CN110400018B - Operation control method, system and device for coal-fired power plant pulverizing system - Google Patents
Operation control method, system and device for coal-fired power plant pulverizing system Download PDFInfo
- Publication number
- CN110400018B CN110400018B CN201910690100.4A CN201910690100A CN110400018B CN 110400018 B CN110400018 B CN 110400018B CN 201910690100 A CN201910690100 A CN 201910690100A CN 110400018 B CN110400018 B CN 110400018B
- Authority
- CN
- China
- Prior art keywords
- coal
- pulverizing system
- data
- rbm
- training
- 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
- 238000010298 pulverizing process Methods 0.000 title claims abstract description 88
- 238000000034 method Methods 0.000 title claims abstract description 56
- 239000003245 coal Substances 0.000 claims abstract description 53
- 238000005265 energy consumption Methods 0.000 claims abstract description 45
- 239000000843 powder Substances 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 22
- 230000005611 electricity Effects 0.000 claims abstract description 16
- 238000011217 control strategy Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 54
- 230000000007 visual effect Effects 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 230000003044 adaptive effect Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims description 5
- 238000012544 monitoring process Methods 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 4
- 238000010248 power generation Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000010977 unit operation Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003303 reheating Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- 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
- 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
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- General Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Feedback Control In General (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an operation control method, an operation control system and an operation control device for a coal-fired power plant coal pulverizing system, wherein the operation method specifically comprises the following steps: acquiring real-time data of a coal-fired power plant pulverizing system; inputting real-time data into a powder process system energy consumption prediction model to respectively obtain prediction data of unit consumption and electricity consumption of the powder process system; and setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system. Compared with the prior art, the method has the advantages that the energy consumption conditions of the coal mill and the primary air motor are predicted and analyzed through the energy consumption prediction model of the coal pulverizing system, the optimal operation control strategy of the coal pulverizing system is selected according to the prediction result, and the coal pulverizing system is controlled, so that the operation mode of the coal pulverizing system is more economical and reasonable, the energy consumption index of the coal pulverizing system is obviously optimized, and the efficiency of the system is improved.
Description
Technical Field
The invention relates to the technical field of information control, in particular to an operation control method, an operation control system and an operation control device for a coal-fired power plant coal pulverizing system.
Background
Along with the rapid development and increasingly strict environmental protection requirements of the new energy power generation industry, the traditional coal-fired thermal power generation faces serious external environmental challenges, and thermal power generation enterprises need to deeply mine the potential of the two aspects of unit optimizing operation and equipment technical transformation, so as to search for technical means for improving the unit efficiency under the full-load working condition. In coal-fired power plants, the main power supply equipment includes boilers, turbines and auxiliary machinery. The pulverizing system is a key auxiliary system of the thermal power unit, and the power consumption accounts for 5-10% of the whole thermal power plant. By changing the operation decision of the pulverizing system, the energy consumption index of the pulverizing system can be obviously optimized, and the efficiency of the system can be improved.
At present, most of the optimal operation schemes of the pulverizing systems of the thermal power plants are only optimized for single equipment. For example, some energy-saving regulation technologies aiming at primary air are proposed, and the power consumption of a primary fan is reduced by adjusting the running air quantity and correcting the primary air-coal ratio, so that the running mode of a powder making system is more economical and reasonable. The energy consumption of the powder making system can be reduced by the operation optimization of the independent equipment, but the influence is smaller, when the powder making system works, equipment such as a coal feeder, a coal mill, a primary fan and the like are required to operate simultaneously, the energy consumption of one of the independent equipment is not necessarily optimal for reducing the energy consumption effect of the whole group of equipment, and an optimal equipment operation combination mode is required to be found. In addition, in the research of the optimized operation parameters of the thermal power plant, the defect of hiding in data can be effectively overcome through data discretization operation, and in order to enable the model structure to be more stable, the relationship between variables can be deeply mined through the discretization operation. At present, the discretization commonly used in a thermal power plant is mainly an equal-width method and an equal-frequency method, which are used for strictly classifying data into certain categories, but do not strictly classify the parameter values of actual operation of the power plant.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an operation control method, an operation control system and an operation control device for a coal-fired power plant coal pulverizing system.
The aim of the invention can be achieved by the following technical scheme:
the operation control method for the coal-fired power plant pulverizing system specifically comprises the following steps:
s1, acquiring real-time data of a coal-fired power plant pulverizing system;
s2, inputting real-time data into a powder process system energy consumption prediction model, and respectively obtaining prediction data of unit consumption and electricity consumption of the powder process system;
s3, setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system.
Further, the pulverizing system energy consumption prediction model is a DBN (Deep Belief Network, deep learning deep belief network) model composed of an RBM (Restricted Boltzmann Machine, boltzmann machine limited) stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM (Fuzzy C-Means) method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence.
Further, the expression of the adaptive function L (c) of the adaptive FCM method in the step A1 is:
wherein,,
represents the overall sample center vector, c represents the number of categories, n represents the sample x= { X 1 ,x 2 ,…,x n Variable number of }，x j Representing observed data, v i Represents an initial matrix of the ith class, u ij Representing a membership matrix of the data, and m represents a fuzzy weighting index.
Further, the state energy function of the RBM stacking network in the step A2 is specifically:
wherein v and h respectively represent a visual layer and a hidden layer, n represents the number of nodes of the v layer, m represents the number of nodes of the h layer, and W ij The weight values from the visual layer node i to the hidden layer node j are represented, θ= { W, a, b } represents the set of all parameters of the system, a represents the set of visual layer biases, and b represents the set of hidden layer biases.
Based on the determined parameters, obtaining a joint probability distribution of the RBM stacked network as follows:
where Z (θ) is a partitioning function used to distribute the joint probability over the [0,1] interval, and e is a natural constant 2.71828.
Further, the gradient update expression of the training parameters of the RBM stacking network in the step A3 is as follows:
wherein DeltaW is ij The weight value from the visual layer node i to the hidden layer node j after gradient update is represented, epsilon represents the learning rate, v i Meaning visual layer inode, Δa i Representing the gradient updated visual layer bias, Δb j Represents hidden layer bias after gradient update, h j Represents hidden j node and k represents sampling times.
An operation control system for a coal fired power plant pulverizing system, comprising:
the monitoring module is used for acquiring real-time data of the coal-fired power plant pulverizing system;
the prediction module is used for inputting real-time data into the powder process system energy consumption prediction model to respectively obtain prediction data of unit consumption and power consumption of the powder process system;
and the control module is used for setting a control strategy of the pulverizing system according to the prediction data and controlling the operation of the pulverizing system.
Further, the pulverizing system energy consumption prediction model is a DBN model formed by an RBM stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence.
An operation control device for a coal-fired power plant coal pulverizing system, the device comprising a processor and a memory, wherein the processor calls a program in the memory for realizing the following steps:
s1, acquiring real-time data of a coal-fired power plant pulverizing system;
s2, inputting real-time data into a powder process system energy consumption prediction model, and respectively obtaining prediction data of unit consumption and electricity consumption of the powder process system;
s3, setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system.
Compared with the prior art, the invention has the following advantages:
1. according to the invention, the energy consumption conditions of the coal mill and the primary air motor are predicted and analyzed through the energy consumption prediction model of the coal pulverizing system, and the optimal operation control strategy of the coal pulverizing system is selected according to the prediction result, so that the coal pulverizing system is controlled, the operation mode of the coal pulverizing system is more economic and reasonable, the energy consumption index of the coal pulverizing system is obviously optimized, and the efficiency of the system is improved.
2. The energy consumption prediction model of the pulverizing system adopts the working condition aggregation of the self-adaptive FCM similar condition to be more accurate than the traditional hard partition classification, and the algorithm can autonomously and quickly determine the clustering number and the clustering center, so that the serious defect artificially given by the traditional FCM clustering parameters is avoided, an operator needs enough experience for artificially giving the clustering parameters to select more proper parameters, and the calculation efficiency and accuracy of the artificial given parameters to the algorithm are high, so that the reliable clustering type is not facilitated to be obtained.
3. According to the invention, the clustered sample data is classified and identified to the original data according to the data characteristics by a DBN model formed by an RBM stacking network and a BP network, and the similar working conditions of the pulverizing system are matched according to the load conditions, so that the optimal operation control strategy is searched. The method can provide an optimized control method for the operation modes of a plurality of coal mills, and is beneficial to reducing the actual operation economic cost of the power plant and reducing the times of the later maintenance schedule.
Drawings
FIG. 1 is a schematic overall flow chart of the present invention;
FIG. 2 is a flow chart of the DBN model structure of the present invention;
FIG. 3 is a schematic diagram of a unit consumption prediction result of a coal mill A in an embodiment;
fig. 4 is a schematic diagram of a unit consumption prediction result of the primary air fan a in the embodiment;
FIG. 5 is a schematic diagram of the unit consumption training error of coal mill A in an embodiment;
fig. 6 is a schematic diagram of a unit consumption training error of the primary air fan a in the embodiment;
FIG. 7 is a graph showing relative error in prediction of unit consumption of coal pulverizer A in an embodiment;
FIG. 8 is a schematic diagram of a relative error of unit consumption prediction of the primary air blower A in the embodiment;
FIG. 9 is a schematic diagram of a predicted result of electricity consumption of coal mill A in an embodiment;
fig. 10 is a schematic diagram of a primary fan a electricity consumption prediction result in the embodiment;
FIG. 11 is a schematic diagram of predicted training errors of power consumption of coal mill A in an embodiment;
fig. 12 is a schematic diagram of a primary fan a power consumption prediction training error in an embodiment;
FIG. 13 is a graph showing relative error in prediction of power consumption of coal pulverizer A according to an embodiment;
fig. 14 is a schematic diagram of a relative error of primary fan a power consumption prediction in the embodiment.
Detailed Description
The invention will now be described in detail with reference to the drawings and specific examples. The present embodiment is implemented on the premise of the technical scheme of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection scope of the present invention is not limited to the following examples.
The embodiment provides an operation control method for a coal-fired power plant coal pulverizing system, which is based on a deep belief network coal pulverizing system prediction model of self-adaptive FCM clustering, wherein a main unit operation parameter (unit power generation amount and total air quantity) and a main unit operation parameter (coal supply amount, each coal mill and primary air machine power consumption) of the coal pulverizing system are clustered by utilizing a self-adaptive FCM algorithm according to historical data and unit load as a clustering standard, and training samples are determined according to a similar working condition principle. And (3) inputting clustered sample data serving as an original data sample of the DBN model, performing positive sequence training learning through the RBM stacking network, and establishing a pulverizing system energy consumption prediction model through reverse fine tuning of the BP neural network, so as to give a pulverizing system operation decision based on the model.
The specific flow chart is shown in fig. 1:
s1, acquiring real-time data of a coal-fired power plant pulverizing system on line;
s2, inputting real-time data into a powder process system energy consumption prediction model, and respectively obtaining prediction data of unit consumption and electricity consumption of the powder process system;
and S3, setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system.
The energy consumption prediction model of the pulverizing system is a DBN model consisting of an RBM stacking network and a BP neural network, and the training method is as follows:
step A1, acquiring historical data of a coal-fired power plant coal pulverizing system, wherein the historical data comprise part of main unit operation parameters and part of relevant operation parameters of the coal pulverizing system, performing discretization pretreatment on each attribute parameter in a data set by adopting a self-adaptive FCM method according to unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
a2, classifying and identifying the clustered training samples through a DBN model according to the data characteristics; and through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence, and the model training is completed.
The self-adaptive network training is that the system is independently operated according to historical data. The self-adaptive system adapts to external environment change through autonomous learning capability, and obtains the most reasonable weighting index m and clustering number C through repeated cyclic training. Adaptive function L (C) of clustering number C of adaptive FCM algorithm for data clustering and total sample center vector
The mathematical formula of (a) is:
wherein,,
represents the overall sample center vector, c represents the number of categories, n represents the sample x= { X 1 ,x 2 ,…,x n Number of variables, x j Representing observed data, v i Represents an initial matrix of the ith class, u ij Representing membership of dataThe degree matrix, m, represents the fuzzy weighting index.
In the expression, the numerator represents the distance between different classes, and the denominator represents the distance from the data in the class to the center of the class, so that the larger L (C), the more reasonable and accurate the clustering, and the obtained maximum value is the optimal solution of C. The adaptive FCM algorithm is based on local finding point minima, so it is only necessary to compare L (c) at local data.
In the embodiment, a pulverizing system of a 350MW unit of a certain power plant is taken as a research object, the certain power plant is provided with two primary reheating, subcritical, steam extraction and condensing type 350MW steam extraction and heat supply generating sets, 5 coal mills with model number MPS212 are configured, and a primary fan is arranged at the downstream of an air preheater, wherein 1 of the primary fans is designed for standby.
The study data adopts historical operation data of 2018, 10 months, 1 day to 12 months and 15 days, the point taking interval is 5 minutes, and 17 parameter data are taken as training samples respectively: unit load, coal mill A, B, C, D, E power usage, coal feed A, B, C, D, E power usage and primary fan A, B, C, D, E power usage.
Firstly, selecting 216-350 MW load working condition cluster data samples under the constraint of a specific coal quality parameter and environmental conditions to carry out self-adaptive FCM test experiments. Each attribute parameter in the dataset is subjected to discretization pretreatment by adopting a self-adaptive FCM algorithm, and the discretization result of the original data of the unit load part is shown in table 1.
Table 1 machine set load discretization membership table (part)
From the experimental results in table 1, it can be determined that in the 220-350 MW load condition cluster, the load of the adaptive FCM algorithm cluster unit is mainly distributed around 250MW, 293MW, 310MW, 331MW and 346 MW. Therefore, the group load working conditions are clustered into 5 types, and working condition 1: (216,286), condition 2 (286,300), condition 3 (300,321), condition 4 (321,342), and condition 5 (342,350).
As shown in fig. 2, the DBN (deep belief network) optimizes weights and thresholds between connection layers according to a sequence through a non-monitored greedy learning algorithm, and the first part of the network is a layer-by-layer pre-training phase, in which the training process is a bottom-up non-supervised learning, and input data can be directly mapped to output, which is why the DBN model can learn nonlinear complex functions. The lowest level RBM of the network is first trained using unlabeled data samples and learned from small to small layer by layer. Adjacent layers are connected in pairs, and the output of the lower layer is the input of the upper layer and is transmitted to the highest layer step by step. After the RBM stacking process, only a few characterization information can be extracted, the data cannot be directly categorized, and thus the resulting data must be sent to the top Softmax supervised classifier.
The state energy function of the RBM is specifically:
wherein v and h respectively represent a visual layer and a hidden layer, n represents the number of nodes of the v layer, m represents the number of nodes of the h layer, and W ij The weight values from the visual layer node i to the hidden layer node j are represented, θ= { W, a, b } represents the set of all parameters of the system, a represents the set of visual layer biases, and b represents the set of hidden layer biases.
Based on the determined parameters, obtaining a joint probability distribution of the RBM stacked network as follows:
where Z (θ) is a partitioning function used to distribute the joint probability over the [0,1] interval, and e is a natural constant 2.71828.
The gradient update expression of the training parameters of the RBM stacking network is as follows:
wherein DeltaW is ij The weight value from the visual layer node i to the hidden layer node j after gradient update is represented, epsilon represents the learning rate, v i Meaning visual layer inode, Δa i Representing the gradient updated visual layer bias, Δb j Represents hidden layer bias after gradient update, h j Represents hidden j node and k represents sampling times.
In the second part of the DBN model, the weight of all hidden layers is regarded as a whole, the weight correction is carried out on the tagged data by using a gradient descent method, and the BP algorithm is used for fine-tuning the whole network so as to improve the reliability of the DBN network classification prediction model.
And determining the DBN topological structure and parameters. In order to ensure the accuracy and the comprehensiveness of a powder process system prediction model, two DBN models are designed through the respective prediction of unit consumption and electricity consumption of the powder process system. The hidden layer number of the DBN algorithm is the RBM stacking layer number, and represents the depth of the DBN.
The number of hidden layer nodes is mainly determined by the model effect in the experimental process, and the initial hidden layer nodes can be set according to an empirical formula and then increased or decreased according to the model effect. According to the training structure of the DBN model network, the learning rate of the network needs to be set in the RBM pre-training stage, and then the iterative training times need to be set. The RBM learning rate is generally set within the [0.1,1] interval. The iterative training frequency setting has no specific rule following, so that the network is not over-fitted or under-fitted, the parameters are not required to be excessively large or excessively small when the parameters are set, and specific numerical values are required to be determined according to the actual training effect adjustment.
And predicting the unit consumption and the electricity consumption of the equipment according to the established DBN model, randomly selecting any working condition of 12 months, 16 days and 31 days in 2018 as the working condition to be predicted, taking sample data as a training set, and predicting the unit consumption and the electricity consumption of the coal mill and the unit consumption and the electricity consumption of the primary fan. In order to reflect the accuracy of the prediction model, the training error, the absolute error and the relative error of the model are tracked, and the accuracy and the reliability of the model can be intuitively and obviously observed.
Because of the large number of objects, the energy consumption prediction situation of the coal mill A and the primary air fan A is analyzed in detail. Fig. 3 and fig. 4 show the unit consumption prediction results of the group a devices. Fig. 5 and 6 are graphs of the unit consumption prediction training error results of the group a devices. Fig. 7 and 8 are graphs of the relative error results of the unit consumption prediction of the group a devices. Fig. 9 and 10 are diagrams of the electricity consumption prediction results of group a devices. Fig. 11 and 12 are graphs of the power consumption prediction training error results of the group a devices. Fig. 13 and 14 are graphs of the relative error of the electricity consumption prediction of group a devices.
It can be seen from fig. 3, fig. 4, fig. 9 and fig. 10 that the DBN energy consumption prediction model can track the original data curve in real time, and the prediction of the unit consumption output and the power consumption output is very stable, and the fluctuation range is small. In the unit consumption prediction, the minimum error can be achieved by observing that the iteration of the coal mill A is performed 10 times in fig. 5, and the iteration number of the primary air fan A in fig. 6 is slightly more, but the final error still can reach the requirement. Fig. 7 and 8 show that the prediction error of group a devices is around ±3%. In the electricity consumption prediction, it can be observed from fig. 11 and 12 that the minimum error can be achieved by iterating the coal mill a and the primary fan a less than 10 times, and fig. 13 and 14 show that the prediction error of the group a equipment is 10 -6 On the order of magnitude. In view of the above, the energy consumption prediction model of the pulverizing system has good reliability and higher accuracy, so that a more accurate and reliable operation control strategy can be obtained to control the pulverizing system.
According to the energy consumption prediction model of the pulverizing system, average energy consumption conditions of the coal mill and the primary air blower under different working conditions are obtained, as shown in tables 2 to 5. In order to achieve the highest economic benefit of the power plant and reduce energy consumption and pollution, the input actual working condition can provide a believed operation control scheme for the pulverizing system according to the predicted value of the similar working condition.
Table 2 predicted values (partial) for unit consumption of coal mills under different working conditions
Table 3 predicted values (portions) of power consumption of coal mills under different conditions
Table 4 predicted values (partial) of unit consumption of primary fans under different working conditions
Table 5 predicted values (portions) of power consumption of primary fans under different working conditions
The operation control strategy of the pulverizing system under different working conditions is obtained through experimental statistics and is shown in table 6. Experiments find that E equipment of a certain power plant is a spare equipment group, and only in a full load condition, the E equipment is rarely started. While A, B, C has little energy consumption, the energy consumption of the D device is slightly higher than that of other D devices, and the selection condition of the D device should be paid attention to when the operation control strategy is selected.
TABLE 6 operational decisions for pulverizing systems
According to the method, according to historical data of a certain 350MW power plant, energy consumption models of 5 coal mills and 5 primary fans are obtained through fitting, on the basis, similar working conditions are clustered by using a self-adaptive FCM algorithm by taking unit load as a standard, a coal pulverizing system energy consumption prediction model based on a deep belief network is established for the similar working conditions, and an example result shows that:
the adaptive FCM similar condition working condition aggregation is more accurate than the traditional hard partition classification, and the algorithm can autonomously and quickly determine the clustering number and the clustering center. Serious drawbacks artificially given by conventional FCM cluster parameters are also avoided. In the DBN power consumption prediction model, the average relative error of unit consumption prediction of the five coal mills is 0.27%, and the average relative error of power consumption prediction is 0.19%. The average relative error of the unit consumption predictions of the 5 primary fans is 0.26%, the average relative error of the electricity consumption predictions is 0.21%, and the maximum relative error is 7.04%, so that the establishment of the DBN pulverizing system energy consumption prediction model provides a guarantee for operation decisions. And (5) matching similar working conditions of the powder making system according to the load conditions, and searching an optimal operation decision scheme. The method can provide a new thought for optimizing the operation modes of a plurality of coal mills, and has a certain reference significance for reducing the actual operation economic cost of the power plant and making a later maintenance plan.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention by one of ordinary skill in the art without undue burden. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.
Claims (4)
1. The operation control method for the coal-fired power plant pulverizing system is characterized by comprising the following steps of:
s1, acquiring real-time data of a coal-fired power plant pulverizing system;
s2, inputting real-time data into a powder process system energy consumption prediction model, and respectively obtaining prediction data of unit consumption and electricity consumption of the powder process system;
s3, setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system;
the pulverizing system energy consumption prediction model is a DBN model formed by an RBM stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence;
the expression of the adaptive function L (c) of the adaptive FCM method in step A1 is:
wherein,,
represents the overall sample center vector, c represents the number of categories, n represents the sample x= { X 1 ,x 2 ,…,x n Number of variables, x j Representing observed data, v i Represents an initial matrix of the ith class, u ij Representing a membership matrix of the data, and m represents a fuzzy weighting index;
the state energy function of the RBM stacking network in the step A2 specifically includes:
wherein v and h respectively represent a visual layer and a hidden layer, n represents the number of nodes of the v layer, m represents the number of nodes of the h layer, and W ij The weight value from the visual layer node i to the hidden layer node j is represented by h j Representing hidden j nodes, θ= { W, a, b } represents a set of all parameters of the system, a represents a set of visual layer biases, and b represents a set of hidden layer biases;
based on the determined parameters, obtaining a joint probability distribution of the RBM stacked network as follows:
wherein Z (θ) is a partitioning function for distributing the joint probability in the [0,1] interval, e is a natural constant 2.71828;
the gradient update expression of the training parameters of the RBM stacking network in the step A3 is as follows:
wherein DeltaW is ij The weight value from the visual layer node i to the hidden layer node j after gradient update is represented, epsilon represents the learning rate, v i Meaning visual layer inode, Δa i Representing the gradient updated visual layer bias, Δb j Represents hidden layer bias after gradient update, h j Represents hidden j node and k represents sampling times.
2. An operation control system for a coal fired power plant pulverizing system, comprising:
the monitoring module is used for acquiring real-time data of the coal-fired power plant pulverizing system;
the prediction module is used for inputting real-time data into the powder process system energy consumption prediction model to respectively obtain prediction data of unit consumption and power consumption of the powder process system;
the control module is used for setting a control strategy of the pulverizing system according to the prediction data and controlling the operation of the pulverizing system;
the pulverizing system energy consumption prediction model is a DBN model formed by an RBM stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence;
the expression of the adaptive function L (c) of the adaptive FCM method in step A1 is:
wherein,,
represents the overall sample center vector, c represents the number of categories, n represents the sample x= { X 1 ,x 2 ,…,x n Number of variables, x j Representing observed data, v i Represents an initial matrix of the ith class, u ij Representing a membership matrix of the data, and m represents a fuzzy weighting index;
the state energy function of the RBM stacking network in the step A2 specifically includes:
wherein v and h respectively represent a visual layer and a hidden layer, n represents the number of nodes of the v layer, m represents the number of nodes of the h layer, and W ij The weight value from the visual layer node i to the hidden layer node j is represented by h j Representing hidden j nodes, θ= { W, a, b } represents a set of all parameters of the system, a represents a set of visual layer biases, and b represents a set of hidden layer biases;
based on the determined parameters, obtaining a joint probability distribution of the RBM stacked network as follows:
wherein Z (θ) is a partitioning function for distributing the joint probability in the [0,1] interval, e is a natural constant 2.71828;
the gradient update expression of the training parameters of the RBM stacking network in the step A3 is as follows:
wherein DeltaW is ij The weight value from the visual layer node i to the hidden layer node j after gradient update is represented, epsilon represents the learning rate, v i Meaning visual layer inode, Δa i Representing the gradient updated visual layer bias, Δb j Represents hidden layer bias after gradient update, h j Represents hidden j node and k represents sampling times.
3. The operation control system for the coal-fired power plant coal pulverizing system according to claim 2, wherein the coal pulverizing system energy consumption prediction model is a DBN model composed of an RBM stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence.
4. An operation control device for a coal-fired power plant coal pulverizing system is characterized by comprising a processor and a memory, wherein the processor calls a program in the memory and is used for realizing the following steps:
s1, acquiring real-time data of a coal-fired power plant pulverizing system;
s2, inputting real-time data into a powder process system energy consumption prediction model, and respectively obtaining prediction data of unit consumption and electricity consumption of the powder process system;
s3, setting a control strategy of the pulverizing system according to the prediction data, and controlling the operation of the pulverizing system;
the pulverizing system energy consumption prediction model is a DBN model formed by an RBM stacking network and a BP neural network, and the training method is as follows:
A1. acquiring historical data of a coal-fired power plant pulverizing system, performing discretization pretreatment on each attribute parameter in a dataset by adopting a self-adaptive FCM method according to the unit load as a clustering standard, and determining a training sample according to a similar working condition principle;
A2. classifying and identifying the clustered training samples according to the data characteristics through a DBN model; through the positive sequence training learning of the RBM stacking network and the reverse fine tuning of the BP neural network, the weight and the threshold between adjacent connection layers in the energy consumption prediction model of the pulverizing system are optimized according to the sequence;
the expression of the adaptive function L (c) of the adaptive FCM method in step A1 is:
wherein,,
represents the overall sample center vector, c represents the number of categories, n represents the sample x= { X 1 ,x 2 ,…,x n Number of variables, x j Representing observed data, v i Represents an initial matrix of the ith class, u ij Representing a membership matrix of the data, and m represents a fuzzy weighting index;
the state energy function of the RBM stacking network in the step A2 specifically includes:
wherein v and h respectively represent a visual layer and a hidden layer, n represents the number of nodes of the v layer, m represents the number of nodes of the h layer, and W ij The weight value from the visual layer node i to the hidden layer node j is represented by h j Representing hidden j nodes, θ= { W, a, b } represents a set of all parameters of the system, a represents a set of visual layer biases, and b represents a set of hidden layer biases;
based on the determined parameters, obtaining a joint probability distribution of the RBM stacked network as follows:
wherein Z (θ) is a partitioning function for distributing the joint probability in the [0,1] interval, e is a natural constant 2.71828;
the gradient update expression of the training parameters of the RBM stacking network in the step A3 is as follows:
wherein DeltaW is ij The weight value from the visual layer node i to the hidden layer node j after gradient update is represented, epsilon represents the learning rate, v i Meaning visual layer inode, Δa i Representing the gradient updated visual layer bias, Δb j Represents hidden layer bias after gradient update, h j Represents hidden j node and k represents sampling times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910690100.4A CN110400018B (en) | 2019-07-29 | 2019-07-29 | Operation control method, system and device for coal-fired power plant pulverizing system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910690100.4A CN110400018B (en) | 2019-07-29 | 2019-07-29 | Operation control method, system and device for coal-fired power plant pulverizing system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110400018A CN110400018A (en) | 2019-11-01 |
| CN110400018B true CN110400018B (en) | 2023-06-27 |
Family
ID=68326380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910690100.4A Active CN110400018B (en) | 2019-07-29 | 2019-07-29 | Operation control method, system and device for coal-fired power plant pulverizing system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110400018B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112181003A (en) * | 2020-08-24 | 2021-01-05 | 中国神华能源股份有限公司国华电力分公司 | Method for controlling unit pressure and electronic equipment |
| CN112130538B (en) * | 2020-09-22 | 2022-02-01 | 京东城市(北京)数字科技有限公司 | Method, device, equipment and medium for control optimization and model training of coal mill |
| CN114332515B (en) * | 2021-11-01 | 2024-06-21 | 江苏方天电力技术有限公司 | On-line judging method for operation safety of coal mill based on intuitionistic fuzzy clustering |
| CN114371619B (en) * | 2021-12-17 | 2024-04-23 | 上海电力大学 | MGT-CCHP variable working condition dynamic energy efficiency optimization control method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102262147A (en) * | 2011-07-15 | 2011-11-30 | 华南理工大学 | Soft measurement method and system for effluent chemical oxygen demand (COD) of waste water treatment system |
| CN103729678A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院信息工程研究所 | Navy detection method and system based on improved DBN model |
| CN105741832A (en) * | 2016-01-27 | 2016-07-06 | 广东外语外贸大学 | Spoken language evaluation method based on deep learning and spoken language evaluation system |
| CN106198909A (en) * | 2016-06-30 | 2016-12-07 | 中南大学 | A kind of aquaculture water quality Forecasting Methodology based on degree of depth study |
| CN107203810A (en) * | 2017-05-22 | 2017-09-26 | 河海大学 | A kind of precipitation Forecasting Methodology based on depth network |
| CN107679671A (en) * | 2017-10-23 | 2018-02-09 | 燕山大学 | A kind of manufacture of cement power consumption Forecasting Methodology of implicit time series depth belief network |
| CN108320016A (en) * | 2018-03-08 | 2018-07-24 | 南京工业大学 | Short-term prediction method for building energy consumption |
-
2019
- 2019-07-29 CN CN201910690100.4A patent/CN110400018B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102262147A (en) * | 2011-07-15 | 2011-11-30 | 华南理工大学 | Soft measurement method and system for effluent chemical oxygen demand (COD) of waste water treatment system |
| CN103729678A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院信息工程研究所 | Navy detection method and system based on improved DBN model |
| CN105741832A (en) * | 2016-01-27 | 2016-07-06 | 广东外语外贸大学 | Spoken language evaluation method based on deep learning and spoken language evaluation system |
| CN106198909A (en) * | 2016-06-30 | 2016-12-07 | 中南大学 | A kind of aquaculture water quality Forecasting Methodology based on degree of depth study |
| CN107203810A (en) * | 2017-05-22 | 2017-09-26 | 河海大学 | A kind of precipitation Forecasting Methodology based on depth network |
| CN107679671A (en) * | 2017-10-23 | 2018-02-09 | 燕山大学 | A kind of manufacture of cement power consumption Forecasting Methodology of implicit time series depth belief network |
| CN108320016A (en) * | 2018-03-08 | 2018-07-24 | 南京工业大学 | Short-term prediction method for building energy consumption |
Non-Patent Citations (1)
| Title |
|---|
| "深度学习—受限玻尔兹曼机详解（RBM）";zsffuture;《https://blog.csdn.net/weixin_42398658/article/details/84279293》;20181123;全文第3页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110400018A (en) | 2019-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110400018B (en) | Operation control method, system and device for coal-fired power plant pulverizing system | |
| CN110929918B (en) | 10kV feeder fault prediction method based on CNN and LightGBM | |
| Zheng et al. | Multi-objective combustion optimization based on data-driven hybrid strategy | |
| CN105069521A (en) | Photovoltaic power plant output power prediction method based on weighted FCM clustering algorithm | |
| CN111461921B (en) | Load modeling typical user database updating method based on machine learning | |
| CN104810860A (en) | Power distribution method and power distribution device in wind power plant | |
| CN110717610A (en) | Wind power prediction method based on data mining | |
| CN106845012A (en) | A kind of blast furnace gas system model membership function based on multiple target Density Clustering determines method | |
| US20230229124A1 (en) | Operation control system and a control method for a gas-steam combined cycle generator unit | |
| CN111158239B (en) | Association rule algorithm and neural network-based pulverizing system performance optimization method | |
| CN117526432A (en) | Distribution network regulation and control system and method for source-load interaction | |
| CN111695666A (en) | Wind power ultra-short term conditional probability prediction method based on deep learning | |
| CN117454093A (en) | Intelligent power module management method and system | |
| CN115730228A (en) | Central air conditioner energy consumption analysis method based on BP neural network | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN113536694B (en) | Robust optimization operation method, system and device for comprehensive energy system and storage medium | |
| Aburiyana et al. | Direct net load forecasting using adaptive neuro fuzzy inference system | |
| CN111160386A (en) | Data-driven power island detection method based on active reactive power disturbance | |
| CN116565877A (en) | Automatic voltage partition control method based on spectral cluster analysis | |
| CN115526433A (en) | Power plant reheat flue gas baffle operation prediction method based on integrated hybrid model | |
| CN116154875A (en) | Photovoltaic power station active power optimization distribution method based on TCN and error function | |
| Zhang et al. | Research on intelligent evaluation and monitoring method for cost of transmission and transformation project | |
| KR102439311B1 (en) | Coordinated optimization method for optimiztion of wind farm using sparsified wake digraph and apparatus performing the same | |
| KR20220076869A (en) | Forecasting system and method for power consumption | |
| Zheng et al. | Combustion process modeling based on deep sparse least squares support vector regression |
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 |