CN115471362A - Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning - Google Patents

Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning Download PDF

Info

Publication number
CN115471362A
CN115471362A CN202211176881.3A CN202211176881A CN115471362A CN 115471362 A CN115471362 A CN 115471362A CN 202211176881 A CN202211176881 A CN 202211176881A CN 115471362 A CN115471362 A CN 115471362A
Authority
CN
China
Prior art keywords
load
power generation
prediction model
historical data
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211176881.3A
Other languages
Chinese (zh)
Inventor
徐青山
甘海庆
杨永标
陈堃
杜姣
张航通
聂卓杰
任禹丞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
State Grid Jiangsu Electric Power Co Ltd
Nanjing Power Supply Co of State Grid Jiangsu Electric Power Co Ltd
Original Assignee
Southeast University
State Grid Jiangsu Electric Power Co Ltd
Nanjing Power Supply Co of State Grid Jiangsu Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University, State Grid Jiangsu Electric Power Co Ltd, Nanjing Power Supply Co of State Grid Jiangsu Electric Power Co Ltd filed Critical Southeast University
Priority to CN202211176881.3A priority Critical patent/CN115471362A/en
Publication of CN115471362A publication Critical patent/CN115471362A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21—Design, administration or maintenance of databases
    • G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Quality & Reliability (AREA)
  • Databases & Information Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Water Supply & Treatment (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Primary Health Care (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Public Health (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a comprehensive energy source-charge prediction method for depth feature-guided two-stage transfer learning, which relates to the technical field of comprehensive energy multi-load prediction, and is characterized in that temperature, season, holidays, wind speed, cloud layer density, illumination intensity, electricity price and current policy influence factors are obtained and converted into a data format; acquiring historical data of photovoltaic, wind power, electric load, heat load and cold load of a park, and cleaning the historical data; constructing a time series prediction model of a long-short term memory cyclic neural network based on a depth residual error network and an attention introducing mechanism; training a time series prediction model by using electric load, heat load and cold load data according to three factors of temperature, season and holidays to obtain a primary model; and on the basis of the primary model, a final prediction model is obtained by adopting a transfer learning strategy.

Description

Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning
Technical Field
The invention belongs to the technical field of comprehensive energy multi-element load prediction, and particularly relates to a comprehensive energy source-load prediction method for deep feature-guided two-stage transfer learning.
Background
As shown in fig. 1, energy service providers of typical regional Integrated Energy Systems (IES) such as actual industrial parks, business centers, residential buildings and the like generally need to provide various Energy requirements such as electricity, heat, cold and the like for users at present, and are obviously affected by factors such as meteorological conditions, human activities, architectural characteristics and the like, in the aspect of meteorological conditions, due to temperature changes, the south and north show obvious seasonal and regional differences in cold and heat load requirements, in the aspect of human activities, different social behaviors can affect the Energy utilization characteristics of the IES, for example, residents are generally out during working days, and the system load is mostly rigid load; the frequent activities of residents on non-working days lead to flexible and diverse energy utilization equipment, energy utilization requirements show randomness and uncertainty, meanwhile, different system function positioning is also an important reason for influencing energy utilization characteristics, and an industrial area is often dominated by power load and assisted by cold and hot loads and is subjected to production schedule arrangement; the electric heating load of the living area is often closely related to the activities of people, different types of loads show certain coupling characteristics, and the IES energy utilization characteristics determine that when certain energy utilization requirement changes, the adjustment of energy service providers to other types of energy utilization requirements is caused; load prediction is a primary premise for management and optimized scheduling of the IES energy consumption requirements, and if a traditional single load prediction method is still adopted, differences, randomness and coupling among different energy consumption requirements cannot be considered, and the load prediction accuracy cannot be ensured. Meanwhile, in view of the fact that a great deal of energy conversion coupling information is stored in an IES energy service provider database during long-time operation of the IES, the energy conversion characteristics hidden in the data are difficult to extract and summarize through establishing a detailed mathematical model.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a comprehensive energy source-load prediction method for deep feature guide two-stage transfer learning, which comprises the following steps:
acquiring influence factors of a source side and a load side of a target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies;
acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period, and cleaning the historical data;
constructing a time series prediction model of a long-short term memory cycle neural network based on a depth residual error network and an attention introducing mechanism, wherein the time series prediction model is used for predicting future time series numerical values according to the change trend of historical time series data;
training a time series prediction model by using historical data of electric load, heat load and cold load based on three influence factors of temperature, season and holidays to obtain a primary model;
on the basis of the primary model, six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies are considered, historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load are used for training, and a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model are obtained.
Preferably, on the basis of the primary model, taking into account six influencing factors including wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies, training by using historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load to obtain a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model, and the method comprises the following steps:
on the basis of the primary model, a migration learning strategy is adopted, influence factors of illumination intensity and cloud layer density are considered, and migration training is carried out by using historical data of photovoltaic power generation to obtain a final photovoltaic power generation prediction model so as to predict photovoltaic output in a future preset time period;
on the basis of the primary model, considering wind speed influence factors, carrying out migration training by using historical data of wind turbine power generation to obtain a final wind turbine power generation prediction model so as to predict the wind turbine output in a future preset time period;
on the basis of the primary model, relevant policies and electricity price influence factors are considered, and historical data of the electric load are used for carrying out migration training to obtain a final electric load prediction model so as to predict the electric load in a future preset time period;
on the basis of the primary model, considering the influence factors of electricity price and gas price, and using historical data of the heat load to carry out migration training to obtain a final heat load prediction model so as to predict the heat load of a future preset time period;
on the basis of the primary model, considering the influence factors of electricity prices, and using historical data of the cold load to perform migration training to obtain a final cold load prediction model so as to predict the cold load of a future preset time period.
Preferably, the acquiring historical data of photovoltaic power generation, fan power generation, electrical load, thermal load and cold load of the target park in the historical preset time period, and the cleaning process of the historical data comprises the following steps:
according to historical operation information of the target park in a historical preset time period, acquiring historical data of photovoltaic power generation, a fan, an electric load, a heat load and a cold load of the target park in the historical preset time period, and respectively expressing as follows:
Figure BDA0003864881190000031
Figure BDA0003864881190000032
Figure BDA0003864881190000033
Figure BDA0003864881190000034
Figure BDA0003864881190000035
wherein, P PV A time series of historical data representing photovoltaic power generation of the target campus,
Figure BDA0003864881190000036
is t i Power of photovoltaic generation at the moment P W A time series of historical data representing the power generation of the fans of the target campus,
Figure BDA0003864881190000041
is t i Power of the wind turbine at all times, P E A time series of historical data representing the electrical load of the target campus,
Figure BDA0003864881190000042
is t i Time of day electrical load power, P C A time series of historical data representing the cooling load of the target campus,
Figure BDA0003864881190000043
is t i Time of day cold load data, P H A time series of historical data representing the thermal load of the target campus,
Figure BDA0003864881190000044
is t i Time of day thermal load data, where i =1,2, ·, n;
to P PV 、P W 、P E 、P C And P H Cleaning the sequence data, firstly calculating the local mean value P of the sequence with the window length n PV(W、E、C、H) The formula is as follows:
Figure BDA0003864881190000045
where x is the window starting point, i is the position from the window starting point, P PV(W、E、C、H)_tx+1 Is composed of tx+i The photovoltaic power generation power value, the fan power generation power value, the electric load power value, the cold load value and the heat load value at the moment.
Preferably, when there are points smaller than the average value point in the window, the window starting point is replaced by a mean value with the length of 3 and taking the window starting point as the midpoint, and when there is sequence missing in the window, the mean value of the window is used for filling.
Preferably, the time series prediction model is constructed by the following steps:
constructing a depth feature extraction network based on the depth residual error network, wherein the depth feature extraction network is used for feature extraction of input data;
and inputting the features extracted by the depth feature extraction network into a long-short term memory cyclic neural network based on an attention mechanism to construct a time sequence prediction model.
Preferably, the memory unit of the long-short term memory recurrent neural network is provided with an input gate i t And an output gate o t And forget door f t To select the correction parameters of the error function of the memory feedback with the gradient descending, the input data at the time t is x t Hidden layer state output value is h t The memory state is c t When the user passes through the forgetting gate, the memory unit state c at the next moment is obtained by discarding the useless information t+1 Wherein, forget the door f t The calculation formula of (a) is as follows:
f t ＝δ(ω f ·[h t-1 ,x t ]+b f )
wherein h is t-1 Hidden layer state at time t-1, ω f And b f Representing a weight matrix and a bias vector in the forgetting gate, wherein delta is an activation function and a sigmoid function is adopted;
the input sequence characteristic information is used for obtaining data required to be input into the memory unit and creating 1 new candidate state through 1 activation function sigmoid function and tanh function respectively
Figure BDA0003864881190000051
The calculation formula is as follows:
i t ＝δ(ω i ·[h t-1 ,x t ]+b i )
Figure BDA0003864881190000052
wherein, ω is i And b i Weight matrix and offset vector, ω, in the input gate, respectively c And b c Are respectively provided withWeight matrix and bias vector in the cell unit state;
cell state value c at time t t Hidden layer state h at time t t And the output value o t The calculation formulas are respectively as follows:
Figure BDA0003864881190000053
o t ＝δ(ω 0 ·[h t-1 ,x t ]+b 0 )
h t ＝o t ·tanhc t
wherein, c t-1 Is the cell state value at time t-1, δ is the activation function, ω 0 And b 0 Respectively, a weight matrix and an offset vector in the output gate.
Preferably, the construction process of the attention mechanism-based long-short term memory cycle neural network comprises the following steps:
hidden layer state h of long-short term memory hidden layer output by introducing attention mechanism in long-short term memory recurrent neural network t Assignment of attention weight by alpha t,k To represent the attention probability distribution value, the attention weight matrix α t,k And a feature vector v, the calculation formula is as follows:
Figure BDA0003864881190000054
Figure BDA0003864881190000055
e t,k ＝μ s ·tanh(ω s h t +b s )
wherein k is the [1,l ]]L is the input data length, α t,k To hide layer state features, α t,k Is h t For the current input { x 1 ,x 2 ,...,x k Assigned attention weight, e t,k Is alpha t,k Non-normalized weight matrix, ω s 、b s And mu s Respectively, an Attention weight matrix, an offset and a time sequence matrix which are initialized randomly.
On the other hand, the comprehensive energy source-load prediction device for depth feature-guided two-stage transfer learning is also provided, and comprises:
the data acquisition module is used for acquiring the influence factors of the source side and the load side of the target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies;
the data processing module is used for acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period and cleaning the historical data;
the time series prediction model is used for predicting future time series numerical values according to the change trend of historical time series data;
the second construction module is used for training the time series prediction model by using historical data of electric load, heat load and cold load based on three influence factors of temperature, season and holidays to obtain a primary model;
and the third construction module is used for training by using historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load in consideration of six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies on the basis of the primary model to obtain a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model.
In another aspect, there is also provided an apparatus comprising:
one or more processors;
a memory for storing one or more computer programs;
when executed by the one or more processors, cause the one or more processors to implement a comprehensive energy source-to-load prediction method for depth feature guided two-stage transfer learning as described in any of the above.
In another aspect, a storage medium is provided, where a computer program is stored, and the computer program is used to execute the method for comprehensive energy source-load prediction of depth feature guided two-stage transfer learning as described in any one of the above.
The invention has the beneficial effects that:
according to the method, a two-stage training process is adopted, a primary model is obtained by adopting common factors under the consideration of various influence factors, then a multivariate load prediction model is obtained by adopting transfer learning based on characteristic factors, the training efficiency of the model is improved, the generalization capability of the model is improved, and the prediction precision of the model is greatly improved by the full-element learning of the first stage and the accurate learning of the second stage; according to the invention, according to the multivariate load prediction data, an energy service provider can accurately predict the multivariate load demand on the basis of summarizing and analyzing various energy demands of different users, and coordinates links such as internal conversion, storage, distribution, consumption and the like of IES to meet the energy demands of different users.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present invention, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a simplified model diagram of an integrated energy interaction architecture according to the present invention;
FIG. 2 is a block diagram of a two-stage integrated energy source-load prediction model according to an embodiment of the present invention;
FIG. 3 is a sequence processing model structure of an LSTM according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an attention mechanism based on LSTM according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a hotel load prediction result according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 2, the depth feature-directed two-stage transfer learning comprehensive energy source-load prediction method includes the following steps:
acquiring influence factors of a source side and a load side of a target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies; wherein, temperature, season, holidays are taken as common influence factors to have certain influence on the source-load of the park, wind speed has strong correlation on the power generation of a fan, cloud layer density and illumination intensity have strong correlation on photovoltaic power generation, electricity price and the current policy have strong correlation on the electricity load, the electricity price directly influences the total amount of the electricity load, and the current policy, for example, the total amount of regional electric loads can be influenced by a double-carbon strategy and an ordered production plan issued by a national network, the electricity price and the gas price have strong correlation to the heat load, the heat load is partially generated by electricity and gas, the electricity price and the gas price can influence the required yield of the heat load, the electricity price has strong correlation to the cold load, the cold load mainly comes from the electricity production, the electricity price can also influence the generation of the cold load, and the overhigh electricity price can lead some users to close cold load generation devices such as a refrigeration air conditioner and the like;
acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period, and cleaning the historical data;
constructing a time series prediction model of a long-short term memory cycle neural network based on a depth residual error network and an attention introducing mechanism, wherein the time series prediction model is used for predicting future time series numerical values according to the change trend of historical time series data;
training a time sequence prediction model by using electric load, heat load and cold load data according to three factors of temperature, season and holidays to obtain a primary model for predicting a future sequence numerical value according to the change trend of historical data;
on the basis of the primary model, six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies are considered, historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load are used for training, and a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model are obtained.
Because photovoltaic power generation is used for electrical load and cold and hot load of a park, an internal coupling relation exists, on the basis of a primary model, a migration learning strategy is adopted, photovoltaic data is used for carrying out migration training to obtain a final photovoltaic power generation prediction model, the photovoltaic power generation prediction model is used for predicting future photovoltaic output by historical power data of photovoltaic power generation, and data support is provided for energy scheduling of subsequent photovoltaic participating in comprehensive energy;
because wind power generation is used for the electrical load and the cold and hot load of a park, an internal coupling relation exists, on the basis of a primary model, wind power data are used for carrying out migration training to obtain a final wind power generation prediction model which is used for predicting the future fan output by the historical power data generated by the fan and providing data support for the subsequent fan to participate in energy scheduling of comprehensive energy;
on the basis of the primary model, migration training is carried out by using the electric load data to obtain a final electric load prediction model, the final electric load prediction model is used for predicting future load electric data according to the historical electric data of the load, and electric load data support is provided for energy scheduling and optimized operation of subsequent comprehensive energy;
on the basis of the primary model, migration training is carried out by using heat load data to obtain a final heat load prediction model, and the final heat load prediction model is used for predicting future heat load according to historical data of the heat load and providing heat load data support for energy scheduling and optimized operation of subsequent comprehensive energy;
on the basis of the primary model, migration training is carried out by using cold load data to obtain a final cold load prediction model, the final cold load prediction model is used for predicting future cold load according to historical data of the cold load, and cold load data support is provided for energy scheduling and optimized operation of subsequent comprehensive energy.
It should be further noted that, in the implementation, the influencing factors include temperature T, season S, holiday F, wind speed W, cloud density C, illumination intensity L, price P and policy information.
It should be further described that, in the specific implementation process, the acquiring historical data of photovoltaic, wind power, electrical load, thermal load and cold load of the park, and the cleaning process of the historical data includes the following steps:
according to historical operation information of the target park in a historical preset time period, acquiring historical data of photovoltaic power generation, a fan, an electric load, a heat load and a cold load of the target park in the historical preset time period, and respectively expressing as follows:
Figure BDA0003864881190000101
Figure BDA0003864881190000102
Figure BDA0003864881190000103
Figure BDA0003864881190000104
Figure BDA0003864881190000105
wherein, P PV A time series of historical data representing photovoltaic power generation of the target campus,
Figure BDA0003864881190000106
is t i Power of photovoltaic generation at all times, P W A time series of historical data representing the power generation of the fans of the target campus,
Figure BDA0003864881190000107
is t i Power generated by the fan at all times, P E A time series of historical data representing the electrical load of the target campus,
Figure BDA0003864881190000108
is t i Time of day electrical load power, P C A time series of historical data representing the cooling load of the target campus,
Figure BDA0003864881190000109
is t i Time of day cold load data, P H A time series of historical data representing the thermal load of the target campus,
Figure BDA00038648811900001010
is t i Time of day thermal load data, where i =1,2, ·, n;
to P PV 、P W 、P E 、P C And P H Cleaning the sequence data, firstly calculating the local mean value P of the sequence with the window length n pv(W、E、C、H) The formula is as follows:
Figure BDA00038648811900001011
where x is the window starting point, i is the position from the window starting point, P PV(W、E、C、H)_tx+1 Is t x+i The photovoltaic power generation power value, the fan power generation power value, the electric load power value, the cold load value and the heat load value at the moment.
It should be further noted that, in the specific implementation process, when there are points smaller than the average value point in the window, the window starting point is replaced with a mean value with a length of 3 and taking the window starting point as a midpoint, and when there is a sequence missing in the window, the mean value of the window is used for filling.
It should be further noted that, in a specific implementation process, the time series prediction model is constructed by the following steps:
constructing a depth feature extraction network based on the depth residual error network, wherein the depth feature extraction network is used for feature extraction of input data;
inputting the features extracted by the depth residual error network into a long-short term memory (LSTM) cyclic neural network based on an Attention mechanism to construct a time sequence prediction model, wherein the structure of a sequence processing model of the LSTM is shown in figure 3, and the structure of the long-short term memory cyclic neural network based on the Attention mechanism is shown in figure 4.
It should be further noted that, in the implementation process, the LSTM recurrent neural network is an improved version of the recurrent neural network RNN, and the LSTM recurrent neural network redesigns the memory unit on the basis of maintaining the basic structure of the recurrent neural network, and the input gate i is arranged t And an output gate o t And forget door f t To select the correction parameters of the error function of the memory feedback with the gradient descending, the input data at the time t is x t The hidden layer state output value is h t The memory state is c t When the user passes through the forgetting gate, the memory unit state c at the next moment is obtained by discarding the useless information t+1 Wherein, forget the door f t The calculation formula of (c) is as follows:
f t ＝δ(ω f ·[h t-1 ,x t ]+b f )
wherein h is t-1 Hidden layer state at time t-1, ω f And b f Representing weight matrix and offset vector in forgetting gate, delta is activation function, sigmoid function is adopted, and f is output t Is between 0 and 1, followed by bitwise and c t-1 When f is multiplied by the element of (b) t When the value of a certain bit is 0, the information of the corresponding bit is completely discarded, and when the value is (0,1), the information of the corresponding bit is partially reserved and discarded, and only when the value is 1, the information of the corresponding bit is completely reserved;
the input sequence characteristic information is used for obtaining data required to be input into a memory unit and creating 1 new candidate state respectively through 1 sigmoid function and tanh function of the activation function
Figure BDA0003864881190000111
The calculation formula is as follows:
i t ＝δ(ω i ·[h t-1 ,x t ]+b i )
Figure BDA0003864881190000112
wherein, ω is i And b i Weight matrix and offset vector, ω, in the input gate, respectively c And b c Respectively a weight matrix and a bias vector in the cell unit state;
cell state value c at time t t Hidden layer state h at time t t And the output value o t The calculation formulas are respectively as follows:
Figure BDA0003864881190000121
o t ＝δ(ω 0 ·[h t-1 ,x t ]+b 0 )
h t ＝o t ·tanhc t
wherein, c t-1 Is the cell state value at time t-1, δ is the activation function, ω 0 And b 0 Respectively, a weight matrix and an offset vector in the output gate.
It should be further noted that, in the implementation process, the Attention mechanism can allocate more Attention to the key part of the input sequence that affects the output result, so as to better learn the information in the sequence, and the Attention mechanism is introduced into LSTM to imply the layer output value h of LSTM t Assign attention weights with alpha t To express the attention probability distribution value, the attention weight matrix α and the feature vector v, the calculation formula is as follows:
Figure BDA0003864881190000122
Figure BDA0003864881190000123
e t,k ＝μ s ·tanh(ω s h t +b s )
wherein k is the [1,l ]]L is the input data length, α t,k To hide layer state features, α t,k Is h t For the current input { x 1 ,x 2 ,...,x k Assigned attention weight, e t,k As an unnormalized weight matrix, ω s 、b s And mu s Respectively an Attention weight matrix, an offset and a time sequence matrix which are initialized randomly.
It should be further explained that, in the specific implementation process, a migration learning strategy is adopted to obtain a final prediction model of photovoltaic, wind power, electric load, heat load and cold load according to the time series prediction primary model;
further considering the influence of illumination intensity and cloud layer density aiming at the photovoltaic on the source side, and carrying out migration training by using photovoltaic data to obtain a final photovoltaic power generation prediction model;
further considering the influence of wind speed factors aiming at the wind power at the source side, carrying out migration training by using wind power data to obtain a final wind power generation prediction model;
further considering policy guidance and the influence of electricity price aiming at the electric load at the charge side, and carrying out migration training by using electric load data to obtain a final electric load prediction model;
further considering the influence of electricity price and gas price aiming at the heat load at the charge side, and carrying out migration training by using heat load data to obtain a final heat load prediction model;
and further considering the influence of the electricity price aiming at the cold load on the charge side, and performing migration training by using the cold load data to obtain a final cold load prediction model.
Example (b): in order to verify the effectiveness of the proposed method, taking a power load prediction model as an example, 35040 pieces of data are obtained at a sampling frequency of 15min from 1/0 to 12/31/24 of 2021 in a hotel, in a neural network, a training set adopts data from 1/0 to 12/30/24 of 2021 as training data, and a testing set adopts 96 points of data of 24 hours in total from 15 points from 12/0 to 31/24 of 2021/31. The LSTM layer takes the reconstructed data as one input at 64 points and the next point as an output, i.e., training in a "many-to-one" mode, and the prediction results are shown in fig. 5:
wherein the virtual straight line is a load prediction curve, and the real straight line is a load real curve. The error evaluation index is defined as:
Figure BDA0003864881190000131
wherein n is the sample size of the test set; a. The i Actual load sequences are test set; f i Load sequences are predicted for the test set. The MAPE can well measure the average size of the prediction error, and as can be seen from FIG. 5, the prediction data is basically consistent with the real load, and the error index MAPE is 0.032. In conclusion, the comprehensive energy of the proposed depth feature for guiding two-stage transfer learning can be seenEffectiveness of the source-load prediction method.
It should be further noted that, based on the same inventive concept, the present invention further provides a comprehensive energy source-load prediction apparatus for deep feature guidance two-stage transfer learning, wherein the method comprises the following steps:
the data acquisition module is used for acquiring the influence factors of the source side and the load side of the target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies;
the data processing module is used for acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period and cleaning the historical data;
the time series prediction model is used for predicting future time series numerical values according to the change trend of historical time series data;
the second construction module is used for training the time series prediction model by using historical data of electric load, heat load and cold load based on three influence factors of temperature, season and holidays to obtain a primary model;
and the third construction module is used for training by using historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load in consideration of six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies on the basis of the primary model to obtain a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model.
Based on the same inventive concept, the present invention also provides a computer apparatus, comprising: one or more processors, and memory for storing one or more computer programs; the program includes program instructions and the processor is configured to execute the program instructions stored by the memory. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal and is configured to implement one or more instructions, and in particular to load and execute one or more instructions in a computer storage medium to implement the method.
It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the above method. The storage medium may take any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of the present invention, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed.

Claims (10)

1. The comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning is characterized by comprising the following steps of:
acquiring influence factors of a source side and a load side of a target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies;
acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period, and cleaning the historical data;
constructing a time series prediction model of a long-short term memory cyclic neural network based on a depth residual error network and an attention introducing mechanism, wherein the time series prediction model is used for predicting future time series numerical values according to the variation trend of historical time series data;
training a time series prediction model by using historical data of electric load, heat load and cold load based on three influence factors of temperature, season and holidays to obtain a primary model;
on the basis of the primary model, six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies are considered, historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load are used for training, and a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model are obtained.
2. The comprehensive energy source-load prediction method for depth feature-guided two-stage migration learning according to claim 1, wherein on the basis of the primary model, taking into account six influencing factors including wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies, historical data of photovoltaic power generation, fan power generation, electrical load, thermal load and cold load are used for training to obtain a final photovoltaic power generation prediction model, wind power generation prediction model, electrical load prediction model, thermal load prediction model and cold load prediction model, and the method comprises the following steps:
on the basis of the primary model, a migration learning strategy is adopted, influence factors of illumination intensity and cloud layer density are considered, and migration training is carried out by using historical data of photovoltaic power generation to obtain a final photovoltaic power generation prediction model so as to predict photovoltaic output in a future preset time period;
on the basis of the primary model, considering wind speed influence factors, carrying out migration training by using historical data of wind turbine power generation to obtain a final wind turbine power generation prediction model so as to predict the wind turbine output in a future preset time period;
on the basis of the primary model, relevant policies and electricity price influence factors are considered, and historical data of the electric load are used for carrying out migration training to obtain a final electric load prediction model so as to predict the electric load in a future preset time period;
on the basis of the primary model, considering the influence factors of electricity price and gas price, and using historical data of the heat load to carry out migration training to obtain a final heat load prediction model so as to predict the heat load of a future preset time period;
on the basis of the primary model, considering the influence factors of electricity prices, and using historical data of the cold load to perform migration training to obtain a final cold load prediction model so as to predict the cold load of a future preset time period.
3. The comprehensive energy source-load prediction method for the depth feature-guided two-stage transfer learning according to claim 1, wherein the process of acquiring historical data of photovoltaic power generation, fan power generation, electrical load, thermal load and cold load of a target park in a historical preset time period and cleaning the historical data comprises the following steps:
according to historical operation information of the target park in a historical preset time period, acquiring historical data of photovoltaic power generation, a fan, an electric load, a heat load and a cold load of the target park in the historical preset time period, and respectively expressing as follows:
Figure FDA0003864881180000021
Figure FDA0003864881180000031
Figure FDA0003864881180000032
Figure FDA0003864881180000033
Figure FDA0003864881180000034
wherein, P PV A time series of historical data representing photovoltaic power generation of the target campus,
Figure FDA0003864881180000035
is t i Power of photovoltaic generation at all times, P W Wind representing a target parkA time series of historical data of the power generation of the machine,
Figure FDA0003864881180000036
is t i Power of the wind turbine at all times, P E A time series of historical data representing the electrical load of the target campus,
Figure FDA0003864881180000037
is t i Time of day electrical load power, P C A time series of historical data representing the cooling load of the target campus,
Figure FDA0003864881180000038
is t i Time of day cold load data, P H A time series of historical data representing the thermal load of the target campus,
Figure FDA0003864881180000039
is t i Time of day thermal load data, where i =1,2, ·, n;
to P PV 、P W 、P E 、P C And P H Cleaning the sequence data, firstly calculating the local mean value P of the sequence with the window length n PV(W、E、C、H) The formula is as follows:
Figure FDA00038648811800000310
where x is the window starting point, i is the position from the window starting point, P PV(W、E、C、H)_tx+1 Is composed of tx+i The photovoltaic power generation power value, the fan power generation power value, the electric load power value, the cold load value and the heat load value at the moment.
4. The comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning according to claim 3, wherein when there are points smaller than the mean point in the window, the window starting point is replaced with a mean value with a length of 3 using the window starting point as a midpoint, and when there is a sequence missing in the window, the mean value of the window is used for filling.
5. The method for comprehensive energy source-load prediction of two-stage migration learning guided by depth features according to claim 1, wherein the construction process of the time series prediction model comprises the following steps:
constructing a depth feature extraction network based on the depth residual error network, wherein the depth feature extraction network is used for feature extraction of input data;
and inputting the features extracted by the depth feature extraction network into a long-short term memory cyclic neural network based on an attention mechanism to construct a time sequence prediction model.
6. The method of claim 5, wherein the memory unit of the long-short term memory recurrent neural network is provided with an input gate i t And an output gate o t And forget door f t To select the correction parameters of the error function of the memory feedback which decrease along with the gradient, and the input data at the time t is x t Hidden layer state output value is h t The memory state is c t When passing through the forgetting gate, the memory unit state c at the next moment is obtained by discarding the useless information t+1 Wherein, forget the door f t The calculation formula of (a) is as follows:
f t ＝δ(ω f ·[h t-1 ,x t ]+b f )
wherein h is t-1 Hidden layer state at time t-1, ω f And b f Representing a weight matrix and a bias vector in the forgetting gate, wherein delta is an activation function and a sigmoid function is adopted;
the input sequence characteristic information is used for obtaining data required to be input into the memory unit and creating 1 new candidate state through 1 activation function sigmoid function and tanh function respectively
Figure FDA0003864881180000041
Calculating maleThe formula is as follows:
i t ＝δ(ω i ·[h t-1 ,x t ]+b i )
Figure FDA0003864881180000042
wherein, ω is i And b i Weight matrix and offset vector, ω, in the input gate, respectively c And b c Respectively a weight matrix and a bias vector in the cell unit state;
cell state value c at time t t Hidden layer state h at time t t And the output value o t The calculation formulas are respectively as follows:
Figure FDA0003864881180000051
o t ＝δ(ω 0 ·[h t-1 ,x t ]+b 0 )
h t ＝o t ·tanhc t
wherein, c t-1 Is the cell state value at time t-1, δ is the activation function, ω 0 And b 0 Respectively, a weight matrix and an offset vector in the output gate.
7. The comprehensive energy source-load prediction method for deep feature guided two-stage transfer learning according to claim 6, wherein the construction process of the long-short term memory recurrent neural network based on the attention mechanism comprises the following steps:
hidden layer state h of long-short term memory hidden layer output by introducing attention mechanism in long-short term memory recurrent neural network t Assignment of attention weight by alpha t,k To represent the attention probability distribution value, the attention weight matrix alpha t,k And a feature vector v, the calculation formula is as follows:
Figure FDA0003864881180000052
Figure FDA0003864881180000053
e t,k ＝μ s ·tanh(ω s h t +b s )
wherein k is the [1,l ]]L is the input data length, α t,k To hide layer state features, α t,k Is h t For the current input { x 1 ,x 2 ,...,x k Assigned attention weight, e t,k Is alpha t,k Unnormalized weight matrix, ω s 、b s And mu s Respectively, an Attention weight matrix, an offset and a time sequence matrix which are initialized randomly.
8. The comprehensive energy source-load prediction device for depth feature guidance two-stage transfer learning is characterized by comprising the following steps:
the data acquisition module is used for acquiring the influence factors of the source side and the load side of the target park and converting the influence factors into a data format; wherein the influencing factors include: temperature, season, holidays, wind speed, cloud density, light intensity, electricity prices, gas prices, and related policies;
the data processing module is used for acquiring historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load of a target park in a historical preset time period and cleaning the historical data;
the time series prediction model is used for predicting future time series numerical values according to the change trend of historical time series data;
the second construction module is used for training the time series prediction model by using historical data of electric load, heat load and cold load based on three influence factors of temperature, season and holidays to obtain a primary model;
and the third construction module is used for training by using historical data of photovoltaic power generation, fan power generation, electric load, heat load and cold load in consideration of six influence factors of wind speed, cloud layer density, illumination intensity, electricity price, gas price and related policies on the basis of the primary model to obtain a final photovoltaic power generation prediction model, a wind power generation prediction model, an electric load prediction model, a heat load prediction model and a cold load prediction model.
9. An apparatus, comprising:
one or more processors;
a memory for storing one or more computer programs;
when executed by the one or more processors, cause the one or more processors to implement the method of integrated energy source-load prediction for depth feature guided two-stage migratory learning of any one of claims 1-7.
10. A storage medium having stored thereon a computer program for executing the method of integrated energy source-load prediction of depth feature guided two-stage migratory learning according to any one of claims 1-7 when executed by a processor.
CN202211176881.3A 2022-09-26 2022-09-26 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning Pending CN115471362A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211176881.3A CN115471362A (en) 2022-09-26 2022-09-26 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211176881.3A CN115471362A (en) 2022-09-26 2022-09-26 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning

Publications (1)

Publication Number Publication Date
CN115471362A true CN115471362A (en) 2022-12-13

Family

ID=84334523

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211176881.3A Pending CN115471362A (en) 2022-09-26 2022-09-26 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning

Country Status (1)

Country Link
CN (1) CN115471362A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115882454A (en) * 2023-02-09 2023-03-31 国网浙江省电力有限公司宁波供电公司 Source load power prediction method and system based on multivariate information fusion
CN116432861A (en) * 2023-05-06 2023-07-14 山东大学 Comprehensive energy system source charge price multitasking combined prediction method and system
CN116565966A (en) * 2023-06-29 2023-08-08 国能青海黄河玛尔挡水电开发有限公司 Intelligent hydraulic power generation double microcomputer automatic synchronous control system
CN117272121A (en) * 2023-11-21 2023-12-22 江苏米特物联网科技有限公司 Hotel load influence factor quantitative analysis method based on Deep SHAP
CN117313958A (en) * 2023-11-28 2023-12-29 山东大学 Low-voltage distributed photovoltaic power prediction method and system based on transfer learning
CN117852721A (en) * 2024-01-15 2024-04-09 国网山东省电力公司潍坊供电公司 EPL-based park medium-long term terminal energy service demand prediction method
CN118014166A (en) * 2024-04-09 2024-05-10 陕西德联新能源有限公司 Load prediction method and system of heating system based on big data

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115882454A (en) * 2023-02-09 2023-03-31 国网浙江省电力有限公司宁波供电公司 Source load power prediction method and system based on multivariate information fusion
CN115882454B (en) * 2023-02-09 2023-08-15 国网浙江省电力有限公司宁波供电公司 Source load power prediction method and system based on multi-element information fusion
CN116432861A (en) * 2023-05-06 2023-07-14 山东大学 Comprehensive energy system source charge price multitasking combined prediction method and system
CN116432861B (en) * 2023-05-06 2024-08-13 山东大学 Comprehensive energy system source charge price multitasking combined prediction method and system
CN116565966A (en) * 2023-06-29 2023-08-08 国能青海黄河玛尔挡水电开发有限公司 Intelligent hydraulic power generation double microcomputer automatic synchronous control system
CN116565966B (en) * 2023-06-29 2023-10-24 国能青海黄河玛尔挡水电开发有限公司 Intelligent hydraulic power generation double microcomputer automatic synchronous control system
CN117272121A (en) * 2023-11-21 2023-12-22 江苏米特物联网科技有限公司 Hotel load influence factor quantitative analysis method based on Deep SHAP
CN117272121B (en) * 2023-11-21 2024-03-12 江苏米特物联网科技有限公司 Hotel load influence factor quantitative analysis method based on Deep SHAP
CN117313958A (en) * 2023-11-28 2023-12-29 山东大学 Low-voltage distributed photovoltaic power prediction method and system based on transfer learning
CN117313958B (en) * 2023-11-28 2024-02-20 山东大学 Low-voltage distributed photovoltaic power prediction method and system based on transfer learning
CN117852721A (en) * 2024-01-15 2024-04-09 国网山东省电力公司潍坊供电公司 EPL-based park medium-long term terminal energy service demand prediction method
CN118014166A (en) * 2024-04-09 2024-05-10 陕西德联新能源有限公司 Load prediction method and system of heating system based on big data

Similar Documents

Publication Publication Date Title
CN115471362A (en) Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning
Niu et al. Short-term multi-energy load forecasting for integrated energy systems based on CNN-BiGRU optimized by attention mechanism
Li et al. Applying support vector machine to predict hourly cooling load in the building
Hu et al. A hybrid model based on CNN and Bi-LSTM for urban water demand prediction
Li et al. Federated learning-based short-term building energy consumption prediction method for solving the data silos problem
Wu et al. An explainable framework for load forecasting of a regional integrated energy system based on coupled features and multi-task learning
Almalaq et al. Parallel building: a complex system approach for smart building energy management
Delcroix et al. Autoregressive neural networks with exogenous variables for indoor temperature prediction in buildings
Xuemei et al. Building cooling load forecasting model based on LS-SVM
Deb et al. A machine learning-based framework for cost-optimal building retrofit
CN112149890A (en) Comprehensive energy load prediction method and system based on user energy label
Liu et al. A short-term probabilistic photovoltaic power prediction method based on feature selection and improved LSTM neural network
CN110796307A (en) Distributed load prediction method and system for comprehensive energy system
CN114819372A (en) Power consumption probability prediction method based on neural network
CN117114438A (en) Building area energy system cold and hot load data driving prediction method with flexibility and interpretability
CN114239991A (en) Building heat supply load prediction method, device and equipment based on data driving
CN107480810B (en) Monthly power consumption prediction method for improving secondary index smoothness by utilizing seasonal index
Alanbar et al. Energy Consumption Prediction Using Deep Learning Technique
CN114239292B (en) Comprehensive evaluation method and system for low-carbon economic operation-oriented multifunctional demand park
CN115759458A (en) Load prediction method based on comprehensive energy data processing and multi-task deep learning
Ahmad et al. Efficient energy planning with decomposition-based evolutionary neural networks
CN109858668B (en) Coordination prediction method for power load region in thunder and lightning climate
Kumar et al. Forecasting indoor temperature for smart buildings with ARIMA, SARIMAX, and LSTM: A fusion approach
Nawar et al. Transfer learning in deep learning models for building load forecasting: Case of limited data
Zou et al. Predicting the electric power consumption of office buildings based on dynamic and static hybrid data analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination