditioning systems in public buildings through existing or newly built automation control systems in public buildings, and obtaining weather forecast data at the same time; S2: Data preprocessing step, the collected data are grouped into one group for historical data and one group for synchronous data in the forecast period, the data are input into the LSTM model as features, the data are preprocessed, the missing data and abnormal data are filled with the previous unit data, and the data are normalized; S3: The step of performing modal decomposition, using ICEEMDAN to decompose the original time series to obtain a series of intrinsic mode functions and a residual sequence; S4: Steps for analyzing the results of the ARIMA model; S5: Steps for result verification.

2. The building air conditioning load forecasting method based on LTSM and ARIMA according to claim 1 is characterized in that: The factor data related to the load prediction of the air conditioning system of the public building in step S1 includes historical external environmental factors, historical internal environmental factors, historical air conditioning system factors, historical time factors, weather forecast factors and forecast available information; Among them, historical external environmental factors include external temperature, external humidity, light intensity, and wind speed; Historical internal environmental factors include indoor temperature, indoor humidity, and human traffic; Historical air conditioning system factors include air conditioning loss load; Historical time factors include time stamps; Weather forecast factors include future maximum and minimum temperatures, future humidity, and future meteorological conditions; The forecasted available information includes the expected time period time stamp.

3. The building air conditioning load forecasting method based on LTSM and ARIMA according to claim 1 is characterized in that: Step S3 is as follows: S31: Decompose the input signal into several eigenmode functions and a residual composition: Where I(n) is the input signal, IMF k (n) is the eigenmode function of Kth, Rex K (n) is the residual; S32: Extract IMF for screening by the following steps: S321: mark local extreme points; S322: connecting the maximum value points to form an upper envelope, and connecting the minimum value points to form a lower envelope; S323: Find the mean signal line of the upper and lower lines; S324: subtract the upper and lower envelope mean signals from the input signal to obtain an intermediate signal; X(t)-m1=h1 S325: judging whether the intermediate signal satisfies the IMF stop screening condition, if so, the signal is an IMF component; if not, re-analyzing STEP 1 to STEP 4 based on the signal; S33: Stop the screening condition. When the SDk obtained in the screening process is less than a given threshold or the number of consecutive screenings is equal to the selected number, the screening process stops; SDK standard deviation calculation formula:

4. The building air conditioning load forecasting method based on LTSM and ARIMA according to claim 1 is characterized in that: Step S4 is specifically as follows: S41: Establish an ARIMA model to conduct forecast analysis on the IMF calculated above; S42: Generate a nonlinear residual sequence Rn; S43: reorder to generate a new sample sequence, and then predict each part of the sample sequence; S44: Finally, the residual Res of the linear prediction result and the ICEEMDAN model prediction result is calculated. K (n) are added together to obtain the final combined residual prediction result.

5. The building air conditioning load forecasting method based on LTSM and ARIMA according to claim 1 is characterized in that: Step S5 is specifically as follows: The trained model is used to predict the building air conditioning load for a certain period in the past and compared with the true value. The prediction model is tested using the mean absolute percentage error indicator, requiring MAPE < 20%; The mean absolute mutual division error is sensitive to relative error and will not change due to the global scaling of the target variable. The prediction obtained by the prediction model output is subtracted from the true value and then divided by the true value to obtain the relative deviation of a single sample, and then summed and averaged. This can better reflect the average consistency between the predicted value and the true value. The MAPE calculation formula is as follows: in: is the predicted value of the prediction model, y i is the true value, and n is the number of samples.

Images