CN115222106A - User day-ahead load prediction method of self-adaptive model - Google Patents
User day-ahead load prediction method of self-adaptive model Download PDFInfo
- Publication number
- CN115222106A CN115222106A CN202210729317.3A CN202210729317A CN115222106A CN 115222106 A CN115222106 A CN 115222106A CN 202210729317 A CN202210729317 A CN 202210729317A CN 115222106 A CN115222106 A CN 115222106A
- Authority
- CN
- China
- Prior art keywords
- load
- user
- data
- day
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000005611 electricity Effects 0.000 claims abstract description 27
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 18
- 230000003044 adaptive effect Effects 0.000 claims description 14
- 238000012706 support-vector machine Methods 0.000 claims description 11
- 238000011156 evaluation Methods 0.000 claims description 9
- 238000004519 manufacturing process Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000000284 resting effect Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000012896 Statistical algorithm Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
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
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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
-
- 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
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Mathematical Physics (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a user day-ahead load prediction method of a self-adaptive model. The method comprises the following steps: 1) Acquiring and preprocessing user data and meteorological data; 2) Classifying users according to the electricity utilization regularity and the historical data quantity of the electricity utilization load; 3) Classifying the user forecast date according to the date type; 4) Based on different user types and user prediction day types, adaptively matching corresponding prediction models; 5) Establishing a prediction model based on the characteristic data such as meteorological data, historical load characteristics and the like; 6) And (4) building a load prediction system and outputting a day-ahead load prediction value. The method and the system can predict the power load at 24 points in the future day based on the historical data, the historical meteorological data, the future meteorological forecast data and other characteristic data of the power load of the user.
Description
Technical Field
The invention relates to the technical field of user day-ahead load prediction.
Technical Field
Different types of user loads have random fluctuation characteristics of different degrees, and the user day-ahead load prediction cannot be avoided in the strategy formulation in the fields of electric power spot market, energy storage scheduling, demand response, energy conservation and efficiency improvement, so that the improvement of the accuracy of the user day-ahead load prediction has great significance.
In the traditional technology, statistical algorithms such as a similar day matching method, a quadratic index smoothing method, a weighting method, a regression analysis method and the like are mainly used, with the rise of artificial intelligence, traditional machine learning and deep learning algorithms are gradually used for load prediction, and a more excellent effect is obtained on load prediction of users with complex rules, but historical data conditions and load regularity of different users are different, and the existing method is difficult to meet the requirement of load prediction of all users in the day and obtain a good prediction effect.
Disclosure of Invention
In view of the problems and deficiencies of the prior art, the present invention aims to provide a method for predicting a user's day-ahead load in an adaptive model, which performs load prediction based on historical data conditions and a load regularity adaptive selection algorithm of different users, and which not only has a good prediction effect, but also can provide a load prediction system as a service to users.
The invention provides a user day-ahead load prediction method of a self-adaptive model, which comprises the following steps:
the method comprises the steps of obtaining and preprocessing user data and meteorological data. Acquiring historical data of the power load of a user at intervals of 1 hour; acquiring historical meteorological data corresponding to historical data of the power load and meteorological data of a future day, wherein the meteorological data comprise a highest daily temperature, a lowest daily temperature, an average daily temperature, a weather type and the like; if the user belongs to the type of the production enterprise, production plan data corresponding to the historical data of the electric load is required to be obtained, wherein the production plan data comprises a working day type and a resting day type.
For each data field, a value exceeding the average plus or minus 3 times the standard deviation is set as an abnormal value.
For outliers and missing values, interpolated fit values can be used instead of processing, since the data fields are time-ordered.
And step two, classifying the users according to the electricity utilization regularity and the historical data volume of the electricity utilization load. Dividing users with regular electricity utilization and electricity utilization load historical data volume more than n days into predictable users; for users with regular electricity utilization and electricity utilization load historical data volume less than n days, dividing the users into potential predictable users, and when the electricity utilization load historical data volume to be used is greater than n days, dividing the potential predictable users into predictable users; and classifying users with no regularity in electricity utilization into unpredictable users.
And step three, classifying the user forecast dates according to date types. Dividing the user prediction date into a holiday type as the holiday type; the forecast date of the user belongs to the working day and is divided into working day types; the forecast date of the user belongs to the rest days and is divided into the rest day type. Particularly, for the holiday type and the holiday type, if the historical data volume of the same type of electrical loads is less than m days, the same type of electrical loads are divided into an unpredictable holiday type and an unpredictable holiday type.
And step four, adaptively matching the corresponding prediction model based on different user types and user prediction day types. For the holiday type, the working day type and the holiday type of the predictable user, a Sequence to Sequence (Sequence 2 Sequence) model which can be fitted with a complex rule is used for predicting; forecasting on the basis of a similar daily load model for an unpredictable holiday type and an unpredictable holiday type of a predictable user, a holiday type and a holiday type of a potential predictable user and all forecast day types of the unpredictable user; for the workday type of a potential predictable user, a support vector machine with strong generalization capability and an XGboost are used for prediction;
and step five, establishing a prediction model based on characteristic data such as time characteristics, meteorological characteristics, historical load characteristics and the like.
Constructing a time feature set, and considering the month, holiday, day of the year, day of the month, day of the week, hour and minute of the time T to form the time feature set T of the time T time,t 。
Constructing a meteorological feature set, and selecting weather types and temperatures as the meteorological feature set T nwp 。
Constructing a historical load characteristic set, and selecting a historical value of the same time T in the previous period (day and week) and a sliding average value of the upper time and the lower time as the historical load characteristic set T of the time T load,t 。
Historical load feature set T load,t Meteorological feature set T at time T nwp And a temporal feature set T time,t Forming a feature vector T t 。
And establishing a similar daily load model. Dates belonging to the same date type are taken as similar days, such as weekdays, saturdays, sundays, holidays. And selecting the latest 5 similar daily loads as load samples for each type of date, wherein the samples with the daily electric quantity lower than 25% of the daily average electric quantity of 5 samples or higher than 200% of the daily average electric quantity of 5 samples are removed, and the residual samples are used for calculating the average value of all the moments to obtain the predicted load.
And establishing a support vector machine or XGboost model. Feature vector T t And carrying out normalization processing on the medium-value type features, and carrying out one-hot coding processing on the discrete type features. Characteristic vector T of each historical time T t And a load P t Forming a data set D, dividing the data set D into a training set and a testing set according to a certain proportion, and for each sample, dividing the characteristic vector T t As input, load P t And as output, establishing a support vector machine or an XGB boost model, training the model on a training set, simultaneously evaluating in a test set, and selecting the model with the optimal evaluation effect as a final prediction model.
And establishing a Seq2Seq model. And modeling the time series of the load by combining a plurality of characteristic factors influencing the load, embedding multiple characteristics, and finally gradually decoding and predicting a future load value to generate a predicted load sequence at 24 moments in the future. Specifically, feature vector T is divided into t And carrying out normalization processing on the medium value type characteristic, and carrying out single-hot coding processing on the discrete type characteristic. For time t, order
H t ＝{[T t-l ,P t-l ],[T t-l+1 ,P t-l+1 ],K,[T t-1 ,P t-1 ]}
F t ＝{T t+1 ,T t+2 ,K,T t+24 }
X t ＝[H t ,F t ]
Y t ＝{P t+1 ,P t+2 ,K,P t+24 }
Wherein X t As model input, Y t As output of the model, P t+k For the k-th time step real load data,
load data is estimated for the kth time step coder, l is the recurrent neural network time window, and Ω (Θ) is a regularization term. The model principle is shown in the figure, the basic unit of the encoder and decoder is LSTM, the encoder and decoder are trained to minimize the objective function as ζ (Θ), and H is t Inputting the encoder, and in the decoder, calculating the hidden state c from the encoder and the estimated load data of the encoder at the last time step
Last time step encoder h t+k-1 And T t+k Input decoder LSTM unit g estimates to obtain estimated load data of k time step encoder
This in turn outputs the predicted load at 24 points in the future.
Inputting X of each time t of history t And load Y t And forming a data set D, and dividing the data set D into a training set and a testing set according to a certain proportion. After the model on the training set is trained, the evaluation is performed on the test set using the following evaluation indexes:
the root-mean-square error of the error,
the average absolute percentage error is calculated as the percentage of the absolute percentage error,
where Y represents the actual value in the test set,
and representing the predicted value obtained by the test set.
And step six, triggering a load prediction system and outputting a day-ahead load prediction value. The system module includes:
the data acquisition module is used for acquiring load data and meteorological data of a user;
and the data processing module is used for calculating and processing the original data to obtain a load value of hour granularity and further obtain a feature vector.
And the model building module is used for building and training a corresponding prediction model in a self-adaptive manner based on the user type and the prediction date type.
And the load prediction module is used for inputting parameters such as prediction date and date type by a system user, adaptively selecting a prediction model based on the user type and the prediction date type, and returning a prediction result to the user.
The implementation of the invention has the following beneficial effects:
the load prediction is carried out based on the historical data volume and the load regularity adaptive selection algorithm of different users, the adaptive prediction algorithm can be selected according to the users and dates of different data conditions, and the problem that the complex algorithm is difficult to train under the condition of insufficient historical data is solved; the Sequence to Sequence model of deep learning is applied to load prediction of 24 points in the day ahead, so that complex rules can be effectively fitted, and the prediction accuracy is improved; the load prediction system provides a service interface, accesses in a URL format and obtains a day-ahead load prediction result.
Drawings
Fig. 1 is an overall schematic diagram of a user day-ahead load prediction method of an adaptive model according to the present invention.
Fig. 2 is a schematic diagram of a prediction system trigger prediction flow.
FIG. 3 is a diagram of the structure of the Seq2Seq model.
Fig. 4 is a diagram illustrating the predicted result of the predicted user working day.
Fig. 5 is a diagram illustrating the predicted result of predicting the user's holiday.
FIG. 6 is a diagram illustrating predicted results of a preset potentially predictable user's work day.
Detailed Description
The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
Fig. 1 shows a user day-ahead load prediction method of an adaptive model according to the present invention, which includes the following steps:
step 2, classifying the users according to the power utilization regularity and the historical data volume of the power utilization load;
step 3, classifying the user prediction dates according to date types;
step 4, based on different user types and user prediction day types, adaptively matching corresponding prediction models;
step 5, establishing a prediction model based on the characteristic data such as meteorological data, historical load characteristics and the like;
and 6, building a load prediction system and outputting a day-ahead load prediction value.
The specific description of the above steps is as follows:
the method comprises the steps of obtaining and preprocessing user data and meteorological data. Acquiring historical data of power loads of users at intervals of 1 hour from a power internet of things platform database; acquiring historical meteorological data corresponding to the historical data of the power load and meteorological forecast data of a future day, wherein the meteorological data comprise a highest daily temperature, a lowest daily temperature, an average daily temperature and a weather type; and if the user belongs to the type of the production enterprise, obtaining production plan data corresponding to the historical data of the electric load, wherein the production plan data comprises a working day type and a rest day.
For each data field, values beyond the mean plus or minus 3 standard deviations were considered outliers.
For outliers and missing values, interpolated fit values can be used instead of processing, since the data fields are time-ordered.
And step two, classifying the users according to the electricity utilization regularity and the historical data volume of the electricity utilization load. Dividing users with regularity in electricity utilization and historical data quantity of electricity utilization load larger than n days into predictable users; for users with regularity in electricity utilization and with historical data quantity of electricity utilization load less than n days, dividing the users into potential predictable users, and when the historical data quantity of the electricity utilization load to be used is greater than n days, dividing the potential predictable users into predictable users; for users without regularity of electricity consumption, dividing the users into unpredictable users;
and step three, classifying the user forecast dates according to date types. Dividing the user prediction date into a holiday type as the holiday type; the forecast date of the user belongs to the working days and is divided into working day types; the forecast date of the user belongs to the rest days and is divided into the rest day type. Particularly, for the holiday type and the holiday type, if the historical data volume of the same type of electrical loads is less than m days, the same type of electrical loads are divided into an unpredictable holiday type and an unpredictable holiday type.
And step four, adaptively matching the corresponding prediction model based on different user types and user prediction day types. For the holiday type, the working day type and the holiday type of the predictable user, a Sequence to Sequence (Sequence 2 Sequence) model which can be fitted with a complex rule is used for predicting; predicting unpredictable holiday types and unpredictable holiday types of predictable users, holiday types and holiday types of potential predictable users and all predicted day types of the unpredictable users based on a similar day load model; for the workday type of a potential predictable user, a support vector machine with strong generalization capability and an XGBoost are used for prediction;
and fifthly, establishing a prediction model based on the characteristic data such as meteorological data and historical load characteristics.
Constructing a time feature set, and considering the month, holiday, day of the year, day of the month, day of the week, hour and minute of the time T to form the time feature set T of the time T time,t 。
Constructing a meteorological feature set, and selecting weather types and temperatures as the meteorological feature set T nwp 。
Constructing a historical load characteristic set, and selecting a historical value of the same time T in the previous period (day and week) and a sliding average value of the upper time and the lower time as the historical load characteristic set T of the time T load,t 。
Historical load feature set T load,t Time of dayMeteorological features T of T time,t And a temporal feature set T time,t Forming a feature vector T t 。
And establishing a similar daily load model. Dates belonging to the same date type are taken as similar days, such as weekdays, saturdays, sundays, holidays. And selecting the latest 5 similar daily loads as load samples for each type of date, wherein the samples with the daily electric quantity lower than 25% of the daily average electric quantity of 5 samples or higher than 200% of the daily average electric quantity of 5 samples are removed, and the residual samples are used for calculating the average value of all the moments to obtain the predicted load.
And establishing a support vector machine or XGboost model. Feature vector T t And carrying out normalization processing on the medium value type characteristic, and carrying out single-hot coding processing on the discrete type characteristic. Characteristic vector T of each historical time T t And a load P t Forming a data set D, dividing the data set D into a training set and a testing set according to a certain proportion, and for each sample, dividing the characteristic vector T t As input, the load P t And as output, establishing a support vector machine or an XGboost model, training the model on a training set, simultaneously evaluating on a test set, and selecting the model with the optimal evaluation effect as a final prediction model.
And establishing a Seq2Seq model. And modeling the time sequence of the load by combining a plurality of characteristic factors influencing the load, embedding multiple characteristics, and finally gradually decoding and predicting future load values to generate a predicted load sequence at 24 moments in the future. Specifically, feature vector T is divided into t And carrying out normalization processing on the medium value type characteristic, and carrying out single-hot coding processing on the discrete type characteristic. For time t, order
H t ＝{[T t-l ,P t-l ],[T t-l+1 ,P t-l+1 ],K,[T t-1 ,P t-1 ]}
F t ＝{T t+1 ,T t+2 ,K,T t+24 }
X t ＝[H t ,F t ]
Y t ＝{P t+1 ,P t+2 ,K,P t+24 }
Wherein X t As model input, Y t As output of the model, P t+k For the k-th time step real load data,
load data is estimated for the kth time step encoder, l is the recurrent neural network time window, and Ω (Θ) is a regularization term. Model principle as shown in fig. 3, the basic unit of the encoder and decoder is LSTM, the encoder and decoder are trained to minimize the objective function as ζ (Θ), and H is t Inputting into encoder, and in decoder, calculating hidden state c from encoder, and estimating load data from last time step encoder
Last time step encoder h t+k-1 And T t+k Input decoder LSTM unit g estimates to obtain estimated load data of k time step encoder
This in turn outputs the predicted load at 24 points in the future.
Inputting X of each time t of history t And load Y t And forming a data set D, and dividing the data set D into a training set and a testing set according to a certain proportion. After the model on the training set is trained, the evaluation is performed on the test set using the following evaluation indexes:
the root-mean-square error of the error,
the average absolute percentage error is calculated as the percentage of the absolute percentage error,
where Y represents the actual value in the test set,
and representing the predicted value obtained by the test set.
And step six, triggering a load prediction system and outputting a day-ahead load prediction value. The system module includes:
the data acquisition module collects and stores load data and meteorological data of a user by depending on an Internet of things platform and an intelligent electric meter;
and the data processing module is used for calculating and processing the original data to obtain a load value of hour granularity and further obtain a characteristic vector.
And the model building module is used for building and training a corresponding prediction model in a self-adaptive manner based on the user type and the prediction date type.
A load prediction module, a schematic diagram of a trigger prediction process of the prediction system is shown in fig. 2, a system user triggers the prediction module in a URL format and inputs parameters such as a prediction date and a date type, the system identifies a user type and a prediction date type, a prediction model is selected in a self-adaptive mode based on identification type matching, and a prediction result is returned to the user.
Specifically, in an example of the present invention, a user is a factory user, three months of historical power load data are provided, a user classification threshold n is set to 90 days, and the power curve of the user has a certain rule through analysis, so that the user is a predictable user, the predicted day type is divided into a working day type and a resting day type, and the load prediction is performed on the working day type and the resting day type of the user respectively. For the forecast day of the working day type, selecting a Seq2Seq model for forecasting, wherein the forecasting result is shown in fig. 4; for the holiday type, because the number of days of the same day type in the user historical data is too small, a similar day model is selected for prediction, and the prediction result is shown in fig. 5. Assuming that the user has only one monthly historical electricity consumption data, the historical electricity consumption load data of the latest month is taken to train the model, the user is a potential predictable user at this moment, and for the forecast day of the working day type, a support vector machine model suitable for small sample training is selected for forecasting, and the forecasting result is as shown in fig. 6.
Selecting different models for training and predicting according to different prediction day types, wherein the statistics of the prediction effects of the different models are as follows:
| preset user type | Predicting day type | Prediction model | RMSE | MAPE |
| Predictive user | Type of workday | Seq2Seq | 883.2 | 0.0207 |
| Predictive user | Type of weekday | Similar day model | 1431.6 | 0.0351 |
| Potential predictable user | Type of workday | Support vector machine | 1142.41 | 0.0276 |
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. This need not be, nor should all embodiments be exhaustive. And obvious variations or modifications of the invention may be made without departing from the scope of the invention.
Claims (9)
1. A user day-ahead load prediction method of an adaptive model is characterized by comprising the following steps:
step 1, acquiring and preprocessing user data and meteorological data;
step 2, classifying the users according to the electricity utilization regularity and the historical data quantity of the electricity utilization load;
step 3, classifying the user prediction dates according to date types;
step 4, based on different user types and user prediction day types, adaptively matching corresponding prediction models;
step 5, establishing a prediction model based on the time characteristic data, the meteorological characteristic data and the historical load characteristic data;
and 6, triggering the load prediction system and outputting the day-ahead load prediction value.
2. The adaptive model user day-ahead load prediction method according to claim 1, wherein in step 1,
the user data acquisition refers to acquiring historical data of the power load of the user at intervals of 1 hour;
acquiring the meteorological data, namely acquiring historical meteorological data corresponding to the historical data of the electric load and meteorological forecast data of a future day, wherein the meteorological data comprises a highest daily temperature, a lowest daily temperature, an average daily temperature and a weather type; if the user belongs to the type of a production enterprise, production plan data corresponding to the historical data of the electric load is required to be acquired, wherein the production plan data comprises a working day type, a holiday type and a holiday type;
the preprocessing refers to setting a value beyond the addition of 3 times of standard deviation to the average value as an abnormal value for each data field; for outliers and missing values, interpolated fit values are used instead of the treatment.
3. The method for predicting the day-ahead load of the user with the adaptive model according to claim 1, wherein in the step 2, the user type division method is as follows:
dividing users with regularity in electricity utilization and historical data quantity of electricity utilization load larger than n days into predictable users;
dividing users with regular electricity utilization and electricity utilization load historical data volume less than n days into potential predictable users; when the historical data volume of the electric load to be used is larger than n days, dividing potential predictable users into predictable users;
and classifying users with no regularity in electricity utilization into unpredictable users.
4. The method for predicting the day-ahead load of the user with the adaptive model according to claim 1, wherein in the step 3, the user prediction day type division method is as follows:
the forecast date of the user belongs to the working day and is divided into working day types;
for the user prediction date belonging to the holidays, dividing the user prediction date into holiday types;
dividing the predicted date of the user belonging to the holiday into holiday types;
particularly, for the holiday type and the holiday type, if the historical data volume of the same type of electrical loads is less than m days, the same type of electrical loads are divided into an unpredictable holiday type and an unpredictable holiday type.
5. The method for predicting the user's day-ahead load of the adaptive model according to claim 1, wherein in the step 4, the specific method for adaptively matching the corresponding prediction model is as follows:
for the holiday type, the workday type and the holiday type of the predictable user, a Sequence to Sequence model which can fit a complex rule is used for predicting;
predicting unpredictable holiday types and unpredictable holiday types of predictable users, holiday types and holiday types of potential predictable users and all predicted day types of the unpredictable users based on a similar day load model;
for the types of workdays of the potentially predictable users, the predictions are made using a support vector machine or an XGboost model.
6. The method for predicting the day-ahead load of the user with the adaptive model according to claim 1, wherein in the step 5, the specific method for establishing the predictive model is as follows:
(1) Constructing a feature vector T t ：
Constructing a time characteristic set, wherein a month, whether a holiday, a day in a year, a day in the month, a day in the week, an hour and a minute at the moment T form the time characteristic set T at the moment T time,t ；
Constructing a meteorological feature set, and selecting weather types and temperatures as the meteorological feature set T nwp ；
Constructing a historical load characteristic set, and selecting a historical value of the same time T in the previous period and a sliding average value of the upper time and the lower time as the historical load characteristic set T of the time T load,t ；
Historical load feature set T load,t Meteorological feature set T nwp And a temporal feature set T time,t Forming a feature vector T t ；
(2) Establishing a similar daily load model:
taking dates belonging to the same date type as similar days; selecting the latest x similar daily loads as load samples for each type of date, wherein the samples with daily electric quantity lower than 25% of daily average electric quantity of x samples or higher than 200% of daily average electric quantity of x samples are removed, and the residual samples are used for obtaining the average value of all the moments to obtain the predicted load;
(3) Establishing a support vector machine or XGboost model:
feature vector T t Carrying out normalization processing on the medium value type characteristics, and carrying out one-hot coding processing on the discrete type characteristics; characteristic vector T of each historical time T t And a load P t Forming a data set D, dividing the data set into a training set and a testing set according to a certain proportion, and for each sample, dividing the characteristic vector T t As input, load P t As output, establishing a support vector machine or an XGboost model;
(4) Establishing a Sequence to Sequence model:
modeling the time sequence of the load by combining a plurality of characteristic factors influencing the load, embedding multiple characteristics, and finally gradually decoding and predicting a future load value to generate a predicted load sequence at 24 moments in the future;
(5) And training the model on a training set, simultaneously evaluating the model on a test set, and selecting the model with the optimal evaluation effect as a final prediction model.
7. The method of claim 1, wherein in step 6, the load prediction system comprises the following modules:
the data acquisition module is used for acquiring load data and meteorological data of a user;
the data processing module is used for calculating and processing the original data to obtain a load value of hour granularity and further obtain a feature vector;
the model building module is used for building and training a corresponding prediction model in a self-adaptive manner based on the user type and the prediction date type;
and the load prediction module is used for inputting parameters by a system user, adaptively selecting a prediction model based on the user type and the prediction date type, and returning a prediction result to the user.
8. The method for predicting the day-ahead load of the user with the adaptive model according to claim 6, wherein the specific method for establishing the Sequence to Sequence model in the step (4) in the step 5 comprises the following steps:
feature vector T t Carrying out normalization processing on the median type features, and carrying out one-hot coding processing on the discrete type features; for time t, let
H t ＝{[T t-l ,P t-l ],[T t-l+1 ,P t-l+1 ],K,[T t-1 ,P t-1 ]}
F t ＝{T t+1 ,T t+2 ,K,T t+24 }
X t ＝[H t ,F t ]
Y t ＝{P t+1 ,P t+2 ,K,P t+24 }
Wherein X t As model input, Y t As output of the model, P t+k For the k-th time step real load data,
estimating load data for a kth time step encoder, wherein l is a cyclic neural network time window, and omega (theta) is a regular term; the basic unit of the encoder and decoder is LSTM, which is trained to minimize the objective function as ζ (Θ), H t Inputting into encoder, and in decoder, calculating hidden state c from encoder, and estimating load data from last time step encoder
Last time step encoder h t+k-1 And T t+k Input decoder LSTM unit g estimates to obtain estimated load data of k time step encoder
This in turn outputs the predicted load at 24 points in the future.
9. The method for predicting the day-ahead load of the user with the adaptive model according to claim 6, wherein the specific method of step (5) in step 5 is as follows:
inputting X of each time t of history t And load Y t Forming a data set D, and dividing the data set D into a training set and a test set according to a certain proportion; after the model on the training set is trained, the evaluation is performed on the test set using the following evaluation indexes:
the root-mean-square error of the error,
the average absolute percentage error is calculated as the percentage of the absolute percentage error,
where Y represents the actual value in the test set,
and expressing the predicted value obtained by the test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210729317.3A CN115222106A (en) | 2022-06-24 | 2022-06-24 | User day-ahead load prediction method of self-adaptive model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210729317.3A CN115222106A (en) | 2022-06-24 | 2022-06-24 | User day-ahead load prediction method of self-adaptive model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115222106A true CN115222106A (en) | 2022-10-21 |
Family
ID=83610222
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210729317.3A Pending CN115222106A (en) | 2022-06-24 | 2022-06-24 | User day-ahead load prediction method of self-adaptive model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115222106A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115798900A (en) * | 2023-01-13 | 2023-03-14 | 江苏凡高电气有限公司 | Intelligent capacity-adjusting transformer convenient to disassemble and assemble |
| CN115936184A (en) * | 2022-11-10 | 2023-04-07 | 国网冀北电力有限公司计量中心 | Load prediction matching method suitable for multi-user types |
| CN117370770A (en) * | 2023-12-08 | 2024-01-09 | 江苏米特物联网科技有限公司 | Hotel load comprehensive prediction method based on shape-XGboost |
| CN117458483A (en) * | 2023-11-22 | 2024-01-26 | 国网新疆电力有限公司乌鲁木齐供电公司 | Electric load prediction method based on Monte Carlo simulation |
-
2022
- 2022-06-24 CN CN202210729317.3A patent/CN115222106A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936184A (en) * | 2022-11-10 | 2023-04-07 | 国网冀北电力有限公司计量中心 | Load prediction matching method suitable for multi-user types |
| CN115936184B (en) * | 2022-11-10 | 2024-09-03 | 国网冀北电力有限公司计量中心 | Load prediction matching method suitable for multi-user types |
| CN115798900A (en) * | 2023-01-13 | 2023-03-14 | 江苏凡高电气有限公司 | Intelligent capacity-adjusting transformer convenient to disassemble and assemble |
| CN117458483A (en) * | 2023-11-22 | 2024-01-26 | 国网新疆电力有限公司乌鲁木齐供电公司 | Electric load prediction method based on Monte Carlo simulation |
| CN117370770A (en) * | 2023-12-08 | 2024-01-09 | 江苏米特物联网科技有限公司 | Hotel load comprehensive prediction method based on shape-XGboost |
| CN117370770B (en) * | 2023-12-08 | 2024-02-13 | 江苏米特物联网科技有限公司 | Hotel load comprehensive prediction method based on shape-XGboost |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Wang et al. | Adaptive learning hybrid model for solar intensity forecasting | |
| CN115222106A (en) | User day-ahead load prediction method of self-adaptive model | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN101383023B (en) | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation | |
| CN108921339B (en) | Quantile regression-based photovoltaic power interval prediction method for genetic support vector machine | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN111079989B (en) | DWT-PCA-LSTM-based water supply amount prediction device for water supply company | |
| CN108053082B (en) | Power grid medium and long term load prediction method based on temperature interval decomposition | |
| CN112329990A (en) | User power load prediction method based on LSTM-BP neural network | |
| CN115496627A (en) | Method and system for evaluating response potential of adjustable resource | |
| CN111680818B (en) | Short-term reactive load prediction method and system | |
| Kofinas et al. | Daily multivariate forecasting of water demand in a touristic island with the use of artificial neural network and adaptive neuro-fuzzy inference system | |
| Tavares et al. | Comparison of PV power generation forecasting in a residential building using ANN and DNN | |
| CN115796915A (en) | Electricity price prediction method and system for electricity trading market | |
| CN115238948A (en) | Method and device for predicting power generation capacity of small hydropower station | |
| CN117407681B (en) | Time sequence data prediction model establishment method based on vector clustering | |
| CN104915727A (en) | Multi-dimensional isomorphic heterogeneous BP neural network optical power ultrashort-term prediction method | |
| CN113570414A (en) | Electricity price prediction method for optimizing deep neural network based on improved Adam algorithm | |
| CN117973953A (en) | Construction method and device of demand response potential prediction model of resident temperature control load | |
| CN116826745B (en) | Layered and partitioned short-term load prediction method and system in power system background | |
| CN113112085A (en) | New energy station power generation load prediction method based on BP neural network | |
| CN112330017A (en) | Power load prediction method, power load prediction device, electronic device, and storage medium | |
| CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm | |
| CN115907228A (en) | Short-term power load prediction analysis method based on PSO-LSSVM |
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 |