Summary of the invention
Technical matters to be solved by this invention provides a kind of method of ultimate analysis of coal.Through neural network method, the recessive error of bringing to system because the ultimate analysis of power plant's coal lacks has been controlled in the foundation of the Nonlinear Mapping model from the coal industry analysis to the ultimate analysis effectively.
In order to solve the problems of the technologies described above, the present invention adopts following technical scheme:
A kind of ultimate analysis of coal method based on numerical model is characterized in that it may further comprise the steps:
1) obtains as received basis carbon (C_ar) content of coal
1.1) at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt carbon is analyzed data conversion analyze data for the as received basis carbon; Secondly data sample is divided into training sample and test sample book;
1.2) set up the artificial nerve network model of as received basis carbon (C_ar) content of coal:
The input layer of network is: as received basis volatile content (V_ar), as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a four-dimension;
The latent layer of network is the neuron node that contains some;
The output layer of network has only an output neuron node promptly to export as received basis carbon element content (C_ar);
1.3) confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis carbon (C_ar) content;
1.4) utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model;
1.5) test sample book is loaded into the artificial nerve network model of as received basis carbon (C_ar) content and tests;
1.6) if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
It further may further comprise the steps:
2) obtain as received basis protium (H_ar) content of coal
2.1) at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt protium is analyzed data conversion analyze data for the as received basis protium; Secondly data sample is divided into training sample and test sample book;
2.2) set up the artificial nerve network model of as received basis protium (H_ar) content of coal;
The input layer of network is: as received basis volatile content (V_ar), as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a four-dimension;
The latent layer of network is the neuron node that contains some;
The output layer of network has only an output neuron node promptly to export as received basis protium content (H_ar);
2.3) confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis protium (H_ar) content;
2.4) utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model;
2.5) test sample book is loaded into the artificial nerve network model of as received basis protium (H_ar) content and tests;
2.6) if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
It further may further comprise the steps:
3) obtain as received basis nitrogen element (N_ar) content of coal
3.1) at first collect the data sample that technical analysis and the elements corresponding of coal are analyzed, and be as received basis nitrogen ultimate analysis data with empty butt nitrogen ultimate analysis data conversion; Secondly data sample is divided into training sample and test sample book;
3.2) set up the artificial nerve network model of as received basis nitrogen element (N_ar) content of coal:
The input layer of network: as received basis volatile content (V_ar), as received basis total moisture content (M_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a three-dimensional;
The latent layer of network is the neuron node that contains some;
The output layer of network has only an output neuron node promptly to export as received basis nitrogen element content (N_ar);
3.3) confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis nitrogen element (N_ar) content;
3.4) utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model;
3.5) test sample book is loaded into the artificial nerve network model of as received basis nitrogen element (N_ar) content and tests;
3.6) if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
It further may further comprise the steps:
4) obtain as received basis element sulphur (S_ar) content of coal
4.1) at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt element sulphur is analyzed data conversion analyze data for the as received basis element sulphur; Secondly data sample is divided into training sample and test sample book.
4.2) set up the artificial nerve network model of as received basis element sulphur (S_ar) content of coal:
The input layer of network: as received basis volatile content (V_ar), as received basis total moisture content (M_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a three-dimensional,
The latent layer of network is the neuron node that contains some,
The output layer of network has only an output neuron node promptly to export as received basis element sulphur content (S_ar);
4.3) confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis element sulphur (S_ar) content;
4.4) utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model;
4.5) test sample book is loaded into the artificial nerve network model of as received basis element sulphur (S_ar) content and tests;
4.6) if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
It further may further comprise the steps:
5) obtain as received basis oxygen element (O_ar) content of coal
5.1) at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt oxygen element is analyzed data conversion analyze data for the as received basis oxygen element; Secondly data sample is divided into training sample and test sample book;
5.2) set up the artificial nerve network model of as received basis oxygen element (O_ar) content of coal:
The input layer of network is: as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis moisture (M_ar) are the input vectors of a three-dimensional;
The latent layer of network is the neuron node that contains some;
The output layer of network has only an output neuron node promptly to export as received basis oxygen element content (O_ar);
5.3) confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis oxygen element (O_ar) content;
5.4) utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model;
5.5) test sample book is loaded into the artificial nerve network model of as received basis oxygen element (O_ar) content and tests;
5.6) if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The present invention compared with prior art has following profitable fruit: the multiple linear regression model from the coal industry analysis to the ultimate analysis, nonlinear multivariable regression model and Neural Network Based Nonlinear mapping model have been carried out deep comparative study.See that from the result of comprehensive comparative analysis the Nonlinear Mapping model of neural network is superior to traditional Linear Regression Model in One Unknown, multiple linear regression model and nonlinear multivariable regression model on precision of prediction.In addition, the Nonlinear Mapping model of neural network also will be optimized traditional multivariate regression model on coal adaptability.Pass through neural network method; The recessive error of bringing to system because the ultimate analysis of power plant's coal lacks has been controlled in the foundation of the Nonlinear Mapping model from the coal industry analysis to the ultimate analysis effectively, for solid foundation has been established in the research and development of large-scale thermal power machine group energy-saving management system.
Embodiment
Modeling method based on numerical value mainly contains multiple linear regression analysis method, multiple nonlinear regression method, neural net method and support vector machine method etc., and the present invention has carried out deep comparative study to the multiple linear regression model from the coal industry analysis to the ultimate analysis, nonlinear multivariable regression model and Neural Network Based Nonlinear mapping model.See that from the result of comprehensive comparative analysis the Nonlinear Mapping model of neural network is superior to traditional Linear Regression Model in One Unknown, multiple linear regression model and nonlinear multivariable regression model on precision of prediction.In addition, the Nonlinear Mapping model of neural network also will be optimized traditional multivariate regression model on coal adaptability.Artificial neural network is a kind of multilayer feedforward neural network, and the principal feature of this network is the transmission of signal forward direction, error back propagation.In forward direction transmitted, input signal was successively handled through hidden layer from input layer, until input layer.The neuron state of each layer only influences one deck neuron state down.If output layer can not get desired output, then change backpropagation over to, constantly approach desired output according to predicated error adjustment network weight and the output of threshold values BP neural network prediction.
The present invention asks the method for ultimate analysis of coal specific as follows:
1. obtain as received basis carbon (C_ar) content of coal
The concrete steps of as received basis (C_ar) carbon element content of obtaining coal are mainly following:
1.1 at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt carbon analyzed data conversion analyze data for the as received basis carbon; Secondly data sample is divided into training sample and test sample book.
Coal industry analysis and ultimate analysis data sample see the following form
1.2 set up the artificial nerve network model of as received basis carbon (C_ar) content of coal.
The topological structure of the artificial neural network of as received basis carbon (C_ar) content of coal is as shown in Figure 1.
The input layer of network is: as received basis volatile content (V_ar), as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a four-dimension.
The latent layer of network is the neuron node that contains some.
The output layer of network has only an output neuron node promptly to export as received basis carbon element content (C_ar).
1.3 confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis carbon (C_ar) content.
Concrete example: the training precision of the artificial nerve network model of as received basis carbon (C_ar) content be made as 0.001 or maximum frequency of training be made as 200000 times.
1.4 utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model.The training error curve of the artificial nerve network model of as received basis carbon (C_ar) content is referring to Fig. 6.
In the accompanying drawing 6,
The realistic accuracy of Performance is 0.242362--sample training is 0.242362, and the aimed at precision of Goal is 0.001-sample training is 0.001;
20000 Epochs-frequency of training (20000 times);
The graph of errors of Training blue-sample training process, the aimed at precision curve of Goal black-sample training.
1.5 test sample book is loaded into the artificial nerve network model of as received basis carbon (C_ar) content and tests.The test sample book graph of errors of the artificial nerve network model of as received basis carbon (C_ar) content is referring to Fig. 7.
In the accompanying drawing 7
The actual error % of as received basis carbon (C_ar) content of dta-C-ar (%)-test sample book;
The actual error number percent % of as received basis carbon (C_ar) content of dta-C-ar-percent (%)-test sample book;
The quantity of test-num--test sample book.
1.6 if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The test sample book of table 1 as received basis carbon (C_ar) error
To be higher than multiple linear regression model and nonlinear multivariable regression model from the precision of the artificial nerve network model of the visible as received basis carbon (C_ar) of table 1.
2. that obtains coal receives protium (H_ar) content
The concrete steps of as received basis protium (H_ar) content of obtaining coal are mainly following:
2.1 at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt protium analyzed data conversion analyze data for the as received basis protium; Secondly data sample is divided into training sample and test sample book.
Coal industry analysis and ultimate analysis data sample see the following form
2.2 set up the artificial nerve network model of as received basis protium (H_ar) content of coal.
The topological structure of the artificial neural network of as received basis protium (H_ar) content of coal is as shown in Figure 2.
The input layer of network is: as received basis volatile content (V_ar), as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a four-dimension.
The latent layer of network is the neuron node that contains some.
The output layer of network has only an output neuron node promptly to export as received basis protium content (H_ar).
2.3 confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis protium (H_ar) content.
Concrete example: the training precision of the artificial nerve network model of as received basis protium (H_ar) content be made as 0.001 or maximum frequency of training be made as 200000 times.
2.4 utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model.The training error curve of the artificial nerve network model of as received basis protium (H_ar) content is referring to Fig. 8.
In the accompanying drawing 8,
The realistic accuracy of Performance is 0.00355929--sample training is 0.00355929, and the aimed at precision of Goal is0.001-sample training is 0.001;
20000 Epochs-frequency of training (20000 times);
The graph of errors of Training blue-sample training process, the aimed at precision curve of Goal black-sample training.
2.5 test sample book is loaded into the artificial nerve network model of as received basis protium (H_ar) content and tests.The graph of errors of the test sample book of the artificial nerve network model of as received basis protium (H_ar) content is referring to Fig. 9.
In the accompanying drawing 9
The actual error % of as received basis protium (H_ar) content of dta-H-ar (%)-test sample book;
The actual error number percent % of as received basis protium (H_ar) content of dta-H-ar-percent (%)-test sample book;
The quantity of test-num--test sample book.
2.6 if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The test sample book of table 2 as received basis protium (H_ar) error
To be higher than multiple linear regression model and nonlinear multivariable regression model from the precision of the artificial nerve network model of the visible as received basis protium (H_ar) of table 2.
3. obtain as received basis nitrogen element (N_ar) content of coal
The concrete steps of as received basis nitrogen element (N_ar) content of obtaining coal are mainly following:
3.1 at first collect the data sample that technical analysis and the elements corresponding of coal are analyzed, and be as received basis nitrogen ultimate analysis data with empty butt nitrogen ultimate analysis data conversion; Secondly data sample is divided into training sample and test sample book.
Coal industry analysis and ultimate analysis data sample see table
3.2 set up the artificial nerve network model of as received basis nitrogen element (N_ar) content of coal.
The topological structure of the artificial neural network of as received basis nitrogen element (N_ar) content of coal is as shown in Figure 3.
The input layer of network: as received basis volatile content (V_ar), as received basis total moisture content (M_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a three-dimensional.
The latent layer of network is the neuron node that contains some.
The output layer of network has only an output neuron node promptly to export as received basis nitrogen element content (N_ar).
3.3 confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis nitrogen element (N_ar) content.
Concrete example: the training precision of the artificial nerve network model of as received basis nitrogen element (N_ar) content be made as 0.001 or maximum frequency of training be made as 200000 times.
3.4 utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model.The training error curve of the artificial nerve network model of as received basis nitrogen element (N_ar) content is referring to Figure 10.
In the accompanying drawing 10,
The realistic accuracy of Performance is 0.0197912--sample training is 0.0197912, and the aimed at precision of Goal is0.001-sample training is 0.001;
20000 Epochs-frequency of training (20000 times);
The graph of errors of Training blue-sample training process, the aimed at precision curve of Goal black-sample training.
3.5 test sample book is loaded into the artificial nerve network model of as received basis nitrogen element (N_ar) content and tests.The graph of errors of the test sample book of the artificial nerve network model of as received basis nitrogen element (N_ar) content is referring to Figure 11.
In the accompanying drawing 11
The actual error % of as received basis nitrogen element (N_ar) content of dta-N-ar (%)-test sample book;
The actual error number percent % of as received basis nitrogen element (N_ar) content of dta-N-ar-percent (%)-test sample book;
The quantity of test-num--test sample book.
3.6 if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The test sample book of table 3 as received basis nitrogen element (N_ar) error
To be higher than multiple linear regression model and nonlinear multivariable regression model from the precision of the artificial nerve network model of the visible as received basis nitrogen element (N_ar) of table 3.
4. obtain as received basis element sulphur (S_ar) content of coal
The concrete steps of as received basis element sulphur (S_ar) content of obtaining coal are mainly following:
4.1 at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt element sulphur analyzed data conversion analyze data for the as received basis element sulphur; Secondly data sample is divided into training sample and test sample book.
Coal industry analysis and ultimate analysis data sample see the following form
4.2 set up the artificial nerve network model of as received basis element sulphur (S_ar) content of coal.
The topological structure of the artificial neural network of as received basis element sulphur (S_ar) content of coal is as shown in Figure 4.
The input layer of network: as received basis volatile content (V_ar), as received basis total moisture content (M_ar), as received basis net calorific value (Q_net_ar) are the input vectors of a three-dimensional.
The latent layer of network is the neuron node that contains some.
The output layer of network has only an output neuron node promptly to export as received basis element sulphur content (S_ar).
4.3 confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis element sulphur (S_ar) content.
Concrete example: the training precision of the artificial nerve network model of as received basis element sulphur (S_ar) content be made as 0.001 or maximum frequency of training be made as 200000 times.
4.4 utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model.The training error curve of the artificial nerve network model of as received basis element sulphur (S_ar) content is referring to Figure 12.
In the accompanying drawing 12,
The realistic accuracy of Performance is 0.00157218--sample training is 0.00157218, and the aimed at precision of Goal is0.001-sample training is 0.001;
20000 Epochs-frequency of training (20000 times);
The graph of errors of Training blue-sample training process, the aimed at precision curve of Goal black-sample training.
4.5 test sample book is loaded into the artificial nerve network model of as received basis element sulphur (S_ar) content and tests.The graph of errors of the test sample book of the artificial nerve network model of as received basis element sulphur (S_ar) content is referring to Figure 13.
In the accompanying drawing 13
The actual error % of as received basis element sulphur (S_ar) content of dta-S-ar (%)-test sample book;
The actual error number percent % of as received basis element sulphur (S_ar) content of dta-S-ar-percent (%)-test sample book;
The quantity of test-num--test sample book.
4.6 if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The test sample book of table 4 as received basis element sulphur (S_ar) error
To be higher than multiple linear regression model and nonlinear multivariable regression model from the precision of the artificial nerve network model of the visible as received basis element sulphur (S_ar) of table 4.
5. obtain as received basis oxygen element (O_ar) content of coal
The concrete steps of as received basis oxygen element (O_ar) content of obtaining coal are mainly following:
5.1 at first collect the technical analysis of coal and the data sample that elements corresponding is analyzed, and empty butt oxygen element analyzed data conversion analyze data for the as received basis oxygen element; Secondly data sample is divided into training sample and test sample book.
Coal industry analysis and oxygen element are analyzed data sample
5.2 set up the artificial nerve network model of as received basis oxygen element (O_ar) content of coal.
The topological structure of the artificial neural network of as received basis oxygen element (O_ar) content of coal is as shown in Figure 5.
The input layer of network is: as received basis fixed carbon content (FC_ar), as received basis ash content (A_ar), as received basis moisture (M_ar) are the input vectors of a three-dimensional.
The latent layer of network is the neuron node that contains some.
The output layer of network has only an output neuron node promptly to export as received basis oxygen element content (O_ar).
5.3 confirm the training precision or the maximum frequency of training of the artificial nerve network model of as received basis oxygen element (O_ar) content.
Concrete example: the training precision of the artificial nerve network model of as received basis oxygen element (O_ar) content be made as 0.001 or maximum frequency of training be made as 200000 times.
5.4 utilize training sample that artificial nerve network model is trained after model satisfies that training precision requires or the training of model reached maximum frequency of training, stop training to model.The training error curve of the artificial nerve network model of as received basis oxygen element (O_ar) content is referring to Figure 14.
In the accompanying drawing 14,
The realistic accuracy of Performance is 0.0489628--sample training is 0.0489628, and the aimed at precision of Goal is0.001-sample training is 0.001;
20000 Epochs-frequency of training (20000 times);
The graph of errors of Training blue-sample training process, the aimed at precision curve of Goal black-sample training.
5.5 test sample book is loaded into the artificial nerve network model of as received basis oxygen element (O_ar) content and tests.The graph of errors of the test sample book of the artificial nerve network model of as received basis oxygen element (O_ar) content is referring to Figure 15.
In the accompanying drawing 15
The actual error % of as received basis oxygen element (O_ar) content of dta-O-ar (%)-test sample book;
The actual error number percent % of as received basis oxygen element (O_ar) content of dta-O-ar-percent (%)-test sample book;
The quantity of test-num--test sample book.
5.6 if the display model as a result of test has satisfied the requirement of precision and generalization ability two aspects, then model training success.
The test sample book of table 5 as received basis oxygen element (O_ar) error
See that from table 5 precision of the artificial nerve network model of as received basis oxygen element (O_ar) will be higher than multiple linear regression model and nonlinear multivariable regression model.