CN114511135A - Artificial intelligence-based short-term load prediction method, computer device and storage medium - Google Patents
Artificial intelligence-based short-term load prediction method, computer device and storage medium Download PDFInfo
- Publication number
- CN114511135A CN114511135A CN202111658345.2A CN202111658345A CN114511135A CN 114511135 A CN114511135 A CN 114511135A CN 202111658345 A CN202111658345 A CN 202111658345A CN 114511135 A CN114511135 A CN 114511135A
- Authority
- CN
- China
- Prior art keywords
- load
- short
- holiday
- decoder
- term
- 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
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 20
- 238000003860 storage Methods 0.000 title claims abstract description 8
- 238000000034 method Methods 0.000 title claims description 55
- 238000013277 forecasting method Methods 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 abstract description 8
- 230000008859 change Effects 0.000 description 14
- 238000004422 calculation algorithm Methods 0.000 description 13
- 238000011160 research Methods 0.000 description 13
- 238000004458 analytical method Methods 0.000 description 10
- 238000013528 artificial neural network Methods 0.000 description 10
- 238000012706 support-vector machine Methods 0.000 description 8
- 238000010276 construction Methods 0.000 description 6
- 238000007405 data analysis Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000010219 correlation analysis Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000003442 weekly effect Effects 0.000 description 2
- 206010026749 Mania Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000009960 carding Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a short-term load forecasting method based on artificial intelligence, computer equipment and a storage medium, wherein the short-term load forecasting method comprises the following steps: firstly, selecting load characteristics, lag characteristics, time characteristics and meteorological characteristics which have influence on the load, and constructing dynamic coding characteristics aiming at holidays with different durations; then, modeling is carried out by adopting a Seq2Seq network, and a short-term load prediction model is established; and finally, predicting the short-term load by adopting a short-term load prediction model. According to the invention, a holiday dynamic coding mode is constructed according to the length of different holidays, and a holiday load model based on LightGBM is established by combining numerical mode prediction, so that the holiday load prediction effect is improved.
Description
Technical Field
The invention relates to a power engineering technology, in particular to a power load prediction technology.
Background
The power grid short-term load prediction is an important component module of a power system and is widely researched and paid attention to in various countries around the world. Traditional power grid short-term load prediction methods include trend extrapolation, regression analysis, exponential smoothing, and the like. However, as power systems are continuously developed, the demand for short-term load prediction is gradually increased, and more modern prediction methods are also introduced into the short-term load prediction, such as machine learning algorithms and artificial intelligence algorithms.
The time series method is one of the more classical methods in the power grid short-term load prediction, and because the power grid short-term load has certain periodic characteristics and is influenced by other factors such as weather, seasons, holidays and the like, in recent years, the research on the short-term load prediction based on the time series method mainly focuses on an autoregressive integrated moving average (ARIMA) model [3] method and an improved version thereof. Chaturvedi [4] researches a prediction method combining ARIMA, wavelet transformation and a neural network, and Hossein Javedani Sadai [5] researches an ARIMA which considers meteorological factors and week and season trends in an improved mode to improve the accuracy of short-term load prediction. The Wanzhi macro [6] provides a regression residual error model based on ARIMA to predict short-term load, and the method improves the traditional ARIMA model and introduces the highest, lowest and week types of each day to improve the accuracy of load prediction in consideration of the fact that load change is greatly influenced by temperature and date types.
With the recent development of machine learning algorithms, short-term load prediction based on machine learning algorithms is increasing. The Wuqian hong [7] carries out systematic carding on the short-term load prediction method based on the traditional machine learning algorithm and the deep learning algorithm at present, and is considered to be roughly divided into a shallow learning method and a deep learning method on the short-term load prediction problem.
The short-term load prediction algorithm based on the shallow learning method mainly comprises traditional machine learning algorithms such as a Support Vector Machine (SVM) [8], a machine forest (RF) [9], a gradient boosting tree (GBDT) [10] and the like. In the short-term load prediction problem in recent years, researchers mainly try to further expand the methods, for example, Chia-NanKo [11] combines SVM, a Radial Basis Function Neural Network (RBFNN) and a Kalman filter to predict the short-term load, the optimal RBFNN structure and initial parameters are selected through the SVM, the Kalman filter is used as a learning algorithm of optimized parameters, and a good prediction effect is achieved. The Biyun sail [10] uses a short-term load prediction method based on GBDT, and original data is sampled in a Bootstrap mode in a training process, so that the generalization performance of a short-term load prediction model is improved. The golden beam [12] provides a combined prediction model based on SVM and LSTM, and the prediction accuracy of the model is improved through combined prediction. Zhongbin et al [13] proposed a LightGBM ultra-short term load prediction model based on particle swarm optimization, which can effectively improve the accuracy of load prediction.
The deep learning method is one of the hot spots of short-term load prediction research due to its strong feature learning and expression ability. Wuchangli [14] constructs DNN considering historical load, weather and time types for short-term load prediction, and Kunjin Chen [15] proposes that the short-term load prediction is carried out by utilizing a deep residual error network structure on the basis of improvement of a DNN network structure, and the prediction accuracy of a model is improved by a model set method. Hushuang [16] also proposes a DNN-based urban power grid load prediction method, and obtains better load prediction accuracy by improving data characteristics. Priyanka Singh [17] the design of data features is thought to have a large impact on DNN prediction results by performing comparative analysis on a variety of DNNs on the short-term load prediction problem.
Because of the superior results of the Recurrent Neural Networks (RNNs) in processing sequence data, there are also a number of researchers exploring the use of RNNs and their variants for the short-term load prediction problem of the power grid. The RNN is popular with a long-short term memory unit (LSTM) and a gated cycle unit (GRU), both of the two unit structures use a gating mechanism to solve the problem of gradient disappearance in the basic RNN training process, and the method has better effect in processing long-time sequence data. A mixed model experiment result based on a Convolutional Neural Network (CNN) and a long-short term memory (LSTM) network is provided aiming at the characteristics of time sequence and nonlinearity of load data in Lu-Relay-flight and the like [18], and the prediction method provided in the text has higher prediction precision than a traditional load prediction method, a random forest model load prediction model method and a standard LSTM network load prediction method. Daniel L.Marino [19] proposed LSTM and LSTM-based Sequence to Sequence (Sequence 2 Sequence) structures on the basis of building-level load prediction, modeled the load prediction, and tested the prediction models with the frequency of minutes and the frequency of hours respectively, and found that the prediction effect of the LSTM-based Sequence 2 Sequence network is better than that of the common LSTM, especially on the prediction of a longer load Sequence.
Because the short-term load of the power grid is influenced by multiple factors such as historical load, weather and time, a plurality of scholars carry out deep research on the construction of the short-term load forecasting characteristics. Generally, only date type characteristics are considered in the short-term load prediction problem, and Magnus Dahl [20] improves the accuracy of load prediction by introducing multiple time characteristics such as hours, weeks, weekends, months and the like and multiple data such as national festivals and holidays, school holidays and the like. Zhu Virwei [21] discusses the research on the influence of meteorological factors on the load characteristics of the power grid, and further refines a load prediction model considering the meteorological factors by analyzing the sensitivity of meteorological sensitive loads to the meteorological factors and the load change rule in each period of Hangzhou city so as to obtain a better short-term load prediction effect. In the feature extraction method, KasunAmarangehe [22] proposes an algorithm based on a Convolutional Neural Network (CNN) to extract historical load features, and the historical load is input as the convolutional neural network, and the network output feature information is combined with the time features of the month, the week, the hour and the like at the moment to serve as the input features of a full connection layer. The Yangkai provided learning based on short-term load of the self-encoder in the stack, and the self-encoder has better learning capability on the characteristics such as load characteristics, meteorological data, date attributes and the like, so as to obtain a more accurate load prediction result. In addition, many scholars extract the characteristics of the load curve by methods such as wavelet analysis and empirical mode decomposition to obtain finer local details in the load curve, and Liuwenbo proposes an improved wavelet neural network prediction model. S.J.Yao [23] proposes a method for predicting the short-term load of a power grid based on wavelet transformation and a neural network. In consideration of the phenomenon of data characteristic missing frequently occurring in the data collected by a power system, the Tpitch [24] provides a GBDT-based power data missing data filling method and a model fusion-based power load prediction method, tests are carried out on global energy data prediction match data, and the accuracy of power load prediction is improved through model fusion.
Holiday load prediction is also one of the difficulties in short-term load prediction. The holiday load presents inherent uncertainty and volatility due to the influence of a plurality of factors such as external weather, self characteristics, national rest policies and the like, and how to dig the internal relation between various relevant factors and the holiday load through a feasible strategy and a forecasting method effectively improves the forecasting precision, so that the holiday load is a research hotspot for forecasting the current short-term load. In the aspect of holiday load prediction, a prediction method [25,26] based on historical similar days is generally adopted, a curve resolution function is established for the plum [25] to search for similar day samples of holidays, and holiday load prediction is improved by improving the selection accuracy of the similar days. Sang Fu Min [26] discusses the daily load prediction of the power grid in the Chongqing area during the national legal holiday, and the accuracy of the load prediction during the holiday is improved mainly by carrying out quantitative analysis on the load of the historical holiday and dividing the daily load curve into different time periods for prediction.
The research result shows that the load characteristics of the holidays in different years are different, so that the strategy of historical similar days is considered in holiday load prediction, and the strategy does not necessarily play a positive role in the accuracy of the prediction result. Zhangqiao elm [27] marks different festivals and holidays by observing the load curve mode of the festivals and holidays, and meanwhile, an SVM-based holiday load prediction model is constructed by combining historical load weather and the like. Zhangqiao elm found by observing holiday loads: 1) during statutory holidays, the load level is significantly reduced; the type of long holiday in 7 days has a wide influence range, particularly in the spring festival, and the spread range is from late April to late April; the influence range of the 3-day small and long false type is relatively small; 2) the load on the day of the holiday is lowest, and the load trends are classified into a "/" type change trend (holiday on the first day of the holiday), a "V" type change trend (holiday in the middle of the holiday), and a "\\" type change trend (holiday on the last day of the holiday) depending on where the holiday is located. According to the above two rules, the holidays are marked in the following way:
a) the holiday characteristics of the New year, the spring festival, the Qingming festival, the labor festival, the end festival, the mid-autumn festival and the national day festival are marked as 1,2, 3, 4, 5, 6 and 7 respectively.
b) For the 3-day small and long false type, "/" type load change trend features are respectively marked as 1,2 and 3; the V-type load change trend characteristics are respectively marked as 3, 1 and 2; the "\\" type load change trend features are labeled 3, 2, and 3, respectively.
c) For the 7-day long vacation type, the load trend signatures for the labeled vacations were 1,2, 3, 4, 5, 6, 7, respectively.
Reference to the literature
[1] Kang Chongqing, Xia Qing, Liu Mei, electric power system load prediction [ M ]. Beijing, China electric power Press, 2007.
[2] Forest and bin, red Tong, realizing carbon peak and carbon neutral and focusing on three ' overall ' J ' price theories and practices, 2021(1):4.
[3] Aixin, Zhou Shi, Wei Yan Nu, et al, transferable load bidding strategy based on autoregressive integral moving average model [ J ] Power System Automation, 2017,41(20):26-31.
[4]Fard A K,Akbari-Zadeh M R.A hybrid method based on wavelet,ANN and ARIMA model for short-term load forecasting[J].Journal of Experimental&Theoretical Artificial Intelligence,2014,26(2):167-182.
[5]Sadaei H J,Frederico Gadelha
Cidiney José da Silva,et al.Short-term load forecasting method based on fuzzy time series,seasonality and long memory process[J].International Journal of Approximate Reasoning,2017,83(C):196-217.
[6] Time series-based short-term load prediction study of electrical power systems [ D ]. south china university, 2012.
[7] Smart grid predictive analysis in "artificial intelligence +" times by wuqianhong, hanbei, von lin, et al [ J ]. university of shanghai university of transportation (natural edition), 2018,52(10): 1206-.
[8] Wuqian red, high military, Houguansong, etc. the short-term load prediction support vector machine algorithm [ J ] for influencing factor multi-source heterogeneous fusion is realized, 2016,40(15):67-72 for power system automation.
[9] Design and implementation of the large data analysis and prediction system for the electrical loads [ D ] Shandong university, 2018.
[10] Research on short-term load prediction models based on fuzzy Bagging-GBDT on Biyunfan, Oyan, Zhangzhi sanden and Sun Wen Hui [ J ]. electric power system and its automated bulletins: 1-6.
[11]Ko C N,Lee C M.Short-term load forecasting using SVR(support vector regression)-based radial basis function neural network with dual extended Kalman filter[J].Energy,2013,49:413-422.
[12] Application of a combination model of a golden beam SVM and a neural network to short-term power load prediction research [ D ].2018.
[13] Zhongbin, Jiangban, Zhao Zhen Yu, etc. LightGBM ultra-short term load prediction research based on particle swarm optimization [ J ] energy and energy conservation, 2021(2):5.
[14] Wuchang et al, research on daily load prediction of power grid based on artificial neural network [ D ]. university of Zhejiang, 2011.
[15]Kunjin C,Kunlong C,Qin W,et al.Short-term Load Forecasting with Deep Residual Networks[J].IEEE Transactions on Smart Grid,2018:1-1.
[16] Hushuang, consider the city power grid load prediction [ D ] of wind power access, Shandong university, 2018.
[17]Singh,Priyanka,Dwivedi,et al.Integration of new evolutionary approach with artificial neural network for solving short term load forecast problem[J].Applied Energy,2018,217:537-549.
[18] Continental descent, sachima culture, poplar macro, and the like, a short-term load prediction method based on a CNN-LSTM hybrid neural network model [ J ] power system automation, 2019(8):7.
[19]Marino D L,Amarasinghe K,Manic M.[IEEE IECON 2016-42nd Annual Conference of the IEEE Industrial Electronics Society-Florence,Italy(2016.10.23-2016.10.26)]IECON 2016-42nd Annual Conference of the IEEE Industrial Electronics Society-Building energy load forecasting using Deep Neural Networks[J].2016:7046-7051.
[20]Dahl M,Brun A,Kirsebom O,et al.Improving short-term heat load forecasts with calendar and holiday data[J].Energies,2018,11(7):1678.
[21] Zhu Zheng Wei, research on influence of meteorological factors on load characteristics of power grid [ D ]. Hangzhou Zhejiang university, 2008.
[22]Amarasinghe K,Marino D L,Manic M.Deep neural networks for energy load forecasting[C]//Industrial Electronics(ISIE),2017 IEEE 26th International,Symposium on.IEEE,2017:1483-1488.
[23]Yao S J,Song Y H,Zhang L Z,et al.Wavelet transform and neural networks for short-term electrical load forecasting[J].Energy conversion and management,2000,41(18):1975-1988.
[24] Study on electric power load prediction algorithm under data loss condition [ J ]. university of science and technology in china, 2016.
[25] Li shun, huangjia, wu yin, et al, a holiday short-term load prediction method based on fractal characteristics to correct meteorological similar days [ J ] grid technology, 2017,41 (6): 1949-1955.
[26] Sang fumin, hu run zi national legal festival, holiday, Chongqing area, Power grid daily load prediction analysis and research [ J ]. Chongqing high electric power department school report, 2015,20 (6): 42-45.
[27] Zhang Qiao elm, Chua Qina, Liu Si Jie, etc. festival holiday short-term load prediction based on sample expansion and signature [ J ]. Guangdong electric power, 2019, 032 (007): 67-74.
The above is mainly the development of a short-term load prediction algorithm in recent years, and with the continuous development of the electric power industry in China, the updating on a short-term load prediction system is urgently needed so as to improve the overall load prediction level of an electric power dispatching department and realize refined electric power dispatching management.
Disclosure of Invention
Aiming at the problems existing in the existing holiday load prediction, the invention aims to provide a short-term power load prediction method based on artificial intelligence, and improve the short-term load prediction precision including holidays.
In order to solve the technical problems, the invention adopts the following technical scheme:
the short-term power load prediction method based on artificial intelligence comprises the following steps:
firstly, selecting load characteristics, lag characteristics, time characteristics and meteorological characteristics which have influence on loads, and constructing dynamic coding characteristics aiming at holidays with different durations, wherein the load characteristics comprise daily sum, daily minimum and daily average load, the lag characteristics are loads or meteorology at a certain moment in the past, the time characteristics are months, days of the week, weekends, hours and minutes, and the meteorological characteristics comprise original values, mean values, maximum values and minimum values of temperature and humidity;
then, modeling is carried out by adopting a Seq2Seq network to establish a short-term load prediction model, wherein the Seq2Seq network consists of an encoder and a decoder, and the encoder inputs a characteristic xtAs shown in the following formula,
xt＝[yt-1，wt，mt]
wt＝[humi，temp，dmax_humi，dmin_humi，davg_humi，dmax_temp，dmin_temp，davg_temp]
mt＝[month,dayofweek,weekend,hour,min]
wherein, ytRepresenting historical real load characteristics at the time t; w is atThe meteorological features at time t include temperature temp, humidity humi and daily maximum value d thereofmax_Minimum daily value dmin_And daily average value davg_；mtThe time series characteristics of the t moment are represented and comprise month codes month, day of week dayofweek, weekend or not, hour codes hour and minute codes min;
the decoder input characteristics are shown as follows,
it is assumed that the current time is t, the basic short-term load prediction task needs the load size at the time of t + M (M is 1,2, …,192) in the future, and when M is 1, the input characteristic of the decoder is xt+1I.e. including the real load y at time ttWeather characteristics of t +1 time mode forecast
time series characteristic m at time t +1t+1(ii) a When M is>1, the input characteristic of the decoder is xt+M；
And finally, predicting the short-term load by adopting a short-term load prediction model.
Preferably, the dynamic coding is performed according to different holiday lengths, and the dynamic coding mode is as follows:
supposing that the holiday duration of the current holiday is n days, the holiday is coded as-n, if the first day after the holiday is also the holiday, the holiday is coded as- (n-1), and if the second day after the holiday also belongs to the holiday, the holiday is coded as- (n-2), and the holiday is deduced in the secondary category;
the code of the current day of the spring festival is 1, the code of the first day after the spring festival is-1, the code of the second day after the spring festival is-2, and so on until the code reaches the 30 th day after the spring festival; the code is 1 for the day before the spring festival, 2 for the second day before the spring festival, and so on until the code is 29 days before the spring festival, and the code is 0 for the non-holiday.
Preferably, the specific steps of modeling by using the Seq2Seq network are as follows:
a) generating a hidden state of an encoder using LSTM for an input sequence XNEncoding is performed such that each input generates a hidden state/output h after passing the input sequence through the encoder LSTMN；
b) The last layer of hidden state C is connected with the input of the decoder and is input into the decoder together with the input of the decoder to obtain a first output YTAnd decoder hidden state ST；
c) The output obtained by the decoder and the hidden state of the decoder are input into the decoder together to obtain the output and the hidden state of the decoder, and the step is repeated until the sequence is finished;
d) and optimizing parameters, namely selecting an optimizer to optimize the model parameters.
Preferably, at the time step setting, the time step of the decoder is 192, i.e. the short-term load of 48 hours in the future is predicted; the time step of the encoder is set to 288, i.e. the load related information of the historical 3 days is used to predict the short-term load of two days in the future.
Preferably, the short-term load prediction model adopts a Teacher-shaping strategy for learning, and in the decoder process, the output of the encoder is randomly selected as the input of the decoder or the true value of the training data is selected as the input of the decoder through probability setting.
The invention also provides a computer device comprising at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executing the computer-executable instructions stored by the memory causes the at least one processor to perform the artificial intelligence based short term power load prediction method.
The invention also provides a computer readable storage medium, wherein the computer readable storage medium stores computer executable instructions, and when a processor executes the computer executable instructions, the method for predicting the short-term power load based on the artificial intelligence is realized.
By adopting the technical scheme, the invention has the following beneficial effects:
1. according to the time sequence of the load and the characteristic of being susceptible to weather, a non-holiday load model based on Seq2Seq is constructed by combining numerical model forecasting, and the load prediction precision is improved.
2. A holiday dynamic coding mode is constructed according to different holiday lengths, a holiday load model based on LightGBM is established by combining numerical mode prediction, and the holiday load prediction effect is improved.
Therefore, the invention greatly improves the accuracy of short-term power load prediction including holidays.
The following detailed description and the accompanying drawings are included to provide a further understanding of the invention.
Drawings
The invention is further described with reference to the following figures and detailed description:
FIG. 1 is a flow chart of a short term load prediction technique of the present invention;
FIG. 2A is a graph of the mean change in the week of the load;
FIG. 2B is a graph of the mean change of the load over the month;
FIG. 2C is a graph of the mean change in hours of load;
FIG. 3 is a diagram of a dynamic holiday coding process;
FIG. 4 is a schematic diagram of an LSTM-based Seq2Seq model;
FIG. 5 is a diagram of the basic elements of an LSTM network;
fig. 6 is a diagram of a Seq2Seq short term load prediction network based on LSTM.
Detailed Description
The technical solutions of the embodiments of the present invention are explained and illustrated below with reference to the drawings of the embodiments of the present invention, but the following embodiments are only preferred embodiments of the present invention, and not all embodiments. Based on the embodiments in the implementation, other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present invention.
The existing load prediction method has partial limitations and mainly comprises the following aspects:
1) the sequential relationship of daily loads is not fully considered. On the basis of the problem of short-term load prediction of a power grid, the principle of 'big-end-up-and-small-end-up' influence is important, the existing load prediction method adopts different training models to predict loads at different moments on the same day, and the sequence relation of the loads is not considered.
2) Current predictions do not adequately account for the effects of meteorological factors. Meteorological factors are the most complex factors affecting short term loads. Different meteorological factors can affect the power load to different degrees, so that the comprehensive and cross effects of the meteorological factors are considered when the load is predicted.
3) The current holiday coding mode is too complicated, the coding length is fixed, the holiday dynamic change cannot be adapted, the expansibility is poor, and people are not considered to be in a rest around a long holiday.
In conclusion, in order to improve the short-term load prediction accuracy, numerical mode prediction data are introduced, a non-holiday load prediction model based on Seq2Seq and a LightGBM holiday load prediction model based on holiday dynamic coding are constructed, and the short-term load prediction accuracy is effectively improved.
As shown in FIG. 1, the invention provides a short-term power load prediction method based on artificial intelligence, which is mainly divided into three parts, wherein the first part is data analysis, the second part is feature engineering, and the third part is model construction.
The main flow of the short-term load prediction technology is as follows:
1. and (3) data analysis: trend analysis, abnormal value analysis, correlation analysis and statistic analysis are carried out aiming at historical load and historical weather;
2. characteristic engineering: selecting characteristics having influence on the load through the data analysis;
3. dynamic coding of festivals and holidays: constructing dynamic coding characteristics aiming at festivals and holidays with different durations;
4. model construction: based on the characteristics, a non-holiday model based on seq2seq and a holiday model based on LightGBM are constructed, and parameters are respectively adjusted for the two models, so that the models can better fit data and have effective generalization capability.
1. Data analysis
The embodiment uses the power load data and the meteorological data between 2014-1-1 to 2019-8-10 in a certain county of Zhejiang province. Wherein the daily load data is 15min data by 15min, and the daily load data is 96 points. The meteorological data are hourly data and mainly include temperature, humidity and the like. When data analysis is carried out, the following aspects are mainly included:
1) timing law analysis
The power grid load curve is closely related to the production and life laws of people, so that the power grid load curve has obvious periodicity in time similar to the life laws of people, and mainly shows that the daily load curve has certain similarity and the weekly load curve also has certain periodicity. By taking a certain county in Zhejiang province as an example, loads of the bijour days such as 2019-1-5 and 2019-1-6 are obviously lower than that of working days by observing the daily load sequence, and load modes of the bijour days are different.
Meanwhile, a load time chart of 2014-plus 2019 is drawn, and the load patterns during the spring festival, the five-one festival, the national day, the New year and other large festivals are completely different from the normal situation. The loads of the holidays are lower than those of the holidays before and after the holidays, and the holidays are of V-shaped structures, so that a foundation is provided for the construction of the characteristics of the holidays. From the annual change point of view, the load is on the whole in an increasing trend. And national statutory holidays exhibit the same regularity each year.
2) Analysis of outliers
According to the load sequence diagram, 2016-9-4 to 2019-9-5 shows different rules from other years, and is detected to be the G20 peak meeting in Hangzhou China, so that data before and after 10 days are deleted. Meanwhile, the special holidays of 2014-2019 are consulted, and it is found that 2015 is 70 weeks of the national anti-solar war and world anti-Fascis war victory, and the holidays are scheduled for rest and are released from 9 months 3 to 5 days, so that the data of 3 days are deleted as abnormal values. In addition, for load and meteorological data, missing values exist at some point in time, and need to be filled. Interpolation methods are various, and mean filling of the time, day, month and year is adopted here.
3) Meteorological factor correlation analysis
The load is influenced by weather besides the regularity of the load. Therefore, the electricity utilization rules of different months of loads in Zhejiang province are checked by plotting the temperature-load correlation and the humidity-load correlation.
The temperature has the highest correlation with load in the summer of 6-9 months, mainly the relatively large peak load caused by high temperature smoldering. Humidity is also the highest but negative correlation in the summer months 6-9. When the humidity is high, the weather is cool, and people can reduce the use of electric appliances such as an air conditioner and the like. Therefore, it is necessary to add meteorological factors in the subsequent modeling process, and take the statistical values thereof such as the mean value and the maximum and minimum values as the characteristic input model.
4) Statistical feature analysis
Graphs showing the changes of the week-average, month-average and hour-average values of the loads 2017-1-1 to 2019-8-11 are plotted, as shown in fig. 2A, 2B and 2C. As can be seen from the weekly mean, sunday to wednesday showed a rising trend, wednesday to saturday showed a falling trend, wednesday was the highest, saturday was the lowest, consistent with the analysis results of the above time chart. As can be seen from the monthly mean value graph, the load is higher in summer, the load is increased in winter and is lowest in spring. From the time-averaged value graph, the power consumption peak period is 9: 00-19: 00, which is consistent with the working time of people; the electricity consumption peak period is 1:00-8:00, which is consistent with the rest time of people. It was further observed that the loads at 11 and 17 are presented as peaks, both at the time of lunch and dinner, so that the loads are highest during the morning and afternoon hours, while the loads are lowest at 13 o' clock between the two peaks, which may be related to the lunch break habit of the country. And constructing load statistic characteristics according to the rules, specifically in a characteristic engineering part.
2. Feature engineering
The load data is time series data, and therefore, the characteristics of the load itself such as the total daily load, the load at the same time of the past day, and the like can be extracted, and at the same time, the load is also affected by weather factors such as temperature, humidity, and the like, and therefore, the following characteristics can be extracted:
TABLE 1 characteristics for load prediction
| Feature(s) | Description of the invention |
| Characteristic of load | Total daily, minimum daily and average dailyUniform load |
| Hysteresis characteristics | Load or weather at a past moment |
| Temporal characteristics | Month, day of week, whether weekend, hour, minute |
| Weather characteristics | Original, mean, maximum, minimum values of temperature, humidity |
A total of 160 features are selected, and are described as follows:
1) and (4) load characteristics. The sum, the maximum and the minimum of the daily load change curves, the average value and the like all influence the load curve, so the sum, the maximum and the average value of the daily load are taken as the statistical characteristics of the load, and the total number of the characteristics is 3.
2) A hysteresis feature. Since future loads are affected by historical loads and historical weather, 288 historical points of load and weather temperature are selected, with a lag being selected every hour to avoid data redundancy. There are 144 features.
3) And (4) time characteristics. Due to the large difference of the load curves in different seasons, the sequence order of 1 month to 12 months is set to be 0 to 11 for 12 months. The loading sequence has a certain periodicity every week. The week number is set to 0 to 6 in the order from Monday to Sunday. Because the working day and non-working day curves are different, whether weekends exist or not is distinguished, and 0-1 coding is adopted. The hour code is directly coded by 1-24, and the minutes codes 1-4 for 15 minutes, 30 minutes, 45 minutes and 0 minute respectively. There are 5 features in total.
4) And (4) meteorological characteristics. The load curve has seasonal variations, which are mainly influenced by two meteorological factors, temperature and humidity. The original values of temperature and humidity and the statistical values thereof such as daily average value, daily maximum value and daily minimum value are selected and used as meteorological features to be input into the model. There are 8 features in total. The meteorological features mainly comprise historical meteorological features and numerical mode forecast meteorological features, wherein the historical meteorological features are used for feature correlation analysis and modeling, and the numerical mode forecast meteorological features are used for model prediction.
3. Dynamic coding of holidays and festivals
When the holiday load is predicted, because the time lengths of different holidays are not uniform, dynamic coding is carried out according to the length of the different holidays. As shown in fig. 3, the dynamic encoding method is as follows:
assuming that the holiday duration of the current holiday is n days, the holiday is coded as-n, if the first day after the holiday is also holiday, the holiday is coded as- (n-1), and if the second day after the holiday also belongs to holiday, the holiday is coded as- (n-2), the next class is deduced. Because the spring festival holidays are long and the holiday times of the work production units are not completely uniform, the work production units are coded separately. The code of the current day of the spring festival is 1, the code of the first day after the spring festival is-1, the code of the second day after the spring festival is-2, and so on until the code reaches the 30 th day after the spring festival. The code is 1 for the day before the spring festival, 2 for the second day before the spring festival, and so on until the code is 29 days before the spring festival. And the non-holiday code is 0.
4. Non-holiday load prediction model based on Seq2Seq
The Seq2Seq network is adopted for modeling in consideration of the time sequence of the load and the characteristic that the future load is influenced by the historical load. The Seq2Seq network consists of two parts, an Encoder (Encoder) and a Decoder (Decoder). The idea is to use an encoder to read the input sequence in time steps, obtain a large fixed-dimension vector representation (context vector), and then use another decoder to extract the output sequence from the vector. The flow chart is shown in fig. 4, and the specific steps are as follows:
a) generating a hidden state of an encoder using LSTM for an input sequence XNAnd (6) coding is carried out. After passing the input sequence through the encoder LSTM, each input produces a hidden state/output hN. We will advance all hidden states generated by the encoder to the next step.
b) The last layer of hidden state C is connected with the input of the decoder and is input into the decoder together with the input of the decoder to obtain a first output YTAnd decoder hidden state ST。
c) The decoder gets the output and the decoder hidden state, and repeats the steps until the sequence is finished.
d) And (4) optimizing parameters, namely selecting a proper optimizer to optimize the model parameters.
The encoders and decoders in the Seq2Seq model employ LSTM neural networks. The LSTM network is an improved Recurrent Neural Network (RNN), effectively solves the problem of gradient disappearance in model training, can learn long-term and short-term dependence information of a time sequence, is the most successful RNN architecture at present, and is applied to a plurality of scenes. The unit of which is shown in fig. 5.
The basic unit of the LSTM network comprises a forgetting gate, an input gate and an output gate. Forget to input X in doortAnd a state memory cell St-1Intermediate output ht-1Jointly determine the forgetting part of the state memory unit. X in input GatetAnd determining the retention vector in the state memory unit together after the sigmoid and the tanh function are changed respectively. Intermediate output htFrom updated stAnd an output otJointly, the calculation formula is as follows:
ft＝σ(WfXxt+Wfhht-1+bf)
it＝σ(WiXxt+Wihht-1+bi)
ot＝σ(WoXxt+Wohht-1+bo)
St＝gt⊙it+St-1⊙ft
based on the technology, the load prediction Seq2Seq network with the following structure is designed. As shown in fig. 6, at the time step setting, the decoder has a time step of 192, i.e., the short term load of 48 hours in the future is predicted. In the encoder setting, through comparative experiments, it is found that the time step of the encoder is set to 288 most appropriately, that is, the load related information of the historical 3 days is used to predict the short-term load of the next two days, so that enough historical reference information can be ensured.
The load prediction Seq2Seq network structure and model construction method of the present invention can refer to the prior art, such as Marino D L, Amarasinger K, Manual M, [ IEEE IECON 2016-42 and annular Conference of the IEEE Industrial Electronics Society-flood, Italy (2016.10.23-2016.10.26) ] IECON 2016-42 and annular Conference of the IEEE Industrial Electronics Society-loading for evaluating usage near Networks [ J ]. 2016: 7046 and 7051, but the application objects are different, the model construction method is also changed:
1) the encoder inputs the features. Encoder input feature xtAs shown in the following formula.
xt＝[yt-1，wt，mt]
wt＝[humi，temp，dmax_humi，dmin_humi，davg_humi，dmax_temp，dmin_temp，davg_temp]
nt＝[month，dayofweek，weekend，hour，min]
Wherein y istRepresenting historical real load characteristics at the time t; w is atThe meteorological features at time t include temperature (temp), humidity (humi), and daily maximum value (d)max_) Daily minimum (d)min_) And daily average value (d)avg_)；mtThe time series characteristics representing the time t mainly include month code (month), day of week (dayofweek), and weekend (wee)kend), hour code (hour), minute code (min).
2) The decoder inputs the features. The decoder input characteristics are shown below.
Assuming that the current time is t, the basic short-term load prediction task needs the load size at the time t + M (M is 1,2, …,192) in the future. When M is 1, the input characteristic of the decoder is xt+1I.e. including the real load y at time ttWeather characteristics of t +1 time mode forecast
time series characteristic m at time t + 1t+1. When M > 1, the input characteristic of the decoder is xt+MThe difference from M-1 is that there is no real load at the previous moment, but the predicted load y at the previous moment of the decoder is usedt。
The short-term load prediction model adopts a Teacher-forcing strategy for learning. In the decoder process, the output of the encoder is randomly selected as the input of the decoder or the true value of the training data is selected as the input of the decoder through probability setting.
Example two
The embodiment provides a computer device comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executing the computer-executable instructions stored by the memory causes the at least one processor to perform the artificial intelligence based short term power load prediction method of embodiment one.
EXAMPLE III
The embodiment provides a computer-readable storage medium, wherein computer-executable instructions are stored in the computer-readable storage medium, and when a processor executes the computer-executable instructions, the method for predicting the artificial intelligence-based short-term power load according to the first embodiment is implemented.
While the invention has been described with reference to specific embodiments thereof, it will be understood by those skilled in the art that the invention is not limited thereto, and may be embodied in many different forms without departing from the spirit and scope of the invention as set forth in the following claims. Any modification which does not depart from the functional and structural principles of the present invention is intended to be included within the scope of the claims.
Claims (7)
1. The short-term power load prediction method based on artificial intelligence is characterized by comprising the following steps:
firstly, selecting load characteristics, lag characteristics, time characteristics and meteorological characteristics which have influence on loads, and constructing dynamic coding characteristics aiming at holidays with different durations, wherein the load characteristics comprise daily sum, daily minimum and daily average load, the lag characteristics are loads or meteorology at a certain moment in the past, the time characteristics are months, days of the week, weekends, hours and minutes, and the meteorological characteristics comprise original values, mean values, maximum values and minimum values of temperature and humidity;
then, modeling is carried out by adopting a Seq2Seq network to establish a short-term load prediction model, wherein the Seq2Seq network consists of an encoder and a decoder, and the encoder inputs a characteristic xtAs shown in the following formula,
xt＝[yt-1，wt，mt]
wt＝[humi，temp，dmax_humi，dmin_humi，davg_humi，dmax_temp，dmin_temp，davg_temp]
mt＝[month，dayofweek，weekend，hour，min]
wherein, ytRepresenting historical real load characteristics at the time t; w is atThe meteorological features at time t include temperature temp, humidity humi and daily maximum value d thereofmax_Minimum daily value dmin_And daily average value davg_；mtTo representtime series characteristics at the time t comprise month code month, day of week dayofweek, weekend or not, hour code hour and minute code min;
the decoder input characteristics are shown as follows,
it is assumed that the current time is t, the basic short-term load prediction task needs the load size at the time of t + M (M is 1,2, …,192) in the future, and when M is 1, the input characteristic of the decoder is xt+1I.e. including the real load y at time ttWeather characteristics of t +1 time mode forecast
time series characteristic m at time t +1t+1(ii) a When M > 1, the input characteristic of the decoder is xt+M；
And finally, predicting the short-term load by adopting a short-term load prediction model.
2. The artificial intelligence based short-term power load forecasting method according to claim 1, wherein dynamic coding is performed according to different holiday lengths, and the dynamic coding mode is as follows:
supposing that the holiday duration of the current holiday is n days, the holiday is coded as-n, if the first day after the holiday is also the holiday, the holiday is coded as- (n-1), and if the second day after the holiday also belongs to the holiday, the holiday is coded as- (n-2), and the holiday is deduced in the secondary category;
the code of the current day of the spring festival is 1, the code of the first day after the spring festival is-1, the code of the second day after the spring festival is-2, and so on until the code reaches the 30 th day after the spring festival; the code is 1 for the day before the spring festival, 2 for the second day before the spring festival, and so on until the code is 29 days before the spring festival, and the code is 0 for the non-holiday.
3. The artificial intelligence based short-term power load forecasting method according to claim 1, characterized in that the concrete steps of modeling by using the Seq2Seq network are as follows:
a) generating a hidden state of an encoder using LSTM for an input sequence XNEncoding is performed such that each input generates a hidden state/output h after passing the input sequence through the encoder LSTMN；
b) The last layer of hidden state C is connected with the input of the decoder and is input into the decoder together with the input of the decoder to obtain a first output YTAnd decoder hidden state ST；
c) The output obtained by the decoder and the hidden state of the decoder are input into the decoder together to obtain the output and the hidden state of the decoder, and the step is repeated until the sequence is finished;
d) and optimizing parameters, namely selecting an optimizer to optimize the model parameters.
4. The artificial intelligence based short term power load prediction method according to claim 1, wherein the decoder has a time step size of 192 at the time step setting, i.e. predicting the short term load 48 hours in the future; the time step of the encoder is set to 288, i.e. the load related information of the historical 3 days is used to predict the short-term load of two days in the future.
5. The artificial intelligence based short-term power load prediction method as claimed in claim 1, wherein the short-term load prediction model is learned by adopting a Teacher-formng strategy, and in the decoder process, the output of the encoder is randomly selected as the input of the decoder or the true value of the training data is selected as the input of the decoder through probability setting.
6. A computer device comprising at least one processor and memory; the memory stores computer-executable instructions; the at least one processor executing the computer-executable instructions stored by the memory cause the at least one processor to perform the artificial intelligence based short term power load prediction method of any one of claims 1 to 5.
7. A computer-readable storage medium having computer-executable instructions stored thereon which, when executed by a processor, implement the artificial intelligence based short-term power load forecasting method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111658345.2A CN114511135A (en) | 2021-12-31 | 2021-12-31 | Artificial intelligence-based short-term load prediction method, computer device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111658345.2A CN114511135A (en) | 2021-12-31 | 2021-12-31 | Artificial intelligence-based short-term load prediction method, computer device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114511135A true CN114511135A (en) | 2022-05-17 |
Family
ID=81548030
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111658345.2A Pending CN114511135A (en) | 2021-12-31 | 2021-12-31 | Artificial intelligence-based short-term load prediction method, computer device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114511135A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115270921A (en) * | 2022-06-22 | 2022-11-01 | 天纳能源科技（上海）有限公司 | Power load prediction method, system and storage medium based on combined prediction model |
| CN116826745A (en) * | 2023-08-30 | 2023-09-29 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
-
2021
- 2021-12-31 CN CN202111658345.2A patent/CN114511135A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115270921A (en) * | 2022-06-22 | 2022-11-01 | 天纳能源科技（上海）有限公司 | Power load prediction method, system and storage medium based on combined prediction model |
| CN116826745A (en) * | 2023-08-30 | 2023-09-29 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
| CN116826745B (en) * | 2023-08-30 | 2024-02-09 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tang et al. | Short‐term power load forecasting based on multi‐layer bidirectional recurrent neural network | |
| Rafi et al. | A short-term load forecasting method using integrated CNN and LSTM network | |
| Xiao et al. | A combined model based on multiple seasonal patterns and modified firefly algorithm for electrical load forecasting | |
| Zhou et al. | An optimized heterogeneous structure LSTM network for electricity price forecasting | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Fan et al. | Development of prediction models for next-day building energy consumption and peak power demand using data mining techniques | |
| Liu et al. | A comparative study of the data-driven day-ahead hourly provincial load forecasting methods: From classical data mining to deep learning | |
| CN109919370B (en) | Power load prediction method and prediction device | |
| Kucukali et al. | Turkey’s short-term gross annual electricity demand forecast by fuzzy logic approach | |
| Jawad et al. | Genetic algorithm‐based non‐linear auto‐regressive with exogenous inputs neural network short‐term and medium‐term uncertainty modelling and prediction for electrical load and wind speed | |
| CN111832809B (en) | Building energy consumption load prediction method and system based on Holt-Winters and extreme learning machine | |
| CN114511135A (en) | Artificial intelligence-based short-term load prediction method, computer device and storage medium | |
| Shi et al. | Short‐term load forecasting based on CNN‐BiLSTM with Bayesian optimization and attention mechanism | |
| Song et al. | Estimate the daily consumption of natural gas in district heating system based on a hybrid seasonal decomposition and temporal convolutional network model | |
| CN114676941B (en) | Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system | |
| Costa | Convolutional-LSTM networks and generalization in forecasting of household photovoltaic generation | |
| Deng et al. | Electricity price prediction based on lstm and lightgbm | |
| Lu et al. | An improved attention-based deep learning approach for robust cooling load prediction: public building cases under diverse occupancy schedules | |
| Bai et al. | Improved Markov‐chain‐based ultra‐short‐term PV forecasting method for enhancing power system resilience | |
| Thejus et al. | Deep learning-based power consumption and generation forecasting for demand side management | |
| Liu et al. | A multi-task learning model for building electrical load prediction | |
| Park et al. | Demand power forecasting with data mining method in smart grid | |
| Nawar et al. | Transfer learning in deep learning models for building load forecasting: Case of limited data | |
| Singh et al. | Load forecasting at distribution transformer using IoT based smart meter data from 6000 Irish homes | |
| Shabani et al. | Hourly prediction of building energy consumption: an incremental ANN approach |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20221107 Address after: 312099 No. 58, Shengli East Road, Shaoxing City, Zhejiang Province Applicant after: STATE GRID ZHEJIANG ELECTRIC POWER CO., LTD. SHAOXING POWER SUPPLY Co. Applicant after: STATE GRID ZHEJIANG ELECTRIC POWER Co.,Ltd. Address before: 312099 No. 58, Shengli East Road, Shaoxing City, Zhejiang Province Applicant before: STATE GRID ZHEJIANG ELECTRIC POWER CO., LTD. SHAOXING POWER SUPPLY Co. Applicant before: STATE GRID ZHEJIANG ELECTRIC POWER Co.,Ltd. Applicant before: HANGZHOU CHENQING HEYE TECHNOLOGY CO.,LTD. |
|
| TA01 | Transfer of patent application right |