CN112348702B - Power load prediction method based on window moving machine learning - Google Patents
Power load prediction method based on window moving machine learning Download PDFInfo
- Publication number
- CN112348702B CN112348702B CN202011231695.6A CN202011231695A CN112348702B CN 112348702 B CN112348702 B CN 112348702B CN 202011231695 A CN202011231695 A CN 202011231695A CN 112348702 B CN112348702 B CN 112348702B
- Authority
- CN
- China
- Prior art keywords
- power load
- data
- power
- prediction model
- prediction
- 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.)
- Active
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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
Abstract
A power load prediction method based on window moving machine learning relates to the field of power load prediction. The method aims to solve the problem that the existing method for predicting the power load is low in accuracy rate of data set prediction with large data volume and strong nonlinearity. The invention includes: acquiring original power load data; step two, the original power load data is subjected to dimensionality increase; thirdly, a power load prediction model is obtained by utilizing the power load training after the dimension is increased; step four, verifying the accuracy of the power load prediction model; and step five, inputting the power load data into the power load prediction model to obtain a prediction result during actual charge prediction. The method is mainly used for predicting the power load.
Description
Technical Field
The invention relates to the field of power load prediction, in particular to a power load prediction method based on window mobile machine learning.
Background
In the process of power production, the load is closely related to safe and stable operation of a power grid, new energy consumption, economic operation of the power grid and optimized dispatching. The power load data is influenced by aspects such as industrial production, daily life of people, climate and the like, and is a data type with extremely high nonlinearity, so a more reasonable and accurate power load prediction method needs to be provided.
In the power load prediction, in the case of a problem in which the relationship between independent variables and dependent variables is relatively clear, polynomial fitting, the least square method, or the like is mainly used, but the above method is not high in data accuracy against strong nonlinearity and large data amount. At present, the intelligent algorithm is also commonly used in the field of load prediction, has the advantages of wide application range, self-learning, robustness, strong universality and the like, and provides a new means for solving the problem of complex optimization. However, due to the limitation of the principle, the purpose of prediction cannot be achieved by aiming at the problem of unary regression analysis. Therefore, the problem of low prediction accuracy exists in the power load prediction problem at present due to large data volume, strong nonlinearity and weak adaptability of machine learning to unary regression.
Disclosure of Invention
The invention aims to solve the problem that the prediction accuracy is not high due to large data size, strong nonlinearity and weak adaptability of machine learning to unitary regression in the conventional method for predicting the power load, and provides a power load prediction method based on window moving machine learning.
The power load prediction method based on the window mobile machine learning comprises the following specific processes:
acquiring original power load data;
the original power load data are a power load data set at the current moment and a moment data set for collecting the power load;
the current time power load Y = { Y = 1 ,y 2 ,…,y n }；
The time X = { X ] of collecting power load 1 ,x 2 ,…,x n }；
Step two, upgrading the dimension of the original power load data:
step two, grouping the power loads at the current moment;
secondly, dividing each group of power loads at the current moment into a characteristic power load and a target power load;
step two, acquiring power load data after dimensionality increase;
thirdly, a power load prediction model is obtained by utilizing the power load training after the dimension is increased;
step four, verifying the accuracy of the power load prediction model;
and step five, inputting the power load data into the power load prediction model to obtain a prediction result when the actual charge is predicted.
The invention has the beneficial effects that:
the method divides the problem of long-term prediction of the power load into a plurality of groups of short-term predictions, finds the relationship between the power load value in the previous part of time period and the power load value in the adjacent next part of time period in a short time period to form a group of data set windows, and realizes the power load prediction in the future part of time period by continuously moving the windows backwards on a time axis. The invention provides a data dimension-increasing method, which solves the problem of extremely large error of a power load prediction result and improves the load prediction precision by converting a dependent variable form.
Drawings
FIG. 1 is a schematic diagram of a power load data dimension promotion;
FIG. 2 is a power load curve for a substation;
FIG. 3 (a) is a load curve for verifying the accuracy of a predicted electrical load using a linear regression algorithm;
FIG. 3 (b) is a graph of error curves of actual values and model calculated values of test data when a linear regression algorithm is used to verify the accuracy of predicted electrical loads;
FIG. 3 (c) is a graph of the error of the actual value and the model calculated value of the predicted data when the accuracy of the predicted power load is verified using a linear regression algorithm;
FIG. 4 (a) is a load curve for verifying the accuracy of a predicted electrical load using the K-nearest neighbor algorithm (KNN);
FIG. 4 (b) is a graph of the error of the actual value and the model calculated value of the test data when the accuracy of the predicted power load is verified by using the K-nearest neighbor algorithm (KNN);
FIG. 4 (c) is a graph of the error between the actual value and the model calculated value of the predicted data when the accuracy of the predicted power load is verified using the K-nearest neighbor algorithm (KNN);
FIG. 5 (a) is a load curve for verifying the accuracy of a predicted electrical load using a decision tree algorithm;
FIG. 5 (b) is a graph of error curves of actual values and model calculated values of test data when a decision tree algorithm is used to verify the accuracy of predicted electrical loads;
FIG. 5 (c) is a graph of the actual values of the predicted data and the error of the model calculated values when the accuracy of the predicted power load is verified by using a decision tree algorithm;
FIG. 6 (a) is a load curve for verifying the accuracy of a predicted power load using a random forest algorithm;
FIG. 6 (b) is an error curve diagram of actual values and model calculated values of test data when a random forest algorithm is used to verify the accuracy of a predicted power load;
FIG. 6 (c) is a graph of the error between the actual value and the model calculated value of the predicted data when the accuracy of the predicted power load is verified by using the random forest algorithm.
Detailed Description
The first embodiment is as follows: the specific process of the power load prediction method based on the window mobile machine learning in the embodiment is as follows:
acquiring original power load data;
the original power load data are a power load data set at the current moment and a moment data set for collecting the power load;
the current time power load Y = { Y = 1 ，y 2 ，…，y n }；
The time of collecting the power load is X = [ X ] 1 ，x 2 ，…，x n }；
Step two, upgrading the dimension of the original power load data, which comprises the following specific processes:
step two, grouping the power loads at the current moment:
performing mobile grouping in the time sequence of the power load by using a mobile window with the size of l:
Y＝{a 1 ，a 2 ，…，a m }
a i ＝{y p(i-1)+1 ，y p(i-1)+2 ，…，y p(i-1)+l }
where Y is the power load at the present time, a i Is the ith group of the current time power load, m is the total number of the power load groups at the current time, i belongs to [1, m ]]L is a i Comprising the number of electrical loads, y being a i P is the step size of the window shift.
Step two, dividing the power load of each group at the current moment into a characteristic power load and a target power load:
wherein the content of the first and second substances,
wherein, a i Is the power load value of the ith group at the current moment, i belongs to [1, m ]]，
Is a characteristic electrical load of the electrical load,
is a target power load, k is
The method comprises the steps of counting the number of electric loads,
the number of (l-k) electric loads;
wherein the division is
And
is determined artificially (in general, k is much larger than l-k).
Step two and step three, obtaining the power load data { a after dimension increasing 1 ，a 2 ，…，a m }。
Wherein the content of the first and second substances,
load data of m rows and k columns is a characteristic power load data set;
step three, obtaining a power load prediction model by utilizing power load training after dimensionality raising, wherein the specific process comprises the following steps:
step three, dividing the power load data after the dimension increasing into a training data set and a testing data set:
n＝[m*η]
wherein m is the total number of power load data packets, η is the percentage of the number of the taken training data sets in the total number of the data sets, n is the number of the training data sets, and n is an integer;
wherein the number of the test data sets is m-n;
wherein eta is set artificially;
the training data set is in an array form;
inputting the training data set into the power load prediction model to be trained to obtain a trained power load prediction model;
step four, verifying the accuracy of the power load prediction model, and the specific process is as follows:
firstly, inputting characteristic power load data in a test data set obtained in the first step into a power load prediction model to obtain a prediction result;
and then, comparing the obtained prediction result with the target power load data in the test data set, and verifying the accuracy of the power load prediction model.
And step five, inputting the power load data into the power load prediction model to obtain a prediction result when the actual power load is predicted.
Example (b):
raw power load data was collected (fig. 2): and acquiring the power load of 35 days in 2019 at a time interval of one hour, wherein the time interval is 840 times and the power load value of a certain transformer substation corresponding to each time.
For the piece of load data, a mobile packet is made, wherein l =20 × 24=480, p =24, k =15 × 24=360, namely, the first 15 days are taken as the characteristic power load
The next 5 days as the target power load
Two adjacent groups a i And a i+1 The interval between them is one day. And (3) performing model training on the load in the first 30 days to obtain 11 groups of load data after dimensionality increase. Here, since the target power load data is 5 days and the moving step is 1 day, in order to prevent false accuracy caused by repeated prediction, the verification process utilizesThe 16 th-30 th day load value is used as the characteristic power load of the test data, and the 31 th-35 th day load value is used as the target power load of the test data.
And performing model training and accuracy verification of test data by using KNN, decision trees, linear regression and random forest algorithm respectively, proving feasibility of the method and calculating model accuracy.
The error of the training data and the test data calculated by the model is shown in the following table, and the power load prediction curve is shown in fig. 3 (a):
TABLE 1 Linear regression verification results
In fig. 3 (a), a total of 20-day load curves are shown, each cycle representing the load fluctuation of one day, with a sampling time of 1h, the abscissa as the number of points, and the ordinate as the load value at that time, in megawatts. The solid line curve in the figure represents the measured load. The curve of the dotted line in the figure is derived from training data 15 days before
And the load curve predicted by the generative model is obtained in the last 5 days. The accuracy of the training data in the previous 15 days represents the fit degree of the data in the training process of the model, and the absolute value of the difference between the actually measured load value and the training data calculated by the model is shown in fig. 3 (b); the accuracy of the data in the last 5 days represents the prediction capability of the model on future data, namely represents the accuracy of the model, and the absolute value of the difference between the actual measurement load value in 31-35 days and the training data calculated by the model is shown in fig. 3 (c). The accuracy of the test data and the training data are compared in the table above, and the comparison is carried out from three aspects of mean square error, root mean square error and mean absolute error.
The verification results by the KNN method are shown in the following table, the power load prediction curve is shown in fig. 4 (a), the absolute value of the difference between the actual measured load value in the previous 15 days and the training data obtained by model calculation is shown in fig. 4 (b), and the absolute value of the difference between the actual measured load value in 31-35 days and the training data obtained by model calculation is shown in fig. 4 (c):
TABLE 2 KNN method validation results
| Training data | Test data | |
| Mean square error of MSE | 9.36 | 14.96 |
| RMSE root mean square error | 3.06 | 3.86 |
| Mean absolute error of MAE | 2.40 | 3.33 |
The verification results by using the decision tree method are shown in the following table, the power load prediction curve is shown in fig. 5 (a), the absolute value of the difference between the actual measured load value in the previous 15 days and the training data obtained by model calculation is shown in fig. 5 (b), and the absolute value of the difference between the actual measured load value in 31-35 days and the training data obtained by model calculation is shown in fig. 5 (c):
TABLE 3 decision Tree method verification results
| Training data | Test data | |
| Mean square error of |
0 | 25.04 |
| RMSE root mean |
0 | 5.00 |
| Mean absolute error of |
0 | 3.98 |
The verification results by using the random forest method are shown in the following table, the power load prediction curve is shown in fig. 6 (a), the absolute value of the difference between the measured load value in the previous 15 days and the training data obtained by model calculation is shown in fig. 6 (b), and the absolute value of the difference between the measured load value in 31-35 days and the training data obtained by model calculation is shown in fig. 6 (c):
table 4 random forest method verification results
| Training data | Test data | |
| Mean square error of MSE | 1.64 | 13.26 |
| RMSE root mean square error | 1.28 | 3.64 |
| Mean absolute error of MAE | 0.90 | 3.07 |
According to the results, the prediction result of the method is high in accuracy, and the method also shows high data reducibility and accuracy of prediction on test data.
Claims (3)
1. The power load prediction method based on the window mobile machine learning is characterized by comprising the following specific processes:
acquiring original power load data;
the original power load data are a power load data set at the current moment and a moment data set for collecting the power load;
the current time power load Y = { Y = 1 ，y 2 ，…，y n }；
The time X = { X ] of collecting power load 1 ，x 2 ，…，x n }；
Step two, upgrading the dimension of the original power load data:
step two, grouping the power loads at the current moment comprises the following specific processes:
performing sliding grouping according to the time sequence of the power load by using a moving window with the size of l:
Y＝{a 1 ，a 2 ，…，a m }
a i ＝{y p(i-1)+1 ，y p(i-1)+2 ，…，y p(i-1)+l }
where Y is the power load at the present time, a i Is the ith group of power loads at the current time, m is the total number of power load groups at the current time, i ∈ [1, m]L is a i The method comprises the steps of counting the number of electric loads, wherein p is the moving step length of a window;
step two, dividing each group of power loads at the current moment into a characteristic power load and a target power load specifically comprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,
wherein, a i Is the power load value of the ith group at the current moment, i belongs to [1, m ]]，
Is a characteristic electrical load of the electric power,
is a target power load, k is
The method comprises the steps of counting the number of electric loads,
the number of (l-k) electric loads;
wherein the division is
And
the proportion of (A) is artificially determined;
step two and step three, obtaining the power load data { a after dimension increasing 1 ，a 2 ，…，a m }；
Wherein, the first and the second end of the pipe are connected with each other,
load data of m rows and k columns is taken as a characteristic power load data set;
thirdly, a power load prediction model is obtained by utilizing the power load training after the dimension is increased;
step four, verifying the accuracy of the power load prediction model;
and step five, inputting the power load data into the power load prediction model to obtain a prediction result when the actual power load is predicted.
2. The power load prediction method based on window moving machine learning according to claim 1, characterized in that: in the third step, a power load prediction model is obtained by utilizing the power load training after the dimensionality is increased:
step three, dividing the power load data after the dimension increasing into a training data set and a testing data set:
n＝[m*η]
wherein m is the total number of power load data packets, η is the percentage of the number of the taken training data sets in the total number of the data sets, and n is the number of the training data sets;
wherein the number of the test data sets is m-n;
the eta is set artificially;
and step two, inputting the training data set into the power load prediction model to train to obtain a trained power load prediction model.
3. The power load prediction method based on window mobile machine learning according to claim 2, characterized in that: the accuracy of the power load prediction model is verified in the fourth step, and the specific process is as follows:
firstly, inputting characteristic power load data in a test data set obtained in the first step into a power load prediction model to obtain a prediction result;
then, the obtained prediction result is compared with the target power load data in the test data set, and the accuracy of the power load prediction model is verified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011231695.6A CN112348702B (en) | 2020-11-06 | 2020-11-06 | Power load prediction method based on window moving machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011231695.6A CN112348702B (en) | 2020-11-06 | 2020-11-06 | Power load prediction method based on window moving machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112348702A CN112348702A (en) | 2021-02-09 |
| CN112348702B true CN112348702B (en) | 2022-11-15 |
Family
ID=74429996
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011231695.6A Active CN112348702B (en) | 2020-11-06 | 2020-11-06 | Power load prediction method based on window moving machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112348702B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631483A (en) * | 2016-03-08 | 2016-06-01 | 国家电网公司 | Method and device for predicting short-term power load |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573854A (en) * | 2014-12-23 | 2015-04-29 | 国家电网公司 | Iron steel electricity consumption forecasting method and device |
| CN110874671B (en) * | 2019-10-24 | 2021-03-16 | 腾讯科技（深圳）有限公司 | Power load prediction method and device of power distribution network and storage medium |
-
2020
- 2020-11-06 CN CN202011231695.6A patent/CN112348702B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631483A (en) * | 2016-03-08 | 2016-06-01 | 国家电网公司 | Method and device for predicting short-term power load |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112348702A (en) | 2021-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876054B (en) | Short-term power load prediction method based on improved genetic algorithm optimization extreme learning machine | |
| CN106529814B (en) | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain | |
| CN107591844B (en) | Active power distribution network robust reconstruction method considering node injection power uncertainty | |
| CN108304623B (en) | Probability load flow online calculation method based on stack noise reduction automatic encoder | |
| CN108336739B (en) | RBF neural network-based probability load flow online calculation method | |
| CN113344288B (en) | Cascade hydropower station group water level prediction method and device and computer readable storage medium | |
| CN105719029A (en) | Combined wind power prediction method based on wind speed fluctuation characteristic extraction | |
| CN113379142B (en) | Short-term wind power prediction method based on wind speed correction and fusion model | |
| CN112668806B (en) | Photovoltaic power ultra-short-term prediction method based on improved random forest | |
| CN114021483A (en) | Ultra-short-term wind power prediction method based on time domain characteristics and XGboost | |
| CN114626487B (en) | Linear transformation relation checking method based on random forest classification algorithm | |
| CN110555784B (en) | Wind power prediction method based on gated cyclic neural network wind speed correction | |
| CN105956708A (en) | Grey correlation time sequence based short-term wind speed forecasting method | |
| CN115577752A (en) | Ultra-short-term photovoltaic power prediction method and system | |
| CN107221933B (en) | Probabilistic load flow calculation method | |
| CN116388184B (en) | Ultra-short-term wind speed revising method and system based on wind speed daily fluctuation characteristics | |
| CN112348702B (en) | Power load prediction method based on window moving machine learning | |
| CN106503828B (en) | Ultra-short term chaotic prediction method for photovoltaic output power | |
| CN115860797B (en) | Electric quantity demand prediction method suitable for new electricity price reform situation | |
| CN104915727A (en) | Multi-dimensional isomorphic heterogeneous BP neural network optical power ultrashort-term prediction method | |
| CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
| CN109193791B (en) | Wind power convergence tendency state-based quantification method based on improved shape value | |
| CN111242371A (en) | Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model | |
| CN114372640A (en) | Wind power prediction method based on fluctuation sequence classification correction | |
| Pravilovic et al. | Wind power forecasting using time series cluster analysis |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |