CN115681597A - Fusion drive-based waste heat valve control optimization method - Google Patents
Fusion drive-based waste heat valve control optimization method Download PDFInfo
- Publication number
- CN115681597A CN115681597A CN202211318479.4A CN202211318479A CN115681597A CN 115681597 A CN115681597 A CN 115681597A CN 202211318479 A CN202211318479 A CN 202211318479A CN 115681597 A CN115681597 A CN 115681597A
- Authority
- CN
- China
- Prior art keywords
- valve
- waste heat
- entity
- value
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P10/00—Technologies related to metal processing
- Y02P10/25—Process efficiency
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses a waste heat valve control optimization method based on fusion drive, which comprises the following steps: s1: data are collected from data monitoring points for arrangement of a grate cooler, an AQC boiler and a generator of a certain cement plant, wherein the collection time interval is 5ms, and the total number of historical data points is 39766; s2: performing concept abstraction on the waste heat recovery text data to obtain waste heat recovery mechanism knowledge; s3: fusing waste heat recovery mechanism knowledge and historical data to construct a knowledge graph model based on a fuzzy set; s4: constructing an LSTM valve opening optimization model based on a time protection mechanism, providing a time protection mechanism algorithm, determining the optimal adjusting frequency of the valve, predicting the change trend of the parameters of the waste heat boiler, and reducing the time delay from the adjustment of the valve to the temperature change; s5: when new data are transmitted, the valve adjusting frequency is judged through a time protection mechanism algorithm, then the parameter change trend is predicted, knowledge reasoning is carried out through a knowledge graph model based on a fuzzy set, and the valve opening degree is recommended.
Description
Technical Field
The invention relates to the technical field of waste heat recovery, in particular to a waste heat valve control optimization method based on fusion drive.
Background
The waste heat is one of the most widely distributed energy sources with the largest potential in industrial production. At present, 30% -60% of industrial waste heat generated by China is discharged to the atmosphere along with waste gas, and the residual heat recovery rate is only 50% of that of developed countries. Therefore, how to more fully and efficiently recover the waste heat is very important.
The traditional waste heat recovery control technology is divided into a mechanism modeling method and a data driving method, wherein the mechanism modeling is to establish a relevant model according to the principles of physical and chemical reactions of an industrial process and theories such as a thermodynamic law, material balance and the like, for example, an article [ Yin Q and the like ] optimizing design of heat recovery systems on scientific kilns using genetic algorithms [ J ]. Applied Energy,2017,202, 153-168] obtains optimal design parameters by deducing a multi-objective Optimization model of a heat recovery system, and reduces heat loss; an article [ Chen Q et al, an alternative Energy flow model for analysis and optimization of Heat Transfer systems [ J ]. International Journal of Heat and Mass Transfer,2017,108 ] proposes An Energy flow model for individual Heat exchangers and Heat exchanger networks, describes system level Heat Transfer characteristics, and optimizes the thermal management system; an article [ Ahmad R and the like, mass and energy balance in grain color of concept plant [ J ]. International Journal of Scientific Engineering and Technology,2013,2 (7): 631-637] establishes a model based on a first principle to simulate the change of gas, solid temperature and wall temperature loss, is used for understanding the influence of various design parameters on cement waste heat recovery, and verifies the effectiveness of the model through simulation experiments. The method achieves better effect, but along with the increase of the complexity of the system, the method is difficult to mine the deep relation between data. With the rapid development of technologies such as industrial internet and the like, massive operation data are generated by equipment, and a system control optimization method based on data driving becomes a research hotspot. The data-driven system control optimization method mainly adjusts and optimizes equipment parameters through internal relations among deep mining historical data. Common data-driven methods are neural networks, genetic algorithms, and the like. For example, in the article [ Liuqiang and the like ] medium and low temperature waste heat power generation system performance optimization based on BP neural network strategy [ J ] chemical engineering management, 2019 (01): 109-110 ] BP neural network is applied to low temperature waste heat system modeling, and the use efficiency of low temperature waste heat can be improved; a regression-based prediction model is developed by using a BPNN (Power prediction of waste recovery system for a part of plant using back prediction neural network and its thermal modeling [ J ]. International Journal of Energy Research,2021,45 (6): 9162-9178 ], and the accuracy is as high as 99.9%; an article [ Liu crystal and the like ] data fusion driven waste heat boiler valve adjusting method [ J ] Yanshan university school report 2021,45 (01): 76-86+94 ] proposes a data fusion driven waste heat boiler valve adjusting method, and the method is based on AQC waste heat boiler valve adjusting historical data modeling so as to achieve the maximization of waste heat recycling. The method achieves better effect, but in practical application, the data coverage condition is required to be comprehensive based on the data driving method, and the time required for effective data accumulation in practical application is long and is difficult to meet.
Disclosure of Invention
Aiming at the problems, a waste heat valve control optimization method based on fusion drive is provided, the method fuses mechanism knowledge and data knowledge to construct a knowledge graph model based on a fuzzy set, materializes valve opening knowledge, establishes an LSTM valve opening optimization model based on a time protection mechanism, provides a time protection mechanism algorithm, determines the optimal adjustment frequency of a valve, and improves the waste heat recovery rate on the premise of protecting equipment.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a waste heat valve control optimization method based on fusion drive comprises the following steps:
s1: data monitoring points are arranged on a grate cooler, an AQC boiler and a generator of a certain cement plant, the opening of a bypass valve, an air mixing flue adjusting valve (cold air valve), the flue gas temperature before a combined superheater, the flue gas pressure of an outlet flue, the flue gas pressure before the combined superheater, the main steam temperature and the high-pressure steam flow are collected, the collection time interval is 5ms, and 397 historical data points are calculated in total;
s2: the method comprises the steps of abstracting concepts of waste heat recovery text data, defining valve opening degrees for some equipment states, and obtaining waste heat recovery mechanism knowledge;
s3: waste heat recovery mechanism knowledge and historical data are fused, and a knowledge graph model based on a fuzzy set is constructed;
s4: constructing an LSTM valve opening optimization model based on a time protection mechanism, determining the optimal adjusting frequency of the valve and predicting the variation trend of parameters;
s5: when new data are transmitted, the valve adjusting frequency is judged, the parameter change trend is predicted, knowledge reasoning is carried out through a knowledge graph model based on a fuzzy set, and the valve opening degree is recommended.
Further, in the step S3, the knowledge and historical data of the waste heat recovery mechanism are fused to construct a knowledge graph model based on a fuzzy set, including the following steps:
1-1) obtaining characteristics influencing the opening of the valve based on a waste heat recovery principle and correlation analysis, sequentially extracting attributes of the characteristics from historical data, and forming an initial entity according to the characteristics and the characteristic attributes;
1-2) determining state sets V = { "high", "higher", "lower", "low" }accordingto words and phrases existing in more expert experiences, determining optimal value sets corresponding to the state sets for different characteristics, determining membership function,
for a variable whose state value is "high", its membership function takes the form:
wherein A (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the state of the current entity being high, a is the optimal value corresponding to the state value high, b is the lowest acceptable characteristic value of the state value high, c is the highest acceptable characteristic value of the state value high, wherein b < a = c,
for variables whose state values are "higher" or "lower," the membership function takes the form:
wherein: a (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the current entity state of higher (lower), and a is the optimal value corresponding to the state value of higher (lower); b is the lowest value of the feature that is acceptable for a state of "higher" ("lower"); c is the highest value of the acceptable feature for which the state is "higher" ("lower"), where b < a < c,
for a variable whose state value is "low", its membership function takes the form:
wherein, a (·) represents a membership function, x is a characteristic value of the current entity, a (x) is a membership degree of the current entity state being "low", a is an optimal value corresponding to the state value "low", b is an acceptable characteristic lowest value of the state value "low", and c is an acceptable characteristic highest value of the state value "low", wherein b = a < c;
1-3) calculating the membership function of the entity, and then determining the state set of the entity. And giving different weights to the characteristics in different state sets according to the influence degree of the characteristics on the valve opening degree of the entity, and calculating the similarity between the entities. Is provided with
A＝(a 1 ,a 2 ,a 3 ,…,a n )
B＝(b 1 ,b 2 ,b 3 ,…,b n )
μ＝(μ 1 ,μ 2 ,μ 3 ,…,μ n )
Wherein A and B are entity feature vectors, a and B are entity features, n is the number of the features of the entity, a 1 、a 2 、a 3 、a n Is a characteristic value of an entity, b 1 、b 2 、b 3 、b n Is the characteristic value of B entity, mu is the weight vector of A and B characteristics, mu 1 、μ 2 、μ 3 、μ n Are the weight values of the corresponding characteristics of A and B, S AB Is the similarity of A and B, a i Is the i-th characteristic value of A, b i Is the ith characteristic value of B, mu i The weight value is corresponding to the ith characteristic of A and B;
1-4) after calculating the similarity between the entities, combining a plurality of entities with the similarity smaller than a set threshold into a set entity, wherein the feature vector of the set entity is the average value of a plurality of old entity vectors;
1-5) calculating the sum of enthalpy in the waste heat exhaust gas
As shown in the following formula:
h＝u+pv
wherein h is enthalpy of the working medium, u is internal energy of the substance, p is pressure, v is volume, and e is the working medium
T is the working medium temperature, T 0 Is the ambient temperature, c p The specific heat capacity of the working medium;
1-6) according to the enthalpy sum of waste heat and exhaust gas
The quality of the data is determined, and the calculation formula is as follows:
Q＝λ 1 H+λ 2 E
wherein Q is mass, λ is weight, λ 1 Is the weight corresponding to the enthalpy, λ 2 
Corresponding weights, H being the enthalpy of the data under similar conditions, E being the enthalpy of the data under similar conditions
To find λ 1 、λ 2 According to the idea of game theory, an objective function is established, the minimum sum of dispersion of index combination weight Q and H and E is taken as the target, and the optimal linear combination coefficient is sought
The index combining weight at this time is the optimal combining weight Q * . The objective function and constraint conditions are as follows:
min||Q-H|| 2 -||Q-E|| 2
s.t.λ 1 +λ 2 ＝1,λ 1 λ 2 ≥0
wherein, | x | | represents the L-2 norm of the calculation vector x, Q is the quality of the data under the similar working condition, H is the enthalpy of the data under the similar working condition, E is the enthalpy of the data under the similar working condition
λ is the weight, λ 1 Is the weight corresponding to the enthalpy, λ 2 
The corresponding weight of the weight is set to be,
based on the differential principle, the first derivative of the minimum can be obtained according to the above formula to obtain λ 1 、λ 2 A value of 1 、λ 2 By performing normalization processing
The final mass calculation formula is as follows:
wherein Q * And is the final quality of the data under similar conditions, λ is the weight,
is the weight corresponding to the enthalpy,
corresponding weights, H being enthalpy of data under similar conditions, E being enthalpy of data under similar conditions
1-7) determining the classification number of continuous variables according to the goodness of fit of the variance and the contour coefficient, carrying out k-meas clustering on the classification number, and then selecting data with higher quality under similar working conditions and putting the data into a decision tree for classification to obtain the relationship between entities;
1-8) establishing a knowledge graph model based on the fuzzy set according to the entity and the relation.
4. Further, in step S4, an LSTM valve opening optimization model based on a time protection mechanism is constructed, an optimal valve adjustment frequency is determined, and a variation trend of the parameter is predicted, and the steps are as follows:
2-1) constructing an LSTM prediction model fused with a convolutional neural network, wherein the model comprises a convolutional layer, a pooling layer, an LSTM layer, a Dropout layer and a BN layer, and training the model;
2-2) determining the time of a protection window according to the field working condition, if the temperature degree and the change speed do not exceed the specified threshold value in the time protection window, continuing monitoring, otherwise, jumping out of the time protection period;
2-3) boiler parameters within a preset time length before the time to be predicted are input into the trained model, and the boiler parameters at the time to be predicted are predicted;
and 2-4) carrying out knowledge reasoning by a knowledge graph model based on a fuzzy set according to the prediction parameters of the boiler, and recommending the opening degree of the valve.
Adopt the produced beneficial effect of above-mentioned technical scheme to lie in:
the invention provides a waste heat valve control optimization method (OWF) based on fusion drive, which comprises the steps of firstly, aiming at the problem that waste heat recovery mechanism knowledge and equipment operation data are difficult to fuse, constructing a knowledge graph model based on a fuzzy set, respectively extracting rules from the mechanism knowledge, extracting attributes from historical data, and forming an initial entity by the rules and the attributes together; meanwhile, in order to accelerate the map retrieval speed, an expert Experience Algorithm (EAF) based on a fuzzy set is provided, and partial entities are aggregated to form a set entity so as to reduce the number of the entities; furthermore, an Evaluation Standard (EST) based on the thermal mechanism knowledge is provided, high-quality data are screened from historical data by relying on the EST, and the relationship among entities is mined, so that a knowledge graph model based on a fuzzy set is established. Secondly, aiming at the problem that frequent adjustment causes easy loss of the boiler valve, an LSTM valve opening optimization model based on a time protection mechanism is constructed, the time protection mechanism algorithm is provided for judging the optimal frequency of valve adjustment, an LSTM prediction algorithm (LSTM-CNN) fused with a convolutional neural network is further provided for predicting the change trend of the parameters of the waste heat boiler, the time delay from the adjustment of the valve to the temperature change is reduced, the valve is adjusted in time, and the equipment risk is reduced.
Compared with the traditional waste heat recovery method, the method comprises the following steps: (1) The model extracts rules from mechanism knowledge, extracts attributes from equipment operation data, combines the rules and the attributes to form an initial entity, establishes an expert experience algorithm based on a fuzzy set aiming at a large number of entities, and finally realizes the construction of a valve opening knowledge graph, thereby solving the problems that the mechanism of the mechanism knowledge is complex, the modeling is difficult, the data knowledge accumulation working condition is slow and the two are difficult to fuse in the traditional waste heat recovery method; (2) The method comprises the following steps of providing an LSTM-based valve opening optimization model, wherein the model firstly provides a time protection mechanism algorithm for judging the optimal frequency of valve adjustment, and further provides an LSTM prediction model fused with a convolutional neural network for predicting the change trend of parameters of the waste heat boiler, adjusting the valve in time and reducing equipment risks, and the model not only reduces the adjustment times of the traditional waste heat recovery valve, but also reduces the influence of time delay on equipment; (3) The evaluation standard based on the knowledge of the thermal mechanism is provided, so that more heat can be generated on the premise of ensuring the safety of equipment, and the 'entropy pollution' to the environment is reduced.
The waste heat valve control optimization method based on fusion drive is substituted into a production data set of a certain cement plant for experiment, the waste heat recovery rate and the equipment safety are effectively improved after the method is used, and the intelligent decision of the opening degree of the waste heat recovery valve is realized.
Drawings
FIG. 1 is a frame structure diagram of a waste heat valve control optimization method based on fusion drive;
FIG. 2 is a knowledge graph model building diagram based on fuzzy sets;
FIG. 3 is a decision tree relationship abstraction graph incorporating EST criteria;
FIG. 4 is a diagram of an LSTM valve opening optimization model construction based on a time protection mechanism;
FIG. 5 is a schematic diagram of a time protection mechanism;
FIG. 6 is data for a portion of a cement plant;
FIG. 7 is a Pearson correlation coefficient;
FIG. 8 is a goodness of fit of variance;
FIG. 9 is a profile factor
FIG. 10 is a line graph comparing OV-LSTM to LSTM predictions;
FIG. 11 is a flow chart of recommended waste heat valve opening;
FIG. 12 is a comparison of enthalpy before and after the application of a fusion drive based waste heat valve control optimization method;
FIG. 13 is a graph comparing bypass valve opening at too low a temperature;
fig. 14 is a comparison graph of the opening degree of the cold air valve when the temperature is too high.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
The invention takes a knowledge graph based on a fuzzy set as a carrier and an LSTM valve opening optimization model based on a time protection mechanism as a main algorithm frame, and the model is shown as figure 1 and comprises the following steps:
s1: data monitoring points are arranged on a grate cooler, an AQC boiler and a generator of a certain cement plant, the opening of a bypass valve, an air mixing flue adjusting valve (cold air valve), the flue gas temperature before a combined superheater, the flue gas pressure of an outlet flue, the flue gas pressure before the combined superheater, the main steam temperature and the high-pressure steam flow are collected, the collection time interval is 5ms, and 397 historical data points are calculated in total;
s2: the method comprises the steps of abstracting concepts of waste heat recovery text data, defining valve opening degrees for some equipment states, and obtaining waste heat recovery mechanism knowledge;
s3: waste heat recovery mechanism knowledge and historical data are fused to construct a knowledge graph model based on a fuzzy set, the flow is shown as a figure 2, and the specific steps are as follows;
1-1) obtaining characteristics influencing the opening of the valve based on a waste heat recovery principle and correlation analysis, sequentially extracting attributes of the characteristics from historical data, and forming an initial entity according to the characteristics and the characteristic attributes;
1-2) determining state sets V = { "high", "higher", "lower", "low" } according to vocabularies with more existence in expert experience, determining optimal value sets corresponding to the state sets for different characteristics, determining membership function,
for a variable whose state value is "high", its membership function takes the form:
wherein, A (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the state of the current entity being 'high', a is the optimal value corresponding to the state value 'high', b is the lowest acceptable characteristic value of the state 'high', c is the highest acceptable characteristic value of the state 'high', wherein b < a = c,
for variables whose state values are "higher" or "lower", the membership function takes the following form:
wherein: a (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the state of the current entity being higher (lower), and a is the optimal value corresponding to the state value higher (lower); b is the lowest value of the feature that is acceptable for the state "higher" ("lower"); c is the highest value of the acceptable feature for which the state is "higher" ("lower"), where b < a < c,
for a variable whose state value is "low", its membership function takes the form:
wherein, a (·) represents a membership function, x is a characteristic value of the current entity, a (x) is a membership degree of the current entity state being "low", a is an optimal value corresponding to the state value "low", b is a lowest acceptable characteristic value of the state value "low", and c is a highest acceptable characteristic value of the state value "low", where b = a < c;
1-3) calculating the membership function of the entity, and then determining the state set of the entity. And giving different weights to the characteristics in different state sets according to the influence degree of the characteristics on the valve opening degree of the entity, and calculating the similarity between the entities. Is provided with
A＝(a 1 ,a 2 ,a 3 ,…,a n )
B＝(b 1 ,b 2 ,b 3 ,…,b n )
μ＝(μ 1 ,μ 2 ,μ 3 ,…,μ n )
Wherein A and B are entity feature vectors, a and B are entity features, n is the number of the features of the entity, a 1 、a 2 、a 3 、a n Is a value of a characteristic of an entity, b 1 、b 2 、b 3 、b n Is the characteristic value of B entity, mu is the weight vector of A and B characteristics, mu 1 、μ 2 、μ 3 、μ n Is the weight value of the corresponding characteristic A and B, S AB Is the similarity of A and B, a i Is the i-th characteristic value of A, b i Is the ith characteristic value of B, mu i The weight value is corresponding to the ith characteristic of A and B;
1-4) after calculating the similarity between the entities, merging a plurality of entities with the similarity smaller than a set threshold into a set entity, wherein the feature vector of the set entity is the average value of a plurality of old entity vectors;
1-5) after obtaining the map entities, extracting the relationship between the entities according to historical data, and calculating the enthalpy sum of the waste heat exhaust gas according to the process shown in figure 3
As shown in the following formula:
h＝u+pv
wherein h is enthalpy of the working medium, u is internal energy of the substance, p is pressure, v is volume, and e is the working medium
T is the working medium temperature, T 0 Is the ambient temperature, c p The specific heat capacity of the working medium;
1-6) according to the enthalpy sum of waste heat and exhaust gas
The quality of the data is determined, and the calculation formula is as follows:
Q＝λ 1 H+λ 2 E
wherein Q is mass, λ is weight, λ 1 Is the weight corresponding to the enthalpy, λ 2 
Corresponding weights, H being the enthalpy of the data under similar conditions, E being the enthalpy of the data under similar conditions
To find lambda 1 、λ 2 Establishing an objective function according to the idea of game theory, and seeking an optimal linear combination coefficient by taking the minimum sum of dispersion of index combination weight Q and H and E as a target
The index combining weight at this time is the optimal combining weight Q * . The objective function and constraint conditions are as follows:
min||Q-H|| 2 -||Q-E|| 2
s.t.λ 1 +λ 2 ＝1,λ 1 λ 2 ≥0
wherein | | | x | | | represents the L-2 norm of the calculation vector x, Q is the quality of the data under the similar working condition, H is the enthalpy of the data under the similar working condition, and E is the enthalpy of the data under the similar working condition
λ is the weight, λ 1 Is the weight corresponding to the enthalpy, λ 2 
The corresponding weight of the weight is set to be,
based on the differential principle, the first derivative of the minimum can be obtained according to the above formula to obtain λ 1 、λ 2 A value of 1 、λ 2 Is normalizedCan solve
The final mass calculation formula is as follows:
wherein Q is * And is the final quality of the data under similar conditions, λ is the weight,
is the weight corresponding to the enthalpy and is,
corresponding weights, H being the enthalpy of the data under similar conditions, E being the enthalpy of the data under similar conditions
1-7) determining the classification number of continuous variables according to the goodness of fit of variance and the contour coefficient, carrying out k-meas clustering on the classification number, then selecting data with higher quality under similar working conditions, putting the data into a decision tree for classification to obtain the relationship between entities,
1-8) establishing a knowledge graph model based on a fuzzy set according to the entity and the relation;
s4: an LSTM valve opening optimization model based on a time protection mechanism is constructed, the optimal adjusting frequency of the valve is determined, and the variation trend of parameters is predicted, wherein a flow chart is shown in figure 4, and the steps are as follows:
2-1) constructing an LSTM prediction model fused with a convolutional neural network, wherein the model comprises a convolutional layer, a pooling layer, an LSTM layer, a Dropout layer and a BN layer, and training the model;
2-2) determining the time of a protection window according to the field working condition, if the temperature degree and the change speed do not exceed the specified threshold value in the time protection window, continuing monitoring, otherwise, jumping out of the time protection period, and the flow chart is shown in FIG. 5;
2-3) boiler parameters within a preset time length before the time to be predicted are input into the trained model, and boiler parameters at the time to be predicted are predicted;
2-4) carrying out knowledge reasoning through a knowledge graph model based on a fuzzy set according to the prediction parameters of the boiler, and recommending the opening degree of the valve;
s5: when new data exist, the valve adjusting frequency is judged, then the parameter change trend is predicted, knowledge reasoning is carried out through a knowledge graph model based on a fuzzy set, and the valve opening degree is recommended.
The invention discloses a test verification of a waste heat valve control optimization method based on fusion drive, which comprises the following steps:
1. description of data
The experiment adopts real data acquired by a waste heat recovery system of a certain cement plant, the acquisition time interval is 8, month and 3 days in 2020 and 27 days in 8, month and 2020, and the acquisition time interval is 5ms, so that 39766 samples are totally acquired. The data is divided into three parts, namely grate cooler data, AQC boiler data and generator data. The grate cooler data comprises the opening of a bypass valve; the data of the AQC boiler mainly comprises the opening degree of a bypass valve, an air mixing flue regulating valve (cold air valve) of the AQC boiler, the front flue gas temperature of a combined superheater of the AQC furnace, the flue gas pressure of an outlet flue of the AQC furnace and the front flue gas pressure of the combined superheater of the AQC; the generator data includes generator power, main steam temperature, AQC high pressure steam flow. The selected portion of data is shown in fig. 6.
2. Experimental procedures
Experiment-knowledge graph construction based on fuzzy set
According to the waste heat recovery principle, relevant characteristics are selected as shown in table 1:
TABLE 1 characteristic parameters
The time delay from the adjusting valve to the saturated steam change can not be directly obtained from the database, so the time delay of the adjusting valve and the saturated steam change is calculated by the Pearson correlation technique, the correlation coefficient curve is shown in FIG. 7, when the time delay is 120 s-150 s, the correlation coefficient is the highest and tends to be flat, and the time delay of the adjusting valve and the saturated steam change is considered to be 150s. And establishing an initial entity according to the characteristics and the characteristic attributes. There were a total of 39766 initial entities.
In order to reduce the number of entities, fuzzy judgment is carried out on the entity characteristics according to an EAF algorithm. The expert experience relating to the flue gas temperature before the AQC combined superheater is summarized as follows:
experience 1: the temperature of flue gas before the AQC combined superheater is higher and changes steadily or when the temperature is reduced, the valve is kept unchanged;
experience 2: when the temperature of the flue gas in front of the AQC combined superheater is high, a cold air valve needs to be opened;
experience 3: when the temperature of the flue gas before the AQC combined superheater is lower and is in an ascending trend, the valve is kept unchanged;
experience 4: when the temperature of flue gas before the AQC combined superheater is low and tends to be stable, the bypass valve is opened.
According to expert experience and membership functions, adding features to entities related to the experience: the characteristic values of the temperature state values are { "high", "higher", "lower" and "}, further integration is carried out according to the similarity between the entities, the number of the entities is reduced by 20420, the number is reduced by 5.1%, and the retrieval efficiency is improved.
And determining the value of a k value according to the goodness of variance fit and the profile coefficient, and as can be seen from fig. 8, the curves of the flue gas temperature and the main steam temperature before the AQC combined superheater tend to be flat at 12, 13, 14 and 15, and the curves of the AQC differential pressure tend to be flat at 14, 15, 16 and 17. Next, the contour coefficients of these several points are calculated, respectively, and the result is shown in fig. 9. Therefore, the front flue gas temperature of the combined superheater can be determined to be classified into 13 types, the main steam temperature is classified into 15 types, and the AQC differential pressure is classified into 16 types.
Samples used for relational extraction were factory data screened by EST standards. The number of samples is 164185, and the samples contain four condition attributes and two decision attributes. The decision attributes are discrete, while the AQC of the conditional attributes is continuous with the main steam temperature and the flue gas temperature before the superheater. The improved algorithm aggregates continuous values of the condition attribute into discrete values through a k-means algorithm.
To verify the validity of the method, the decision tree fused with EST criteria was compared with the conventional ID3 decision tree, and the results are shown in table 2. Experiments prove that the accuracy of the training set and the accuracy of the testing set of the algorithm are higher than those of the traditional algorithm.
TABLE 2 decision Tree accuracy comparison
Experiment two LSTM valve opening optimization model construction based on time protection mechanism
Evaluation indexes are typically used in regression tasks to calculate prediction errors. Some evaluation criteria were used to determine how the predicted data deviated from the target (actual) data during fitting to measure the performance of the model in this study. Prediction error measurement techniques used herein are Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
To verify the effectiveness of OV-LSTM time series prediction, an experiment is designed to compare the prediction results of the model with those of other models, and as shown in FIG. 10, it can be seen that the predicted value of OV-LSTM is obviously closer to the true value than the predicted value of LSTM.
To further determine the accuracy of OV-LSTM prediction, the MAE and RMSE values for OV-LSTM and LSTM were calculated and the results are shown in Table 3. As can be seen from the table, the MAE and RMSE values of OV-LSTM are lower than that of LSTM, and the effectiveness of OV-LSTM time series prediction can be further determined.
TABLE 3 OV-LSTM vs LSTM prediction comparison
Next, in order to verify the effectiveness of the time protection mechanism in practical application, a simulation experiment is performed on the 8-month data, and the experimental results are shown in table 4. The number of valve adjustments is significantly reduced after the addition of the time protection mechanism. After a time protection mechanism is added, the input data is analyzed and predicted, whether the valve needs to be adjusted or not is judged, and unnecessary adjusting times of the valve are reduced.
TABLE 4 number of valve adjustments before and after time protection mechanism
Experiment three comparative experiments
The flow of the OWF waste heat valve control is shown in FIG. 11, a data set of a cement plant in a half month is selected as input in a comparison experiment, the sampling time is 8 months from No. 3 to 8 months from 16 days, data preprocessing is carried out on the sampling segment, and then the sampling segment is input into an OWF model, so that the recommended valve opening degree is obtained.
In order to verify the effectiveness of the OWF in improving the residual heat recovery rate (here represented by the enthalpy value) in practical application, the residual heat recovery rate after applying the method is compared with the real regulation condition, the abscissa is the date, and the ordinate is the average value of the enthalpy value of the day, and the comparison result is shown in fig. 12. According to the comparison result, except that the enthalpy values of 8 months and 5 days, 8 months and 10 days, 8 months and 11 days and 8 months and 13 days are close, the enthalpy values of other dates after the OWF is applied are obviously higher than the original enthalpy value, namely the enthalpy value after the method is applied for 12 days in 16 days of comparison is obviously higher than the original enthalpy value, the probability that the enthalpy value is obviously higher than the original enthalpy value is considered to be 75%, and the effectiveness of the method on improving the waste heat recovery rate can be demonstrated. In order to verify the effectiveness of the OWF protection equipment, the difference between the valve opening recommended by the method and the valve opening of the regulating valve of an employee is compared in a comparison experiment. As can be seen from fig. 13, when the temperature is lower than the safety threshold, except for 8 months and 4 days, the probability of opening the valve and raising the temperature by the method is higher than that of the staff, that is, when the temperature is lower than the safety threshold, the effective rate of improving the safety of the equipment by the OWF method is 94%, and the safety probability of the equipment is averagely improved by 52.12%; according to fig. 14, except 8 months and 5 days, when the temperature is higher than the safety threshold, the probability of opening the valve to cool by the method is greater than or equal to that of the staff, that is, when the temperature is higher than the safety threshold, the effective rate of improving the safety of the equipment by the OWF method is 94%, and the safety probability of the equipment is improved by 43.88% on average. This experiment therefore verifies the effectiveness of the OWF protection device.
3. Conclusion
Aiming at the problems that mechanism knowledge and data knowledge are difficult to fuse and the like in the traditional waste heat valve control technology, the invention provides a waste heat valve control optimization method based on fusion drive, and the method is divided into two parts. The first part is to extract classification rules according to industrial mechanism knowledge, extract node attributes according to historical data knowledge, form initial entities together with the rules and the attributes, establish an expert experience algorithm based on a fuzzy set to reduce the number of the entities and realize the construction of a valve opening knowledge graph. The second part provides a time protection mechanism algorithm for judging the optimal frequency of valve adjustment and reducing the valve adjustment times on one hand, and predicts the change trend of parameters through an LSTM prediction model fused with a convolutional neural network on the other hand, so that the valve is adjusted in time and the equipment risk is reduced. Experimental results show that the method effectively fuses mechanism knowledge and data knowledge, and achieves high efficiency and greening of management decision and production and manufacturing.
The present invention has been described in detail with reference to the examples, but the present invention is only preferred examples of the present invention and should not be construed as limiting the scope of the present invention. All equivalent changes and modifications made within the scope of the present invention shall fall within the scope of the present invention.
Claims (3)
1. A waste heat valve control optimization method based on fusion drive is characterized by comprising the following steps:
s1: data monitoring points are arranged on a grate cooler, an AQC boiler and a generator of a certain cement plant, the opening of a bypass valve, an air mixing flue adjusting valve, the front flue gas temperature of a combined superheater, the flue gas pressure of an outlet flue, the front flue gas pressure of the combined superheater, the main steam temperature and the high-pressure steam flow are collected, the collection time interval is 5ms, and the total number of historical data points is 39766;
s2: the method comprises the steps of abstracting the concept of waste heat recovery text data, defining the opening degree of a valve for some equipment states, and obtaining waste heat recovery mechanism knowledge;
s3: waste heat recovery mechanism knowledge and historical data are fused, and a knowledge graph model based on a fuzzy set is constructed;
s4: constructing an LSTM valve opening optimization model based on a time protection mechanism, determining the optimal adjusting frequency of the valve and predicting the variation trend of parameters;
s5: when new data exist, the valve adjusting frequency is judged, then the parameter change trend is predicted, knowledge reasoning is carried out through a knowledge graph model based on a fuzzy set, and the valve opening degree is recommended.
2. The fusion drive-based waste heat valve control optimization method according to claim 1, characterized in that: in the step S3, waste heat recovery mechanism knowledge and historical data are fused to construct a knowledge graph model based on a fuzzy set, and the method comprises the following steps:
1-1) obtaining characteristics influencing the opening of the valve based on a waste heat recovery principle and correlation analysis, sequentially extracting attributes of the characteristics from historical data, and forming an initial entity according to the characteristics and the characteristic attributes;
1-2) determining state sets V = { "high", "higher", "lower", "low" }accordingto words and phrases existing in more expert experiences, determining optimal value sets corresponding to the state sets for different characteristics, determining membership function,
for a variable whose state value is "high", its membership function takes the form:
wherein, A (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the state of the current entity being 'high', a is the optimal value corresponding to the state value 'high', b is the lowest acceptable characteristic value of the state 'high', c is the highest acceptable characteristic value of the state 'high', wherein b < a = c,
for variables whose state values are "higher" or "lower", the membership function takes the following form:
wherein: a (-) represents a membership function, x is the characteristic value of the current entity, A (x) is the membership of the current entity state of higher (lower), and a is the optimal value corresponding to the state value of higher (lower); b is the lowest value of the feature that is acceptable for a state of "higher" ("lower"); c is the highest value of the acceptable feature for which the state is "higher" ("lower"), where b < a < c,
for a variable whose state value is "low", its membership function takes the form:
wherein, a (·) represents a membership function, x is a characteristic value of the current entity, a (x) is a membership degree of the current entity state being "low", a is an optimal value corresponding to the state value "low", b is a lowest acceptable characteristic value of the state value "low", and c is a highest acceptable characteristic value of the state value "low", where b = a < c;
1-3) calculating the membership function of the entity, and then determining the state set of the entity. And giving different weights to the characteristics in different state sets according to the influence degree of the characteristics on the valve opening degree of the entity, and calculating the similarity between the entities. Is provided with
A＝(a 1 ,a 2 ,a 3 ,…,a n )
B＝(b 1 ,b 2 ,b 3 ,…,b n )
μ＝(μ 1 ,μ 2 ,μ 3 ,…,μ n )
Wherein, A and B are entity feature vectors, a and B are entity features, n is the number of the features of the entity, a 1 、a 2 、a 3 、a n Is a characteristic value of an entity, b 1 、b 2 、b 3 、b n Is the characteristic value of B entity, mu is the weight vector of A and B characteristics, mu 1 、μ 2 、μ 3 、μ n Are the weight values of the corresponding characteristics of A and B, S AB Is the similarity of A and B, a i Is the ith characteristic value of A, b i Is the i-th characteristic value of B, mu i The weight values corresponding to the ith characteristics of A and B;
1-4) after calculating the similarity between the entities, combining a plurality of entities with the similarity smaller than a set threshold into a set entity, wherein the feature vector of the set entity is the average value of a plurality of old entity vectors;
1-5) calculating the sum of enthalpies in the waste heat exhaust gas
As shown in the following formula:
h＝u+pv
wherein h is the enthalpy of the working medium, u is the internal energy of the substance, p is the pressure, v is the volume, and e is the enthalpy of the working medium
T is the working medium temperature, T 0 Is the ambient temperature, c p The specific heat capacity of the working medium;
1-6) according to the enthalpy sum of waste heat and exhaust gas
The quality of the data is determined, and the calculation formula is as follows:
Q＝λ 1 H+λ 2 E
wherein Q is mass, λ is weight, λ 1 As a weight corresponding to the enthalpy, λ 2 
Corresponding weights, H being enthalpy of data under similar conditions, E being enthalpy of data under similar conditions
To find λ 1 、λ 2 Establishing an objective function according to the idea of game theory, and seeking an optimal linear combination coefficient by taking the minimum sum of dispersion of index combination weight Q and H and E as a target
The index combining weight at this time is the optimal combining weight Q * . The objective function and constraint conditions are as follows:
min||Q-H|| 2 -||Q-E|| 2
s.t.λ 1 +λ 2 ＝1,λ 1 λ 2 ≥0
wherein | | | x | | | represents the L-2 norm of the calculation vector x, Q is the quality of the data under the similar working condition, H is the enthalpy of the data under the similar working condition, and E is the enthalpy of the data under the similar working condition
λ is the weight, λ 1 As a weight corresponding to the enthalpy, λ 2 
The corresponding weight of the weight is set to be,
based on the differential principle, the first derivative of the minimum can be obtained according to the above formula to obtain λ 1 、λ 2 A value of 1 、λ 2 By performing normalization processing
The final mass calculation formula is as follows:
wherein Q is * Which is the final quality of the data under similar conditions, lambda is the weight,
is the weight corresponding to the enthalpy,
corresponding weights, H being the enthalpy of the data under similar conditions, E being the enthalpy of the data under similar conditions
1-7) determining the classification number of continuous variables according to the goodness of fit of the variance and the contour coefficient, carrying out k-meas clustering on the classification number, and then selecting data with higher quality under similar working conditions and putting the data into a decision tree for classification to obtain the relationship between entities;
1-8) establishing a knowledge graph model based on the fuzzy set according to the entity and the relation.
3. The fusion drive-based waste heat valve control optimization method according to claim 1, characterized in that: in the step S4, an LSTM valve opening optimization model based on a time protection mechanism is constructed, the optimal adjustment frequency of the valve is determined, and the variation trend of the parameters is predicted, which includes the following steps:
2-1) constructing an LSTM prediction model fused with a convolutional neural network, wherein the model comprises a convolutional layer, a pooling layer, an LSTM layer, a Dropout layer and a BN layer, and training the model;
2-2) determining the time of a protection window according to the field working condition, if the temperature degree and the change speed do not exceed the specified threshold value in the time protection window, continuing monitoring, otherwise, jumping out of the time protection period;
2-3) boiler parameters within a preset time length before the time to be predicted are input into the trained model, and boiler parameters at the time to be predicted are predicted;
and 2-4) carrying out knowledge reasoning by a knowledge graph model based on a fuzzy set according to the prediction parameters of the boiler, and recommending the opening degree of the valve.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211318479.4A CN115681597A (en) | 2022-10-26 | 2022-10-26 | Fusion drive-based waste heat valve control optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211318479.4A CN115681597A (en) | 2022-10-26 | 2022-10-26 | Fusion drive-based waste heat valve control optimization method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115681597A true CN115681597A (en) | 2023-02-03 |
Family
ID=85098652
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211318479.4A Pending CN115681597A (en) | 2022-10-26 | 2022-10-26 | Fusion drive-based waste heat valve control optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115681597A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117032348A (en) * | 2023-10-10 | 2023-11-10 | 郯城众一科环化工有限公司 | Chemical reaction kettle temperature control method, system, equipment and storage medium |
-
2022
- 2022-10-26 CN CN202211318479.4A patent/CN115681597A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117032348A (en) * | 2023-10-10 | 2023-11-10 | 郯城众一科环化工有限公司 | Chemical reaction kettle temperature control method, system, equipment and storage medium |
| CN117032348B (en) * | 2023-10-10 | 2024-01-09 | 郯城众一科环化工有限公司 | Chemical reaction kettle temperature control method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11755976B2 (en) | Method for predicting oxygen load in iron and steel enterprises based on production plan | |
| CN106920006B (en) | Subway station air conditioning system energy consumption prediction method based on ISOA-LSSVM | |
| CN114692265B (en) | Zero-carbon building optimization design method based on deep reinforcement learning | |
| CN112066355B (en) | Self-adaptive adjusting method of waste heat boiler valve based on data driving | |
| CN110689171A (en) | Turbine health state prediction method based on E-LSTM | |
| CN113107626B (en) | Load prediction method of combined cycle generator set based on multivariable LSTM | |
| CN110837223A (en) | Combustion optimization control method and system for gas turbine | |
| CN113240184B (en) | Building space unit cold load prediction method and system based on federal learning | |
| CN110400018B (en) | Operation control method, system and device for coal-fired power plant pulverizing system | |
| CN115681597A (en) | Fusion drive-based waste heat valve control optimization method | |
| CN113762387A (en) | Data center station multi-load prediction method based on hybrid model prediction | |
| Wang et al. | Multi-criteria comprehensive study on predictive algorithm of heating energy consumption of district heating station based on timeseries processing | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN117515802A (en) | Fixed-frequency central air conditioner daily-load prediction method considering running state of air conditioner | |
| CN115526433A (en) | Power plant reheat flue gas baffle operation prediction method based on integrated hybrid model | |
| CN115392125A (en) | Temperature prediction method for rotary cement kiln | |
| CN114861555A (en) | Regional comprehensive energy system short-term load prediction method based on Copula theory | |
| CN114004146A (en) | Method for optimizing and determining flexible operation sliding pressure curve of heat supply unit based on all working conditions | |
| CN114202106A (en) | Air conditioning system load prediction method based on deep learning | |
| CN112992285A (en) | IPSO-HKELM-based blast furnace molten iron silicon content prediction method | |
| Li et al. | Research and application of data mining in power plant process control and optimization | |
| Li et al. | Optimization for Boiler Based on Data Mining and Multi-Condition Combustion Model | |
| Fan et al. | Short-term Heating Load Prediction of Heat Exchange Station Based on DAIGA-LSTM Neural Network | |
| CN117077508B (en) | Performance prediction and optimization method for thermoelectric generator |
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 |