CN112712189A - Heat supply demand load prediction method - Google Patents
Heat supply demand load prediction method Download PDFInfo
- Publication number
- CN112712189A CN112712189A CN201911021471.XA CN201911021471A CN112712189A CN 112712189 A CN112712189 A CN 112712189A CN 201911021471 A CN201911021471 A CN 201911021471A CN 112712189 A CN112712189 A CN 112712189A
- Authority
- CN
- China
- Prior art keywords
- data
- load prediction
- demand load
- prediction model
- input
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000010438 heat treatment Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 34
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- 238000012937 correction Methods 0.000 claims abstract description 24
- 230000006403 short-term memory Effects 0.000 claims abstract description 15
- 230000007787 long-term memory Effects 0.000 claims abstract description 14
- 238000004140 cleaning Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 19
- 230000015654 memory Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000003032 molecular docking Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000003672 processing method Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012821 model calculation Methods 0.000 claims description 2
- 230000001502 supplementing effect Effects 0.000 claims description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 13
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000005265 energy consumption Methods 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000007812 deficiency Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 210000001503 joint Anatomy 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a heat supply load prediction method, which comprises the following steps: step 1, determining an input variable and an output variable of a demand load prediction model based on relevant factors of a heating system load; step 2, butting operation data, cleaning missing and abnormal data, giving correction factors according to operation experience data under the condition of no room temperature data, correcting sample load, and determining load prediction model training data; step 3, training and generating a heat supply system demand load prediction model based on a deep long-term and short-term memory network algorithm; step 4, providing correction factors by combining the empirical values, and improving the accuracy of the prediction model; and 5, detecting the model precision on line, and retraining to generate a demand load prediction model according to rules. The heat supply demand load prediction method provided by the invention can realize accurate prediction of the demand load of the heat supply system.
Description
Technical Field
The invention relates to the field of heating systems, in particular to a heating demand load prediction method applying an LSTM algorithm and without room temperature data correction.
Background
By 2016, the heating area of northern towns is about 206 hundred million square meters, and the energy consumption of a heating system accounts for about 20 percent of the energy consumption of buildings, however, the overall energy consumption level is higher due to the fact that the operation management mode of heating enterprises is relatively extensive. According to statistics of Chinese building energy-saving annual development research reports, the over-supply load of a heating system in China is about 35%, and the energy consumption waste is serious. The prediction of the demand load is one of the main ways to realize the energy-saving operation of the heating system. The heat load of the heating system is related to a plurality of factors, including the building envelope, the heat consumption habits of users, the regulation and control level of operators, the automation conditions and the feedback condition of actual users, and the complex factors are the difficulty in realizing accurate load prediction. Secondly, the data of the room temperature measuring points are the most key parameters for evaluating and predicting the load, but because the information level of the existing heat supply network is limited, most of secondary networks do not have the data of the room temperature measuring points, and the load prediction is difficult to evaluate. The neural network algorithm is a machine learning algorithm of basic data, and is more suitable for establishing a 'black box' model with fuzzy relation among multiple factors aiming at the characteristics of the complex incidence relation.
In recent years, a Recurrent Neural Networks (RNNs) algorithm is taken as a representative example, and has been successfully applied to various fields. RNNs are networks with cycles, which are characterized by connecting previous occurrence information to current targets, allowing information to exist continuously, and using the previous occurrence information to infer the current occurrence content, which is not possessed by traditional neural networks, therefore, the RNNs have succeeded in applications of natural language processing, image recognition, speech recognition, power grid prediction, and industrial process modeling, however, as time span increases, the perception of the previous time node information by the following time node decreases, and the information in historical data cannot be effectively used, so that the problem of model distortion occurs.
Disclosure of Invention
Aiming at the problems, the invention aims to provide a heat supply demand load prediction method applying a deep Long Short Term Memory algorithm and no room temperature data correction, and the method realizes the accurate prediction of the demand load of a heat supply system by combining an online applied deep Long Short Term Memory (LSTM) network algorithm and a heat supply system corrected by an operator according to experience parameters, so as to overcome the problem of information loss of an RNNs algorithm when the time span is increased and realize the accurate prediction of the demand load of the heat supply system.
In order to solve the above problems, the present invention provides a method for predicting a heat demand load, comprising the steps of:
step 2, butting operation data, performing data cleaning on missing and abnormal data, giving correction factors according to operation experience data under the condition of no room temperature data, correcting sample load, and determining load prediction model training data;
and 5, detecting the model precision on line, and retraining to generate a demand load prediction model according to rules.
According to some embodiments of the present invention, in step 1, input variables of the demand load prediction model are various factors affecting the heat supply load, output data of the output variables are demand load data at a time to be predicted, and the input data and the output data form a sample set for load prediction model training.
According to some embodiments of the present invention, the docking operation data in step 2 is real-time access input data, the obtained input data is stored in the load prediction model database, an input data set of the load prediction model is established according to the input data, and a certain number of groups of input data sets form an input data set matrix.
According to some embodiments of the present invention, in step 2, a data cleaning process is performed on missing and abnormal data, a missing proportion threshold for determining whether the data is selected as sample data is set, an abnormal data threshold for determining abnormal data is set, and the data is classified according to the missing proportion and processed: the method comprises the steps that in the first type, when the proportion of the number of missing data in the current data record to the total number of data is larger than a missing proportion threshold value, the group of data is discarded, and the number of sample data pieces is automatically complemented;
in the second type, the proportion of the number of the missing data in the current data record to the total number of the data is less than or equal to a missing proportion threshold value, and the data is subjected to completion processing;
when the current data abnormity judgment exceeds the abnormal data threshold value, discarding the abnormal data and then regarding the abnormal data as missing data, supplementing the data according to the first or second method, and completing replacement processing on the abnormal data;
and in the fourth class, for the data without the room temperature measuring point, a load correction factor is added, the load of the sample is corrected by combining the matched correction factor, and the corrected data is used as the input data and the output data of the sample.
According to some embodiments of the invention, the second type of processing method comprises:
when the missing characteristic is a discontinuous value, the completion method adopts a front and back numerical arithmetic mean method;
when the missing features are continuous values, the completion method adopts an interpolation method.
According to some embodiments of the invention, the input data and the output data are normalized, and the normalized data is used as sample data of the load prediction training model.
According to some embodiments of the present invention, in the step of constructing the demand load prediction model in step 3, the deep long-term and short-term memory network structure is composed of an input layer, a hidden layer and an output layer, the input layer, the hidden layer and the output layer respectively correspond to respective coefficients, the relationship among the input layer, the hidden layer and the output layer is established according to the coefficients, and the demand load prediction model is formed by optimizing the coefficients through an algorithm.
According to some embodiments of the invention, the deep long and short term memory network is formed by linking long and short term memory units, each long and short term memory unit comprising:
an input gate for controlling whether the information of the long and short term memory unit needs to be updated;
a forgetting gate for controlling whether the information of the long-term and short-term memory unit needs to be deleted;
and the output gate controls whether the information of the long-term and short-term memory unit needs to be reflected in the activation vector, wherein the used activation function adopts a sigmoid function.
According to some embodiments of the present invention, in step 3, a mean square error is used as a loss function for training the demand load prediction model, an iterative convergence coefficient is set, and when the loss function is smaller than the iterative convergence coefficient, the training of the demand load prediction model is completed and the demand load prediction model for predicting the demand load is obtained.
According to some embodiments of the invention, in step 5, when the accuracy of the demand load prediction model calculation value and the actual measurement value deviates by more than an iteration convergence coefficient, the model is retrained, and the step of retraining the model comprises:
step 5.1, based on the demand load prediction model, accessing input data at the current moment, and calculating a predicted load value of the demand load prediction model at the current moment;
step 5.2, capturing the operation data of each site at the current moment, and calculating the consumption load of the site at the current moment;
step 5.3, calculating the deviation precision of the demand load prediction model;
step 5.4, when the deviation precision is less than or equal to the iterative convergence coefficient, the model demand load prediction is not corrected;
step 5.5, when the deviation precision is larger than the iterative convergence coefficient, starting from the time node of the current moment, reading input and output historical data in a time interval, and retraining the model, wherein the time interval is the interval of the time interval with the same group number as the group number of the input data sets in the input data set matrix; when missing data occurs in the time interval, the interval moves forward until the number of the input data set groups is unchanged.
The invention has the beneficial effects that: the invention adopts a special gate structure in a deep LSTM network method to effectively memorize or screen historical data information, and solves the problem that the traditional RNNs neural network can not effectively utilize long-term data information; and secondly, the invention combines the data fed back by the operation experience as the correction of the load prediction model, so that the predicted load is more in line with the actual requirement, and the precision of the load prediction model is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other embodiments can be obtained by using the drawings without creative efforts.
FIG. 1 is a block diagram of the main steps of the heating demand load prediction method provided by the present invention;
FIG. 2 is a schematic diagram of data docking of the heat demand load prediction method provided by the present invention;
fig. 3 is a schematic structural diagram of an LSTM network of the heating demand load prediction method provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following embodiments of the present invention are described in further detail with reference to the accompanying drawings.
It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are used for distinguishing two entities with the same name but different names or different parameters, and it is understood that "first" and "second" are only used for convenience of expression and should not be construed as limitations to the embodiments of the present invention, and the descriptions thereof in the following embodiments are omitted.
Based on the above purpose, the embodiment of the invention provides a heat supply demand load prediction method.
Fig. 1 is a block diagram illustrating the main steps of the heating demand load prediction method provided by the present invention. Referring to fig. 1, the method for predicting a heat demand load using a deep long and short term memory algorithm without room temperature data correction according to the present invention mainly includes: step 1, shown in block S100, determining an input variable and an output variable of a demand load prediction model based on relevant factors of a heating system load; step 2, as shown in a block S200, the operation data is butted, missing and abnormal data are subjected to data cleaning, under the condition of no room temperature data, correction factors are given according to operation experience data, sample load is corrected, and load prediction model training data are determined; step 3, shown in a block S300, training and generating a heat supply system demand load prediction model based on a deep long-term and short-term memory network algorithm; step 4 shown in the block S400, load correction factors are given by combining the empirical values, and the accuracy of the prediction model is improved; and step 5, detecting the model precision on line, and retraining to generate the demand load prediction model according to the rule in the block S500.
Further, in step 1 shown in block S100, input and output variables of the load prediction model are determined based on the relevant factors of the heating system load, where: the LSTM network algorithm belongs to a machine learning algorithm, the first step of establishing a load prediction model based on the machine learning algorithm is to analyze relevant factors related to the load model, determine input variables and output variables, and then establish a 'black box' model for load prediction by adopting algorithm training based on a data set of historical operating conditions of a heat supply system.
The demand load prediction in the heating system means that the indoor temperature of the user side is maintained to reach the demand load of 18 ℃, and therefore the output of the load prediction model is the load value to be predicted.
Output of load prediction modelThe input data are various factors influencing the heating load, and the input data X of the load in the invention comprises the following components: the meteorological data U at the next moment, whether the current day is a holiday L and other factors, and the actual load Q at the previous momentsThe output data of the user side room temperature data T is the demand load data Q at the moment to be predictedx. Therefore, the input data X and the output data Q in the historical operation condition of the heating systemxA sample set of load prediction model training is constructed.
The object of demand load prediction in the present invention may be a heat source, a heat station, or a heat consumer. Wherein, the meteorological data, the holiday data and the room temperature data are public data conditions, the load data is individual data, and when the object is a heat source, Q issAll loads produced for that heat source; when the object is a heating power station, then QsSupplying the output load of the secondary network to the site; when the object is a single hot user, Qs is the thermal load that the user has consumed at the last moment. The following description takes a thermal station as an example.
Further, step 2 shown in block S200, docking the operation data, performing data cleaning on missing and abnormal data, giving a correction factor according to the operation experience under the condition of no room temperature data, correcting the sample load, and determining the training data of the load prediction model, includes: step 2.1, butting with operation data; step 2.2, data cleaning is carried out on missing and abnormal data; and 2.3, carrying out data standardization processing to determine an input data sample set and an output data sample set of the training model.
Further, step 2.1, docking with operational data, wherein: the invention provides the on-line prediction of the heat supply load, so that the real-time access of the operation data is the premise work of processing the operation data and then modeling.
Fig. 2 is a schematic data docking diagram of the heating demand load prediction method provided by the invention. The input data involved in the data interfacing comprises meteorological data, holiday data, actual load at the last moment and user side room temperature data. The meteorological data and the holiday data are obtained by reading an appointed third-party data interface 2 of a third-party database 1, wherein the meteorological data comprise: parameters such as outdoor temperature, air humidity, wind direction, illumination conditions and the like at the next moment; the actual load and the user side room temperature data can be read from the heat supply enterprise operation database by adopting OPC or other modes. Calculating relevant data of the actual load, wherein the relevant data comprises the water supply temperature of the primary network, the water return temperature of the primary network and the flow of the primary network of each station, which are acquired by the heat supply automation system; and the output data of the room temperature data of the user side, namely the measuring point data of the indoor temperature of the heat user is the heat supply load Q data of each heating power station in the historical operating condition.
The purpose of data butt joint is to establish a data set of a load prediction model, the realization mode is that according to a specified time interval, third-party data interfaces 2 or an automatic control system database 4 are read through data butt joint software 3, required data are captured and stored in a load prediction model database 5, the characteristics of data captured at a time are n, a model training data set is m groups, and then an input data set matrix X can be expressed as:
wherein x isijThe jth characteristic data representing the ith sample input data.
Further, step 2.2, data cleaning is carried out on missing and abnormal data, and the method comprises the following steps: in the process of data cleaning, missing data and abnormal data are mainly processed, and the purposes of removing abnormality, correcting errors and complementing missing are achieved.
In the process of data acquisition and transmission, data loss or abnormal phenomena can occur. For the data missing condition, classifying according to the data missing proportion, and the processing method is as follows:
let Y be [ Y1, Y2.., yi.,. yn ],
(1) if the proportion of the number of the missing data pieces of the current data record to the total number of the data pieces exceeds theta%, the following steps are carried out:
discarding the data, and automatically complementing the number of the sample data to meet the standard of the m groups of data after discarding the group of data;
(2) when the missing ratio is less than theta%, the data needs to be completed, and when the missing characteristic is a discontinuous value yk, a front-back numerical arithmetic mean method is adopted, so that the missing characteristic yk:
when the deficiency ratio is less than θ%, for example, and the deficiency characteristics are continuous values yk, yk + p, since the measurement time interval of the heating system data is short, typically several tens of seconds or several minutes, the characteristics of the operation data are regarded as a continuously changing process, and the complement method employs an interpolation method. Then:
(3) for abnormal data, i.e. data that suddenly changes at a certain time, replacement processing is also required. Judging abnormal data ypThe method comprises the following steps:
when in use
And (3) discarding the abnormal data and then regarding the discarded abnormal data as missing data, and performing abnormal data replacement by adopting the method (1) and the method (2).
Wherein:
lithe number of the missing data features of the ith piece of data;
θ: judging whether the data is selected as the missing proportion threshold of the sample data;
y*the mean value of all characteristic data in the first column;
beta: and judging the threshold value of the abnormal data.
(4) And adding a load correction factor to the data without the room temperature measuring point, and processing the sample load data. Supplying water to the secondary network under the condition of no room temperatureThe temperature and the return water temperature are used as the basis for operators to judge whether station heat supply reaches the standard, corresponding to each thermal power station, the water supply temperature and the return water temperature which are required by each operator exist, when the actual temperature exceeds the required temperature, the over-supply is represented, otherwise, the under-supply is represented, therefore, whether station loads meet the requirements is judged according to the parameters, correction parameters of a load prediction model are established, and the correction is carried out aiming at the loads of the thermal power stations, and the method comprises the following steps: according to the outdoor temperature interval [ Tlow，Tup]Divided into z outdoor conditions, i.e. U ═ U1,U2,...,Uj,...Uz]TAn operator gives an empirical value of the required water supply temperature Ts and an empirical value of the required water return temperature Tr of the secondary network of each station under z working conditions, wherein the empirical values are mainly based on the relation between room temperature data measured by households and the water supply and return temperatures, namely, for example, the indoor temperature measured under a certain working condition is 18 ℃, the water supply temperature of the secondary network is 50 ℃, and the temperature is the required water supply temperature of the secondary network of the station. And obtaining the correction factor lambda of the water supply temperature under the corresponding working conditions，λs＝[λs1,λs2,...,λsj,...λsz]TCorrection parameter alpha with return water temperature of secondary networkr，αr＝[αr1,αr2,...,αrj,...αrz]TAnd then:
the working condition Ui and the corresponding secondary network supply return temperature correction factor are [ lambda ]sj、λrj]And correcting the station load data under the condition of no room temperature according to the correction factors as follows according to the working condition: qs'j＝Qsj×λsj×λrj
And if no measured quantity data exists, correcting the load of the sample by combining the outdoor working condition U of the sample data, and taking the corrected data as the input and output data of the sample.
Step 2.3, standardizing the data by adopting a normalization method, and determining an input data sample set and an output data sample set of the training model, wherein the method comprises the following steps:
inputting a data set matrix X, and carrying out normalization processing on each column of data, wherein the processing method comprises the following steps:
and X is carried out on the data sequence, and the transformation rule is as follows:
the processed data matrix Z:
wherein,
s is the standard deviation of the first column of feature data.
And taking the normalized data as sample data of the load prediction training model.
Further, in step 3 shown in block S300, based on the deep LSTM network algorithm, training and generating a heat supply system demand load prediction model, including: step 3.1, constructing a load prediction model based on a deep LSTM network algorithm; and 3.2, training to generate a load prediction model.
Further, step 3.1, applying a deep LSTM network algorithm to construct a load prediction model, including: the deep LSTM network structure is composed of an input layer, a hidden layer, and an output layer. The input layer is used as an input condition interface of the model, namely various factors influencing the load; the hidden layer is an intermediate layer of the LSTM and generally comprises a plurality of layers, wherein the input of the current hidden layer is the output of the previous hidden layer, and the output of the current hidden layer is the input of the next hidden layer; and the output layer calculates the load predicted value under the current input condition through a model. Each layer of the network has a corresponding coefficient by which to establish a connection between the input layer, the hidden layer and the output layer. And (3) optimizing coefficients of each layer by adopting an algorithm, evaluating the convergence effect of the network by using a loss function to establish the characteristics between input data and output data, and forming a demand load prediction model after the calibration of model parameters meets the conditions.
Fig. 3 is a schematic structural diagram of an LSTM network of the heating demand load prediction method provided by the present invention. Similar to RNNs networks, LSTM networks are formed by linking identical LSTM units 6, but LSTM units are more complex than RNNs units and are more conducive to information updating. Each LSTM unit comprises 3 special gate structures, and deletion or updating of long-term memory information is realized through application of different gate structures, so that a better training effect is achieved. The specific calculation process of the LSTM unit is as follows:
forget the door 7:
the input gate 8:
memory cell state:
an output gate 9:
wherein σ is an activation function; x is the number of<t＞Inputting data of the network when the time step is t; a is<t＞Output data of the LSTM unit hidden layer when the time step is t; wf、Wu、Wc、WoAnd bf、bu、bc、boValues generated by network training for parameters common to all LSTM units at the same layer.
Therefore, forget to pass through the door
And c<t＞Performing dot multiplication to determine whether to reserve the original state of the t-1 time step memory unit; input gate through
And
performing dot multiplication to determine whether to update the state of the t time step memory unit; memory unit state c when forgetting to calculate time step t<t＞Then, the output is processed by the output of tanh nonlinear function and output gate
Obtaining hidden layer a under t time step by dot multiplication<t＞. Determination after algorithm training
I.e. a gate structure that controls whether the information of the LSTM network element needs to be updated (input gate), deleted (forget gate), and reflected in the activation vector (output gate).
The activation function of the gate adopts a sigmoid function, and the function value can be fully smooth in the (0, 1) interval. The network adopts the full connection layer as the model output value of the load prediction value
Input as hidden layer data a<t＞Namely:
wherein Wdense is a weight coefficient of the last layer of the network, b is a Bayesian parameter, and both are obtained through network training; actv () is an activation function, and the present invention takes a linear function as the activation function. Based on the LSTM network structure, a load prediction model can be constructed
Further, step 3.2, training and generating a load prediction model, including: the coefficient of the model is trained by adopting an algorithm, the mean square error is taken as a 'loss function' of the model training, and the smaller the loss function is, the better the model training is represented. The mean square error expression is:
setting an iterative convergence coefficient epsilon when satisfying
When the model training is finished, the model for predicting the demand can be obtained
Further, step S500, detecting the model accuracy on line, and retraining the generated load prediction model according to the rule, including: when the calculated values of the model deviate from the actual measured values, namely the model begins to be misaligned, when the precision deviation exceeds epsilon, the model needs to be retrained. The steps for retraining the model are as follows:
(1) load-based prediction model
Inputting the input data at the current moment, and calculating the model predicted load value at the current moment t1
(2) Capturing the operation data of each station at the current moment, and calculating the consumption load Q of the station at the current momentx。
(3) Calculating deviation accuracy of model
(4) When in use
When the model is not corrected, the model is not corrected; when in use
And (5) turning to the step (5).
(5) Selecting t1 time node 1; selecting a t2 time node 2, wherein the time interval between t1 and t2 is m × Δ t; reading input and output historical data in a time interval [ t1, t2] and retraining the model; if missing data appears in the time interval, the time interval is moved forward until the requirements of m groups of model training sample data are met.
The foregoing is an exemplary embodiment of the present disclosure, but it should be noted that various changes and modifications could be made herein without departing from the scope of the present disclosure as defined by the appended claims. The functions, steps and/or actions of the method claims in accordance with the disclosed embodiments described herein need not be performed in any particular order. Furthermore, although elements of the disclosed embodiments of the invention may be described or claimed in the singular, the plural is contemplated unless limitation to the singular is explicitly stated.
It should be understood that, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly supports the exception. It should also be understood that "and/or" as used herein is meant to include any and all possible combinations of one or more of the associated listed items.
The numbers of the embodiments disclosed in the embodiments of the present invention are merely for description, and do not represent the merits of the embodiments.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, of embodiments of the invention is limited to these examples; within the idea of an embodiment of the invention, also technical features in the above embodiment or in different embodiments may be combined and there are many other variations of the different aspects of the embodiments of the invention as described above, which are not provided in detail for the sake of brevity. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present invention are intended to be included within the scope of the embodiments of the present invention.
Claims (10)
1. A heating demand load prediction method, characterized by comprising the steps of:
step 1, determining an input variable and an output variable of a demand load prediction model based on relevant factors of a heating system load;
step 2, butting operation data, cleaning missing and abnormal data, correcting sample load based on correction factors under the condition of no room temperature data, and determining load prediction model training data;
step 3, training and generating a heat supply system demand load prediction model based on a deep long-term and short-term memory network algorithm;
step 4, combining the correction factors to improve the accuracy of the prediction model;
and 5, detecting the model precision on line, and retraining to generate a demand load prediction model according to rules.
2. The heating demand load prediction method according to claim 1, wherein the input variables of the demand load prediction model in step 1 are various factors affecting a heating load, the output data of the output variables are demand load data at a time to be predicted, and the input data and the output data form a sample set for load prediction model training.
3. A heating demand load prediction method according to claim 1, wherein the docking operation data in step 2 is input data in real time, the obtained input data is stored in a load prediction model database, an input data set of a load prediction model is established according to the input data, and a certain number of sets of the input data set form an input data set matrix.
4. The method according to claim 1, wherein in step 2, performing a data cleaning process on missing and abnormal data includes setting a missing proportion threshold for determining whether at least one piece of data is selected as sample data, setting an abnormal data threshold for determining abnormal data, and performing processing according to data missing proportion classification: when the proportion of the number of the missing data of the current data record to the total number of the data is larger than the missing proportion threshold value, discarding the group of data, and automatically complementing the number of the sample data;
in the second type, the ratio of the number of the missing data of the current data record to the total number of the data is less than or equal to the missing ratio threshold value, and the data is subjected to completion processing;
when the current data abnormity judgment exceeds the abnormal data threshold value, discarding the abnormal data and regarding the abnormal data as missing data, supplementing the data according to the first-class or second-class method, and completing replacement processing on the abnormal data;
and in the fourth class, load correction factors are added to the data without the room temperature measuring points, the load of the sample is corrected by combining the matched correction factors, and the corrected data are used as the input data and the output data of the sample.
5. A heating demand load prediction method according to claim 4, wherein the second type of processing method includes:
when the missing characteristic is a discontinuous value, the completion method adopts a front and back numerical arithmetic mean method;
when the missing features are continuous values, the completion method adopts an interpolation method.
6. A heating demand load prediction method according to claim 5, wherein the input data and the output data are normalized, and the normalized data are used as sample data of a load prediction training model.
7. A heating demand load prediction method according to claim 4, wherein in the step of constructing a demand load prediction model in step 3, the deep long-short term memory network structure is composed of an input layer, a hidden layer, and an output layer, the input layer, the hidden layer, and the output layer correspond to respective coefficients, the relationships among the input layer, the hidden layer, and the output layer are established according to the coefficients, and the coefficients are optimized through an algorithm to form the demand load prediction model.
8. A heating demand load prediction method according to claim 1, wherein a deep long-short term memory network is formed by linking long-short term memory units, each of the long-short term memory units comprising:
an input gate, which controls whether the information of the long-term and short-term memory unit needs to be updated;
a forgetting door, wherein the forgetting door controls whether the information of the long-short term memory unit needs to be deleted;
and the output gate controls whether the information of the long-short term memory unit needs to be reflected in the activation vector or not, wherein the used activation function adopts a sigmoid function.
9. A heating demand load prediction method according to claim 1, wherein in the step 3, an iterative convergence coefficient is set with a mean square error as a loss function of a demand load prediction model training, and when the loss function is smaller than the iterative convergence coefficient, the demand load prediction model training is completed and a demand load prediction model for predicting a demand load is obtained.
10. A heating demand load prediction method according to claim 9, wherein in the step 5, when the accuracy of the demand load prediction model calculation value and the actual measurement value deviates by more than the iterative convergence coefficient, the model is retrained, and the step of retraining the model includes:
step 5.1, based on the demand load prediction model, accessing input data at the current moment, and calculating a predicted load value of the demand load prediction model at the current moment;
step 5.2, capturing the operation data of each site at the current moment, and calculating the consumption load of the site at the current moment;
step 5.3, calculating the deviation precision of the demand load prediction model;
step 5.4, when the deviation precision is less than or equal to the iterative convergence coefficient, the model demand load prediction is not corrected;
step 5.5, when the deviation precision is larger than the iterative convergence coefficient, reading input and output historical data in a time interval from a time node of the current moment, and retraining a model, wherein the time interval is an interval of a time interval with the same group number as that of input data sets in the input data set matrix; and when missing data appears in the time interval, the interval is moved forward until the number of the input data sets is unchanged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911021471.XA CN112712189B (en) | 2019-10-25 | 2019-10-25 | Heat supply demand load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911021471.XA CN112712189B (en) | 2019-10-25 | 2019-10-25 | Heat supply demand load prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112712189A true CN112712189A (en) | 2021-04-27 |
| CN112712189B CN112712189B (en) | 2024-06-14 |
Family
ID=75541391
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911021471.XA Active CN112712189B (en) | 2019-10-25 | 2019-10-25 | Heat supply demand load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112712189B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222231A (en) * | 2021-04-30 | 2021-08-06 | 洛阳双瑞精铸钛业有限公司 | Wisdom heating system based on internet of things |
| CN113623719A (en) * | 2021-06-23 | 2021-11-09 | 国家电投集团东北电力有限公司大连开热分公司 | Heat exchange station prediction control method based on effective room temperature detection |
| CN113887797A (en) * | 2021-09-29 | 2022-01-04 | 南京用能互联网研究院有限公司 | Building electric heating load prediction model establishing method, device and equipment |
| CN114841410A (en) * | 2022-03-31 | 2022-08-02 | 大连海心信息工程有限公司 | Heat exchange station load prediction method and system based on combination strategy |
| CN114857656A (en) * | 2022-05-25 | 2022-08-05 | 京东城市(北京)数字科技有限公司 | Heat supply load prediction method and device |
| CN114881382A (en) * | 2022-07-13 | 2022-08-09 | 天津能源物联网科技股份有限公司 | Optimization and adjustment method for autonomous prediction of heat supply demand load |
| CN115793558A (en) * | 2022-11-25 | 2023-03-14 | 上海时链节能科技有限公司 | Heating power station secondary pipe network intelligence control system |
| CN117574290A (en) * | 2023-11-02 | 2024-02-20 | 天津大学 | Thermodynamic data anomaly detection and repair method based on mechanism and data collaborative driving |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20130017368A (en) * | 2011-08-10 | 2013-02-20 | 아주대학교산학협력단 | Heating supply determination method for district heating network |
| CN104515194A (en) * | 2013-09-26 | 2015-04-15 | 珠海格力电器股份有限公司 | control method and device for heating system |
| WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
| CN106813293A (en) * | 2017-03-25 | 2017-06-09 | 余宝法 | Central heating system whole-network heat balance control method |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| 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 |
| CN109297086A (en) * | 2018-09-10 | 2019-02-01 | 常州英集动力科技有限公司 | Thermal substation load rolling forecast and adaptive corrective method and system at times |
| CN109670629A (en) * | 2018-11-16 | 2019-04-23 | 浙江蓝卓工业互联网信息技术有限公司 | Coal-burning boiler thermal efficiency forecast method based on shot and long term Memory Neural Networks |
| CN110298501A (en) * | 2019-06-21 | 2019-10-01 | 河海大学常州校区 | Electric load prediction technique based on long Memory Neural Networks in short-term |
| CN110334726A (en) * | 2019-04-24 | 2019-10-15 | 华北电力大学 | A kind of identification of the electric load abnormal data based on Density Clustering and LSTM and restorative procedure |
-
2019
- 2019-10-25 CN CN201911021471.XA patent/CN112712189B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20130017368A (en) * | 2011-08-10 | 2013-02-20 | 아주대학교산학협력단 | Heating supply determination method for district heating network |
| CN104515194A (en) * | 2013-09-26 | 2015-04-15 | 珠海格力电器股份有限公司 | control method and device for heating system |
| WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
| CN106813293A (en) * | 2017-03-25 | 2017-06-09 | 余宝法 | Central heating system whole-network heat balance control method |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| 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 |
| CN109297086A (en) * | 2018-09-10 | 2019-02-01 | 常州英集动力科技有限公司 | Thermal substation load rolling forecast and adaptive corrective method and system at times |
| CN109670629A (en) * | 2018-11-16 | 2019-04-23 | 浙江蓝卓工业互联网信息技术有限公司 | Coal-burning boiler thermal efficiency forecast method based on shot and long term Memory Neural Networks |
| CN110334726A (en) * | 2019-04-24 | 2019-10-15 | 华北电力大学 | A kind of identification of the electric load abnormal data based on Density Clustering and LSTM and restorative procedure |
| CN110298501A (en) * | 2019-06-21 | 2019-10-01 | 河海大学常州校区 | Electric load prediction technique based on long Memory Neural Networks in short-term |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222231A (en) * | 2021-04-30 | 2021-08-06 | 洛阳双瑞精铸钛业有限公司 | Wisdom heating system based on internet of things |
| CN113623719B (en) * | 2021-06-23 | 2022-08-19 | 国家电投集团东北电力有限公司大连开热分公司 | Heat exchange station prediction control method based on effective room temperature detection |
| CN113623719A (en) * | 2021-06-23 | 2021-11-09 | 国家电投集团东北电力有限公司大连开热分公司 | Heat exchange station prediction control method based on effective room temperature detection |
| CN113887797A (en) * | 2021-09-29 | 2022-01-04 | 南京用能互联网研究院有限公司 | Building electric heating load prediction model establishing method, device and equipment |
| CN114841410A (en) * | 2022-03-31 | 2022-08-02 | 大连海心信息工程有限公司 | Heat exchange station load prediction method and system based on combination strategy |
| CN114857656A (en) * | 2022-05-25 | 2022-08-05 | 京东城市(北京)数字科技有限公司 | Heat supply load prediction method and device |
| CN114857656B (en) * | 2022-05-25 | 2024-07-16 | 京东城市(北京)数字科技有限公司 | Heat supply load prediction method and device |
| CN114881382A (en) * | 2022-07-13 | 2022-08-09 | 天津能源物联网科技股份有限公司 | Optimization and adjustment method for autonomous prediction of heat supply demand load |
| CN114881382B (en) * | 2022-07-13 | 2022-10-11 | 天津能源物联网科技股份有限公司 | Optimized adjusting method for autonomous prediction of heat supply demand load |
| CN115793558A (en) * | 2022-11-25 | 2023-03-14 | 上海时链节能科技有限公司 | Heating power station secondary pipe network intelligence control system |
| CN115793558B (en) * | 2022-11-25 | 2023-10-03 | 上海时链节能科技有限公司 | Intelligent control system for secondary pipe network of heating power station |
| CN117574290A (en) * | 2023-11-02 | 2024-02-20 | 天津大学 | Thermodynamic data anomaly detection and repair method based on mechanism and data collaborative driving |
| CN117574290B (en) * | 2023-11-02 | 2024-05-10 | 天津大学 | Thermodynamic data anomaly detection and repair method based on mechanism and data collaborative driving |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112712189B (en) | 2024-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112712189A (en) | Heat supply demand load prediction method | |
| CN109270842B (en) | Bayesian network-based regional heat supply model prediction control system and method | |
| CN113112077B (en) | HVAC control system based on multi-step prediction deep reinforcement learning algorithm | |
| CN111365828A (en) | Model prediction control method for realizing energy-saving temperature control of data center by combining machine learning | |
| CN107909220A (en) | Electric heating load prediction method | |
| CN114484557B (en) | Building group heat supply load regulation and control method based on target energy consumption management and control | |
| CN110504676B (en) | Power distribution network state estimation method based on APSO-BP | |
| CN104239982A (en) | Method for predicting energy consumption of buildings during holidays and festivals on basis of time series and neural networks | |
| CN110796307A (en) | Distributed load prediction method and system for comprehensive energy system | |
| CN104102211B (en) | Scheduling method, server and system of combined heat and power generation system of fire coal thermal power plant | |
| CN111598225B (en) | Air conditioner cold load prediction method based on self-adaptive deep confidence network | |
| CN112365056A (en) | Electrical load joint prediction method and device, terminal and storage medium | |
| CN116167531A (en) | Photovoltaic power generation prediction method based on digital twin | |
| CN114444660A (en) | Short-term power load prediction method based on attention mechanism and LSTM | |
| CN111340305A (en) | Building operation energy consumption prediction method | |
| CN114757330A (en) | Urban instantaneous water consumption prediction method based on LSTM | |
| CN113919545A (en) | Photovoltaic power generation power prediction method and system with integration of multiple data models | |
| CN112541633A (en) | Ultrahigh-precision photovoltaic power generation power prediction method and system based on neural network | |
| Wang et al. | Multi-criteria comprehensive study on predictive algorithm of heating energy consumption of district heating station based on timeseries processing | |
| CN111582588B (en) | Building energy consumption prediction method based on triple convolution fusion GRU | |
| CN114415503B (en) | Temperature big data internet of things detection and intelligent control system | |
| Yang et al. | Prediction research on the energy consumption of public building based on mlr-bp neural network | |
| CN113609768A (en) | Bidirectional LSTM network-based distribution room line loss rate calculation method | |
| Qin et al. | Exploring new building energy saving control strategy application under the energy internet of things | |
| CN111626465A (en) | New energy power short-term interval prediction method and system |
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 |