CN115271198A - Net load prediction method and device of photovoltaic equipment - Google Patents

Net load prediction method and device of photovoltaic equipment Download PDF

Info

Publication number
CN115271198A
CN115271198A CN202210871446.6A CN202210871446A CN115271198A CN 115271198 A CN115271198 A CN 115271198A CN 202210871446 A CN202210871446 A CN 202210871446A CN 115271198 A CN115271198 A CN 115271198A
Authority
CN
China
Prior art keywords
data
net load
photovoltaic
wavelet
regression model
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
Application number
CN202210871446.6A
Other languages
Chinese (zh)
Inventor
宫飞翔
陈宋宋
龚桃荣
石坤
邓志东
田诺
黄秀彬
刘鲲鹏
李子乾
周颖
袁金斗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Corp of China SGCC
China Electric Power Research Institute Co Ltd CEPRI
State Grid Jiangsu Electric Power Co Ltd
Original Assignee
State Grid Corp of China SGCC
China Electric Power Research Institute Co Ltd CEPRI
State Grid Jiangsu Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Corp of China SGCC, China Electric Power Research Institute Co Ltd CEPRI, State Grid Jiangsu Electric Power Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN202210871446.6A priority Critical patent/CN115271198A/en
Publication of CN115271198A publication Critical patent/CN115271198A/en
Pending legal-status Critical Current

Links

Images

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
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • 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
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Operations Research (AREA)
  • Algebra (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Power Engineering (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to the technical field of power system photovoltaic net load prediction, and particularly provides a net load prediction method and a net load prediction device of photovoltaic equipment, wherein the net load prediction method and the net load prediction device comprise the following steps: forecasting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model; performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment; wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients. According to the technical scheme provided by the invention, the dimensionality reduction of data and the high-precision prediction of photovoltaic load can be effectively realized, the analysis of various external characteristics and power utilization behaviors can be realized, and the high-dimensional data variable screening and high-precision prediction are realized, so that the operation plan of a power system is reasonably arranged.

Description

Net load prediction method and device of photovoltaic equipment
Technical Field
The invention relates to the technical field of power system photovoltaic net load prediction, in particular to a net load prediction method and device of photovoltaic equipment.
Background
The net load of the power distribution network refers to a difference value between a user power load and a renewable energy power generation load, namely a load value provided by a main power grid of a power system to the power distribution network, and particularly refers to a difference value between a household power load and household photovoltaic equipment for converting electric energy by utilizing sunlight. At present, aiming at relatively less analysis and prediction of net load data, two methods are provided for predicting net load, one is to predict electric load and renewable energy power generation respectively, then difference is obtained, namely indirect prediction is carried out, and the other is to select a proper model to carry out prediction directly according to a net load historical sequence, namely direct prediction is carried out. According to the classification of prediction results, the load prediction can be classified into short-term load prediction and medium-and long-term load prediction. For short-term load prediction, domestic and foreign research methods are roughly divided into two categories: one is a conventional method represented by a time series method; the other is a new artificial intelligence method represented by an artificial neural network.
The method comprises the following processes of auto-regression (AR) process; a sliding-averaging (MA) process; an autoregressive moving average (ARMA) process; an integral autoregressive sliding (ARIMA) process; sequences modeled with a Transfer Function (TF). The classical time series method has relatively less requirements on historical load data, manual intervention is not needed much, so that workload is relatively less, the whole prediction process is excellent in calculation speed and can be automatically completed, and the advantages of the method are achieved. The method has the defects that the method depends on the data of the power load data, and other variable factors are not sufficiently processed, so that higher prediction precision cannot be achieved, and the method is generally only suitable for load prediction which is stable and uniform per se. The regression analysis method is also one of the classical prediction methods based on the historical data of the load, and only the regression analysis method also takes into consideration the external factors influencing the load on the basis of the historical data. Its advantages are simple and convenient operation, and no real reflection of actual load and influence factor. The wavelet analysis and prediction method is to select and classify wavelets, distinguish loads with different properties, select a certain load according to different properties, analyze the rule of the certain load to determine to adopt a corresponding prediction method, respectively predict the selected load, decompose a sequence, and then reconstruct the sequence, thereby finally achieving the purpose of prediction. The wavelet prediction method can observe details in signals, particularly some singular signals in the signals, the response of the wavelet prediction method is particularly sensitive, and abrupt or some weak signals can be well processed.
The latter method mainly uses machine learning and deep learning methods to predict load at present, and can predict nonlinear and non-stationary data more accurately. Among them, artificial Neural Networks (ANN) and Recurrent Neural Networks (RNN) are typical representatives of this type of method. The ANN theory is widely used for the research of short-term load prediction, and has the outstanding advantages of self-adapting function to a large number of non-structural and non-precise laws, and the characteristics of information memory, autonomous learning, knowledge reasoning and optimized calculation. The ANN has strong self-learning and complex nonlinear function fitting capability and is very suitable for the problem of power load prediction, but the research process also shows that the ANN method has the problems of local optimization, large generalization error, difficult determination of hidden unit number and the like. Compared with the traditional neural network, the RNN model has great advantages by introducing a time sequence concept, can store information learned before time in the network, can utilize the information of the previous time when processing data at each time, and has continuity in information transmission, so that the RNN can well process the problem of modeling of periodic data such as time sequences and the like. However, if information different from the current time is used, the RNN often cannot make full use of the information, which is the problem of gradient disappearance of the RNN, and in practice the RNN can only process information for a very limited time. To solve the problem of RNN gradient disappearance, variants of RNN such as long-short-term memory neural network (LSTM), gated cycle unit (GRU), etc. were proposed one after another, and consequently subsequently, variants of LSTM such as Q-RNN, etc. were proposed, the above algorithm proved to have better accuracy or faster speed in load prediction than RNN. As can be seen from the above description, the application ranges and performance advantages of different prediction methods are different, and in addition, since the net load is affected by factors such as temperature and illumination, a more accurate result cannot be obtained by using a single algorithm to predict the net load.
Disclosure of Invention
In order to overcome the defects, the invention provides a method and a device for predicting the net load of photovoltaic equipment.
In a first aspect, a method for predicting a net load of a photovoltaic device is provided, and the method for predicting the net load of the photovoltaic device includes:
forecasting the sparseness of wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model;
performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
Preferably, the calculation formula of the low-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000021
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000022
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000023
is composed of
Figure BDA0003760904260000024
The corresponding coefficients of the decomposition are,
Figure BDA0003760904260000025
a wavelet function selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic installation,
Figure BDA0003760904260000026
is psimnCorresponding decomposition coefficient, #mnIs prepared by reacting with
Figure BDA0003760904260000031
Complementary wavelet functions of (1).
Further, the
Figure BDA0003760904260000032
The corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000033
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000034
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
Preferably, the obtaining process of the pre-constructed regression model includes:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional characteristic space by adopting a kernel method to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until model indexes of the trained Lasso regression model meet convergence conditions to obtain the pre-constructed regression model.
Further, in the process of training the initial regression model by using the training data, the regression model loss function is calculated as follows:
LReg(β)＝LOLS(β)+P
in the above formula, LRegAs a penalty function, LOLSIs a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
Further, the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is an X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
Further, the model index includes at least one of: mean absolute percentage error, mean square error.
Further, the average absolute percentage error is calculated as follows:
Figure BDA0003760904260000035
the mean square error is calculated as follows:
Figure BDA0003760904260000041
in the above formula, MSE is the average absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure BDA0003760904260000042
and MAPE is a predicted value of the ith sample data in the verification data, and is a mean square error.
In a second aspect, there is provided a net load prediction device for a photovoltaic apparatus, including:
the prediction module is used for predicting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional characteristic space by utilizing a pre-constructed regression model;
the reconstruction module is used for performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
Preferably, in the prediction module, the calculation formula of the low-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000043
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000044
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000045
is composed of
Figure BDA0003760904260000046
The corresponding coefficients of the decomposition are,
Figure BDA0003760904260000047
a wavelet function selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic installation,
Figure BDA0003760904260000048
is psimnCorresponding decomposition coefficient,/mnIs and is
Figure BDA0003760904260000049
Complementary wavelet functions of (1).
Further, the
Figure BDA00037609042600000410
The corresponding decomposition coefficients are calculated as follows:
Figure BDA00037609042600000411
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure BDA00037609042600000412
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
Preferably, in the prediction module, the obtaining process of the pre-constructed regression model includes:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional feature space by adopting a kernel device to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until the model index of the trained Lasso regression model meets the convergence condition to obtain the pre-constructed regression model.
Further, in the process of training the initial regression model by using the training data, the regression model loss function is calculated as follows:
LReg(β)＝LOLS(β)+P
in the above formula, LRegAnd the LOLS is a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
Further, the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is a X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
Further, the model index includes at least one of: mean absolute percentage error, mean square error.
Further, the average absolute percentage error is calculated as follows:
Figure BDA0003760904260000051
the mean square error is calculated as follows:
Figure BDA0003760904260000052
in the above formula, MSE is the average absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure BDA0003760904260000053
and MAPE is the predicted value of the ith sample data in the verification data and is the mean square error.
In a third aspect, a computer device is provided, comprising: one or more processors;
the processor to store one or more programs;
the one or more programs, when executed by the one or more processors, implement the method for net load prediction for a photovoltaic device.
In a fourth aspect, a computer-readable storage medium is provided, having stored thereon a computer program that, when executed, performs the method of predicting the net load of a photovoltaic plant.
One or more technical schemes of the invention at least have one or more of the following beneficial effects:
the invention provides a net load prediction method and a net load prediction device for photovoltaic equipment, wherein the net load prediction method comprises the following steps: forecasting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model; performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment; wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients. According to the technical scheme provided by the invention, the dimensionality reduction of data and the high-precision prediction of photovoltaic load can be effectively realized, the analysis of various external characteristics and power utilization behaviors can be realized, and the high-dimensional data variable screening and high-precision prediction are realized, so that the operation plan of a power system is reasonably arranged;
furthermore, the technical scheme provided by the invention introduces a prediction model combining wavelet transformation and a Lasso regression model, wherein the wavelet transformation exchanges the time domain and the frequency domain of time series data, focuses on the details of the data, and is more suitable for describing the intrinsic characteristics of a photovoltaic net load, and a kernel method is introduced into the Lasso regression model to map the original data to a suitable high-dimensional characteristic space, so that the Lasso regression model is applied to nonlinear photovoltaic net load data, the accurate prediction of the photovoltaic load is finally realized, and the stable operation of a large power grid is maintained.
Drawings
FIG. 1 is a schematic flow chart of the main steps of a method for predicting the net load of a photovoltaic plant according to an embodiment of the present invention;
fig. 2 is a main block diagram of a net load prediction device of a photovoltaic apparatus according to an embodiment of the present invention.
Detailed Description
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
As disclosed in the background art, in order to achieve the dual-carbon target and alleviate the current situation of shortage of electricity in various areas, the use of household photovoltaic devices is currently carried out in a plurality of areas, and households with photovoltaic devices can generate electricity through the photovoltaic devices under the illumination situation, so that self-sufficiency of electricity can be realized when conditions are appropriate, and even the electricity is reversely transmitted to a power distribution network to alleviate the power supply pressure of the power grid.
In order to improve the above problem, the method and apparatus for predicting the net load of a photovoltaic device provided by the present invention include: forecasting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model; performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment; wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients. According to the technical scheme provided by the invention, the dimensionality reduction of data and the high-precision prediction of photovoltaic load can be effectively realized, the analysis of various external characteristics and power utilization behaviors can be realized, and the high-dimensional data variable screening and high-precision prediction are realized, so that the operation plan of a power system is reasonably arranged;
furthermore, the technical scheme provided by the invention introduces a prediction model combining wavelet transformation and a Lasso regression model, wherein the wavelet transformation exchanges the time domain and the frequency domain of time series data, focuses on the details of the data, and is more suitable for describing the intrinsic characteristics of a photovoltaic net load, and a kernel method is introduced into the Lasso regression model to map the original data to a suitable high-dimensional characteristic space, so that the Lasso regression model is applied to nonlinear photovoltaic net load data, the accurate prediction of the photovoltaic load is finally realized, and the stable operation of a large power grid is maintained. The above scheme is explained in detail below.
Example 1
Referring to fig. 1, fig. 1 is a schematic flow chart of main steps of a method for predicting a net load of a photovoltaic device according to an embodiment of the present invention. As shown in fig. 1, the method for predicting the net load of a photovoltaic device in the embodiment of the present invention mainly includes the following steps:
step S101: forecasting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model;
step S102: performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
The wavelet transform mainly includes a continuous wavelet transform and a discrete wavelet transform. Since the payload at the present time is predicted from the past data at the same time in the present embodiment, the present embodiment decomposes the original payload time-series data using discrete wavelet transform.
Decomposing the input photovoltaic net load time series, wherein after decomposition, an approximate series and a plurality of detail series can be generated from the original signal. The approximate series represents low-frequency coefficients (appoximate coefficients) containing trend information of photovoltaic net loads; the detail series represents high frequency coefficients (coefficients) that contain characteristics of the influencing factors related to the photovoltaic net load. Wherein the calculation formula of the low-frequency coefficient of the net load of the photovoltaic equipment is as follows:
Figure BDA0003760904260000071
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000072
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000073
is composed of
Figure BDA0003760904260000074
The corresponding coefficients of the decomposition are then compared,
Figure BDA0003760904260000075
wavelet functions selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000076
is phimnCorresponding decomposition coefficient, #mnIs prepared by reacting with
Figure BDA0003760904260000077
Complementary wavelet functions of (1).
Further, the
Figure BDA0003760904260000078
The corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000081
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000082
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
In the time series processing, daubechies (db) is better applied, so in this embodiment, daubechies wavelet transform is selected to decompose the original payload data, and after the Lasso regression model is predicted, wavelet reconstruction is further required to restore the coefficient data to the original payload data.
In this embodiment, a linear regression method of kernel Lasso is used to predict the low-frequency coefficient and the high-frequency coefficient of the photovoltaic payload obtained by wavelet transform. Compared with the common linear regression, the Lasso can enhance the fitting effect of the low-frequency coefficient and the high-frequency coefficient of the photovoltaic net load, and the Lasso can be popularized to the nonlinear photovoltaic net load data by using the kernel function.
The pre-constructed regression model obtaining process comprises the following steps:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional characteristic space by adopting a kernel method to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until model indexes of the trained Lasso regression model meet convergence conditions to obtain the pre-constructed regression model.
Further, the basis of selecting the function by the conventional regression method is that the function has the smallest sum of squares (i.e., variance) of the difference between the low and high frequency coefficients obtained by the independent variable and the actual value, but this method is easy to cause overfitting. In this embodiment, a penalty function is introduced on the basis of the standard loss function, a shrinkage parameter is multiplied before the absolute value of the coefficient, the coefficient value is shrunk by this method to reduce the variance, and the regression model loss function is calculated in the process of training the initial regression model by using the training data as follows:
LReg(β)＝LOLS(β)+P
in the above formula, LRegIs punished byPenalty loss function, LOLSIs a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
Wherein the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is a X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
The low and high frequency coefficients of the photovoltaic net load obtained by wavelet transformation are predicted through the Lasso model, overfitting can be effectively prevented, and continuous or discrete Lasso can be processed no matter in the condition of less training data or over-high dimension.
On the other hand, the low and high frequency coefficients of the photovoltaic net load obtained by wavelet transformation may be nonlinear, while the conventional Lasso method cannot process nonlinear data, and for the data types that cannot be processed by linear regression, the original data can be expanded to a high-dimensional space, and linear regression can be performed in the high-dimensional space.
The Kernel Method (KMs) is a class of pattern recognition algorithms. The purpose is to find and learn relationships to each other in a set of data. The kernel methods with wide application range include support vector machine, gaussian process, etc.
By the nuclear method, the nonlinear photovoltaic net load low-high frequency coefficient can be embedded into a proper high-dimensional characteristic space; the patterns are then analyzed and processed in this new space using a general linear learner. And the existence of the Kernel function (Kernel function) can make synchronous computation by hiding the nonlinear mapping in a linear learner, so that the computation complexity is independent of the dimension of the high-dimensional feature space.
And expanding the photovoltaic net load low-high frequency coefficient data to the data of the high-dimensional feature space through a kernel function to perform Lasso fitting, thereby obtaining the sparse solution of the high-dimensional feature space.
In an optimal implementation mode, kernel Lasso regression prediction is mainly performed on each dimension of 96-dimensional net load data, discrete wavelet transform is required to be performed on the photovoltaic net load data of each dimension to obtain a high-frequency coefficient D and a low-frequency coefficient A, then training and prediction are performed on the D and A of each dimension to obtain predicted D and A, wavelet reconstruction is performed to obtain a predicted value of one dimension, then the predicted values of 96 dimensions are combined to obtain a one-day prediction condition, and then the prediction condition is compared with an actual value, and a prediction effect is analyzed.
In this embodiment, the model index includes at least one of the following: mean absolute percentage error, mean square error.
In one embodiment, the mean absolute percentage error is calculated as follows:
Figure BDA0003760904260000091
the mean square error is calculated as follows:
Figure BDA0003760904260000092
in the above formula, MSE is the average absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure BDA0003760904260000093
and MAPE is a predicted value of the ith sample data in the verification data, and is a mean square error.
In an optimal implementation mode, the overall prediction effect is good when the method for predicting the net load of the photovoltaic equipment is adopted for prediction, and the weekly prediction indexes of a certain distribution area aiming at 2021/5/12-2021/5/18 are shown as shown in table 1:
TABLE 1
Figure BDA0003760904260000101
Therefore, the prediction effect of the net load prediction method of the photovoltaic equipment is good.
Then, selecting wavelet transform and kernel Lasso regression algorithm to predict each station area, taking historical data before the week to be predicted as a training set, performing week prediction on the week after training, wherein the result index values of the week prediction from 2021/1/1-2021/1/7 are shown in table 2:
TABLE 2
Figure BDA0003760904260000102
Therefore, the kernel Lasso regression algorithm has a good prediction effect on most of the transformer areas, but the effect on the transformer areas 5 and 9 is poor, and analysis shows that the difference between the prediction values and the actual values of the transformer areas 5 and 9 is mainly reflected on the peak value of the photovoltaic net load, and the difference between the peak value and the values in the vicinity of the peak value is large, so that the prediction effect is finally caused.
Further, using the effect of wavelet transform
Finally, a comparison of predictions using only the kernel Lasso regression model and using wavelet transform in combination with the kernel Lasso regression model revealed that the latter in some data regions had much improved accuracy over the former, and the weekly prediction index results at 2021/5/5-2021/5/11 in a certain region are shown below in table 3:
Figure BDA0003760904260000111
the method has good effect, and is mainly characterized in that the wavelet transformation energy conversion divides a time sequence into a more stable and smooth coefficient sequence, so that the prediction of a subsequent kernel Lasso regression model is facilitated. Therefore, by combining the kernel Lasso regression model based on wavelet transformation with the kernel Lasso regression model only, that is, selecting a model with smaller index value according to the last predicted index value for prediction, and continuing, the accuracy of the overall prediction can be improved.
Example 2
Based on the same inventive concept, the present invention further provides a device for predicting a net load of a photovoltaic device, as shown in fig. 2, the device for predicting a net load of a photovoltaic device includes:
the prediction module is used for predicting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional characteristic space by utilizing a pre-constructed regression model;
the reconstruction module is used for performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
Preferably, in the prediction module, the calculation formula of the low-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000112
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure BDA0003760904260000121
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000122
is composed of
Figure BDA0003760904260000123
The corresponding coefficients of the decomposition are,
Figure BDA0003760904260000124
wavelet functions selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic device,
Figure BDA0003760904260000125
is psimnCorresponding decomposition coefficient, #mnIs prepared by reacting with
Figure BDA0003760904260000126
Complementary wavelet functions of (a).
Further, the
Figure BDA0003760904260000127
The corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000128
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure BDA0003760904260000129
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
Preferably, in the prediction module, the obtaining process of the pre-constructed regression model includes:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional feature space by adopting a kernel device to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until model indexes of the trained Lasso regression model meet convergence conditions to obtain the pre-constructed regression model.
Further, in the process of training the initial regression model by using the training data, the regression model loss function is calculated as follows:
LReg(β)＝LOLS(β)+P
in the above formula, LRegAnd for the penalized loss function, LOLS is a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
Further, the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is a X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
Further, the model index includes at least one of: mean absolute percentage error, mean square error.
Further, the average absolute percentage error is calculated as follows:
Figure BDA0003760904260000131
the mean square error is calculated as follows:
Figure BDA0003760904260000132
in the above formula, MSE is the mean absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure BDA0003760904260000133
and MAPE is the predicted value of the ith sample data in the verification data and is the mean square error.
Example 3
Based on the same inventive concept, the present invention also provides a computer apparatus comprising a processor and a memory, the memory being configured to store a computer program comprising program instructions, the processor being configured to execute the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is specifically adapted to implement one or more instructions, and to load and execute one or more instructions in a computer storage medium to implement a corresponding method flow or a corresponding function, so as to implement the steps of the net load prediction method of the photovoltaic device in the above embodiments.
Example 4
Based on the same inventive concept, the present invention further provides a storage medium, in particular, a computer-readable storage medium (Memory), which is a Memory device in a computer device and is used for storing programs and data. It is understood that the computer readable storage medium herein can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer readable storage medium may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the steps of a method for predicting a net load of a photovoltaic device in the above embodiments.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention 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 invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams 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 solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (18)

1. A method of net load prediction for a photovoltaic device, the method comprising:
forecasting the sparseness of wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space by utilizing a pre-constructed regression model;
performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
2. The method of claim 1, wherein the low frequency coefficient of the net load of the photovoltaic device is calculated as follows:
Figure FDA0003760904250000011
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure FDA0003760904250000012
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure FDA0003760904250000013
is composed of
Figure FDA0003760904250000014
The corresponding coefficients of the decomposition are then compared,
Figure FDA0003760904250000015
a wavelet function selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic installation,
Figure FDA0003760904250000016
is psimnCorresponding decomposition coefficient,/mnIs and is
Figure FDA0003760904250000017
Complementary wavelet functions of (1).
3. The method of claim 2, wherein the method is as set forth in claim 2
Figure FDA0003760904250000018
The corresponding decomposition coefficients are calculated as follows:
Figure FDA0003760904250000019
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure FDA00037609042500000110
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
4. The method of claim 1, wherein the pre-constructed regression model obtaining process comprises:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional characteristic space by adopting a kernel method to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until model indexes of the trained Lasso regression model meet convergence conditions to obtain the pre-constructed regression model.
5. The method of claim 4, wherein in the training of the initial regression model using the training data, the regression model loss function is calculated as follows:
LReg(β)＝LOLS(β)+P
in the above formula, LRegAs a penalized loss function, LOLSIs a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
6. The method of claim 5, wherein the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is an X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
7. The method of claim 4, wherein the model metrics include at least one of: mean absolute percentage error, mean square error.
8. The method of claim 7, wherein the average absolute percentage error is calculated as follows:
Figure FDA0003760904250000021
the mean square error is calculated as follows:
Figure FDA0003760904250000022
in the above formula, MSE is the average absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure FDA0003760904250000023
and MAPE is a predicted value of the ith sample data in the verification data, and is a mean square error.
9. An apparatus for predicting a net load of a photovoltaic device, the apparatus comprising:
the prediction module is used for predicting sparse solution of wavelet data of net load of the photovoltaic equipment in a high-dimensional characteristic space by utilizing a pre-constructed regression model;
the reconstruction module is used for performing wavelet reconstruction on sparse solution of the wavelet data of the net load of the photovoltaic equipment in a high-dimensional feature space to obtain net load prediction data of the photovoltaic equipment;
wherein the wavelet data comprises: low frequency coefficients and high frequency coefficients.
10. The apparatus of claim 9, wherein the low frequency coefficients of the net load of the photovoltaic device are calculated in the prediction module as follows:
Figure FDA0003760904250000024
the calculation formula of the high-frequency coefficient of the net load of the photovoltaic device is as follows:
Figure FDA0003760904250000025
in the above formula, A is the low frequency coefficient of the net load of the photovoltaic device,
Figure FDA0003760904250000031
is composed of
Figure FDA0003760904250000032
The corresponding coefficients of the decomposition are then compared,
Figure FDA0003760904250000033
a wavelet function selected for the corresponding scaling constant m and translation constant n, t being the current moment, D being the high frequency coefficient of the net load of the photovoltaic installation,
Figure FDA0003760904250000034
is psimnCorresponding decomposition coefficient, #mnIs prepared by reacting with
Figure FDA0003760904250000035
Complementary wavelet functions of (1).
11. The apparatus of claim 10, wherein the apparatus is characterized in that
Figure FDA0003760904250000036
The corresponding decomposition coefficients are calculated as follows:
Figure FDA0003760904250000037
the psimnThe corresponding decomposition coefficients are calculated as follows:
Figure FDA0003760904250000038
in the above formula, T is the net load sequence length of the photovoltaic device, ptThe net load of the photovoltaic device at time t.
12. The apparatus of claim 9, wherein the prediction module is configured to obtain the pre-constructed regression model by:
performing wavelet transformation on the historical net load data of the photovoltaic equipment to obtain wavelet data of the historical net load data of the photovoltaic equipment;
expanding wavelet data of historical net load data of the photovoltaic equipment to a high-dimensional characteristic space by adopting a kernel device to obtain linear data corresponding to the wavelet data of the historical net load data of the photovoltaic equipment;
constructing training data and verification data by using the linear data;
and training an initial Lasso regression model by using the training data, and verifying the trained Lasso regression model by using the verification data until model indexes of the trained Lasso regression model meet convergence conditions to obtain the pre-constructed regression model.
13. The apparatus of claim 12, wherein the regression model loss function is calculated as follows during the training of the initial regression model using the training data:
LReg(β)＝LOLS(β)+P
in the above formula, LRegAs a penalty function, LOLSIs a standard loss function, P is a penalty function value, and beta is a regression coefficient vector.
14. The apparatus of claim 13, wherein the standard loss function is calculated as follows:
LOLS(β)＝||Y-Xβ||2
in the above formula, Y is a X × 1-dimensional predictor variable matrix, X is an X × p-dimensional result variable vector, X is the number of observed values, and p is the number of predictor variables.
15. The apparatus of claim 12, wherein the model metrics comprise at least one of: mean absolute percentage error, mean square error.
16. The apparatus of claim 15, wherein the mean absolute percentage error is calculated as follows:
Figure FDA0003760904250000041
the mean square error is calculated as follows:
Figure FDA0003760904250000042
in the above formula, MSE is the average absolute percentage error, I is the total number of sample data in the verification data, yiFor the actual value of the ith sample data in the validation data,
Figure FDA0003760904250000043
and MAPE is the predicted value of the ith sample data in the verification data and is the mean square error.
17. A computer device, comprising: one or more processors;
the processor to store one or more programs;
the one or more programs, when executed by the one or more processors, implement a method of net load prediction for a photovoltaic device as recited in any of claims 1 to 8.
18. A computer-readable storage medium, on which a computer program is stored which, when executed, carries out a method of net load prediction of a photovoltaic plant according to any one of claims 1 to 8.
CN202210871446.6A 2022-07-22 2022-07-22 Net load prediction method and device of photovoltaic equipment Pending CN115271198A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210871446.6A CN115271198A (en) 2022-07-22 2022-07-22 Net load prediction method and device of photovoltaic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210871446.6A CN115271198A (en) 2022-07-22 2022-07-22 Net load prediction method and device of photovoltaic equipment

Publications (1)

Publication Number Publication Date
CN115271198A true CN115271198A (en) 2022-11-01

Family

ID=83770223

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210871446.6A Pending CN115271198A (en) 2022-07-22 2022-07-22 Net load prediction method and device of photovoltaic equipment

Country Status (1)

Country Link
CN (1) CN115271198A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117742135A (en) * 2024-02-09 2024-03-22 石家庄学院 Photovoltaic energy-saving control method and system for communication machine room

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117742135A (en) * 2024-02-09 2024-03-22 石家庄学院 Photovoltaic energy-saving control method and system for communication machine room

Similar Documents

Publication Publication Date Title
Moreno et al. Wind speed forecasting approach based on singular spectrum analysis and adaptive neuro fuzzy inference system
Yin et al. Multi-temporal-spatial-scale temporal convolution network for short-term load forecasting of power systems
Dong et al. Data-driven scenario generation of renewable energy production based on controllable generative adversarial networks with interpretability
Chen et al. Short-term wind speed forecasting based on long short-term memory and improved BP neural network
Catalão et al. Short-term electricity prices forecasting in a competitive market by a hybrid intelligent approach
Li et al. A hybrid short-term building electrical load forecasting model combining the periodic pattern, fuzzy system, and wavelet transform
CN111079989B (en) DWT-PCA-LSTM-based water supply amount prediction device for water supply company
CN110264012B (en) Renewable energy power combination prediction method and system based on empirical mode decomposition
CN109886464A (en) The low information loss short-term wind speed forecasting method of feature set is generated based on optimization singular value decomposition
Zhu et al. Hybrid of EMD and SVMs for short-term load forecasting
Hafeez et al. Short term load forecasting based on deep learning for smart grid applications
CN116843083A (en) Carbon emission prediction system and method based on hybrid neural network model
Alamaniotis Synergism of deep neural network and elm for smart very-short-term load forecasting
Li et al. Short-term wind power prediction based on modal reconstruction and CNN-BiLSTM
Al-Ja’afreh et al. An enhanced CNN-LSTM based multi-stage framework for PV and load short-term forecasting: DSO scenarios
Wu et al. Explainable temporal dependence in multi-step wind power forecast via decomposition based chain echo state networks
CN115271198A (en) Net load prediction method and device of photovoltaic equipment
Huang et al. Short-term load forecasting based on a hybrid neural network and phase space reconstruction
Zhang et al. Deep reinforcement learning based interpretable photovoltaic power prediction framework
CN117592615A (en) Short-term power load prediction method based on ICEEMDAN and integrated LSTM-TCN model
Arshad et al. Wind power prediction using genetic programming based ensemble of artificial neural networks (GPeANN)
CN117520818A (en) Power fingerprint identification method and device, electronic equipment and storage medium
CN116780506A (en) Household micro-grid energy management method, device, equipment and storage medium
Gobu et al. Two-stage machine learning framework for simultaneous forecasting of price-load in the smart grid
CN117134315A (en) Distribution transformer load prediction method and device based on BERT algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication