CN111158239A - Association rule algorithm and neural network-based pulverizing system performance optimization method - Google Patents
Association rule algorithm and neural network-based pulverizing system performance optimization method Download PDFInfo
- Publication number
- CN111158239A CN111158239A CN202010021684.9A CN202010021684A CN111158239A CN 111158239 A CN111158239 A CN 111158239A CN 202010021684 A CN202010021684 A CN 202010021684A CN 111158239 A CN111158239 A CN 111158239A
- Authority
- CN
- China
- Prior art keywords
- data
- association rule
- historical
- neural network
- making system
- 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
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention provides a powder making system performance optimization method based on an association rule algorithm and a neural network, and belongs to the technical field of information control. The method comprises the following steps: s1, establishing a powder process system database according to the historical data of the powder process system; s2, screening stable operation parameters; s3, performing working condition clustering on the historical data in the pulverizing system database through a supervised self-organizing neural network; s4, discretizing data, constraining latitude and compressing samples; s5, mining the operation optimization parameters of each working condition cluster based on the improved association rule algorithm; s6, judging and classifying the operation condition; and S7, accumulating the new working condition data to a certain degree and mining again. The method has the advantages of saving computing resources, improving the operation efficiency and performance of the association rule algorithm, making data mining more targeted, reducing redundant items and improving the mining efficiency.
Description
Technical Field
The invention relates to a powder making system performance optimization method based on an association rule algorithm and a neural network, and belongs to the technical field of information control.
Background
With the rapid development of the new energy power generation industry and increasingly strict environmental requirements, the traditional coal-fired thermal power generation faces a severe external environmental challenge. In order to adapt to the new trend of the industry, thermal power generation enterprises start internal management, deeply dig potentials about the optimized operation of a unit and the technical improvement of equipment, and find technical means for improving the efficiency of the unit under the full-load working condition. In a coal-fired power plant, main power supply equipment comprises a boiler, a steam turbine and an auxiliary machine, wherein the relevant research on energy conservation and consumption reduction of the boiler and the steam turbine is mature, but the optimization operation research of the auxiliary machine is fresh. The powder process system is a key auxiliary system of the thermal power unit, and the power consumption accounts for 5 to 10 percent of the whole thermal power plant. If the operation of the powder process system is guided according to manual experience, the efficiency of the system is improved by optimizing the energy consumption of the powder process system, and the power consumption is reduced, and guidance opinions can be provided for power generation enterprises on the operation decision of the powder process system.
At present, most of the optimized operation research of the pulverizing system of the thermal power plant only aims at single equipment to be optimized. For example, by summarizing the CO precipitation rule of the coal mill and analyzing the precipitation temperature, an optimization scheme of the outlet temperature of the coal mill is provided. If the phenomenon that unit consumption of unit equipment is high is analyzed, main reasons are found and effective countermeasures are made, and although certain economic benefit is obtained, the consideration factor is single, and further research space still exists. Therefore, most of the existing research results aiming at the optimization of the powder process system mainly aim at the optimization of the parameters of a specific device of the powder process system, and no overall energy consumption index of the powder process system is observed.
Due to the fact that the energy consumption factors of the pulverizing system are numerous, the historical operation database of the power plant is high in dimensionality and large in capacity, the mining efficiency is low, and the algorithm occupies more computing resources. The method mainly comprises the steps of mining strong association rules of all working conditions of the thermal power generating unit to obtain the optimal value of a target parameter, wherein the thermal power big data mining method is provided aiming at the thermal power optimization big data mining condition in recent years, but the optimization effect of the parameter still has a liftable space; in the process of data mining of the power plant, the power supply coal consumption is taken into consideration as an optimized target value, but no clear method is provided for further optimizing each parameter index.
Disclosure of Invention
The invention is carried out to solve the problems, and aims to provide a powder process system performance optimization method based on an association rule algorithm and a neural network.
The invention provides a method for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network, which is characterized by comprising the following steps: step 1, establishing a powder process system database according to historical data of a powder process system, wherein the historical data comprises: the main operation parameters of the excavating equipment unit and the related operation parameters of the pulverizing system enter the step 2; step 2, judging whether the historical data in the powder making system database is stable operation data, if so, retaining the data in the powder making system database, otherwise, discarding the data in the powder making system database, and entering step 3; step 3, clustering the working conditions of the historical data in the pulverizing system database through a supervised self-organizing neural network, counting the attribute characteristics of each category, determining the specific division standard of the working condition clustering, and entering step 4; step 4, discretizing historical data in a pulverizing system database by adopting a quantile grouping method to obtain discretization historical data, determining the clean raw coal power generation amount as a mining target, processing the discretization historical data by adopting a dimension reduction compression mode based on target guidance, establishing a target function, and entering step 5; step 5, adopting an improved association rule algorithm based on target guidance to dig out an optimal value of a related operation parameter of the powder making system when a target function is optimal in a powder making system database, storing the optimal value in a historical knowledge base of the powder making system, and entering step 6; step 6, collecting real-time related operation parameters of the powder making system, comparing the real-time related operation parameters with data in a historical knowledge base of the powder making system, judging whether the real-time related operation parameters are superior to historical related operation parameters when the historical related operation parameters of the powder making system under similar working conditions are obtained through matching, guiding the powder making system to operate by using the historical related operation parameters when the real-time related operation parameters are not superior to the historical related operation parameters, otherwise guiding the powder making system to operate by using the real-time related operation parameters, and determining that a newly added working condition cluster occurs when the historical related operation parameters of the powder making system under similar working conditions cannot be obtained through matching, and entering; and 7, accumulating the operation data of the newly-added working condition cluster to a certain amount, excavating the optimal value of the related operation parameter of the powder making system under the newly-added working condition cluster when the objective function is optimal by adopting an improved association rule algorithm based on target guidance, and storing the optimal value in a historical knowledge base of the powder making system.
In the method for optimizing the performance of the pulverizing system based on the association rule algorithm and the neural network, the method can also have the following characteristics: the method for clustering the working conditions of the historical data in the powder process system database by the supervised self-organizing neural network comprises the following steps: step 3-1, normalizing the data, and entering step 3-2; step 3-2, initializing the supervised self-organizing neural network, and initializing the weight w of the input and competition layerijAnd the weights w of the competition layer and the output layerjkIs [0,1 ]]Internal random number, learning rate η (t) is set to initial value of (0)1) range value taking, determining a learning radius r and the total learning times, and entering the step 3-3; step 3-3, establishing a training sample X ═ X1,x2,…，xn) Calculating the Euclidean distance between the training sample X input and the neuron of the competition layer, wherein the calculation formula is as follows:
entering the step 3-4; 3-4, searching for a winning neuron, selecting a shortest distance competition layer neuron c as a superior neuron in the obtained Euclidean distance, and entering into the step 3-5; step 3-5, adjusting the weight w of the input and competition layerijThe calculation formula is as follows:
wij＝wij+η1(X-wij)，
adjusting weight w between competition layer and output layerjkThe calculation formula is as follows:
wik＝wjk+η2(Yk-wjk)
in the formula, η1、η2The learning rate is in linear inverse proportion correlation with the learning times, YkEntering step 3-6 for the category to which the sample belongs; and 3-6, judging whether the algorithm is finished, if not, returning to the step 3-2, otherwise, finishing the algorithm.
In the method for optimizing the performance of the pulverizing system based on the association rule algorithm and the neural network, the method can also have the following characteristics: the improved association rule algorithm comprises the following steps:
step 5-1, set of terms is I ═ I1,i2,…,im}(m∈N+) Transaction D is composed of the set { i }1,i2,…,inDenotes and each transaction is a collection of items such that T ∩ I, the identifier of each transaction is the TID.
Any transaction tjThe vector of (d) is represented as:
Dj＝(d1j,d2j,…,dmj) Wherein
1-the set of items is as follows:
R＝(D1,D2,…,Dn)
and scanning the Boolean matrix R, merging non-zero items to obtain a two-dimensional array Z [ R ] [2], wherein R represents the number of non-zero items in R and is arranged in ascending order of rows and columns. Scanning the value of a matrix R corresponding to the Z-row coordinate position of the array, counting the candidate 1-item, deleting the whole row of the array corresponding to the obtained infrequent item, keeping the frequent 1-item count, and entering the step 5-2; step 5-2, candidate 2-entries are generated from frequent 1-entries, and the transposed array Z is Z'. Scanning Z' according to row sequence arrangement and column sequence, counting the R matrix values of the corresponding positions of the candidate 2-item sets, keeping frequent 2-item counting, eliminating the whole row of the non-frequent items, namely deleting the positions with the support degree less than min, and entering the step 5-3; step 5-3, when the frequent (k-1) -item dynamically generates a candidate k-item, scanning a transfer matrix Z' and realizing dynamic pruning until the frequent k-item is an empty set, ending the algorithm, and entering step 5-4; step 5-4, generating an association rule: for frequent term L, a subset S is generated in which L is all non-empty, and if the ratio of the number of supports of S and L is greater than minsup, the rule S ═ L-S is generated.
Action and Effect of the invention
According to the association rule algorithm and neural network-based powder making system performance optimization method, the scanned initial database matrix is dumped into the effective item coordinate array, so that the traditional association rule algorithm mining algorithm is improved, therefore, the calculation resources are saved, and the operation efficiency and performance of the association rule algorithm are improved.
According to the method for optimizing the performance of the pulverizing system based on the association rule algorithm and the neural network, the data mining is more targeted, redundant items are reduced, and the mining efficiency is improved because the dimension constraint and the sample compression are carried out on the database in a target guidance mode.
Drawings
FIG. 1 is a flow chart of a method for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network according to an embodiment of the present invention;
FIG. 2 is a schematic view of a coal pulverizer assembly according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a single-condition classification result of a unit supervised self-organizing neural network according to an embodiment of the present invention;
FIG. 4 is a comparison graph of unit net power generation before and after optimization of condition 1 in the embodiment of the present invention;
FIG. 5 is a comparison graph of unit net power generation before and after optimization of condition 2 in the embodiment of the present invention;
FIG. 6 is a comparison graph of unit net power generation before and after optimization of condition 3 in the embodiment of the present invention;
FIG. 7 is a comparison graph of unit net power generation before and after optimization of condition 4 in the embodiment of the present invention;
FIG. 8 is a comparison graph of the accuracy of the modified and conventional association rule algorithms in an embodiment of the present invention;
FIG. 9 is a comparison graph of improved and conventional association rule algorithm detection rates in an embodiment of the present invention;
FIG. 10 is a graph comparing run time of an improved and conventional association rule algorithm in an embodiment of the present invention;
FIG. 11 is a diagram illustrating the unit consumption optimization results of the coal mill under the working condition 1 according to the embodiment of the present invention;
FIG. 12 is a diagram illustrating the unit consumption optimization results of the coal mill under the working condition 2 according to the embodiment of the present invention;
FIG. 13 is a diagram illustrating the unit consumption optimization results of a coal pulverizer under working condition 3 according to an embodiment of the present invention; and
FIG. 14 is a diagram of the unit consumption optimization result of the working condition 4 coal mill in the embodiment of the invention.
Detailed Description
In order to make the technical means, the creation features, the achievement purposes and the effects of the invention easy to understand, the invention is specifically described below by combining the embodiment and the attached drawings.
< example >
Fig. 1 is a flowchart of a method for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network according to an embodiment of the present invention.
As shown in figure 1, the performance optimization method of the powder making system based on the improved association rule algorithm and the supervised self-organizing neural network has the advantages that the coverage of the historical operating conditions of the power plant unit is wide, the database is increased in real time, the actual operating state of the unit is truly reflected, and equivalently, a huge historical knowledge base is stored for operators. Firstly, judging whether the data is stable operation data, then dividing working conditions through a supervised self-organized neural network, and mining by using an improved association rule algorithm based on target guidance to obtain a powder making system historical knowledge base. And comparing the current real-time operation data of the unit with a historical knowledge base, matching historical operation data of similar working conditions, and searching an optimal target value to guide the operation of the pulverizing system. If the current operating parameter is better than the historical optimal target value, the historical knowledge base is immediately updated. And if a newly added working condition cluster occurs, after the operating data of the working condition cluster is accumulated, the working condition cluster is excavated again and new knowledge is recorded into the historical database.
Specifically, the method for optimizing the performance of the pulverizing system based on the association rule algorithm and the neural network in the embodiment includes the following steps:
s0, firstly, determining the research object. The method is characterized in that a pulverizing system of a 350MW unit of a certain power plant is taken as a research object, two single reheating, subcritical, steam extraction and condensing 350MW steam extraction heat supply generator sets are arranged in the certain power plant, 5 coal mills (respectively numbered as A-E coal mills) with models of MPS212 are configured, a primary fan is arranged at the downstream of an air preheater, and 1 of the primary fans is designed for standby. The traditional method for starting and stopping the coal mills according to the unit load is rough, and the attention to the characteristic difference among the coal mills is less. According to the method, the net raw coal power generation amount of the unit is improved through the operation parameters and the operation mode of the deep excavation powder process system, the optimized operation of the powder process system is realized, and important references are provided for field operation guidance.
And S1, determining an optimized model database of the pulverizing system. And according to historical data of the power plant pulverizing system, the historical data comprises part of main operating parameters of the unit and part of related operating parameters of the pulverizing system. The operation parameters and the operation mode of the deep excavation pulverizing system improve the clean power generation quantity of raw coal of the unit.
Under the rated output of the coal mill, the fineness R of the coal powder is between A and E90Is between 18.0 and 18.8 percent (the designed coal powder fineness R)9018%) was obtained. The opening of the coal mill is 45%, the content of the fly ash combustible in the boiler is considered to be low, the heat loss of unburned carbon is lower than the design value, and the fineness of pulverized coal is not adjusted. The study data used was historical steady state operating data from 3 months 1 day to 5 months 30 days in 2019, with 1 minute intervals. The total number of the optimization parameters is 9, and the number of the optimization parameters is 143,616 groups in total, and the specific optimization parameters are shown in table 1.
TABLE 1 optimization parameters
And determining the excavating equipment combination of the pulverizing system. From the collected data, the combined operation of the coal mill under various operating conditions is summarized according to the active power of the ABCDE coal mill as shown in FIG. 2. As can be seen from fig. 2, the main operation combination modes of the coal pulverizing system are three modes, namely ABC, BCD and BCDE, wherein the ABC coal mill combination mode mainly operates in a medium-low load range, the BCD coal mill combination mainly and intensively distributes in a medium-high load range, the BCDE coal mill combination mainly operates in a high load range, and the operation times of other coal mill combination modes are shorter, so that the coal mill combination mode is omitted in the embodiment, and the three coal mill operation combination modes of ABC, BCD and BCDE are reserved for subsequent excavation;
and S2, screening and judging the data in the powder making system optimization model database, and if the data are judged to be stably operated, keeping the data in the powder making system optimization model database, otherwise, discarding the data from the powder making system optimization model database.
And S3, clustering the working conditions of the historical data in the pulverizing system database through the supervised self-organizing neural network, counting the attribute characteristics of each category, and determining the specific division standard of the working condition clustering.
The self-organizing neural network is an unsupervised self-organizing competitive neural network, and can automatically cluster according to the recognized environmental characteristics. In the learning process of the network, a specific neuron identification can be used as a detector of an input pattern due to unsupervised competitive learning. Trained network neurons are divided into different regions and correspondDifferent response characteristics. The self-organizing neural network is composed of an input layer and a competition layer, wherein the first layer is the input layer, and the number of neurons is m. The second layer is a competition layer distributed in a two-dimensional array, and the number of nodes of the competition layer is n. The nodes of the input layer and the competition layer are connected by variable weight, and the connection weight is wij(i＝1,2,…,m；j＝1,2,…,n)
The supervised self-organizing neural network adds an output layer after a competition layer of the self-organizing neural network to realize supervised learning classification. The output layer is connected with the weights of the nodes of the competition layer, so that the node weights between three layers of the network need to be considered to adjust the whole network.
The supervised self-organizing neural network clustering steps are as follows:
s3-1, normalizing data;
and S3-2, initializing a supervised self-organizing neural network. Initializing weights w of input and competing layersijWeights w of competition layer and output layerjkIs [0,1 ]]The initial value of the internal random number, learning rate η (t), is set to be in the range of (0,1), and the learning radius r and the total learning number are determined.
S3-3, changing the training sample X into (X)1,x2,…，xn) Calculating the X input layer and the competition layer
Euclidean distance over yuan:
s3-4, searching for the dominant neuron. And selecting the shortest distance competition layer neuron c as an upper neuron in the obtained Euclidean distance.
And S3-5, adjusting the weight. Adjusting the weight w of the input layer and the competition layerijThe formula is as follows:
wij＝wij+η1(X-wij)
adjusting weight w between competition layer and output layerjkThe formula is as follows:
wik＝wjk+η2(Yk-wjk)
in the formula, η1、η2The learning rate is in linear inverse proportion correlation with the learning times, YkIs the category to which the sample belongs.
And S3-6, judging whether the algorithm is finished or not, and returning to the loop of S3-2 if the algorithm is not finished.
The method selects the unit load to carry out single working condition division of the supervised self-organizing neural network. Firstly, selecting 230-350 MW load working condition cluster data samples under the constraints of certain coal quality parameters and environmental conditions for testing experiments, collecting historical operation data of a power plant from 3 months, 1 day to 5 months, 30 days in 2019 by using a supervised self-organized neural network algorithm, and dividing stable operation working conditions of a pulverizing system. The total number of samples is 4375 groups, wherein 4000 samples are used as training samples, 875 samples are used as testing samples, the division result is shown in fig. 3 and table 2, fig. 3 shows the overall classification result of the data, and table 2 shows the specific attribute of each working condition class.
As can be seen from fig. 3, the historical data is classified by the supervised self-organizing neural network, the weights between neurons are calculated after iteration, and the internal similarities of the data are counted and reflected on the neurons of each layer. The historical data is divided into 4 types, namely the working condition of the unit load is divided into 4 types. The attribute characteristics of each category are counted in sequence, and the specific division standard of the working condition cluster can be obtained, as shown in table 2.
TABLE 2 clustering of unit load conditions
According to the classification results, the operation conditions of the pulverizing system are classified into 4 types, the 1-4 types are arranged according to the ascending order of the unit load, and the observation shows that the pulverizing system mainly operates in a middle-high load section and sample data in a low-load section is rare. And most test data classification results are the same as expected classes, and 875 groups of test data are classified correctly and are 842 groups, and the network accuracy is 96.23%. Therefore, the supervised self-organizing neural network has a good clustering effect, and lays a good foundation for subsequent data mining and system optimization.
And S4, discretizing data in the optimized model database of the coal pulverizing system by adopting a quantile grouping method to obtain discretization historical data, determining the clean power generation quantity of raw coal as a mining target, processing the discretization historical data by adopting a dimension reduction compression mode based on target guidance, and establishing a target function.
The data discretization realizes the quantization processing of the unit operation parameters, the median or the mean of the interval parameters is selected as an interval representative value, the mean can accurately reflect the data characteristics, and the median can reflect the distribution form of the data to a certain extent;
since the association rule algorithm is applied to a boolean database, the processing should be discrete data, while the raw data is continuous, and thus, the data needs to be discretized. In the process of dispersing data, how to reasonably customize the size of the division areas becomes a main challenge difficulty. If the discrete interval is too large, a "minimum confidence problem" may occur; if the discrete interval is too small, a "minimum support problem" may occur. A common discretization approach is grouping, dividing the continuous data into multiple non-overlapping intervals according to a priori knowledge. There are two grouping criteria: one at the group distance and the other at the quantile. However, improper group distance selection causes large difference between sample sizes between groups, and has great influence on subsequent modeling and algorithm implementation. The quantile grouping rule does not have the problems, and the method is based on the principle that the sum of variables of each group is approximate on the premise of specifying the number of the groups, so that the grouped sample size can be ensured to be more average.
In this embodiment, a quantile grouping method is adopted to discretize data, so as to realize quantitative processing of unit operation parameters. In the discretization process, the fact that the distribution of actual operation parameters has more or less deviation is found, so that the selection mode of a general interval representative value is changed, the original interval midpoint representative value is changed into the mode of selecting the median or the mean of the interval parameters as the interval representative value, the mean can accurately reflect the characteristics of data, the median can reflect the distribution form of the data to a certain degree, and meanwhile, the median is also an operation reachable value of actual operation, so that the significance of guiding the actual operation is achieved.
The method comprises the steps that a large amount of measuring point monitoring data are generated in actual operation of a power plant and stored according to the second level, and for data with high dimensionality and large data volume, in order to avoid a large amount of redundant items in an excavation result, improve excavation efficiency, enable data excavation to be more targeted and achieve compression and dimensionality reduction of power plant sample data, steady-state data are processed in a mode of combining dimensionality reduction compression based on target guidance.
First, the excavation target is clarified. Since the final objective of the research is to optimize the economy of the coal pulverizing system, the comprehensive relationship between the optimization and the running cost of the coal pulverizing system is considered, and the relationship function F of the net generated electricity of each unit of raw coal is calculated by fitting the collected datajDetermining FjFor evaluation index, the expression is as follows:
in the formula, bgRepresenting standard coal consumption, g/(kW.h); k represents the unit consumption of the pulverizing system (kW.h)/t.
By established FjThe target model is excavated by adopting an improved association rule algorithm FjAnd (5) the optimal value of the related operation parameter of the pulverizing system is higher. Under the combination mode of divided working conditions and specific coal mills, the coal pulverizing method takes FjSelecting an optimized parameter data sample set of a pulverizing system optimization model database as a target index, and realizing the dimensionality reduction and compression of the data set in a target guidance mode, so that when mining is carried out by using an improved association rule algorithm in the follow-up process, the support degree of the sample space is only required to be limited as a mining condition, and a support degree calculation method is determined by a large number of experimental results and has the following calculation formula:
s5, mining by adopting an improved association rule algorithm based on target guidance to obtain a historical knowledge base of the powder making system, mining the optimal value of the related operation parameter of the powder making system when the target function is optimal, and storing the optimal value in the historical knowledge base of the powder making system;
the coal quality factor is considered, the unit load is two main external influence factors, and the coal quality condition is fixed because the content of combustible substances in the fly ash of the boiler under the current experimental data working condition is low, the heat loss of unburned carbon is lower than the design value, and the fineness of pulverized coal is not adjusted. Under the condition, the unit working conditions are divided by utilizing the supervised self-organized neural network to form a stable-state operation working condition cluster of the powder preparation system. Then, based on a target guidance method, determining an optimization index as a unit load, a coal feeder speed, a smoke exhaust oxygen content, a smoke exhaust temperature, coal mill active power, primary fan active power, primary air volume, coal mill inlet and outlet air pressure and raw clean coal power generation quantity as evaluation indexes to form an excavation sample space.
And S6, collecting the real-time relevant operation parameters of the powder making system, comparing the real-time relevant operation parameters with the data in the historical knowledge base of the powder making system, judging whether the similar working conditions can be obtained by matching, judging whether the real-time relevant operation parameters are superior to the historical relevant operation parameters when the historical relevant operation parameters of the powder making system under the similar working conditions are obtained by matching, guiding the powder making system to operate by using the historical relevant operation parameters when the historical relevant operation parameters are not obtained by matching, and guiding the powder making system to operate by using the real-time relevant operation parameters and immediately updating the historical knowledge base if the historical relevant. When the history related operation parameters of the pulverizing system under the similar working conditions can not be obtained by matching, the newly added working condition cluster is determined to appear, the step S7 is judged,
raw coal net power generation quantity F by improving association rule algorithmjObtaining a preferred F for the objective functionjThe association rule is shown in table 3, and the optimal value intervals of all important parameters of different coal mill combination modes under all the optimized working conditions are given. In the experiment, due to the fact that the data volume is limited, the coal mill combination mode cannot only select the large data volume for research. The full working condition comprises 4 working conditions, the working mode of the coal mill under the working condition 1(230MW-260MW) is ABC, the working condition 2(260MW-290MW) has two combination modes of ABCD and BCD respectively, the coal mill under the working condition 3(290MW-320MW) has a combination mode of BCD, the working condition 4(320MW-350MW) is BCDE, the optimization range of each parameter is obtained by improving the association rule algorithm, and important reference is provided for guiding the optimization of the field operation.
TABLE 3 Association rules after optimization of all conditions
And fitting a curve of the net generated energy per unit under the full working condition according to the optimized data of the full working condition obtained in the table 3, and comparing the curve with a curve of the fitted average value of the net generated energy per unit under the original working condition, wherein the comparison graphs are shown in fig. 5 to 8. The observation shows that the net generated coal amount is improved by about 1.92% under the working condition 1, and the net generated coal amount is improved by about 1.38% under the working condition 2. The net generated coal amount is improved by about 1.52% under the working condition 3, and the net generated coal amount is improved by about 2.79% under the working condition 4, so that the optimization effect is satisfactory under all the working conditions, and the optimized unit net generated energy is at a relatively high level. Meanwhile, as is apparent from fig. 6, under the condition that the performance of the coal mill is improved by various combination modes, the BCD combination mode is more excellent in economy than the ABCD combination mode, that is, the number of coal mills running is not positively correlated with the unit net power generation amount of the coal pulverizing system.
Optimal adjustment of the parameters of the pulverizing system can affect the combustion state of the boiler, the economic operation of the power plant and the NOXAnd (4) discharging. According to statistics of parameter optimization experiment results of the coal pulverizing system, the operation number of the coal mills is reduced under medium and low loads, the total active power of the coal mills is reduced, the power consumption and the primary air quantity are reduced, the power and the current of a primary air fan are reduced accordingly, and the energy-saving effect is achieved. Meanwhile, because the output of the coal mill is improved, in order to ensure the pressure difference and the temperature of the inlet and the outlet of the coal mill, the opening degree of a primary air door valve is enlarged, the smoke exhaust temperature is reduced, and the boiler efficiency is improved. From the above analysis, it can be derivedAnd (4) conclusion: within the range of medium and low load, the number of running coal mills is reduced, the output is increased, the coal consumption is reduced, and simultaneously the primary air quantity and NO can be reducedXThe discharge amount is increased, and the economical efficiency and the environmental protection performance of the operation are improved.
And S7, after accumulating the operation data of the newly added working condition cluster to a certain amount, excavating the optimal value of the related operation parameter of the powder making system under the newly added working condition cluster when the objective function is optimal by adopting an improved association rule algorithm based on target guidance, and storing the optimal value in a historical knowledge base of the powder making system.
S8, analyzing and improving the overall performance of the association rule algorithm mining method compared with the traditional association rule algorithm mining method;
in order to better verify that the performance of the improved association rule algorithm is superior to that of the traditional association rule algorithm, three performance indexes of data mining are compared, wherein the three performance indexes comprise a correct rate, a detection rate and the running time of the algorithm. In order to ensure the accuracy and reliability of the performance indexes, multiple simulation experiments are carried out, because the classification result of each experiment and the characteristics of the algorithm are not constant, a comparison graph of the accuracy, the detection rate and the algorithm running time is drawn by selecting 10 simulation results, the average value of the corresponding performance indexes of the 10 experiments is obtained, and the result is shown in table 4.
TABLE 4 comparison of the efficiency of operation of the conventional association rule algorithm with the improved association rule algorithm
As is evident from the performance index data in Table 4, the improved association rule algorithm improves the accuracy of the conventional association rule algorithm by 3.74%, improves the detection rate by 0.6%, and significantly reduces the operation time of the algorithm by about 50%. By observing fig. 8 to 10, each performance index was analyzed one by one. FIG. 8 shows that the improved association rule algorithm is significantly better than the conventional association rule algorithm in terms of accuracy, both in stability and accuracy per se; analyzing fig. 9, it can be known that the detection rate change amplitude of the association rule algorithm before and after the improvement is smaller, but the overall detection rate of the improved association rule algorithm is slightly improved and is more stable; fig. 10 makes it clear that the running time of the improved association rule algorithm is rapidly reduced, and the operation time of each experiment is basically maintained at the same level. The improved association rule algorithm has better overall performance by integrating 3 performance indexes.
And S9, excavating the optimal running state of the coal pulverizing system, and carrying out comparative analysis on unit consumption of the coal mill before and after optimization under all working conditions by utilizing the optimal interval of each parameter of the coal mill obtained by improving the association rule algorithm.
In order to excavate the optimal running state of the coal pulverizing system, the optimal interval of each parameter of the coal mill obtained by improving the association rule algorithm is utilized to compare and analyze the unit consumption of the coal mill before and after optimization under all working conditions, and the range values of the original unit consumption and the unit consumption of the coal mill after optimization under all working conditions are shown in table 5.
TABLE 5 comparison of unit consumption of pulverizing system under all operating conditions
Due to research data limitations, table 5 only shows the coal mill unit consumption optimization comparison range of the specific coal mill combination operation mode under each working condition, and fig. 11 to 14 are unit consumption optimization results under the specific coal mill combination mode under all working conditions. By combining table 5 and fig. 11 to 14, it can be found that the unit consumption test verification is performed by improving the parameters optimized by the association rule algorithm, and under any working condition, the optimized unit consumption of the coal mills is reduced by about 1kWh/t, and because the calculation formula of the unit consumption is the ratio of the electricity consumption of a single coal mill to the total coal supply, the electricity consumption of 1kWh per ton of coal of each coal mill is reduced, the plant electricity consumption rate is directly reduced, and thus the production efficiency of the power plant is improved.
Effects and effects of the embodiments
According to the association rule algorithm and neural network-based powder making system performance optimization method, the scanned initial database matrix is dumped into the effective item coordinate array, so that the traditional association rule algorithm mining algorithm is improved, therefore, the calculation resources are saved, and the operation efficiency and performance of the association rule algorithm are improved.
According to the association rule algorithm and neural network-based powder making system performance optimization method, the data mining is more targeted, redundant items are reduced, and the mining efficiency is improved because the dimension constraint and the sample compression are performed on the database in a target guidance mode.
According to the method for optimizing the performance of the coal pulverizing system based on the association rule algorithm and the neural network, the unit is divided into single working conditions by adopting the supervised self-organized neural network clustering algorithm according to the historical operating data of the 350MW unit coal pulverizing system of the power plant, the logarithm discretization is carried out by the quantile grouping method, the improved association rule algorithm mining algorithm with the improved overall performance is used by comparing the traditional association rule algorithm with the improved association rule algorithm through experiments, each important parameter of the coal pulverizing system is used as a mining sample, the optimal parameter value of the coal pulverizing system under the full working condition after optimization is obtained by taking the net coal power generation quantity as an evaluation index, so the economy of the optimized coal pulverizing system in the embodiment is improved, the unit net power generation quantity of the full working condition is improved, and the operating quantity and the output of coal mills are reduced and increased aiming at the middle-low load section, can effectively reduce coal consumption, primary air quantity and nitrogen oxide emission. Improving the economy and environmental protection of the unit. In addition, unit consumption experiments verify that unit consumption of the coal mill with optimized parameters is obviously reduced, the plant power rate can be effectively reduced, and reference is provided for field operation guidance.
The above embodiments are preferred examples of the present invention, and are not intended to limit the scope of the present invention.
Claims (3)
1. A method for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network is characterized by comprising the following steps:
step 1, establishing a powder process system database according to historical data of a powder process system, wherein the historical data comprises: the main operation parameters of the excavating equipment unit and the related operation parameters of the pulverizing system enter the step 2;
step 2, judging whether the historical data in the powder making system database is stable operation data or not, if so, retaining the data in the powder making system database, otherwise, discarding the data in the powder making system database, and entering step 3;
step 3, clustering the working conditions of the historical data in the pulverizing system database through a supervised self-organizing neural network, counting the attribute characteristics of each category, determining the specific division standard of the working condition clustering, and entering step 4;
step 4, discretizing the historical data in the pulverizing system database by adopting a quantile grouping method to obtain discretization historical data, determining the clean raw coal power generation amount as a mining target, processing the discretization historical data by adopting a dimension reduction compression mode based on target guidance, establishing a target function, and entering step 5;
step 5, adopting an improved association rule algorithm based on target guidance to dig out an optimal value of a related operation parameter of the powder making system when the target function is optimal in the powder making system database, storing the optimal value in a powder making system historical knowledge base, and entering step 6;
step 6, collecting real-time related operation parameters of the powder making system, comparing the real-time related operation parameters with data in a powder making system historical knowledge base, judging whether the real-time related operation parameters are superior to the historical related operation parameters when the historical related operation parameters of the powder making system under similar working conditions are obtained through matching, if not, using the historical related operation parameters to guide the powder making system to operate, otherwise, using the real-time related operation parameters to guide the powder making system to operate, and if the historical related operation parameters of the powder making system under similar working conditions cannot be obtained through matching, determining that a newly added working condition cluster occurs, and entering step 7;
and 7, after accumulating the operation data of the newly increased working condition cluster to a certain amount, excavating the optimal value of the related operation parameter of the powder making system under the newly increased working condition cluster when the target function is optimal by adopting an improved association rule algorithm based on target guidance, and storing the optimal value in a powder making system historical knowledge base.
2. The method of claim 1 for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network, wherein:
the method for clustering the working conditions of the historical data in the powder process system database by the supervised self-organized neural network comprises the following steps:
step 3-1, normalizing the data, and entering step 3-2;
step 3-2, initializing the supervised self-organizing neural network, and initializing the weight w of the input and competition layerijAnd the weights w of the competition layer and the output layerjkIs [0,1 ]]Taking values of an initial value of an internal random number and a learning rate η (t) in a range of (0,1), determining a learning radius r and a total learning frequency, and entering a step 3-3;
step 3-3, establishing a training sample X ═ X1,x2,…，xn) Calculating the Euclidean distance between the training sample X input and the neuron of the competition layer, wherein the calculation formula is as follows:
entering the step 3-4;
3-4, searching for a winning neuron, selecting a shortest distance competition layer neuron c as a superior neuron in the obtained Euclidean distance, and entering into the step 3-5;
step 3-5, adjusting the weight w of the input and competition layerijThe calculation formula is as follows:
wij＝wij+η1(X-wij)，
adjusting weight w between competition layer and output layerjkThe calculation formula is as follows:
wik＝wjk+η2(Yk-wjk)
in the formula, η1、η2The learning rate is in linear inverse proportion correlation with the learning times, YkEntering step 3-6 for the category to which the sample belongs;
and 3-6, judging whether the algorithm is finished, if not, returning to the step 3-2, otherwise, finishing the algorithm.
3. The method of claim 1 for optimizing the performance of a pulverizing system based on an association rule algorithm and a neural network, wherein:
wherein the improved association rule algorithm comprises the following steps:
step 5-1, set of terms is I ═ I1,i2,…,im}(m∈N+) Transaction D is composed of the set { i }1,i2,…,inDenotes, and each transaction is a collection of items, such that T ∩ I, the identifier of each transaction is TIDjThe vector of (d) is represented as:
Dj＝(d1j,d2j,…,dmj) Wherein
1-the set of items is as follows:
R＝(D1,D2,…,Dn)
and scanning the Boolean matrix R, merging non-zero items to obtain a two-dimensional array Z [ R ] [2], wherein R represents the number of non-zero items in R and is arranged in ascending order of rows and columns. Scanning the value of a matrix R corresponding to the Z-row coordinate position of the array, counting the candidate 1-item, deleting the whole row of the array corresponding to the obtained infrequent item, keeping the frequent 1-item count, and entering the step 5-2;
step 5-2, candidate 2-entries are generated from frequent 1-entries, and the transposed array Z is Z'. Scanning Z' according to row sequence arrangement and column sequence, counting the R matrix values of the corresponding positions of the candidate 2-item sets, keeping frequent 2-item counting, eliminating the whole row of the non-frequent items, namely deleting the positions with the support degree less than min, and entering the step 5-3;
step 5-3, when the frequent (k-1) -item dynamically generates a candidate k-item, scanning a transfer matrix Z' and realizing dynamic pruning until the frequent k-item is an empty set, ending the algorithm, and entering step 5-4;
step 5-4, generating an association rule: for frequent term L, a subset S is generated in which L is all non-empty, and if the ratio of the number of supports of S and L is greater than minsup, the rule S ═ L-S is generated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010021684.9A CN111158239B (en) | 2020-01-09 | 2020-01-09 | Association rule algorithm and neural network-based pulverizing system performance optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010021684.9A CN111158239B (en) | 2020-01-09 | 2020-01-09 | Association rule algorithm and neural network-based pulverizing system performance optimization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111158239A true CN111158239A (en) | 2020-05-15 |
| CN111158239B CN111158239B (en) | 2022-06-21 |
Family
ID=70562286
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010021684.9A Active CN111158239B (en) | 2020-01-09 | 2020-01-09 | Association rule algorithm and neural network-based pulverizing system performance optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111158239B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113504765A (en) * | 2021-06-18 | 2021-10-15 | 汉谷云智(武汉)科技有限公司 | Method for optimizing operation of circulating fluidized bed boiler of power plant |
| CN113554153A (en) * | 2021-07-23 | 2021-10-26 | 潍柴动力股份有限公司 | Method and device for predicting emission of nitrogen oxides, computer equipment and medium |
| CN114488798A (en) * | 2022-01-10 | 2022-05-13 | 国能蚌埠发电有限公司 | Method for monitoring performance and optimizing operation of secondary reheating unit based on data coordination |
| CN115814686A (en) * | 2023-02-14 | 2023-03-21 | 博纯材料股份有限公司 | State monitoring method and system for laser gas mixing production system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20020018520A (en) * | 2000-09-02 | 2002-03-08 | 이문권 | association rule |
| CN104376380A (en) * | 2014-11-17 | 2015-02-25 | 北京工业大学 | Ammonia concentration predicting method based on recursion self-organization neural network |
| CN107330517A (en) * | 2017-06-14 | 2017-11-07 | 华北电力大学 | One kind is based on S_Kohonen non-intrusion type resident load recognition methods |
| CN109695892A (en) * | 2018-12-03 | 2019-04-30 | 中冶南方都市环保工程技术股份有限公司 | A kind of boiler heating surface wall temperature guard method and device |
| US20190347590A1 (en) * | 2014-03-12 | 2019-11-14 | Bahwan CyberTek Private Limited | Intelligent Decision Synchronization in Real Time for both Discrete and Continuous Process Industries |
-
2020
- 2020-01-09 CN CN202010021684.9A patent/CN111158239B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20020018520A (en) * | 2000-09-02 | 2002-03-08 | 이문권 | association rule |
| US20190347590A1 (en) * | 2014-03-12 | 2019-11-14 | Bahwan CyberTek Private Limited | Intelligent Decision Synchronization in Real Time for both Discrete and Continuous Process Industries |
| CN104376380A (en) * | 2014-11-17 | 2015-02-25 | 北京工业大学 | Ammonia concentration predicting method based on recursion self-organization neural network |
| CN107330517A (en) * | 2017-06-14 | 2017-11-07 | 华北电力大学 | One kind is based on S_Kohonen non-intrusion type resident load recognition methods |
| CN109695892A (en) * | 2018-12-03 | 2019-04-30 | 中冶南方都市环保工程技术股份有限公司 | A kind of boiler heating surface wall temperature guard method and device |
Non-Patent Citations (1)
| Title |
|---|
| 罗志疆等: "基于Apriori改进关联规则算法的发电厂运行参数基准值挖掘", 《工业控制计算机》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113504765A (en) * | 2021-06-18 | 2021-10-15 | 汉谷云智(武汉)科技有限公司 | Method for optimizing operation of circulating fluidized bed boiler of power plant |
| CN113554153A (en) * | 2021-07-23 | 2021-10-26 | 潍柴动力股份有限公司 | Method and device for predicting emission of nitrogen oxides, computer equipment and medium |
| CN114488798A (en) * | 2022-01-10 | 2022-05-13 | 国能蚌埠发电有限公司 | Method for monitoring performance and optimizing operation of secondary reheating unit based on data coordination |
| CN114488798B (en) * | 2022-01-10 | 2024-01-12 | 国能蚌埠发电有限公司 | Method for monitoring performance and optimizing operation of secondary reheating unit based on data coordination |
| CN115814686A (en) * | 2023-02-14 | 2023-03-21 | 博纯材料股份有限公司 | State monitoring method and system for laser gas mixing production system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111158239B (en) | 2022-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111158239B (en) | Association rule algorithm and neural network-based pulverizing system performance optimization method | |
| CN104534507B (en) | A kind of boiler combustion optimization control method | |
| CN106094744A (en) | The determination method of thermoelectricity factory owner's operational factor desired value based on association rule mining | |
| Zheng et al. | Multi-objective combustion optimization based on data-driven hybrid strategy | |
| CN110400018B (en) | Operation control method, system and device for coal-fired power plant pulverizing system | |
| CN110532674A (en) | A kind of coal-fired power station boiler fire box temperature measurement method | |
| CN101893877A (en) | Optimization operational method based on energy consumption analysis for power plant and system thereof | |
| CN107038334A (en) | Circulating fluid bed domestic garbage burning boiler CO emitted smoke system and methods | |
| CN108954375A (en) | Saving coals from boiler control method | |
| CN105005878B (en) | A kind of comprehensive estimation method of strong intelligent grid | |
| CN111461921B (en) | Load modeling typical user database updating method based on machine learning | |
| CN115102170A (en) | Coordination optimization method for wind power photovoltaic energy storage ratio | |
| Dong et al. | Evaluation of TPGU using entropy-improved TOPSIS-GRA method in China | |
| CN113504765A (en) | Method for optimizing operation of circulating fluidized bed boiler of power plant | |
| CN115526433A (en) | Power plant reheat flue gas baffle operation prediction method based on integrated hybrid model | |
| CN112836920A (en) | Coal electric unit energy efficiency state evaluation method and device and coal electric unit system | |
| CN115681597A (en) | Fusion drive-based waste heat valve control optimization method | |
| CN111199304A (en) | Multi-target combustion optimization method based on data-driven fusion strategy | |
| Li et al. | Research and application of data mining in power plant process control and optimization | |
| Liu et al. | Study on prediction model of diesel engine with regulated two-stage turbocharging system based on hybrid genetic algorithm-particle swarm optimization method at different altitudes | |
| Guo et al. | Data Optimization Analysis of Integrated Energy System Based on-Means Algorithm | |
| Yuxin et al. | Prediction of Boiler Control Parameters Based on LSTM Neural Network | |
| Liao et al. | A method of bird damage risk assessment based on K-Means clustering | |
| Zhou et al. | Prediction Model of Coal Consumption Based on Random Forest Variable Selection and Random-Grid Hyperparametric Optimization Algorithm | |
| CN111860923A (en) | Boiler flue gas NO based on multi-model clustering integrationxEmission prediction algorithm |
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 |