CN112862213B - Heat supply demand prediction method, system and equipment based on periodic feedback LSTM - Google Patents
Heat supply demand prediction method, system and equipment based on periodic feedback LSTM Download PDFInfo
- Publication number
- CN112862213B CN112862213B CN202110256564.1A CN202110256564A CN112862213B CN 112862213 B CN112862213 B CN 112862213B CN 202110256564 A CN202110256564 A CN 202110256564A CN 112862213 B CN112862213 B CN 112862213B
- Authority
- CN
- China
- Prior art keywords
- state
- lstm
- cell state
- heat supply
- current
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000000737 periodic effect Effects 0.000 title claims abstract description 26
- 239000013598 vector Substances 0.000 claims abstract description 52
- 238000010438 heat treatment Methods 0.000 claims abstract description 20
- 238000001914 filtration Methods 0.000 claims abstract description 5
- 238000012216 screening Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 36
- 238000004590 computer program Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 8
- 230000015654 memory Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000005265 energy consumption Methods 0.000 claims description 5
- 238000009423 ventilation Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000000452 restraining effect Effects 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 2
- 230000005540 biological transmission Effects 0.000 abstract description 4
- 239000010410 layer Substances 0.000 description 21
- 238000013528 artificial neural network Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 11
- 230000000306 recurrent effect Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 238000005338 heat storage Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Operations Research (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a heat supply demand prediction method, a heat supply demand prediction system and a heat supply demand prediction device based on periodic feedback LSTM, wherein the heat supply demand prediction method comprises the following steps: according to the input of the current moment, forgetting the cell state by combining with the hidden state, filtering the historical misuse information, and screening the cell state; the hidden state and the input feature vector are processed by a full-connection layer, the value domain is constrained by a tanh function, the current newly-added state is subjected to gate control constraint, and the newly-added state is fused with the cell state screened in the first step to obtain an updated cell state; the value domain of the updated cell state is transformed by using a tanh function, and information constraint is carried out by an output door to obtain the current state, namely, the current heating quantity predicted value required by the system; the LSTM controls the transmission state through the gating state, can memorize information which needs to be memorized for a long time and is not important, and in the periodic feedback LSTM, the heat requirement at the same time of the previous day is used as an important parameter, so that the prediction accuracy is greatly improved.
Description
Technical Field
The invention belongs to the technical field of heat supply optimization, and particularly relates to a heat supply demand prediction method, a heat supply demand prediction system and heat supply demand prediction equipment based on periodic feedback LSTM.
Background
In order to save energy and reduce emission, enough heat supply quantity in winter can be ensured, the heat supply demand quantity of some buildings and parks needs to be estimated in advance, and then heat supply is carried out according to the estimated quantity. Currently, the heat supply demand is estimated based on building area or by referring to the average of the heat supply over the last years. Although the methods are simple, along with the continuous change of global air temperature and the change of building structures, the prediction is often inaccurate, the heat supply quantity is not matched with the actual demand quantity, and the energy loss is often caused by insufficient heat supply or excessive heat supply.
And the LSTM recurrent neural network algorithm is utilized to predict the heat supply requirement, so that big data can be applied to the traditional heat supply industry, and intelligent heat supply is realized. Recurrent neural networks are networks with loops that allow information to persist, classifying events that occur at each point in time, and reasoning about previous events to arrive at a later event. Compared with the prior rough estimation method, the method has the advantages that the estimation by using the recurrent neural network algorithm is very accurate, the most efficient utilization of energy can be truly realized, and the energy conservation and emission reduction are realized.
In the heat supply demand estimation, according to the characteristics of the building and the indoor and outdoor temperatures, we can acquire a lot of useful information: the method comprises the steps of building area, building ventilation quantity, average people flow in a building, current outdoor air temperature, current indoor air temperature, electric appliance energy consumption in the building and the like. The characteristic information is the characteristic which is strongly related to the required heating amount, and is very suitable for predicting the heating amount; on the other hand, these feature amounts are relatively easy to acquire, the acquisition cost is low, and real-time acquisition and updating can be realized. After the characteristic information is obtained, the characteristic information is used as an input sequence in a recurrent neural network, the input sequence is encoded into a characteristic vector, and finally the characteristic vector is decoded as an output sequence, so that a predicted value of the heat supply requirement can be obtained.
In the recurrent neural network algorithm, the conventional LSTM model learns an input sequence by using LSTM units, then represents the input sequence by using vectors with fixed lengths, and then reads the vectors by using LSTM units and decodes the vectors as an output sequence. The model with the structure has good results on many prediction problems, is a current mainstream prediction method, and has good prediction effects in many other fields.
However, there is a problem with the conventional LSTM model for the prediction of heating demand: these characteristic information (building area, indoor and outdoor temperature, traffic, etc.) are used as input sequences, which are encoded into a fixed length vector representation, and are limited to this fixed length vector representation when decoding. This limits the performance of the heating prediction model, especially if the input sequence is long with increasing number of prediction days, the information is too much to keep all necessary information, in which case the conventional LSTM model may become poor. Reference is made to paper Sequence to Sequence Learning with Neural Networks and Long Short-Term Memory.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a heat supply demand prediction method, a heat supply demand prediction system and heat supply demand prediction equipment based on periodic feedback LSTM, which break the limitation that the traditional coding and decoding structure depends on a fixed length vector inside, and keep the intermediate output result of an LSTM coder on an input sequence.
In order to achieve the above purpose, the technical scheme of the invention is as follows: a heat supply demand estimation method based on periodic feedback LSTM comprises the following steps:
forgetting the cell state according to the feature vector at the current moment and combining the hidden state, filtering the historical misuse information, and screening the cell state;
processing the hidden state and the characteristic vector at the current moment through a full-connection layer, restraining the value range of the hidden state and the characteristic vector by a tanh function, then carrying out gate control restraint on the current newly-added state, and fusing the newly-added state with the screened cell state to obtain an updated cell state;
and transforming the value domain of the updated cell state by using a tanh function, and carrying out information constraint by an output door to obtain the current hidden state, namely the current required heating quantity predicted value of the system.
The input feature vector comprises building area, building ventilation quantity, average people flow in the building, current outdoor air temperature, current indoor air temperature and electric appliance energy consumption in the building, the building area in the input feature vector is fixed, and other parameters in the input feature vector are measured and updated once every preset time.
According to the input of the current moment, forgetting the cell state by combining the hidden state, and realizing the cell state through a forgetting door, wherein the method comprises the following steps of:
f t ＝z(W f ·[h t-1 ,h t-24 ,x t ]+b f )；
wherein h is t-1 Is the hidden state of LSTM at the last moment, x t Is the characteristic vector of the current moment, W f And b f The weight and the bias of the full connection layer of the forgetting gate are respectively h t-24 The feedback state is obtained by collecting heat information required by the same time of the previous day and encoding and transmitting the heat information into a system, and z is a sigmoid function.
Processing the hidden state and the input feature vector through a full-connection layer; restricting the value domain by using the tanh function, then performing gate control restriction on the current newly-added state, and fusing with the screened cell state to obtain an updated cell state C t The specific calculation formula is as follows:
i t ＝z(W i ·[h t-1 ,h t-24 ,x t ]+b i )
wherein W is i And b i Weights and biases of all connection layers of forgetting gate, x t Is the characteristic vector of the current moment, W c And b c Is the weight and bias of the cell state update layer,is C t-1 Is the result.
Hidden state O according to cell state and feedback state t The updating is carried out as follows:
O t ＝z(W o [h t-1 ,h t-24 ,x t ]+b o )；
h t ＝O t *tanh(C t )；
W o and b o The weight and the bias cell state of the gate full-connection layer are respectively output, and the feedback state and the hidden state are used as the expression of the cell state at the moment and are used as the input of the LSTM of the next layer or the time sequence estimation network; h obtained by calculation t The heating quantity predicted value at the current moment of the system is obtained.
And acquiring heat supply demand data of a set time in the past, training the LSTM by taking the minimum mean square error as a loss function, wherein the loss function is specifically as follows:
wherein y is t Is the real heat supply demand of each hour, W l For outputting weight vectors, the elements in hidden states are weighted and summed, the weight of the LSTM is derived by using a back propagation mode, and the LSTM is updated by a random gradient descent method.
The heat supply demand pre-estimating system based on the LSTM comprises a forgetting module, a cell state updating module and a hidden state updating module; forgetting module: forgetting the cell state according to the characteristic vector at the current moment and combining the hidden state, filtering the historical misuse information, and screening the cell state;
the cell state updating module is used for processing the hidden state and the characteristic vector at the current moment through a full-connection layer, restricting the value range of the feature vector by a tanh function, then carrying out gate control restriction on the current newly-added state, and fusing the newly-added state with the screened cell state to obtain an updated cell state;
and the hidden state updating module transforms the value domain of the updated cell state by using a tanh function, and then the output gate performs information constraint to obtain the current hidden state, namely, the current required heating quantity predicted value of the system.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the periodic feedback LSTM based heat supply demand estimation method of the present invention when the computer program is executed.
A computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the periodic feedback LSTM based heat supply demand estimation method of the present invention.
The invention has the beneficial effects that:
the LSTM controls the transmission state through the gating state, can memorize information which needs to be memorized for a long time and forgets unimportant continuously, is very suitable for being used in heat demand prediction, has good prediction effect, takes the heat demand at the same moment of the previous day as a very important parameter in the periodic feedback LSTM, greatly improves prediction accuracy, and the heat demand prediction method based on the periodic feedback LSTM breaks the limitation that the traditional coding and decoding structure depends on a fixed length vector inside, selectively learns the inputs by retaining the intermediate output result of an LSTM coder and then trains a model and associates the output sequence with the input result when the model is output.
Drawings
FIG. 1 is a schematic diagram of a smart heating architecture.
FIG. 2 is a flow chart of a heat demand prediction process.
Fig. 3 is a schematic diagram of a sigmoid function.
FIG. 4 is a schematic diagram of a tanh function.
Fig. 5 is a block diagram of the repetition module of the LSTM.
FIG. 6 is a block diagram of a repetition module for LSTM based heating prediction.
Fig. 7a is a schematic diagram of the time-by-time heat demand of a base for different time periods of 1 month and 5 days.
Fig. 7b is a schematic diagram of the time-by-time heat demand of a base at different time periods of 2 months and 5 days before.
Fig. 7c is a schematic diagram of the time-by-time heat demand of a base at different time periods of 3 months and 5 days before.
FIG. 7d is a schematic diagram of the time-by-time heat demand of a base for different time periods of 11 months 15-11 months 25 days.
FIG. 7e is a schematic diagram of the time-by-time caloric requirement of a base for different periods of 12 months 11-12 months 15 days.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings.
The heat supply demand prediction method based on the periodic feedback LSTM breaks through the limitation that the traditional coding and decoding structure depends on a fixed length vector inside. The intermediate output result of the LSTM encoder on the input sequence is reserved, then a model is trained to selectively learn the inputs and correlate the output sequence with the input sequence when the model is output, the intermediate output result is the heat supply demand at the same time of the previous day, and the data is strongly related to the heat supply demand at the moment, so that the performance of the LSTM model is greatly improved.
An intelligent heat supply technology architecture is shown in figure 1, and a complete intelligent heat supply system comprises a control method, an information acquisition system, an information transmission system and a physical system, wherein the physical system also comprises a heat source, a heat supply network, a heat load and a heat storage system. The invention is based on the information acquisition, information transmission and physical system of the existing heating system, and only needs to improve the method of the control system, so that the invention does not need to change the structure of the whole heating system.
The heat demand prediction process may be simplified as represented by the flow chart of fig. 2: and transmitting the acquired information set to a computer terminal, and predicting the heat demand by using a prediction algorithm through the computer by utilizing big data. And after prediction, using the prediction information at each heat demand end and re-collecting demand end data, and then combining the collected demand end data with other collected data to form a new information set for the next heat demand prediction.
The improvement point of the invention is to reconstruct a memory feedforward mechanism of the LSTM by introducing a periodic feedback mechanism and considering the specific attribute of a heat supply demand pre-estimated scene, thereby improving the periodic perception of the LSTM on time-series pre-estimation and improving the pre-estimation precision.
The conventional LSTM recurrent neural network is composed of successive repeating modules, and each repeating module may be represented by fig. 5, where there are mainly three core steps inside the LSTM as shown, and four single-layer fully-connected layers, and each fully-connected network follows a set of nonlinear functions with gating properties, and constrains the output of the fully-connected layers element by element. Each core is described in detail below:
first, the cell state is forgotten. LSTM as one of the feedback neural networks, the core is that the feedback neural network is controlled by cell states (C t ) And the hidden state (h t ) Two vector pairs t i The time sequence state before the moment is recorded and is used as the input of the LSTM cell at the next moment. Therefore, firstly, according to the input of the current moment, the cell state is forgotten by combining with the hidden state, and the history misuse information is filtered. This step is performed by "forget door" f t The process is carried out:
f t ＝z(W f ·[h t-1 ,x t ]+b f )；
wherein h is t-1 Is the hidden state of LSTM at the last moment, x t Is the characteristic vector of the current moment, W f And b f The weight and bias of the forgetting gate full connection layer are respectively.
z is a sigmoid function, expressed asAs shown in fig. 3, when the argument is less than-5, the function value is close to 0, called a forgetting area; thus, sigmoid is used in a neural network as a gating function, see fig. 3.
And secondly, updating the cell state according to the feature vector at the current moment. The method comprises the steps of dividing the hidden state and the input feature vector of the previous step into two links, processing the hidden state and the input feature vector by a full-connection layer, and restraining a value range by a tanh function:
wherein W is c And b c Is the weight and bias of the cell state update layer; similar to a forgetful door, h t-1 Is the hidden state of LSTM at the last moment, x t Is the feature vector of the current moment.
the tanh function has the expression ofThe function shape is shown in figure 4, and the tanh function can transform the independent variable into a (-1, 1) interval, so that the extreme value of the feedback neural network can be restrained, and the self-excitation is avoided; the second link is to gate control constraint on the current newly added state and fuse with the cell state screened by the forgetting gate, specifically:
i t ＝sigmoid(W i ·[h t-1 ,x t ]+b i )；
W i and b i Respectively are provided withIs the weight and bias of the forget gate full connection layer.
And thirdly, updating the hidden state according to the cell state. Firstly, converting the value domain of the cell state updated in the last step by using a tanh function, and then carrying out information constraint by an output gate:
O t ＝z(W o [h t-1 ,x t ]+b o )
h t ＝O t *tanh(C t )；
W o and b o The weight and bias of the gate full connection layer are respectively output, the cell state and the hidden state are used as the expression of the cell state at the moment, and on one hand, the cell state is used as the input of the LSTM of the next layer or the time sequence estimation network.
In a heating system, h is used for heat which needs to be acquired at a certain moment of building t Indicating that the heat required at the last moment and the next moment is h t-1 And h t+1 The prediction time interval is set to be predicted once an hour. By x t Representing the obtained characteristic vector, including building area, building ventilation quantity, average people flow in the building, current outdoor air temperature, current indoor air temperature and electric appliance energy consumption in the building; the building area information can be obtained by one-time measurement and cannot be changed along with time; indoor and outdoor air temperature, average people flow in the building, and electric appliance energy consumption in the building are measured and updated once every other time unit (1 hour). The feature vector x to be acquired t As input of LSTM system, combine hidden variable h of LSTM at last moment t-1 Updating the state of LSTM according to the output h of LSTM at this moment t And estimating the heat supply requirement.
The estimated number of steps of LSTM is set to 72, i.e., the historical time series state of the past 72 hours is processed. However, although LSTM has some ability to retain long-term history of timing, studies have shown that LSTM still cannot capture states before multiple steps, limited by single-layer neural networks and the expressive power of state vectors. Therefore, "attention mechanisms" were introduced in an attempt to enhance the ability of LSTM to capture long-term memory information. In the heat supply demand pre-estimation scene, the method comprises the following steps ofIt can be easily seen from the historical data of past heating actual demand that the heat demand at the same time of the previous day is an important reference value, because the variation between two consecutive days is relatively small, and the heating demand at the same time of the day is closest. Therefore, a new feedback state is introduced in the heat demand assessment method: h is a t-24 The utilization of the weather-to-weather signal by LSTM is enhanced.
The repetition of the LSTM heating prediction after introducing new parameters is shown in fig. 6.
The periodic feedback LSTM after adding new parameters still consists of three steps:
first, forget cell state:
f t ＝z(W f ·[h t-1 ,h t-24 ,x t ]+b f )；
wherein h is t-1 Is the hidden state of LSTM at the last moment, x t Is the characteristic vector of the current moment, W f And b f The weight and the bias of the full connection layer of the forgetting gate are respectively h t-24 The feedback state is obtained by collecting heat information required by the same time of the previous day and encoding and transmitting the heat information into a system, and z is a sigmoid function.
Secondly, updating the cell state according to the current feature vector:
i t ＝z(W i ·[h t-1 ,h t-24 ,x t ]+b i )
wherein W is i And b i Weights and biases of the full connection layers of the forgetting gate, W c And b c Is the weight and bias of the cell state update layer.
Thirdly, updating the hidden state according to the cell state and the feedback state:
O t ＝z(W o [h t-1 ,h t-24 ,x t ]+b o )
h t ＝O t *tanh(C t )；
W o and b o The weight and bias of the gate full-connection layer are respectively output, the cell state, the feedback state and the hidden state are taken as the expression of the current cell state together and are taken as the input of the LSTM of the next layer or the time sequence estimating network;
h obtained by calculation t The estimated heating amount required at this time of the system.
Introducing a new parameter h t-24 Compared with the traditional heat supply demand prediction method, the periodic feedback LSTM system is more accurate and intelligent; compared with the traditional LSTM system, the system has stronger robustness, greatly improves the system stability, and can accurately and timely predict the heat demand of a certain area at the current moment by periodically feeding back the LSTM, thereby achieving good energy-saving and emission-reducing effects.
Periodic feedback LSTM model training process:
heat supply demand data was collected over the last 5 years, training LSTM with minimum mean square error as a loss function:
wherein y is t Is the real heat supply demand of each hour, W l To output weight vectors, the elements of hidden states are weighted and summed, the weight of LSTM is derived by back propagation and updated by random gradient descent, specifically, each h is calculated from the loss function t These comprise two components, namely a gradient directly conducted by a loss function and a loss function propagated by a later LSTM:
wherein LSTM t+1 Input gate representing next step LSTMGuiding hidden states by output gates, forget gates and state update items
The sum of the numbers, after which the derivative of the loss function with respect to the cell state is as follows:
from the chain law, the gradient of LSTM weights can be found from the loss function versus hidden state and cell state:
according to the random gradient descent method, the gradient of the weight is updated based on the loss function:
wherein alpha is the learning rate, which is set to 0.01 by the invention.
Referring to fig. 7a, 7b, 7c, 7d and 7e, a schematic diagram of time-by-time thermal load demands in different time periods of a base is obtained based on the method of the present invention.
The heat supply demand estimation method based on the periodic feedback LSTM can be in the form of an embodiment of complete hardware, an embodiment of complete software or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The heat supply demand prediction method based on the periodic feedback LSTM can be stored in a computer readable storage medium if the heat supply demand prediction method is realized in the form of a software functional unit and sold or used as an independent product.
Based on such understanding, in an exemplary embodiment, a computer readable storage medium is also provided, where the present invention implements all or part of the flow of the method of the above embodiment, and may also be implemented by a computer program to instruct related hardware, where the computer program may be stored in the computer readable storage medium, where the computer program, when executed by a processor, may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. Computer-readable storage media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. The computer storage media may be any available media or data storage device that can be accessed by a computer, including, but not limited to, magnetic storage (e.g., floppy disks, hard disks, magnetic tape, magneto-optical disks (MOs), etc.), optical storage (e.g., CD, DVD, BD, HVD, etc.), and semiconductor storage (e.g., ROM, EPROM, EEPROM, nonvolatile storage (NANDFLASH), solid State Disk (SSD)), etc.
In an exemplary embodiment, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the periodic feedback LSTM based heat supply demand estimation method when executing the computer program. The processor may be a central processing unit (CentralProcessingUnit, CPU), but may also be other general purpose processors, digital signal processors (DigitalSignalProcessor, DSP), application specific integrated circuits (ApplicationSpecificIntegratedCircuit, ASIC), off-the-shelf programmable gate arrays (Field-ProgrammableGateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like.
The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (6)
1. A heat supply demand estimation method based on periodic feedback LSTM is characterized by comprising the following steps:
forgetting the cell state according to the feature vector at the current moment and combining the hidden state, filtering the historical misuse information, and screening the cell state;
processing the hidden state and the characteristic vector at the current moment through a full-connection layer, restraining the value range of the hidden state and the characteristic vector by a tanh function, then carrying out gate control restraint on the current newly-added state, and fusing the newly-added state with the screened cell state to obtain an updated cell state;
transforming the value domain of the updated cell state by using a tanh function, and then carrying out information constraint by an output door to obtain a current hidden state, namely a current heating quantity predicted value required by the system;
according to the feature vector at the current moment, forgetting the cell state by combining the hidden state, and realizing the cell state through a forgetting door, wherein the method comprises the following steps of:
f t =z(W f ·[h t-1 ,h t-24 ,x t ] +b f )；
wherein h is t-1 Is the hidden state of LSTM at the last moment, x t Is the characteristic vector of the current moment, W f And b f The weight and the bias of the full connection layer of the forgetting gate are respectively h t-24 The feedback state is obtained by collecting heat information required by the same time of the previous day and encoding and transmitting the heat information into a system, and z is a sigmoid function; processing the hidden state and the input feature vector through a full-connection layer; restricting the value domain by using the tanh function, then performing gate control restriction on the current newly-added state, and fusing with the screened cell state to obtain an updated cell state C t The specific calculation formula is as follows:
i t =z(W i ·[h t-1 ,h t-24 ,x t ] +b i )
Ĉ t =tanh(W c ·[h t-1 ,h t-24 ,x t ] + b c )
C t =f t *C t-1 +i t *Ĉ t ；
wherein W is i And b i Weights and biases of all connection layers of forgetting gate, x t Is the characteristic vector of the current moment, W c And b c Is the weight and bias of the cell state update layer;
hidden state O according to cell state and feedback state t The updating is carried out as follows:
O t = z (W o [h t-1 ,h t-24 ,x t ] + b o )；
h t = O t * tanh(C t )；
W o and b o The weight and the bias cell state of the gate full-connection layer are respectively output, and the feedback state and the hidden state are used as the expression of the cell state at the moment and are used as the input of the LSTM of the next layer or the time sequence estimation network; h obtained by calculation t The heating quantity predicted value at the current moment of the system is obtained.
2. The heat supply demand estimating method based on the periodic feedback LSTM according to claim 1, wherein the feature vector at the present moment includes a building area, a building ventilation amount, an average people flow in the building, a present outdoor air temperature, a present indoor air temperature, and an electric appliance energy consumption in the building, the building area in the input feature vector is fixed, and the remaining parameters in the input feature vector are measured and updated once every preset time.
3. The heat supply demand estimation method based on periodic feedback LSTM according to claim 1, wherein heat supply demand data of a set time in the past is collected, and the LSTM is trained with a minimum mean square error as a loss function, and the loss function is specifically:
loss =
wherein y is t Is the real heat supply demand of each hour, W l For outputting weight vectors, the elements in hidden states are weighted and summed, the weight of the LSTM is derived by using a back propagation mode, and the LSTM is updated by a random gradient descent method.
4. The heat supply demand prediction system based on the LSTM is characterized by being used for realizing the heat supply demand prediction method based on the periodic feedback LSTM according to any one of claims 1-3, and comprising a forgetting module, a cell state updating module and a hidden state updating module; the forgetting module is used for forgetting the cell state according to the characteristic vector at the current moment and combining the hidden state, filtering the historical misuse information and screening the cell state;
the cell state updating module is used for processing the hidden state and the characteristic vector at the current moment through a full-connection layer, restricting the value range of the feature vector by a tanh function, then carrying out gate control restriction on the current newly-added state, and fusing the newly-added state with the screened cell state to obtain an updated cell state; according to the feature vector at the current moment, forgetting the cell state by combining the hidden state, and realizing the cell state through a forgetting door, wherein the method comprises the following steps of:
f t =z(W f ·[h t-1 ,h t-24 ,x t ] +b f )；
wherein h is t-1 Is the hidden state of LSTM at the last moment, x t Is the characteristic vector of the current moment, W f And b f The weight and the bias of the full connection layer of the forgetting gate are respectively h t-24 The feedback state is obtained by collecting heat information required by the same time of the previous day and encoding and transmitting the heat information into a system, and z is a sigmoid function;
and the hidden state updating module transforms the value domain of the updated cell state by using a tanh function, and then the output gate performs information constraint to obtain the current hidden state, namely, the current required heating quantity predicted value of the system.
5. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, carries out the steps of the heat supply demand estimation method based on periodic feedback LSTM according to any one of claims 1 to 3.
6. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the heat supply demand estimation method based on periodic feedback LSTM as claimed in any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110256564.1A CN112862213B (en) | 2021-03-09 | 2021-03-09 | Heat supply demand prediction method, system and equipment based on periodic feedback LSTM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110256564.1A CN112862213B (en) | 2021-03-09 | 2021-03-09 | Heat supply demand prediction method, system and equipment based on periodic feedback LSTM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112862213A CN112862213A (en) | 2021-05-28 |
| CN112862213B true CN112862213B (en) | 2024-02-20 |
Family
ID=75993613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110256564.1A Active CN112862213B (en) | 2021-03-09 | 2021-03-09 | Heat supply demand prediction method, system and equipment based on periodic feedback LSTM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112862213B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361829B (en) * | 2021-08-09 | 2021-11-02 | 新风光电子科技股份有限公司 | Time adjustment method for contract energy management based on artificial intelligence and big data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921341A (en) * | 2018-06-26 | 2018-11-30 | 国网山东省电力公司电力科学研究院 | A kind of steam power plant's short term thermal load forecasting method encoded certainly based on gate |
| CN110619389A (en) * | 2019-09-23 | 2019-12-27 | 山东大学 | Load prediction method and system of combined cooling heating and power system based on LSTM-RNN |
| CN110782093A (en) * | 2019-10-26 | 2020-02-11 | 福州大学 | PM fusing SSAE deep feature learning and LSTM network 2.5Hourly concentration prediction method and system |
| CN111523730A (en) * | 2020-04-23 | 2020-08-11 | 中国华能集团有限公司 | Method for predicting coal consumption of cogeneration unit |
| CN111797979A (en) * | 2020-07-17 | 2020-10-20 | 中国人民解放军海军工程大学 | Vibration transmission system based on LSTM model |
-
2021
- 2021-03-09 CN CN202110256564.1A patent/CN112862213B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921341A (en) * | 2018-06-26 | 2018-11-30 | 国网山东省电力公司电力科学研究院 | A kind of steam power plant's short term thermal load forecasting method encoded certainly based on gate |
| CN110619389A (en) * | 2019-09-23 | 2019-12-27 | 山东大学 | Load prediction method and system of combined cooling heating and power system based on LSTM-RNN |
| CN110782093A (en) * | 2019-10-26 | 2020-02-11 | 福州大学 | PM fusing SSAE deep feature learning and LSTM network 2.5Hourly concentration prediction method and system |
| CN111523730A (en) * | 2020-04-23 | 2020-08-11 | 中国华能集团有限公司 | Method for predicting coal consumption of cogeneration unit |
| CN111797979A (en) * | 2020-07-17 | 2020-10-20 | 中国人民解放军海军工程大学 | Vibration transmission system based on LSTM model |
Non-Patent Citations (3)
| Title |
|---|
| "District Heating Load Prediction Algorithm Based on Feature Fusion LSTM Model";Guixiang Xue等;《energies》;全文 * |
| "基于LSTM 模型的短期负荷预测";张亚宁;《科技与创新》(第12期);第3节，图4 * |
| "基于改进BP神经网络的供热负荷预测模型";李思琦 等;《软件导刊》;第18卷(第7期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112862213A (en) | 2021-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Wind power prediction based on PSO-SVR and grey combination model | |
| CN110705743B (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
| Liu et al. | Short-term load forecasting using a long short-term memory network | |
| CN103903452B (en) | Forecasting Approach for Short-term Traffic Flow | |
| Alencar et al. | Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil | |
| CN108711847B (en) | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network | |
| CN110047291B (en) | Short-term traffic flow prediction method considering diffusion process | |
| CN111027772A (en) | Multi-factor short-term load prediction method based on PCA-DBILSTM | |
| Shi et al. | Short‐term load forecasting based on CNN‐BiLSTM with Bayesian optimization and attention mechanism | |
| CN111160659B (en) | Power load prediction method considering temperature fuzzification | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN112862213B (en) | Heat supply demand prediction method, system and equipment based on periodic feedback LSTM | |
| CN115222106A (en) | User day-ahead load prediction method of self-adaptive model | |
| CN116049764A (en) | Cross-scale time sequence data fusion method and system for Internet of things | |
| Ghassemi et al. | Optimal surrogate and neural network modeling for day-ahead forecasting of the hourly energy consumption of university buildings | |
| CN112767692A (en) | Short-term traffic flow prediction system based on SARIMA-GA-Elman combined model | |
| Fatema et al. | Analysing and forecasting electricity demand and price using deep learning model during the COVID-19 pandemic | |
| Perez et al. | Performance analysis of bagging feed-forward neural network for forecasting building energy demand | |
| Azad et al. | Grid serving deployment of smart meter data in the context of distribution grid automation | |
| CN109146191B (en) | Short-term power load prediction method and system based on generic implication model | |
| CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
| Shabani et al. | Hourly prediction of building energy consumption: an incremental ANN approach | |
| CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm | |
| CN110009132A (en) | A kind of short-term electric load fining prediction technique based on LSTM deep neural network | |
| JP3650304B2 (en) | Demand forecasting device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |