CN117313990A - Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment - Google Patents
Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment Download PDFInfo
- Publication number
- CN117313990A CN117313990A CN202311154933.1A CN202311154933A CN117313990A CN 117313990 A CN117313990 A CN 117313990A CN 202311154933 A CN202311154933 A CN 202311154933A CN 117313990 A CN117313990 A CN 117313990A
- Authority
- CN
- China
- Prior art keywords
- air conditioner
- conditioner load
- sub
- prediction
- meta
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 97
- 238000004364 calculation method Methods 0.000 claims abstract description 22
- 230000009466 transformation Effects 0.000 claims abstract description 15
- 230000008878 coupling Effects 0.000 claims abstract description 12
- 238000010168 coupling process Methods 0.000 claims abstract description 12
- 238000005859 coupling reaction Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 39
- 230000008569 process Effects 0.000 claims description 37
- 238000004378 air conditioning Methods 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 28
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 12
- 238000003860 storage Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000009966 trimming Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 239000010410 layer Substances 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 238000004590 computer program Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013486 operation strategy Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000007430 reference method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
- G06F17/148—Wavelet transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Mathematical Physics (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Pure & Applied Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Computational Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Primary Health Care (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Educational Administration (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Water Supply & Treatment (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an air conditioner load prediction method, an air conditioner load prediction system and electronic equipment based on meta-integrated learning, and belongs to the technical field of air conditioner load prediction. According to the air conditioner load prediction method based on the meta-integrated learning, the time sequence characteristics of the air conditioner load sub-signals are extracted by constructing a wavelet transformation model, a point prediction model, a meta-integrated learning model and a result prediction model, so that the depth invariant characteristics in the air conditioner load can be accurately learned; the weight of the single-point prediction model can be adaptively adjusted; and finally, carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integration learning, so that pre-adjustment of an air conditioner unit can be realized, the prediction precision of the air conditioner load is effectively improved, the use experience of a user is improved, and popularization and use are facilitated.
Description
Technical Field
The invention relates to an air conditioner load prediction method, an air conditioner load prediction system and electronic equipment based on meta-integrated learning, and belongs to the technical field of air conditioner load prediction.
Background
The central air conditioning system brings comfortable indoor environment for people and consumes huge electric energy. The load demands of the air conditioner are continuously changed due to the changes of factors such as outdoor weather, indoor heat and the like, and at the moment, the load demands are met by changing the operation strategy of the water chilling unit, and meanwhile, the high-efficiency operation of the water chilling unit is maintained. Because of the high delay and large hysteresis characteristics of the air conditioning system, the air conditioning system which is regulated by only negative feedback may not ensure indoor thermal comfort and energy-saving operation of the water chiller.
Further, china patent (bulletin number: CN 115879590A) discloses a load prediction method based on wavelet feature extraction and an ensemble learning model, which adopts a wavelet transformation method to carry out frequency decomposition on a load sequence, adopts an XGBoost algorithm to carry out feature selection on the load sequence, and combines a Stacking ensemble learning model to carry out load prediction. In the first stage of load prediction, load data, temperature, humidity, rainfall and holiday information at a plurality of historical moments are adopted to preprocess and normalize related data; the second stage uses wavelet decomposition to obtain a stable load sequence of multiple frequency components based on the processed data in the previous stage, uses XGBoost algorithm to select the characteristics of the decomposed load sequence, and reduces the input dimension while removing irrelevant characteristics. And the third stage is to input the previously constructed feature set into a Stacking deep learning model for training.
However, the scheme is mainly suitable for short-term prediction of the power load, and because the air conditioner load has more complicated nonlinear, polytropic and dynamic characteristics compared with the power load, the scheme cannot be directly suitable for prediction of the air conditioner load, further the scheme and the existing scheme cannot accurately obtain the load predicted value at the next moment, the air conditioner unit cannot be helped to realize pre-adjustment, the use experience of a user is influenced, and popularization and use are not facilitated.
The information disclosed in this background is only for the understanding of the background of the inventive concept and therefore it may comprise information that does not form the prior art.
Disclosure of Invention
In view of the above-mentioned problems or one of the above-mentioned problems, an object of the present invention is to provide a method for obtaining a plurality of air conditioner load sub-signals by processing time-series air conditioner load data; based on a stacked coding algorithm, extracting time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data; and then a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data are output; and finally, coupling calculation is carried out on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period, and air conditioner load prediction based on meta-integrated learning is completed, so that pre-adjustment of an air conditioner unit can be realized, the use experience of a user is improved, and the popularization and the use of the air conditioner load prediction method, the system and the electronic equipment based on the meta-integrated learning are facilitated.
In view of the above-mentioned problems or one of the above-mentioned problems, a second object of the present invention is to provide a method for extracting time-series characteristics of air-conditioning load sub-signals, from which depth invariant characteristics in air-conditioning loads can be accurately learned; the method and the system can adaptively adjust the weight of the single-point prediction model, so that the prediction precision of the air conditioner load can be further improved, the scheme is scientific, reasonable and feasible, and the method and the system are particularly suitable for the air conditioner load prediction method, the system and the electronic equipment based on meta-integrated learning for the air conditioner load prediction.
In order to achieve one of the above objects, a first technical solution of the present invention is:
an air conditioner load prediction method based on meta-integrated learning comprises the following steps:
firstly, acquiring time series air conditioner load data;
secondly, processing time sequence air conditioner load data by utilizing a pre-constructed wavelet transformation model to obtain a plurality of air conditioner load sub-signals;
thirdly, constructing a point prediction model with the same number according to the number of the air conditioner load sub-signals;
the point prediction model is based on a stacked coding algorithm, and extracts time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data;
fourth, a pre-built meta-integrated learning model is utilized to output a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data;
and fifthly, performing coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data by utilizing a pre-constructed result prediction model to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integrated learning.
Through continuous exploration and experiment, the invention processes the time series air conditioner load data by constructing a wavelet transformation model, a point prediction model, a meta-integrated learning model and a result prediction model to obtain a plurality of air conditioner load sub-signals; based on a stacked coding algorithm, extracting time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data; and then a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data are output; and finally, carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integration learning, so that pre-adjustment of an air conditioner unit can be realized, the use experience of a user is improved, and popularization and use are facilitated.
According to the invention, the time sequence characteristics of the air conditioner load sub-signals are extracted through the point prediction model, so that the depth invariant characteristics in the air conditioner load can be accurately learned; the weight of the single-point prediction model can be adaptively adjusted by utilizing the meta-integrated learning model, so that the prediction accuracy of the air conditioner load can be further improved, and the scheme is scientific, reasonable and feasible, and is particularly suitable for the prediction of the air conditioner load.
As a preferred technical measure:
the method for obtaining the plurality of air conditioner load sub-signals by the wavelet transformation model comprises the following steps:
decomposing the time series air conditioning load data by adopting Mallat discrete wavelet transformation to obtain a plurality of air conditioning load sub-signals, wherein the plurality of air conditioning load sub-signals at least comprise an approximate sub-signal and a plurality of detail sub-signals;
the Mallat discrete wavelet transform uses two low-pass filters and two high-pass filters to calculate wavelet coefficients, whose calculation formula is as follows:
wherein the method comprises the steps ofRepresents a mother wavelet, m and n are +.>T and T represent the discrete time index and the length of the original signal f (T), respectively.
As a preferred technical measure:
the method for obtaining the plurality of pieces of air conditioner load prediction data by the point prediction model comprises the following steps:
step 31, obtaining an approximate sub-signal and a plurality of detail sub-signals;
step 32, mapping an approximate sub-signal and a plurality of detail sub-signals to a feature space respectively, and coding to obtain sub-signal coding data with corresponding quantity;
step 33, mapping the sub-signal coding data back to the sample space, and decoding to obtain a plurality of predicted sub-signals;
and step 34, combining the plurality of prediction sub-signals to form a plurality of pieces of air conditioner load prediction data.
As a preferred technical measure:
the approximate sub-signal and the detail sub-signals are respectively mapped to the characteristic space, and the calculation formula for coding is as follows:
y＝f θ (Wx+b)
wherein y represents the output sub-signal coding data, x is an approximate sub-signal and a plurality of detail sub-signals, and θ is a point prediction model parameter composed of a weight W and a bias b; f is an activation function;
the sub-signal encoded data is mapped back to the sample space and decoded according to the following calculation formula:
y′＝g θ′ (W′x+b′)
wherein θ 'represents a decoding parameter, and is composed of a weight W' and a bias b ', g represents an activation function, y' is a plurality of prediction sub-signals, and x is sub-signal encoded data.
As a preferred technical measure:
the method for outputting a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data by using the meta-integrated learning model is as follows:
screening the external environment variable and the characteristic information in the air conditioner load prediction data by utilizing a pre-constructed characteristic selector;
the characteristic information comprises the ambient humidity, the ambient temperature, the slope of the air conditioner load change and the current air conditioner load level;
and updating the weight coefficient of each piece of air conditioner load prediction data in real time through a pre-constructed weight updater according to the characteristic information.
As a preferred technical measure:
the method for updating the weight coefficient of each air conditioner load prediction data in real time by the weight updater is as follows:
the characteristic information is first input in a linear combination to form an intermediate variable z i (a p ) The calculation formula is as follows:
wherein a is p (j) Representing the input characteristic information; omega ij Representing the weight between the ith characteristic information and the jth intermediate variable; m is the number of feature information;
the intermediate variables were then normalized using a softmax activation function, the calculation formula of which is shown below:
wherein a is p Representing the input characteristic information; c (C) i (x) A weight coefficient representing the i-th air conditioner load prediction data; e represents an exponential function; m is the number of air conditioner load prediction data; zi (a) p ) As intermediate variable z i (a p )。
As a preferred technical measure:
the result prediction model is used for carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data according to the following formula:
wherein F (x) p ) Representing a final air conditioner load prediction result;represents the i-th air conditioner load prediction data, C i (x p ) Representing the i-th weight coefficient.
As a preferred technical measure:
the method for constructing the point prediction model is as follows:
using the approximate sub-signal or detail sub-signal samples as inputs; taking a predicted value sample of the next time step as output, and calculating the weight and deviation parameters of the predicted value sample; the method is divided into a pre-training process and a fine-tuning process;
the pre-training process is to initialize all automatic encoder parameters in a point prediction model to obtain a parameter initial value;
in an automatic encoder, the purpose of the training process is to approximate the output to the input, the expression of its objective function is as follows:
wherein J represents an objective function; n is the number of samples; x is x i For the i-th entry item,for the ith denoising entry, g is the activation function, θ' is the decoder parameter; d represents a weight decay term, λ represents a coefficient, y' is a decoder, W is a weight, and L is a reconstruction error; θ is a point prediction model parameter; f is an activation function;
after obtaining the parameter initial value of the point prediction model from the pre-training process, performing parameter fine adjustment by using a gradient descent method until an objective function reaches the minimum, so that the parameters of the point prediction model are converged and a prediction task is realized;
the objective function expression in the trimming process is as follows:
wherein h (x) (i) ) And t (i) The predicted result and the actual output of the regressor are represented, D represents the weight decay term, lambda represents the coefficient, and L is the reconstruction error.
In order to achieve one of the above objects, a second technical solution of the present invention is:
an air conditioner load prediction system based on meta-integrated learning is applied to the air conditioner load prediction method based on meta-integrated learning;
the method comprises a wavelet transformation model block, a point prediction module, a meta-integrated learning module and a result prediction module:
the wavelet transformation model block is used for processing the time sequence air conditioner load data to obtain a plurality of air conditioner load sub-signals;
the point prediction module is used for extracting time sequence characteristics of the air conditioner load sub-signals based on a stacked coding algorithm to obtain a plurality of pieces of air conditioner load prediction data;
the meta-integrated learning module is used for outputting a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data;
and the result prediction module is used for carrying out coupling calculation on the plurality of weight coefficients and the plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period and predicting the air conditioner load based on meta-integration learning.
The invention is continuously explored and tested, and a plurality of air conditioner load sub-signals are obtained by constructing a wavelet transformation model block, a point prediction module, a meta-integrated learning module and a result prediction module and processing time sequence air conditioner load data; based on a stacked coding algorithm, extracting time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data; and then a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data are output; and finally, carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integration learning, so that pre-adjustment of an air conditioner unit can be realized, the use experience of a user is improved, and popularization and use are facilitated.
According to the invention, the time sequence characteristics of the air conditioner load sub-signals are extracted through the point prediction module, so that the depth invariant characteristics in the air conditioner load can be accurately learned; the weight of the single-point prediction module can be adaptively adjusted by utilizing the meta-integrated learning module, so that the prediction accuracy of the air conditioner load can be further improved, and the scheme is scientific, reasonable and feasible, and is particularly suitable for the prediction of the air conditioner load.
In order to achieve one of the above objects, a third technical solution of the present invention is:
an electronic device, comprising:
one or more processors;
a storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement an air conditioning load prediction method based on meta-integrated learning as described above.
Compared with the prior art, the invention has the following beneficial effects:
the invention obtains a plurality of air conditioner load sub-signals through continuous exploration and test and processing of time sequence air conditioner load data; based on a stacked coding algorithm, extracting time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data; and then a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data are output; and finally, carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integration learning, so that pre-adjustment of an air conditioner unit can be realized, the use experience of a user is improved, and popularization and use are facilitated.
The invention can extract the time sequence characteristics of the air conditioner load sub-signals, and can accurately learn the depth invariant characteristics in the air conditioner load; the weight of the single-point prediction model can be adaptively adjusted, so that the prediction accuracy of the air conditioner load can be further improved, and the scheme is scientific, reasonable and feasible, and is particularly suitable for the prediction of the air conditioner load.
Drawings
FIG. 1 is a schematic flow chart of the air conditioner load prediction method of the present invention;
FIG. 2 is a block flow diagram of a method for predicting air conditioning load according to the present invention;
FIG. 3 is a flow chart of a point prediction model based on wavelet transform and stacked coding SAE of the present invention;
FIG. 4 is a block flow diagram of a weight updater according to the present invention;
FIG. 5 is a schematic flow chart of the air conditioning load accurate prediction training process of the present invention;
fig. 6 is a schematic diagram of load scattering points of an air conditioner in spring and summer according to an embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
On the contrary, the invention is intended to cover any alternatives, modifications, equivalents, and variations as may be included within the spirit and scope of the invention as defined by the appended claims. Further, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. The present invention will be fully understood by those skilled in the art without the details described herein.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.
As shown in fig. 1, a first specific embodiment of the air conditioner load prediction method based on meta-integrated learning of the present invention:
an air conditioner load prediction method based on meta-integrated learning comprises the following steps:
firstly, acquiring time series air conditioner load data;
secondly, processing time sequence air conditioner load data by utilizing a pre-constructed wavelet transformation model to obtain a plurality of air conditioner load sub-signals;
thirdly, constructing a point prediction model with the same number according to the number of the air conditioner load sub-signals;
the point prediction model is based on a stacked coding algorithm, and extracts time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data;
fourth, a pre-built meta-integrated learning model is utilized to output a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data;
and fifthly, performing coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data by utilizing a pre-constructed result prediction model to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integrated learning.
The second specific embodiment of the air conditioner load prediction method based on meta-integrated learning of the invention is as follows:
an air conditioner load prediction method based on meta-integrated learning comprises the following steps:
(1) The basic structure of the prediction model based on the meta-integrated learning is composed of two parts, namely a deterministic prediction part and a meta-integrated learning part, and can be seen in fig. 2.
The deterministic prediction part consists of a wavelet transform based multi-point prediction and stacked coding SAE. The inputs of the point predictors are time series air conditioning load data, and the outputs are predicted values of loads at the next moment. The wavelet transform is responsible for converting the time-series air conditioning load signal into a plurality of smoother sub-signals. The stacked code SAE is used to extract the time series characteristics of each sub-signal, and adding the prediction results of each sub-signal is the prediction result of each point predictor.
The meta-integration learning is used to update the weight coefficients of the output of each point predictor. It consists of a feature selector and a weight updater. The feature selector is used for selecting main feature variables in the air conditioner load data, including environment variables and time sequence feature variables. The weight updater updates the weight of each point predictor in real time, thereby improving prediction accuracy.
(2) A stacked-coding SAE based point predictor, which is a wavelet transform and stacked-coding SAE based point prediction model, can be seen in fig. 3.
The wavelet transformation decomposes the time series air conditioner load data into an approximate sub-signal and a plurality of detail sub-signals, then establishes an independent stacked automatic coding predictor for each sub-signal, is used for extracting the time series characteristics of the sub-signals and realizing the prediction of the sub-signals, and finally, the result of each sub-signal can be synthesized to obtain the prediction result of the point prediction structure.
One specific embodiment of the wavelet transform of the present invention:
the application uses Mallat Discrete Wavelet Transform (DWT) to decompose the air load. The DWT uses two low pass filters and two high pass filters to calculate wavelet coefficients. The formula is as follows:
wherein the method comprises the steps ofThe 4 th Daubechies function is chosen as the mother wavelet in this application because it provides a good balance between wavelength and smoothness. m and n are->And a translation variable. Further, T and T represent the discrete time index and the length of the original signal f (T), respectively.
The invention is based on a specific embodiment of a stacked coded SAE point predictor:
the stacked coding SAE is mainly used for prediction of the approximation sub-signal and the detail sub-signal. Stacked coding SAE is a typical feature extraction deep learning algorithm, which consists of a plurality of automatic encoders. The basic structure of an automatic encoder is similar to a single layer hidden layer sensor, with the input layer dimension being the same as the output layer dimension, and it includes an encoding process (input layer to hidden layer) and a decoding process (hidden layer to output layer). The encoding process maps the input data to a feature space and the decoding process maps the encoded data back to a sample space.
For an input signal, the encoder is expressed mathematically as:
y＝f θ (Wx+b) (2)
here, y represents the output, x is the input, θ is the stacked code SAE parameter consisting of the weight W and the bias b, and f is the activation function. Sigmoid is generally used and is regarded as an activation function.
The decoder is expressed mathematically as:
y′＝g θ′ (W′x+b′) (3)
where θ 'represents a decoder parameter, is composed of a weight W' and a bias b ', g represents an activation function, y' is the output of the decoder, and x is the input of the decoder.
The invention relates to a specific embodiment of meta-integrated learning:
the meta-integrated learning section includes a feature selector and a weight updater. The feature selector is mainly used for selecting features in the input data. These characteristics include environmental variables such as humidity and temperature, and data characteristic variables such as slope of air conditioning load change and current air conditioning load level. In addition, the feature selector is also used to increase or decrease the feature variable during the error adjustment process, so as to keep the prediction error at a low level. The weight updater is mainly used for updating the weight of each point predictor in real time.
The weight updater first combines input features in a linear fashion to form an intermediate variable z i (a p ) As shown in the following formula:
wherein a is p Representing a feature input; omega ij Representing the weight between the ith feature and the jth z; m is the number of featuresAmount of the components.
The variable z is then normalized using the softmax activation function as follows:
wherein a is p Representing a feature input; c (C) i (x) A weight coefficient representing an i-th point predictor; e represents an exponential function; m is the number of point predictors.
At this time, the variable C is normalized i (x) May be regarded as weights based on the point predictor of the stacked coded SAE. The structure of the weight updater is shown in fig. 4.
The present invention is based on one specific embodiment of a stacked coded SAE point predictor training procedure:
the training process of the point predictor mainly trains a prediction structure based on stacked coding SAE, and obtains weight and deviation parameters thereof. Each stacked coded SAE prediction structure is trained separately. The approximation sub-signal or detail sub-signal samples are used as inputs. The output is the predicted value of the sub-signal at the next time step. The whole training process based on the predictive model of the stacked-coding SAE is divided into a pre-training process and a fine-tuning process.
The a pre-training process includes the following:
the pre-training procedure is to initialize all auto-encoder parameters in the stacked encoding SAE. In an automatic encoder, the purpose of the training process is to make the output as close as possible to the input, whose objective function can be expressed as:
wherein J represents an objective function; n is the number of samples; x is x i For the i-th entry item,for the ith denoising entry, g is the activation function, θ' is the decoder parameter; d represents the weight decay term, λ represents its coefficient, y' is the decoder, W is the weight, and L is the reconstruction error.
The fine tuning process b includes the following:
initial values of parameters in the stacked code SAE are obtained during the pre-training process. The supervised fine tuning of these stacked encoded SAE parameters then requires the use of a gradient descent method until its loss function is minimized, thereby converging the parameters of the stacked encoded SAE and achieving the predictive task. Here, the Back Propagation (BP) algorithm is used to fine tune the parameters of the stacked encoded SAE. During the fine tuning process, the BP algorithm will periodically update the stack-coded SAE parameters. In each update procedure, the prediction result of the stacked code SAE is updated using training samples, and a prediction error can be obtained. The stacked encoded SAE parameters are then updated in a back-propagation manner from top layer to bottom layer. When the loss function cannot be reduced, this means that the stacked coding SAE network has converged. At this point, the entire training process for the stacked code SAE is completed. The objective function in the trimming process is generally represented as follows
Wherein h (x) (i) ) And t (i) The predicted result and the actual output of the regressor are represented, D represents the weight decay term, λ represents its coefficient, and L is the reconstruction error.
One specific embodiment of the training process of meta-integrated learning of the present invention:
the purpose of the training process of meta-integrated learning is to determine parameters in the weight updater to best fit the predicted outcome and the corresponding expected output.
The final prediction result of the proposed prediction model can be described as follows:
wherein F (x) p ) Representing the final prediction result;and C i (x p ) Forecast result representing ith point forecaster and weight thereof
Considering the predictive model in fig. 2, the present invention obtains weight parameters for meta-ensemble learning by minimizing the energy function of the training samples. In this application, the energy function is defined as the sum of the mean square error of the training samples and the weight decay term, as follows:
wherein E is f Is an energy function; y is i Represents the actual air conditioning load of the ith sample, F (x p ) Represents the final prediction result, λ represents its coefficient, and W is the weight.
A flow embodiment of the predictive model of the present invention:
first, temperature, humidity, the number of people in a building, air conditioning load level, first/second order load variation, and solar radiation are used as data features. Then, the air conditioning load data for more than one year is divided into 12 groups, one group per month. Each set of training samples includes samples from 1 to 25 days per month, with the remaining samples being test samples. Subsequently, 12 independent point predictors were designed, where the number of input data was 12 or 24, the number of wavelet decomposition layers was 3 or 4, and the number of stacked coding SAE layers was 4, 7, or 10, respectively. The pre-training process and the fine tuning process are used to optimally adjust the model parameters of the 12 independent point predictors. Subsequently, the weight of the weight updater is obtained according to equations (10) - (11). The entire training process of the predictive model has been completed so far. Finally, the test sample is input into a trained predictive model, and the overall output is the predicted result. An overall flow of the air conditioning load accurate prediction training process based on meta-integrated learning is shown in fig. 5.
A specific embodiment of the invention is used for prediction:
the method predicts the air conditioning load of a certain province, and comprises the following steps:
(1) Data selection
And selecting a certain spring and summer air conditioner load data set as experimental data, and testing the prediction effect of different deep learning algorithms. The data set 2020 includes various types of detailed data such as air conditioning load and weather data.
(2) Data preprocessing
In order to obtain meaningful data characteristics, air conditioning load history data, temperature data and weather condition data most relevant to air conditioning load are selected. Therefore, these 3 features are selected as the main features of the input model.
(3) Model selection
2 different algorithms are selected for comparison, and the influence of different prediction models on the prediction effect is explored. In the experiment, temperature, humidity, the number of people in a building, air conditioning load level, first/second order load change and solar radiation are selected as data characteristics, a counter propagation algorithm and a support vector machine are used for comparison analysis, and 3 performance evaluation indexes are used: average absolute error (MAE), root Mean Square Error (RMSE), average absolute percent error (MAPE) to measure the prediction.
Table 1 evaluation of different prediction methods
According to fig. 6 and table 1, it can be seen that the scatter diagram of the prediction result converges well to the central line, which indicates that the prediction model provided by the application has better prediction capability and prediction stability, and has better prediction performance than the other two reference methods.
An embodiment of a device for applying the method of the invention:
an electronic device, comprising:
one or more processors;
a storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement an air conditioning load prediction method based on meta-integrated learning as described above.
A computer medium embodiment to which the method of the invention is applied:
a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements an air conditioning load prediction method based on meta-integrated learning as described above.
It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method, system, computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described in terms of methods, apparatus (systems), computer program products, flowcharts, and/or block diagrams in accordance with embodiments of the present application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (10)
1. An air conditioner load prediction method based on meta-integrated learning is characterized in that,
the method comprises the following steps:
firstly, acquiring time series air conditioner load data;
secondly, processing time sequence air conditioner load data by utilizing a pre-constructed wavelet transformation model to obtain a plurality of air conditioner load sub-signals;
thirdly, constructing a point prediction model with the same number according to the number of the air conditioner load sub-signals;
the point prediction model is based on a stacked coding algorithm, and extracts time sequence characteristics of air conditioner load sub-signals to obtain a plurality of pieces of air conditioner load prediction data;
fourth, a pre-built meta-integrated learning model is utilized to output a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data;
and fifthly, performing coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data by utilizing a pre-constructed result prediction model to obtain an air conditioner load prediction result of the next period, and completing air conditioner load prediction based on meta-integrated learning.
2. The method for predicting air conditioner load based on meta-integrated learning according to claim 1, wherein,
the method for obtaining the plurality of air conditioner load sub-signals by the wavelet transformation model comprises the following steps:
decomposing the time series air conditioning load data by adopting Mallat discrete wavelet transformation to obtain a plurality of air conditioning load sub-signals, wherein the plurality of air conditioning load sub-signals at least comprise an approximate sub-signal and a plurality of detail sub-signals;
the Mallat discrete wavelet transform uses two low-pass filters and two high-pass filters to calculate wavelet coefficients, whose calculation formula is as follows:
wherein the method comprises the steps ofRepresents a mother wavelet, m and n are +.>T and T represent the discrete time index and the length of the original signal f (T), respectively.
3. The method for predicting air conditioner load based on meta-integrated learning according to claim 2, wherein,
the method for obtaining the plurality of pieces of air conditioner load prediction data by the point prediction model comprises the following steps:
step 31, obtaining an approximate sub-signal and a plurality of detail sub-signals;
step 32, mapping an approximate sub-signal and a plurality of detail sub-signals to a feature space respectively, and coding to obtain sub-signal coding data with corresponding quantity;
step 33, mapping the sub-signal coding data back to the sample space, and decoding to obtain a plurality of predicted sub-signals;
and step 34, combining the plurality of prediction sub-signals to form a plurality of pieces of air conditioner load prediction data.
4. The air conditioner load prediction method based on meta-integrated learning according to claim 3, wherein,
the approximate sub-signal and the detail sub-signals are respectively mapped to the characteristic space, and the calculation formula for coding is as follows:
y＝f θ (Wx+b)
wherein y represents the output sub-signal coding data, x is an approximate sub-signal and a plurality of detail sub-signals, and θ is a point prediction model parameter composed of a weight W and a bias b; f is an activation function;
the sub-signal encoded data is mapped back to the sample space and decoded according to the following calculation formula:
y′＝g θ′ (W′x+b′)
wherein θ 'represents a decoding parameter, and is composed of a weight W' and a bias b ', g represents an activation function, y' is a plurality of prediction sub-signals, and x is sub-signal encoded data.
5. The method for predicting air conditioner load based on meta-integrated learning according to claim 1, wherein,
the method for outputting a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data by using the meta-integrated learning model is as follows:
screening the external environment variable and the characteristic information in the air conditioner load prediction data by utilizing a pre-constructed characteristic selector;
the characteristic information comprises the ambient humidity, the ambient temperature, the slope of the air conditioner load change and the current air conditioner load level;
and updating the weight coefficient of each piece of air conditioner load prediction data in real time through a pre-constructed weight updater according to the characteristic information.
6. The method for predicting air conditioner load based on meta-integrated learning of claim 5, wherein,
the method for updating the weight coefficient of each air conditioner load prediction data in real time by the weight updater is as follows:
the characteristic information is first input in a linear combination to form an intermediate variable z i (a p ) The calculation formula is as follows:
wherein a is p (j) Representing the input characteristic information; omega ij Representing the weight between the ith characteristic information and the jth intermediate variable; m is the number of feature information;
the intermediate variables were then normalized using a softmax activation function, the calculation formula of which is shown below:
wherein a is p Representing the input characteristic information; c (C) i (x) A weight coefficient representing the i-th air conditioner load prediction data; e represents an exponential function; m is the amount of air conditioning load prediction data, zi (a p ) As intermediate variable z i (a p )。
7. The method for predicting air conditioner load based on meta-integrated learning according to claim 1, wherein,
the result prediction model is used for carrying out coupling calculation on a plurality of weight coefficients and a plurality of pieces of air conditioner load prediction data according to the following formula:
wherein F (x) p ) Representing a final air conditioner load prediction result;represents the i-th air conditioner load prediction data, C i (x p ) Representing the i-th weight coefficient.
8. The method for predicting air conditioner load based on meta-integrated learning according to claim 1, wherein,
the method for constructing the point prediction model is as follows:
using the approximate sub-signal or detail sub-signal samples as inputs; taking a predicted value sample of the next time step as output, and calculating the weight and deviation parameters of the predicted value sample; the method is divided into a pre-training process and a fine-tuning process;
the pre-training process is to initialize all automatic encoder parameters in a point prediction model to obtain a parameter initial value;
in an automatic encoder, the purpose of the training process is to approximate the output to the input, the expression of its objective function is as follows:
wherein J represents an objective function; n isNumber of samples; x is x i For the i-th entry item,for the ith denoising entry, g is the activation function, θ' is the decoder parameter; d represents a weight decay term, λ represents a coefficient, y' is a decoder, W is a weight, and L is a reconstruction error; θ is a point prediction model parameter; f is an activation function;
after obtaining the parameter initial value of the point prediction model from the pre-training process, performing parameter fine adjustment by using a gradient descent method until an objective function reaches the minimum, so that the parameters of the point prediction model are converged and a prediction task is realized;
the objective function expression in the trimming process is as follows:
wherein h (x) (i) ) And t (i) The predicted result and the actual output of the regressor are represented, D represents the weight decay term, lambda represents the coefficient, and L is the reconstruction error.
9. An air conditioner load prediction system based on meta-integrated learning is characterized in that,
applying the air conditioner load prediction method based on meta-integrated learning as claimed in any one of claims 1 to 8;
the method comprises a wavelet transformation model block, a point prediction module, a meta-integrated learning module and a result prediction module:
the wavelet transformation model block is used for processing the time sequence air conditioner load data to obtain a plurality of air conditioner load sub-signals;
the point prediction module is used for extracting time sequence characteristics of the air conditioner load sub-signals based on a stacked coding algorithm to obtain a plurality of pieces of air conditioner load prediction data;
the meta-integrated learning module is used for outputting a plurality of weight coefficients corresponding to a plurality of pieces of air conditioner load prediction data;
and the result prediction module is used for carrying out coupling calculation on the plurality of weight coefficients and the plurality of pieces of air conditioner load prediction data to obtain an air conditioner load prediction result of the next period and predicting the air conditioner load based on meta-integration learning.
10. An electronic device, characterized in that,
it comprises the following steps:
one or more processors;
a storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement a meta-ensemble learning-based air conditioning load prediction method as set forth in any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311154933.1A CN117313990A (en) | 2023-09-07 | 2023-09-07 | Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311154933.1A CN117313990A (en) | 2023-09-07 | 2023-09-07 | Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117313990A true CN117313990A (en) | 2023-12-29 |
Family
ID=89272831
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311154933.1A Pending CN117313990A (en) | 2023-09-07 | 2023-09-07 | Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117313990A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117909928A (en) * | 2024-03-19 | 2024-04-19 | 国网四川省电力公司成都供电公司 | Air conditioner load prediction method and system based on big data analysis |
-
2023
- 2023-09-07 CN CN202311154933.1A patent/CN117313990A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117909928A (en) * | 2024-03-19 | 2024-04-19 | 国网四川省电力公司成都供电公司 | Air conditioner load prediction method and system based on big data analysis |
| CN117909928B (en) * | 2024-03-19 | 2024-05-28 | 国网四川省电力公司成都供电公司 | Air conditioner load prediction method and system based on big data analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109214592B (en) | Multi-model-fused deep learning air quality prediction method | |
| CN112488415A (en) | Power load prediction method based on empirical mode decomposition and long-and-short-term memory network | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN107704966A (en) | A kind of Energy Load forecasting system and method based on weather big data | |
| CN111210082A (en) | Optimized BP neural network algorithm-based precipitation prediction method | |
| CN114119273B (en) | Non-invasive load decomposition method and system for park comprehensive energy system | |
| CN107358318A (en) | Based on GM（1,1）The urban power consumption Forecasting Methodology of model and Grey Markov chain predicting model | |
| CN112488396A (en) | Wavelet transform-based electric power load prediction method of Holt-Winters and LSTM combined model | |
| CN103793887B (en) | Short-term electric load on-line prediction method based on self-adaptive enhancement algorithm | |
| CN111915092B (en) | Ultra-short-term wind power prediction method based on long-short-term memory neural network | |
| CN109978283A (en) | Photovoltaic power generation power prediction method based on branch evolution neural network | |
| CN112116162A (en) | Power transmission line icing thickness prediction method based on CEEMDAN-QFAOA-LSTM | |
| CN111079989A (en) | Water supply company water supply amount prediction device based on DWT-PCA-LSTM | |
| CN112001537B (en) | Short-term wind power prediction method based on gray model and support vector machine | |
| CN117313990A (en) | Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment | |
| CN106600041A (en) | Restricted Boltzmann machine-based photovoltaic power generation short-term power probability prediction method | |
| CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
| CN109583645A (en) | A kind of public building short-term load forecasting method | |
| CN112288137A (en) | LSTM short-term load prediction method and device considering electricity price and Attention mechanism | |
| CN111931983A (en) | Precipitation prediction method and system | |
| CN115186923A (en) | Photovoltaic power generation power prediction method and device and electronic equipment | |
| CN116702831A (en) | Hybrid short-term wind power prediction method considering massive loss of data | |
| CN116169670A (en) | Short-term non-resident load prediction method and system based on improved neural network | |
| El Bakali et al. | Data-Based Solar Radiation Forecasting with Pre-Processing Using Variational Mode Decomposition | |
| CN117407660B (en) | Regional sea wave forecasting method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |