CN112990556A - User power consumption prediction method based on Prophet-LSTM model - Google Patents

User power consumption prediction method based on Prophet-LSTM model Download PDF

Info

Publication number
CN112990556A
CN112990556A CN202110209867.8A CN202110209867A CN112990556A CN 112990556 A CN112990556 A CN 112990556A CN 202110209867 A CN202110209867 A CN 202110209867A CN 112990556 A CN112990556 A CN 112990556A
Authority
CN
China
Prior art keywords
data
model
prediction
prophet
time
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
CN202110209867.8A
Other languages
Chinese (zh)
Inventor
汪洋
张慧
刘超
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.)
Jiangsu University
Original Assignee
Jiangsu University
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 Jiangsu University filed Critical Jiangsu University
Priority to CN202110209867.8A priority Critical patent/CN112990556A/en
Publication of CN112990556A publication Critical patent/CN112990556A/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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • 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)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Educational Administration (AREA)
  • Operations Research (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a Prophet-LSTM model-based user power consumption prediction method, which comprises the following steps: s1, acquiring historical data of power consumption of a user through an intelligent electric meter, wherein the historical data comprises time sequence data, weather temperature data and holiday data; s2, preprocessing and normalizing historical data: the original power consumption data is as follows: x ═ X1，x2，...，xnPreprocessing the original data, including processing missing values, abnormal values, repeated values and invalid values; s3, constructing a Prophet prediction model, and processing the historical power consumption energy consumptionData X ═ X'1，x′2，...，x′nInputting the predicted data into a Prophet model to predict the Prophet; s4, in order to prevent prediction overfitting, combining an improved long-time memory network LSTM model to perform combined prediction; and S5, measuring and verifying the fitting degree and the prediction effect of the combined model, and using a common evaluation index. The method analyzes the characteristics and the rules of the power consumption data, improves the accuracy of the prediction model, and has important guiding significance for national power grids and various power supply companies to formulate effective power supply services.

Description

User power consumption prediction method based on Prophet-LSTM model
Technical Field
The invention relates to the field of time series analysis and energy consumption prediction, in particular to a Prophet-LSTM model-based user power consumption prediction method.
Background
The power consumption of the user is analyzed and predicted, whether the abnormal condition of the power consumption of the user occurs or not can be judged and a corresponding solution is provided for a national power grid or a power supply company, the related power supply company can adjust a power supply decision scheme plan in time according to the prediction trend of the power consumption, the efficiency and the reliability of power supply service are improved, the development of energy conservation and emission reduction consciousness is promoted, and a power-saving society is constructed. Many scholars have studied in this respect, but the prediction of the power consumption and energy consumption of users is comprehensively influenced by a plurality of factors such as power consumption behaviors of users, load changes, holidays, seasonal changes and the like, so that the trend of unbalanced time series changes, and the commonly used prediction model does not finely decompose data, so that the prediction result is poor. Therefore, establishing an efficient user electricity consumption prediction model is one of the hotspots of research in the power field. Prophet is a time series prediction model, and is initially used for business prediction, such as the economic and financial industry. The method has the advantages of simple operation, low complexity of a parameter model, short calculation prediction time, good prediction effect and the like, is rapidly popular in various fields, but the Prophet model has the defect of getting into overfitting at a special time point and has the defect of showing the composite characteristic of a time sequence, so that the prediction effect is poor aiming at the defect of a single prediction model, and the strategy of combining the improved LSTM model and the Prophet model is provided, the advantages of the two models are fused, and the prediction error is reduced.
Disclosure of Invention
The invention aims at the technical problems in the prior art, processes, analyzes and predicts the historical energy consumption data of the user electricity consumption, and provides a method for predicting the user electricity consumption based on a Prophet-LSTM model.
In order to achieve the purpose of the invention, the invention adopts the following technical scheme:
a Prophet-LSTM model-based user electricity consumption prediction method comprises the following steps:
s1, acquiring historical data of power consumption of a user through an intelligent electric meter, wherein the historical data comprises time sequence data, weather temperature data and holiday data; the time sequence data comprises power consumption data at different times and is used for describing the condition that the power supply demand changes along with the time.
S2, preprocessing and normalizing historical data
The original power consumption data is as follows: x ═ X1，x2，...，xnPreprocessing the original data, including processing missing values, abnormal values, repeated values and invalid values;
further, the step 2 is realized specifically:
(1) the missing data and the repeated data adopt an average value and maximum and minimum value calculation method to replace missing values or deleting repeated values;
(2) calculating abnormal values and invalid values by adopting a statistical method for deleting or replacing the abnormal values and the invalid values;
(3) carrying out data normalization processing on the processed data: using formulas
Figure BDA0002950788060000021
Performing data preprocessing, wherein xiIs the actual value of the historical data, xmaxIs a historical data value, xminIs the minimum value of the historical data, xi *Is the normalized data.
S3, constructing a Prophet prediction model, and setting the processed historical electricity consumption data X '({ X'i，x′2，...，x′nInputting the predicted data into a Prophet model to predict the Prophet;
further, the specific implementation of step 3:
(1) selecting historical power consumption data sample data, and dividing the selected data into training set dataXr′＝{x′r1，x′r2，...，x′rnAnd test set data Xt′＝{x′r1，x′t2，...，x′tn}。
(2) And (3) checking a Prophet model established by the training data, and comprising the following steps:
a) the Prophet model decomposes the electricity consumption time-series data into three parts: trend changes, holidays and seasonal trends; the decomposition function is formulated as follows: y (t) ═ g (t) + h (t) + s (t) + epsilontWherein g (t) is a trend change function for processing aperiodic changes in the predicted values; h (t) is that the holiday term represents the influence of holiday holidays on the time series data; s (t) is a period term for processing periodic variations in time series data; epsilontIt is the error term that represents the fluctuations in the model that are not predicted.
b) From the training data, a trend change function is calculated by a trend formula:
Figure BDA0002950788060000022
c) wherein c represents the capacity of the prediction model, k is the trend growth rate, n is the offset parameter, and t is time;
d) through the data information of the holidays, it can be known that different holidays are independent models at different time points, so that the holiday model formula expresses that:
Figure BDA0002950788060000023
wherein k isiShowing the effect of holidays on predicted values, i is a holiday, DiIs the time t contained in the window period (if time t is a virtual variable, D i1, otherwise 0:
e) from seasonal feature information, the periodic component is approximated by a period term s (t), typically using a fourier series, as:
Figure BDA0002950788060000031
where S is the period of the time series, 2N is the number of periods expected in the prediction model, N is the order of the Fourier transform, an，bnIs a parameter to be estimated.
(3) Training set data to perform model training, primarily evaluating a model by the test set data, adjusting parameters of a test result, and determining a trend function model, the number of cycles, the growth rate, the seasonality and the holiday fitting degree of a final model so as to analyze and predict the change of the power consumption sequence;
(4) the time series development process can be well shown through a Prophet prediction model, the trend change of the time series is described, and different quantized values are obtained; finally, a Prophet model predicted value P (t) is obtained.
S4, in order to prevent prediction overfitting, combined prediction is carried out by combining an improved LSTM model;
further, the specific implementation of the improved LSTM model prediction:
(1) aiming at the problem that the traditional LSTM model has prediction lag due to insufficient memory capacity, the LSTM model is improved, before the LSTM model processes data, a Convolutional Neural Network (CNN) is used for extracting high-order characteristic information, useless information is removed at the same time, the improved LSTM model reduces the processed data volume and improves the processing speed of the LSTM model, and the LSTM model and the improved LSTM model use the same weight, so that not only is the network load increment reduced, but also the memory capacity of the LSTM network is improved.
(2) The LSTM (long-short time memory network) solves the problem of gradient disappearance or gradient expansion by increasing a threshold, and three logic control units are added on the basis of a recurrent neural network: the system comprises an input gate, an output gate and a forgetting gate; there is an improved LSTM neural network formula at time t:
ft＝(Wf·[ht-1，xt]+bf)
it＝σ(Wi·[ht-1，xt]+bi)
ot＝σ(Wo·[ht-1，xt]+bo)
Figure BDA0002950788060000032
Figure BDA0002950788060000033
ht＝ot*tanhct
wherein xtIs the input vector at time t, xi' is the vector output after being processed by the convolution neural network, sigma is sigmoid function and tanh is hyperbolic tangent function, sigma and tanh are both activation functions, and forgetting gate ftInput door itAnd an output gate otThe weighting matrix of each corresponding threshold is Wf，WiAnd WoEach converted deviation value bf，biAnd boDenotes matrix multiplication, ctIs the output of the state of the network element at time t, i.e. the memory element, the weighting matrix of which is WcConversion of the offset value bc，ct-1Is the information that the forgetting gate determines the memory cell was discarded at the last moment,
Figure BDA0002950788060000041
is to update the memory cell information, htIs implicit information output by the memory cell, ht-1Is the implicit information input by the memory unit.
Improved LSTM model training step:
a. inputting the power consumption value at the time t into an input layer, and calculating an output result through an excitation function;
b. training data by using an improved LSTM network model, selecting a super parameter with the minimum error as a prediction model parameter after multiple times of training, determining the super parameter of the model by using a network search algorithm, and selecting an optimal parameter with the best performance from each parameter combination of cyclic traversal;
c. and (3) obtaining a predicted value L (t) of the power consumption of the user on the ith day by using an improved LSTM prediction model for the data of the test set.
(3) And finally, after the Prophet model predicted value P (t) at the time t and the improved LSTM model predicted value L (t) t (1, 2), comparing the Prophet model predicted value P (t) with the improved LSTM model predicted value L (t) t with the real value, and judging the prediction effect of the combined model.
S5, measuring and verifying the fitting degree and the prediction effect of the combined model, and using the following two common evaluation indexes: root mean square error RMSE and mean absolute percentage MAPE, the formula is as follows:
Figure BDA0002950788060000042
Figure BDA0002950788060000043
wherein xiRepresenting the true value of the time series at the ith time, diIndicating the time-series prediction values at the same time.
Compared with the prior art, the invention has the beneficial effects that:
(1) according to the traditional time sequence prediction method, a mathematical model is established to fit a historical time trend curve, the data is simple and has a hysteresis problem, and the Prophet model adopted by the method not only fits the historical time data curve, but also adapts to periodic trend change, holiday effect and seasonal trend in the data, and particularly plays an effective role in robustness of abnormal values and missing values.
(2) The method analyzes the characteristics and the rules of the power consumption data, improves the accuracy of the prediction model, and has important guiding significance for national power grids and various power supply companies to formulate effective power supply services.
(3) The single prediction method has poor capturing capability on the composite characteristics of the time series, and in order to optimize the prediction effect of the model, different time series processing and analyzing methods are used for fusing the composite model and performing prediction analysis on the power consumption data.
Drawings
FIG. 1: Prophet-LSTM combined model based algorithm flow chart
FIG. 2: prophet model prediction workflow diagram
FIG. 3: improved LSTM neural network structure diagram
Detailed Description
In this embodiment, a method for predicting power consumption of a user based on a Prophet-LSTM model, as shown in fig. one, includes:
s1, acquiring historical data of power consumption of a user through an intelligent electric meter, wherein the historical data comprises time sequence data, weather temperature data and holiday data; the time sequence data comprises power consumption data at different times and is used for describing the condition that the power supply demand changes along with the time.
S2, preprocessing and normalizing historical data
The original power consumption data is as follows: x ═ X1，x2，...，xnPreprocessing the original data, including processing missing values, abnormal values, repeated values and invalid values;
further, the step 2 is realized specifically:
(1) the missing data and the repeated data adopt an average value and maximum and minimum value calculation method to replace missing values or deleting repeated values;
(2) calculating abnormal values and invalid values by adopting a statistical method for deleting or replacing the abnormal values and the invalid values;
(3) carrying out data normalization processing on the processed data: using formulas
Figure BDA0002950788060000051
Performing data preprocessing, wherein xtIs the actual value of the historical data, xmaxIs a historical data value, xminIs the minimum value of the historical data, xi *Is the normalized data.
S3, a Prophet prediction model is constructed, and the post-processing historical electric energy consumption data X 'is { X'1，x′2，...，x′nInputting the predicted data into a Prophet model to predict the Prophet;
further, the specific implementation of step 3:
(1) selecting historical power consumption data sample data, and dividing the selected data into training set data Xr′＝{x′r1，x′r2，...，x′rn}
And test set data Xt′＝{x′t1，x′t2，...，x′tn}。
(2) And (3) checking a Prophet model established by the training data, and comprising the following steps:
a) the Prophet model decomposes the electricity consumption time-series data into three parts: trend changes, holidays and seasonal trends; the decomposition function is formulated as follows: y (t) ═ g (t) + h (t) + s (t) + epsilontWherein g (t) is a trend change function for processing aperiodic changes in the predicted values; h (t) is that the holiday term represents the influence of holiday holidays on the time series data; s (t) is a period term for processing periodic variations in time series data; epsilontIt is the error term that represents the fluctuations in the model that are not predicted.
b) From the training data, a trend change function is calculated by a trend formula:
Figure BDA0002950788060000061
wherein c represents the capacity of the prediction model, k is the trend growth rate, n is the offset parameter, and t is time;
c) through the data information of the holidays, it can be known that different holidays are independent models at different time points, so that the holiday model formula expresses that:
Figure BDA0002950788060000064
wherein k isiShowing the effect of holidays on predicted values, i is a holiday, DiIs the time t contained in the window period (if time t is a virtual variable, DiIs 1, otherwise is 0;
d) from seasonal feature information, the periodic component is approximated by a period term s (t), typically using a fourier series, as:
Figure BDA0002950788060000063
wherein S isPeriod of time series, 2N is the number of periods expected in the prediction model, N is the order of the Fourier transform, an，bnIs a parameter to be estimated.
(3) Training set data to perform model training, primarily evaluating a model by the test set data, adjusting parameters of a test result, and determining a trend function model, the number of cycles, the growth rate, the seasonality and the holiday fitting degree of a final model so as to analyze and predict the change of the power consumption sequence;
(4) the time series development process can be well shown through a Prophet prediction model, the trend change of the time series is described, and different quantized values are obtained; finally, a Prophet model predicted value P (t) is obtained.
S4, in order to prevent prediction overfitting, combining the improved LSTM model to perform combined prediction, as shown in the third figure;
further, the specific implementation of the improved LSTM model prediction:
(1) aiming at the problem that the traditional LSTM model has prediction lag due to insufficient memory capacity, the LSTM model is improved, before the LSTM model processes data, a Convolutional Neural Network (CNN) is used for extracting high-order characteristic information, useless information is removed at the same time, the improved LSTM model reduces the processed data volume and improves the processing speed of the LSTM model, and the LSTM model and the improved LSTM model use the same weight, so that not only is the network load increment reduced, but also the memory capacity of the LSTM network is improved.
(2) The LSTM (long-short time memory network) solves the problem of gradient disappearance or gradient expansion by increasing a threshold, and three logic control units are added on the basis of a recurrent neural network: the system comprises an input gate, an output gate and a forgetting gate; modified LSTM neural network formula at time t:
ft＝(Wf·[ht-1，xt]+bf)
it＝σ(Wi·[ht-1，xt]+bi)
ot＝σ(Wo·[ht-1，xt]+bo)
Figure BDA0002950788060000071
Figure BDA0002950788060000072
ht＝ot*tanhct
wherein xtIs the input vector at time t, xt' is the vector output after being processed by the convolution neural network, sigma is sigmoid function and tanh is hyperbolic tangent function, sigma and tanh are both activation functions, and forgetting gate ftInput door itAnd an output gate otThe weighting matrix of each corresponding threshold is Wf，WiAnd WoEach converted deviation value bf，biAnd boDenotes matrix multiplication, ctIs the output of the state of the network element at time t, i.e. the memory element, the weighting matrix of which is WcConversion of the offset value bc，ct-1Is the information that the forgetting gate determines the memory cell was discarded at the last moment,
Figure BDA0002950788060000073
is to update the memory cell information, htIs implicit information output by the memory cell, ht-1Is the implicit information input by the memory unit.
Improved LSTM model training step:
a) inputting the power consumption value at the time t into an input layer, and calculating an output result through an excitation function;
b) training the training data by using an improved LSTM network model, selecting the super-parameter with the minimum error as a prediction model parameter after training for multiple times, determining the super-parameter of the model by a network search algorithm, and selecting the optimal parameter with the best performance from each parameter combination of cyclic traversal: taking the data set of the present invention as an example, three layers of LSTM models are established, which respectively include 128, 64, and 16 neurons, the learning rate learning _ rate is 0.3, the activation function activation _ function is "tank", the batch _ size is 1000, the epochs is 10, and the loss function loss is "MAE";
c) and (3) obtaining a predicted value L (t) of the power consumption of the user on the ith day by using an improved LSTM prediction model for the data of the test set.
(3) And finally, after the Prophet model predicted value P (t) at the time t and the improved LSTM model predicted value L (t) t (1, 2), comparing the Prophet model predicted value P (t) with the improved LSTM model predicted value L (t) t with the real value, and judging the prediction effect of the combined model.
S5, measuring and verifying the fitting degree and the prediction effect of the combined model, and using the following two common evaluation indexes: root mean square error RMSE and mean absolute percentage MAPE, the formula is as follows:
Figure BDA0002950788060000081
Figure BDA0002950788060000082
wherein xiRepresenting the true value of the time series at the ith time, diIndicating the time-series prediction values at the same time.
The invention combines the Prophet model and the improved LSTM model, gives full play to the advantages of the two prediction models, utilizes the network search algorithm to optimally select the parameters of the models, can improve the accuracy of the prediction models, and can provide certain reference in the fields of time series analysis and energy consumption prediction.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an illustrative embodiment," "an example," "a specific example," or "some examples" or the like mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims (5)

1. A Prophet-LSTM model-based user electricity consumption prediction method is characterized by comprising the following steps:
s1, acquiring historical data of power consumption of a user through an intelligent electric meter, wherein the historical data comprises time sequence data, weather temperature data and holiday data; the time sequence data comprises power consumption data at different times and is used for describing the condition that the power supply demand changes along with the time;
s2, preprocessing and normalizing historical data: the original power consumption data is as follows: x ═ X1，x2，...，xnPreprocessing the original data, including processing missing values, abnormal values, repeated values and invalid values;
s3, constructing a Prophet prediction model, and setting the processed historical electricity consumption data X '({ X'1，x′2，...，x′nInputting the predicted data into a Prophet model to predict the Prophet;
s4, in order to prevent prediction overfitting, combining an improved long-time memory network LSTM model to perform combined prediction;
s5, measuring and verifying the fitting degree and the prediction effect of the combined model, and using the following two common evaluation indexes: root mean square error RMSE and mean absolute percentage MAPE, the formula is as follows:
Figure FDA0002950788050000011
Figure FDA0002950788050000012
wherein xiRepresenting the true value of the time series at the ith time, diIndicating the time-series prediction values at the same time.
2. The Prophet-LSTM model-based method for predicting power consumption of users according to claim 1, wherein the step 2 is implemented as follows:
2.1, replacing a missing value or a deleted duplicate value by the missing data and the duplicate data by adopting an average value and maximum and minimum value calculation method;
2.2, calculating abnormal values and invalid values by adopting a statistical method to delete or replace the abnormal values and the invalid values;
2.3, carrying out data normalization processing on the processed data: using formulas
Figure FDA0002950788050000013
Performing data preprocessing, wherein xiIs the actual value of the historical data, xmaxIs a historical data value, xminIs the minimum value of the historical data, xi *Is the normalized data.
3. The Prophet-LSTM model-based prediction method for power consumption of users according to claim 1, wherein the step S3 is implemented by:
3.1, selecting historical power consumption data sample data, and dividing the selected data into training set data Xr′＝{x′r1，x′r2，...，x′rnAnd test set data Xt′＝{x′r1，x′t2，...，x′tn}；
3.2, checking a Prophet model established by the training data;
3.3, training the training set data to perform model training, primarily evaluating the model by the test set data, adjusting parameters of the test result, and determining a trend function model, the number of cycles, the growth rate, the seasonality and the holiday fitting degree of the final model so as to analyze and predict the power consumption sequence change; the time series development process can be shown through a Prophet prediction model, the trend change of the time series is described, and different quantized values are obtained; finally, a Prophet model predicted value P (t) is obtained.
4. The Prophet-LSTM model-based power consumption prediction method for users according to claim 3, wherein the step of checking the Prophet model established by the training data is as follows:
a. the Prophet model decomposes the electricity consumption time-series data into three parts: trend changes, holidays and seasonal trends; the decomposition function is formulated as follows: y (t) ═ g (t) + h (t) + s (t) + epsilontWherein g (t) is a trend change function for processing aperiodic changes in the predicted values; h (t) is that the holiday term represents the influence of holiday holidays on the time series data; s (t) is a period term for processing periodic variations in time series data; epsilontIs that the error term represents fluctuations in the model that are not predicted;
b. from the training data, a trend change function is calculated by a trend formula:
Figure FDA0002950788050000021
wherein c represents the capacity of the prediction model, k is the trend growth rate, n is the offset parameter, and t is time;
c. through the data information of the holidays, it can be known that different holidays are independent models at different time points, so that the holiday model formula expresses that:
Figure FDA0002950788050000022
wherein k isiShowing the effect of holidays on predicted values, i is a holiday, DiIs the time t contained in the window period, if the time t is a virtual variable, DiIs 1, otherwise is 0;
d. from seasonal feature information, the periodic component is approximated by a period term s (t), typically using a fourier series, as:
Figure FDA0002950788050000023
where S is the period of the time series, 2N is the number of periods expected in the prediction model, N is the order of the Fourier transform, an，bnIs a parameter to be estimated.
5. The Prophet-LSTM model-based method for predicting power consumption of users according to claim 1, wherein the step S4 is implemented as follows:
4.1, aiming at the problem that the traditional LSTM model has prediction lag due to insufficient memory capacity, the LSTM model is improved, before the LSTM model processes data, a convolutional neural network CNN is used for extracting high-order characteristic information, useless information is removed at the same time, the improved LSTM model reduces the processed data volume, the processing speed of the LSTM model is improved, and the improved LSTM model use the same weight, so that not only is the network load increment reduced, but also the memory capacity of the LSTM model is improved;
4.2, the LSTM solves the problem of gradient disappearance or gradient expansion by increasing a threshold, and three logic control units are added on the basis of a recurrent neural network: the system comprises an input gate, an output gate and a forgetting gate; there is an improved LSTM neural network formula at time t:
ft＝(Wf·[ht-1，xt′]+bf)
it＝σ(Wi·[ht-1，xt′]+bi)
ot＝σ(Wo·[ht-1，xt′]+bo)
Figure FDA0002950788050000031
Figure FDA0002950788050000032
ht＝ot*tanhct
wherein xtIs the input vector at time t, xt' is the vector output after being processed by the convolution neural network, sigma is sigmoid function and tanh is hyperbolic tangent function, sigma and tanh are both activation functions, and forgetting gate ftInput door itAnd an output gate otThe weighting matrix of each corresponding threshold is Wf，WiAnd WoEach converted deviation value bf，biAnd boDenotes matrix multiplication, ctIs the output of the state of the network element at time t, i.e. the memory element, the weighting matrix of which is WcConversion of the offset value bc，ct-1Is the information that the forgetting gate determines the memory cell was discarded at the last moment,
Figure FDA0002950788050000034
is to update the memory cell information, htIs implicit information output by the memory cell, ht-1Is the implicit information input by the memory unit; improved LSTM model training step:
a. inputting the power consumption value at the time t into an input layer, and calculating an output result through an excitation function;
b. training data by using an improved LSTM network model, selecting a super parameter with the minimum error as a prediction model parameter after multiple times of training, determining the super parameter of the model by using a network search algorithm, and selecting an optimal parameter with the best performance from each parameter combination of cyclic traversal;
c. and (3) obtaining a predicted value L (t) of the power consumption of the user on the ith day by using an improved LSTM prediction model for the data of the test set.
4.3, the Prophet model predicted value P (t) at the time t and the improved LSTM model predicted value
Figure FDA0002950788050000033
And after the combined prediction is carried out, comparing with the true value, and judging the prediction effect of the combined model.
CN202110209867.8A 2021-02-24 2021-02-24 User power consumption prediction method based on Prophet-LSTM model Pending CN112990556A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110209867.8A CN112990556A (en) 2021-02-24 2021-02-24 User power consumption prediction method based on Prophet-LSTM model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110209867.8A CN112990556A (en) 2021-02-24 2021-02-24 User power consumption prediction method based on Prophet-LSTM model

Publications (1)

Publication Number Publication Date
CN112990556A true CN112990556A (en) 2021-06-18

Family

ID=76350440

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110209867.8A Pending CN112990556A (en) 2021-02-24 2021-02-24 User power consumption prediction method based on Prophet-LSTM model

Country Status (1)

Country Link
CN (1) CN112990556A (en)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537591A (en) * 2021-07-14 2021-10-22 北京琥珀创想科技有限公司 Long-term weather prediction method and device, computer equipment and storage medium
CN113554464A (en) * 2021-07-22 2021-10-26 四川中烟工业有限责任公司 Method and device for realizing cigarette demand prediction based on data analysis
CN113644834A (en) * 2021-10-14 2021-11-12 广东电网有限责任公司珠海供电局 Loss assessment method and system for modular multilevel converter
CN113705867A (en) * 2021-08-16 2021-11-26 浙江浙大中控信息技术有限公司 Equipment abnormity diagnosis method based on energy consumption interval prediction
CN113743665A (en) * 2021-09-03 2021-12-03 国网湖南省电力有限公司 Power load prediction method based on typical user group of power supply grid
CN113835341A (en) * 2021-09-18 2021-12-24 中邮科通信技术股份有限公司 Energy consumption analysis and diagnosis method based on intelligent building AI
CN113837487A (en) * 2021-10-13 2021-12-24 国网湖南省电力有限公司 Power system load prediction method based on combined model
CN114065667A (en) * 2021-11-19 2022-02-18 国网上海市电力公司 SF6 equipment gas pressure prediction method based on Prophet-LSTM model
CN114169568A (en) * 2021-11-03 2022-03-11 国网浙江省电力有限公司瑞安市供电公司 Prophet model-based power distribution line current prediction and heavy overload early warning and system
CN114756604A (en) * 2022-06-13 2022-07-15 西南交通大学 Monitoring time sequence data prediction method based on Prophet combination model
CN115412567A (en) * 2022-08-09 2022-11-29 浪潮云信息技术股份公司 Cloud platform storage capacity planning system and method based on time series prediction
CN115795255A (en) * 2022-09-21 2023-03-14 深圳大学 Method, device, medium and terminal for detecting time series change of wetland
CN116011312A (en) * 2022-06-15 2023-04-25 盐城师范学院 Bridge strain stress data prediction method based on multiple models
CN116090637A (en) * 2023-01-19 2023-05-09 国网江苏省电力有限公司苏州供电分公司 Short-time electricity consumption prediction method and system
CN116780658A (en) * 2023-08-17 2023-09-19 国网浙江省电力有限公司金华供电公司 Multi-energy complementary optimization scheduling method considering source-load bilateral uncertainty
CN117057488A (en) * 2023-10-12 2023-11-14 江南大学附属医院 Method and system for predicting workload of radiation medical equipment based on artificial intelligent model
CN117370330A (en) * 2023-12-07 2024-01-09 浙江有数数智科技有限公司 Prophet-SVM model-based data monitoring and early warning method of big data acquisition system
CN117977815A (en) * 2024-03-29 2024-05-03 杭州互为综合能源服务有限公司 Electric energy metering and collecting system with electricity consumption prediction function

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108830487A (en) * 2018-06-21 2018-11-16 王芊霖 Methods of electric load forecasting based on long neural network in short-term
CN109992608A (en) * 2019-03-26 2019-07-09 浙江大学 A kind of multi-model fusion forecasting method and system based on frequency domain
CN110111036A (en) * 2019-03-28 2019-08-09 跨越速运集团有限公司 Logistics goods amount prediction technique and system based on LSTM Model Fusion
CN111241755A (en) * 2020-02-24 2020-06-05 国网(苏州)城市能源研究院有限责任公司 Power load prediction method
CN111784043A (en) * 2020-06-29 2020-10-16 南京工程学院 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108830487A (en) * 2018-06-21 2018-11-16 王芊霖 Methods of electric load forecasting based on long neural network in short-term
CN109992608A (en) * 2019-03-26 2019-07-09 浙江大学 A kind of multi-model fusion forecasting method and system based on frequency domain
CN110111036A (en) * 2019-03-28 2019-08-09 跨越速运集团有限公司 Logistics goods amount prediction technique and system based on LSTM Model Fusion
CN111241755A (en) * 2020-02-24 2020-06-05 国网(苏州)城市能源研究院有限责任公司 Power load prediction method
CN111784043A (en) * 2020-06-29 2020-10-16 南京工程学院 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵英等: "基于LSTM-Prophet非线性组合的时间序列预测模型", 《计算机与现代化》, no. 09, 30 September 2020 (2020-09-30), pages 6 - 11 *

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537591A (en) * 2021-07-14 2021-10-22 北京琥珀创想科技有限公司 Long-term weather prediction method and device, computer equipment and storage medium
CN113554464A (en) * 2021-07-22 2021-10-26 四川中烟工业有限责任公司 Method and device for realizing cigarette demand prediction based on data analysis
CN113705867A (en) * 2021-08-16 2021-11-26 浙江浙大中控信息技术有限公司 Equipment abnormity diagnosis method based on energy consumption interval prediction
CN113743665A (en) * 2021-09-03 2021-12-03 国网湖南省电力有限公司 Power load prediction method based on typical user group of power supply grid
CN113835341B (en) * 2021-09-18 2024-05-17 中邮科通信技术股份有限公司 Intelligent building AI-based energy consumption analysis and diagnosis method
CN113835341A (en) * 2021-09-18 2021-12-24 中邮科通信技术股份有限公司 Energy consumption analysis and diagnosis method based on intelligent building AI
CN113837487A (en) * 2021-10-13 2021-12-24 国网湖南省电力有限公司 Power system load prediction method based on combined model
CN113644834B (en) * 2021-10-14 2022-02-15 广东电网有限责任公司珠海供电局 Loss assessment method and system for modular multilevel converter
CN113644834A (en) * 2021-10-14 2021-11-12 广东电网有限责任公司珠海供电局 Loss assessment method and system for modular multilevel converter
CN114169568A (en) * 2021-11-03 2022-03-11 国网浙江省电力有限公司瑞安市供电公司 Prophet model-based power distribution line current prediction and heavy overload early warning and system
CN114065667A (en) * 2021-11-19 2022-02-18 国网上海市电力公司 SF6 equipment gas pressure prediction method based on Prophet-LSTM model
CN114756604A (en) * 2022-06-13 2022-07-15 西南交通大学 Monitoring time sequence data prediction method based on Prophet combination model
CN114756604B (en) * 2022-06-13 2022-09-09 西南交通大学 Monitoring time sequence data prediction method based on Prophet combination model
CN116011312A (en) * 2022-06-15 2023-04-25 盐城师范学院 Bridge strain stress data prediction method based on multiple models
CN115412567A (en) * 2022-08-09 2022-11-29 浪潮云信息技术股份公司 Cloud platform storage capacity planning system and method based on time series prediction
CN115412567B (en) * 2022-08-09 2024-04-30 浪潮云信息技术股份公司 Cloud platform storage capacity planning system and method based on time sequence prediction
CN115795255B (en) * 2022-09-21 2024-03-26 深圳大学 Method, device, medium and terminal for detecting time sequence change of wetland
CN115795255A (en) * 2022-09-21 2023-03-14 深圳大学 Method, device, medium and terminal for detecting time series change of wetland
CN116090637A (en) * 2023-01-19 2023-05-09 国网江苏省电力有限公司苏州供电分公司 Short-time electricity consumption prediction method and system
CN116780658A (en) * 2023-08-17 2023-09-19 国网浙江省电力有限公司金华供电公司 Multi-energy complementary optimization scheduling method considering source-load bilateral uncertainty
CN116780658B (en) * 2023-08-17 2023-11-10 国网浙江省电力有限公司金华供电公司 Multi-energy complementary optimization scheduling method considering source-load bilateral uncertainty
CN117057488A (en) * 2023-10-12 2023-11-14 江南大学附属医院 Method and system for predicting workload of radiation medical equipment based on artificial intelligent model
CN117057488B (en) * 2023-10-12 2023-12-26 江南大学附属医院 Method and system for predicting workload of radiation medical equipment based on artificial intelligent model
CN117370330A (en) * 2023-12-07 2024-01-09 浙江有数数智科技有限公司 Prophet-SVM model-based data monitoring and early warning method of big data acquisition system
CN117977815A (en) * 2024-03-29 2024-05-03 杭州互为综合能源服务有限公司 Electric energy metering and collecting system with electricity consumption prediction function

Similar Documents

Publication Publication Date Title
CN112990556A (en) User power consumption prediction method based on Prophet-LSTM model
Ke et al. Short-term electrical load forecasting method based on stacked auto-encoding and GRU neural network
Zhang et al. Wind speed forecasting based on quantile regression minimal gated memory network and kernel density estimation
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
CN111260136A (en) Building short-term load prediction method based on ARIMA-LSTM combined model
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN112434848B (en) Nonlinear weighted combination wind power prediction method based on deep belief network
CN110598854A (en) GRU model-based transformer area line loss rate prediction method
CN111680786B (en) Time sequence prediction method based on improved weight gating unit
CN112329990A (en) User power load prediction method based on LSTM-BP neural network
CN111241755A (en) Power load prediction method
CN114218872B (en) DBN-LSTM semi-supervised joint model-based residual service life prediction method
CN113780684A (en) Intelligent building user energy consumption behavior prediction method based on LSTM neural network
CN112803398A (en) Load prediction method and system based on empirical mode decomposition and deep neural network
CN113868938A (en) Short-term load probability density prediction method, device and system based on quantile regression
Bian et al. Load forecasting of hybrid deep learning model considering accumulated temperature effect
CN115456287A (en) Long-and-short-term memory network-based multi-element load prediction method for comprehensive energy system
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN114819395A (en) Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN113128666A (en) Mo-S-LSTMs model-based time series multi-step prediction method
CN112232570A (en) Forward active total electric quantity prediction method and device and readable storage medium
CN117151770A (en) Attention mechanism-based LSTM carbon price prediction method and system
Wang et al. Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device

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