CN103235885B - Thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule goes out force prediction method - Google Patents

Abstract

A kind of thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule goes out force prediction method, collection signal data form on-the-spot historical data base, and this database comprises six variablees: mill load BML, grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, mill ventilation amount BMV, pulverizer adequacy BMPC.To exert oneself forecast model according to obtained Database bowl mill, and training optimization is carried out to model parameter, the Accurate Prediction that pulverized coal preparation system is exerted oneself is realized finally by obtained forecast model, thus pulverized coal preparation system under unit current loads is estimated working time, so not only can reduce the workload of system maintenance, and the serviceable life of extension device, thus under the prerequisite of safe operation, improve the economic benefit of thermal power plant.

Description

Thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule goes out force prediction method

Technical field

The present invention relates to thermal power plant ball mill pulverized coal preparation system and go out force prediction method, be specifically related to a kind of thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule and go out force prediction method.

Background technology

In Power Plant in China, Coal-pulverizing System with Ball Mill is the main auxiliary equipment of boiler, applies very extensive.Meanwhile, Coal-pulverizing System with Ball Mill or highly energy-consuming equipment, power consumption accounts for 15% ~ 25% of station-service electricity, and its operational efficiency directly has influence on the economic benefit of thermal power plant.Exerting oneself of pulverized coal preparation system refers to that it prepares the ability of coal dust, weighed, and the size of powder process amount decides the height of powder level in coal powder banker by the number of powder process amount.When powder position reach a certain height, illustrate that the coal dust of preparation has met the demand of boiler combustion under current unit load, then stop the operation of pulverized coal preparation system, so both can reduce pulverized coal preparation system energy consumption can also serviceable life of extension device.

For the measurement of Mill output, classic method is all indirectly reflected by coal powder density.At present for the measurement of coal powder density, the more method of domestic and international application has heat balance method of, weight method, microwave method, photoelectric detection method, laser method, isokinetic sampling's method, mixing pressure differential method, velocity contrast platen press, fibre-optical probe method, induced potential method etc.Wherein, heat balance method of is the temperature of coal dust and air after the temperature of coal dust before temperature by measuring air before the mixing of wind powder, the mixing of wind powder, the mixing of wind powder, then law of conservation of energy is utilized to calculate the ratio of coal dust and air in pulverized coal channel, the method principle is simple, but measuring period is longer, real-time is poor.Weight method utilizes the powder-feeding amount in the impact flow meter measuring unit time of machine supplying powder outlet installation, adopt pressure differential method measuring wind simultaneously, thus the air capacity obtained in the unit interval and powder amount, finally obtain coal powder density, although the method principle is simple, impact flow meter wearing and tearing are comparatively serious.The different principle of the influence degree that microwave method utilizes the pulverized coal flow of variable concentrations to produce microwave signal is to measure coal powder density, although the method precision is higher, its equipment cost is high, but also needs the problem solving actual installation position.Photoelectric detection method is that moving particles is calculated coal powder density to the induced signal of light after opto-electronic conversion, analog to digital conversion.Laser method utilizes coal dust to the impact of laser luminous flux to measure coal powder density.Although above two kinds of precision are higher, the grey problem of probe knot limits the application of the method.Isokinetic sampling's method have measure accurately, the feature of highly versatile, but it can not be measured in real time and causes its limitation in application.Although mixing pressure reduction ratio juris is simple, and easily realizes, need the data of measurement more, calculated amount is larger.Velocity contrast platen press cannot be suitable for the situation that fluid has acceleration.Although fibre-optical probe method signal to noise ratio (S/N ratio) is high, simplicity of design, its measurement range has certain restriction.Induced potential method is not perfect in software and hardware.

As previously mentioned, for the measurement of coal powder density, although method is more, the shortcoming of various method and applicable elements cause they application on limitation.And said method is all adopt single signal to reflect Mill output, but pulverized coal preparation system is the object of a multivariate, non-linear, strong coupling, large dead time, its measurement model is not unalterable, in actual moving process, measurement model parameter also can be drifted about because of the change of ature of coal, coal, steel ball loading capacity, utilizes merely a kind of signal to exert oneself existing defects to reflect pulverized coal preparation system.

Summary of the invention

In order to overcome the deficiency that above-mentioned prior art exists, a kind of thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule is the object of the present invention is to provide to go out force prediction method, realize the Accurate Prediction to Mill output, thus pulverized coal preparation system under unit current loads is estimated working time, so not only system energy consumption can be reduced, and the serviceable life of extension device, thus under the prerequisite of safe operation, improve the economic benefit of thermal power plant.

In order to achieve the above object, the technical solution adopted in the present invention is:

Thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule goes out a force prediction method, and step is as follows:

Step 1: first thermal power plant ball mill pulverized coal preparation system collection signal data form on-the-spot historical data base D, this database D comprises six variablees: mill load BML, grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, mill ventilation amount BMV and pulverizer adequacy BMPC; Like this, database D is a sextuple database, wherein by mill load BML, and grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, set { BML, BMDP that mill ventilation amount BMV forms, BMIP, BMOT, BMV} are the set of Takagi-Sugeno Fuzzy rule former piece, and pulverizer adequacy BMPC is Takagi-Sugeno Fuzzy consequent;

Step 2: thermal power plant ball mill pulverized coal preparation system adopts cuclear density clustering method to carry out cluster analysis to database D, specific as follows:

The object p that first extraction one is untreated from database D is as initial object, and search for its epsilon neighborhood N ε (p) according to the measuring similarity between object, if the object that N ε (p) comprises is no less than Tm, then p is a kernel object and sets up new bunch of Cv, and is included in N ε (p) a little in Cv; ε and Tm is the threshold value of setting, and the measuring similarity expression formula of object is:

$d_H(a,b)=\sqrt{2(1-K(a,b\left)\right)}$

In formula, kernel function a, b are the object in database D, and δ is the width parameter of gaussian kernel function;

Then, using a certain object p' in the epsilon neighborhood of p as EXPANDING DISPLAY AREA, if p' is kernel object, the object in the epsilon neighborhood of p' and p' is classified as bunch Cv created by p, and claims p' directly can reach from p density, otherwise only p' is classified as bunch Cv created by p; If p' directly can reach from p density, and another object p " directly can reach from p' density, then claim p " can reach from p density; All objects that can reach from p density are all classified as bunch Cv created by p; Then, from input space database, again extract an object not being classified as certain bunch as initial object, repeat process above, until all objects are all included into certain bunch in database D;

Step 3: thermal power plant ball mill pulverized coal preparation system obtains C bunch by step 2, and namely the number of fuzzy language value and Takagi-Sugeno Fuzzy rule is C, and the barycenter of definition bunch is bunch heart of this bunch; Suppose { x ₁, x ₂, x ₃, x ₄, x ₅be the former piece input variable set of bowl mill, { y} is the consequent output variable set of bowl mill; Adopt Gaussian function as membership function, then the i-th rule be subordinate to angle value μ _ican be expressed as:

${\mathrm{\μ}}_i=\underset{j=1}{\overset{5}{\Π}}{e}^{-\frac{{(x_j-c_{ij})}^2}{2{{\mathrm{\σ}}_{ij}}^2}},i\∈\{1,2,...,C\},j\∈\{1,2,3,4,5\}$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, σ _ijdegree of membership width, degree of membership width cs _ijexpression formula be $\[max\left(x_j\right)-min\left(x_j\right)\]/\sqrt{8};$

Step 4: thermal power plant ball mill pulverized coal preparation system is according to being subordinate to angle value μ according to step 3 gained _icalculate linear coefficient k _ijand b _i:

$\left\{\begin{array}{c}{k}_{ij}=\frac{y}{x_j}\·\frac{\underset{i=1}{\overset{C}{\Σ}}{\mathrm{\μ}}_i}{{\mathrm{\μ}}_i}\\ b_i=y\·\frac{\underset{i=1}{\overset{C}{\Σ}}{\mathrm{\μ}}_i}{{\mathrm{\μ}}_i}\end{array}\right.$

Step 5: thermal power plant ball mill pulverized coal preparation system obtains being subordinate to angle value μ required for Modling model _iand linear coefficient k _ijand b _i, and set up initial Takagi-Sugeno Fuzzy Rule-Based Modeling; Thus, the prediction of model is exerted oneself output , i.e. pulverizer adequacy BMPC, can be expressed as:

$\hat{y}=\frac{\underset{i=1}{\overset{C}{\Σ}}{\mathrm{\μ}}_i{y}_i^{*}}{\underset{i=1}{\overset{C}{\Σ}}{\mathrm{\μ}}_i}$

In formula, $y_i^{*}=k_{i1}{x}_1+k_{i2}{x}_2+...+k_{i5}{x}_5+b_i;$

Step 6: thermal power plant ball mill pulverized coal preparation system is exerted oneself closer to actual pulverizer adequacy to make the prediction of model, adopts iteration optimization algorithms forecast model parameters to be carried out training optimization; If matrix P and S is as follows:

In formula, N=6C;

Step 7: thermal power plant ball mill pulverized coal preparation system is according to one group of training data setup parameter matrix $a={\[{\mathrm{\μ}}_1{x}_1^l,{\mathrm{\μ}}_1{x}_2^l,...,{\mathrm{\μ}}_1,{\mathrm{\μ}}_2{x}_1^l,{\mathrm{\μ}}_2{x}_2^l,...,{\mathrm{\μ}}_2,...,{\mathrm{\μ}}_C{x}_1^l,{\mathrm{\μ}}_C{x}_2^l,...,{\mathrm{\μ}}_C\]}^T,$ B=y ^l, wherein l ∈ 1,2 ..., Nt}, Nt are the group numbers of training data, then matrix P and S can be optimized by following methods:

$\left\{\begin{array}{c}P\←P+S\·\[a\·\left(b^T-a^{T}P\right)\]\\ S\←S-\frac{(S\·a)(a^T\·S)}{1+a^T\·(S\·a)}\end{array}\right.$

Matrix P initial value is set as null matrix, through Nt iteration, and the linear coefficient k after being finally optimized _ijand b _i; So far, the parameter k of forecast model _ijand b _ibe optimized;

Step 8: thermal power plant ball mill pulverized coal preparation system adopts root-mean-square error RMSE as the numerical indication weighing pulverizer adequacy model accuracy, and to set RMSE initial value be Th _e; The RMSE value of gained forecast model in calculation procedure 6, if it is greater than Th _e, then the parameter c continuing optimal prediction model is needed _ijand σ _ij;

Step 9: thermal power plant ball mill pulverized coal preparation system is according to training data l ∈ 1,2 ..., Nt}, obtains intermediate variable de _l:

${\mathrm{de}}_l=(-6\·C)\·\underset{i=1}{\overset{C}{\Σ}}\frac{-2{\mathrm{\μ}}_i{y}_i^{*}\·(y^l-{\hat{y}}^l)}{\underset{i=1}{\overset{C}{\Σ}}{\mathrm{\μ}}_i}$ In formula, be input as pulverizer adequacy forecast model export;

Step 10: thermal power plant ball mill pulverized coal preparation system hypothesis initial matrix T _cand T _σbe all 6 × C matrix, and the initial value of matrix element is zero; Then there is following formula:

$\left\{\begin{array}{c}{T}_c(i,j)=T_c(i,j)+\underset{l=1}{\overset{Nt}{\mathrm{\Σ}}}{\mathrm{de}}_l.\frac{(x_j^l-c_{ij})}{{{\mathrm{\σ}}_{ij}}^2}\·e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\σ}}_{ij}}^2}}\\ T_{\mathrm{\σ}}(i,j)=T_{\mathrm{\σ}}(i,j)+\underset{l=1}{\overset{Nt}{\mathrm{\Σ}}}{\mathrm{de}}_l.\frac{{(x_j^l-c_{ij})}^2}{{{\mathrm{\σ}}_{ij}}^3}\·e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\σ}}_{ij}}^2}}\end{array}\right.$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, σ _ijit is degree of membership width;

Step 11: thermal power plant ball mill pulverized coal preparation system is calculated by following formula, the prediction model parameters c be finally optimized _ijand σ _ij;

$\left\{\begin{array}{c}{c}_{ij}=c_{ij}-\frac{ss\·T_c(i,j)}{\sqrt{\underset{q=1}{\overset{6\·C}{\Σ}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\σ}}{\left(i,j\right)}^2)}}\\ {\mathrm{\σ}}_{ij}={\mathrm{\σ}}_{ij}-\frac{ss\·T_{\mathrm{\σ}}(i,j)}{\sqrt{\underset{q=1}{\overset{6\·C}{\Σ}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\σ}}{\left(i,j\right)}^2)}}\end{array}\right.$

In formula, ss is training pace, and its initial value is set as 0.1;

Step 12: in the training process, if RMSE changes according to certain ad hoc rules, then thermal power plant ball mill pulverized coal preparation system needs to revise the size needs of ss value; If RMSE (k) is the root-mean-square error of gained model after the training of kth step, then have:

If $\left\{\begin{array}{c}\mathrm{RMSE}\left(k\right)<\mathrm{RMSE}(k-1)\\ \mathrm{RMSE}(k-1)>\mathrm{RMSE}(k-2)\\ \mathrm{RMSE}(k-2)<\mathrm{RMSE}(k-3)\\ \mathrm{RMSE}(k-3)>\mathrm{RMSE}(k-4)\end{array}\right.,$ Then ss ← ssRd

If $\left\{\begin{array}{c}\mathrm{RMSE}\left(k\right)<\mathrm{RMSE}(k-1)\\ \mathrm{RMSE}(k-1)<\mathrm{RMSE}(k-2)\\ \mathrm{RMSE}(k-2)<\mathrm{RMSE}(k-3)\\ \mathrm{RMSE}(k-3)<\mathrm{RMSE}(k-4)\end{array}\right.,$ Then ss ← ssRi

Wherein, Rd is the attenuation rate of step-length ss, and its value is set as 0.9; Ri is the rate of growth of step-length ss, and its value is set as 1.1;

Step 13: thermal power plant ball mill pulverized coal preparation system is by iteration training step 5 ~ step 12, until RMSE reaches initial set value Th _eor train epochs reaches predetermined value; So far, thermal power plant ball mill pulverized coal preparation system finally obtains the prediction mould after parameter optimization

Step 14: thermal power plant ball mill pulverized coal preparation system obtains on the basis of current mill load, grinding machine gateway pressure reduction, mill entrance negative pressure, grinding machine outlet temperature and mill ventilation amount runtime value by measurement module, according to gained forecast model, realize the Accurate Prediction that pulverized coal preparation system is exerted oneself.

Described thermal power plant ball mill pulverized coal preparation system adopts the DDC system of PLC and computing machine composition, and gathers related process variable, and acquisition rate is more than 500ms.

From field notes data, according to mill load, grinding machine gateway pressure reduction, mill entrance negative pressure, grinding machine outlet temperature and mill ventilation amount, set up pulverizer adequacy forecast model, and model is optimized, realize the Accurate Prediction that pulverized coal preparation system is exerted oneself, thus pulverized coal preparation system is estimated working time, achieve pulverized coal preparation system and save energy and reduce the cost, improve the economic benefit of thermal power plant.

Embodiment

Below in conjunction with embodiment, the present invention will be described in more detail.

The thermal power plant ball mill pulverized coal preparation system that the present invention is based on Takagi-Sugeno Fuzzy rule goes out force prediction method, and step is as follows:

Step 1: first thermal power plant ball mill pulverized coal preparation system collection signal data form on-the-spot historical data base D, this database D comprises six variablees: mill load BML, grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, mill ventilation amount BMV and pulverizer adequacy BMPC.Like this, database D is a sextuple database, as shown in table 1, wherein by mill load BML, grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, set { the BML that mill ventilation amount BMV forms, BMDP, BMIP, BMOT, BMV} is the set of Takagi-Sugeno Fuzzy rule former piece, and pulverizer adequacy BMPC is Takagi-Sugeno Fuzzy consequent.

Table 1

No.	BML(％)	BMDP(Pa)	BMIP(Pa)	BMOT(℃)	BMV(Km 3/h)	BMPC(ton/h)
							1	55.25	2672.94	-687.57	87.3	144.88	44.47
2	55.51	2689.04	-658.25	87.3	144.92	44.47
							3	55.46	2700.39	-668.51	87.3	144.92	44.47
4	55.46	2700.39	-668.51	87.4	144.95	44.47
							5	55.33	2728.54	-674.15	87.4	144.93	44.47
6	55.87	2730.16	-675.87	87.4	144.93	44.47
							7	55.87	2730.16	-675.87	87.5	144.91	44.47
8	55.08	2743.88	-672.7	87.5	144.92	44.47
							9	55.34	2712.28	-691.35	87.5	144.92	44.47
10	55.34	2712.28	-691.35	87.5	145	44.47
							11	54.88	2725.89	-697.93	87.5	145.02	44.47
12	55.06	2745.32	-692.05	87.5	145.02	44.47
							13	55.06	2745.32	-692.05	87.6	145	44.47
14	55.22	2753.34	-694.15	87.6	144.93	44.47
							15	55.08	2785.38	-681.49	87.6	144.93	44.47
16	55.08	2785.38	-681.49	87.7	144.86	44.47
							17	55.87	2765.47	-690.9	87.7	144.77	44.47

18	55.02	2787.42	-676.8	87.7	144.77	44.47
							19	55.02	2787.42	-676.8	87.7	144.66	44.47
20	55.34	2778.81	-667.18	87.7	144.51	44.47
							21	55.54	2783.72	-662.72	87.7	144.51	44.47
22	55.54	2783.72	-662.72	87.7	144.34	44.47
							23	55.71	2729.54	-670.45	87.8	144.14	44.47
24	56.05	2726.98	-674.37	87.8	144.14	44.47
							25	56.05	2726.98	-674.37	87.8	143.98	44.47
26	55.26	2696.82	-688.14	87.9	143.92	44.47
							27	55.47	2726.88	-687.18	87.9	143.92	44.47
28	55.47	2726.88	-687.18	87.9	143.93	44.47
							29	55.36	2737.76	-685.44	87.9	144	44.47
30	55.42	2765.36	-685.41	87.9	144	44.47
							31	55.42	2765.36	-685.41	87.9	144.06	44.47
32	55.58	2766.21	-687.58	87.9	144.11	44.47
							33	54.8	2797.98	-682.56	87.9	144.11	44.47
34	54.8	2797.98	-682.56	88	144.23	44.47
							35	55.38	2783.22	-688.56	88	144.31	44.47
36	55.13	2792.63	-683.7	88	144.31	44.47
							37	55.13	2792.63	-683.7	88	144.33	44.47
38	55.34	2793.34	-681.08	88	144.31	44.47
							39	55.36	2786.09	-673.91	88	144.31	44.47

40	55.36	2786.09	-673.91	88.1	144.21	44.47
							41	55.1	2754.33	-686.86	88.1	144.05	44.47
42	55.85	2743.57	-684.63	88.1	144.05	44.47
							43	55.85	2743.57	-684.63	88.1	143.9	44.47
44	55.13	2742.07	-682.09	88.2	143.81	44.47
							45	55.51	2760.03	-681.84	88.2	143.81	44.47
46	55.51	2760.03	-681.84	88.2	143.81	44.47
							47	55.87	2728.92	-690.87	88.2	143.75	44.47
48	55.68	2727.49	-700.74	88.2	143.75	44.47
							49	55.68	2727.49	-700.74	88.2	143.65	44.47
50	55.94	2722.21	-710.29	88.2	143.67	44.47

Step 2: thermal power plant ball mill pulverized coal preparation system adopts cuclear density clustering method to carry out cluster analysis to database D, specific as follows:

The object p that first extraction one is untreated from database D is as initial object, and search for its epsilon neighborhood N ε (p) according to the measuring similarity between object, if the object that N ε (p) comprises is no less than Tm, then p is a kernel object and sets up new bunch of Cv, and is included in N ε (p) a little in Cv.ε and Tm is set as 0.6 and 12 respectively, and the measuring similarity expression formula of object is:

$d_H(a,b)=\sqrt{2(1-K(a,b\left)\right)}$

In formula, kernel function a, b are the object in database D, and δ is the width parameter of gaussian kernel function, and δ value is set as 150.

Then, using a certain object p' in the epsilon neighborhood of p as EXPANDING DISPLAY AREA, if p' is kernel object, the object in the epsilon neighborhood of p' and p' is classified as bunch Cv created by p, and claims p' directly can reach from p density, otherwise only p' is classified as bunch Cv created by p.If p' directly can reach from p density, and another object p " directly can reach from p' density, then claim p " can reach from p density.All objects that can reach from p density are all classified as bunch Cv created by p.Then, from input space database, again extract an object not being classified as certain bunch as initial object, repeat process above, until all objects are all included into certain bunch in database D.

Step 3: thermal power plant ball mill pulverized coal preparation system obtains 10 bunches by step 2, namely the number of fuzzy language value and Takagi-Sugeno Fuzzy rule is C, and the barycenter of definition bunch is bunch heart of this bunch.Bunch heart of each bunch is as shown in table 2:

Table 2

No.	BML(％)	BMDP(Pa)	BMIP(Pa)	BMOT(℃)	BMV(Km 3/h)
						Bunch 1	55.09	2746.19	-692.02	87.57	144.97
Bunches 2	55.25	2673.64	-687.33	87.31	144.86
						Bunches 3	55.33	2711.59	-691.18	87.52	144.90
Bunches 4	55.79	2727.98	-673.89	87.69	144.31
						Bunches 5	55.64	2765.04	-688.24	87.79	144.40
Bunches 6	55.05	2786.34	-679.14	87.67	144.80
						Bunches 7	55.46	2698.53	-667.16	87.35	144.92
Bunches 8	55.81	2730.74	-675.79	87.48	144.86
						Bunches 9	55.49	2782.40	-663.92	87.70	144.45
Bunches 10	55.41	2727.09	-687.98	87.86	144.03

Suppose { x ₁, x ₂, x ₃, x ₄, x ₅be the former piece input variable set of bowl mill, { y} is the consequent output variable set of bowl mill.Adopt Gaussian function as membership function, then the i-th rule be subordinate to angle value μ _ican be expressed as:

${\mathrm{\&mu;}}_i=\underset{j=1}{\overset{5}{\&Pi;}}{e}^{-\frac{{(x_j-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}},i\&Element;\{1,2,...,10\},j\&Element;\{1,2,3,4,5\}$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, degree of membership width cs _ijexpression formula be $\&lsqb;max\left(x_j\right)-min\left(x_j\right)\&rsqb;/\sqrt{8}.$

Step 4: thermal power plant ball mill pulverized coal preparation system is subordinate to angle value μ according to step 3 gained _icalculate linear coefficient k _ijand b _i:

$\left\{\begin{array}{c}{k}_{ij}=\frac{y}{x_j}\&CenterDot;\frac{\underset{i=1}{\overset{10}{\&Sigma;}}{\mathrm{\&mu;}}_i}{{\mathrm{\&mu;}}_i}\\ b_i=y\&CenterDot;\frac{\underset{i=1}{\overset{10}{\&Sigma;}}{\mathrm{\&mu;}}_i}{{\mathrm{\&mu;}}_i}\end{array}\right.$

Step 5: thermal power plant ball mill pulverized coal preparation system obtains being subordinate to angle value μ required for Modling model _iand linear coefficient k _ijand b _i, set up initial Takagi-Sugeno Fuzzy Rule-Based Modeling.Thus, the prediction of model is exerted oneself output , i.e. pulverizer adequacy BMPC, can be expressed as:

$\hat{y}=\frac{\underset{i=1}{\overset{10}{\&Sigma;}}{\mathrm{\&mu;}}_i{y}_i^{*}}{\underset{i=1}{\overset{10}{\&Sigma;}}{\mathrm{\&mu;}}_i}$

In formula, $y_i^{*}=k_{i1}{x}_1+k_{i2}{x}_2+...+k_{i5}{x}_5+b_i.$

Step 6: thermal power plant ball mill pulverized coal preparation system is exerted oneself closer to actual pulverizer adequacy to make the prediction of model, adopts iteration optimization algorithms forecast model parameters to be carried out training optimization.If matrix P and S is as follows:

In formula, N=60.

Step 7: thermal power plant ball mill pulverized coal preparation system is according to one group of training data setup parameter matrix $a={\&lsqb;{\mathrm{\&mu;}}_1{x}_1^l,{\mathrm{\&mu;}}_1{x}_2^l,...,{\mathrm{\&mu;}}_1,{\mathrm{\&mu;}}_2{x}_1^l,{\mathrm{\&mu;}}_2{x}_2^l,...,{\mathrm{\&mu;}}_2,...,{\mathrm{\&mu;}}_{10}{x}_1^l,{\mathrm{\&mu;}}_{10}{x}_2^l,...,{\mathrm{\&mu;}}_{10}\&rsqb;}^T,b=y^l,$ Wherein l ∈ 1,2 ..., 50}, 50 is group numbers of training data.Then matrix P and S can be optimized by following methods:

$\left\{\begin{array}{c}P\&LeftArrow;P+S\&CenterDot;\&lsqb;a\&CenterDot;\left(b^T-a^{T}P\right)\&rsqb;\\ S\&LeftArrow;S-\frac{(S\&CenterDot;a)(a^T\&CenterDot;S)}{1+a^T\&CenterDot;(S\&CenterDot;a)}\end{array}\right.$

Matrix P initial value is set as null matrix, through 50 iteration, and the linear coefficient k after being finally optimized _ijand b _i.So far, the linear coefficient k of forecast model _ijand b _ibe optimized.

Step 8: thermal power plant ball mill pulverized coal preparation system adopts root-mean-square error RMSE as the numerical indication weighing pulverizer adequacy model accuracy, and sets RMSE initial value Th _e=0.The RMSE value of gained forecast model in calculation procedure 6, if it is greater than Th _e, then the parameter c continuing optimal prediction model is needed _ijand σ _ij.

Step 9: thermal power plant ball mill pulverized coal preparation system is according to training data l ∈ 1,2 ..., 50}, obtains intermediate variable de _l:

${\mathrm{de}}_l=(-60)\&CenterDot;\underset{i=1}{\overset{10}{\&Sigma;}}\frac{-2{\mathrm{\&mu;}}_i{y}_i^{*}\&CenterDot;(y^l-{\hat{y}}^l)}{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i}$

In formula, be input as pulverizer adequacy forecast model export.

Step 10: thermal power plant ball mill pulverized coal preparation system hypothesis initial matrix T _cand T _σbe all 6 × 10 matrixes, and the initial value of matrix element is zero.Then there is following formula:

$\left\{\begin{array}{c}{T}_c(i,j)=T_c(i,j)+\underset{l=1}{\overset{50}{\mathrm{\&Sigma;}}}{\mathrm{de}}_l.\frac{(x_j^l-c_{ij})}{{{\mathrm{\&sigma;}}_{ij}}^2}\&CenterDot;e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}}\\ T_{\mathrm{\&sigma;}}(i,j)=T_{\mathrm{\&sigma;}}(i,j)+\underset{l=1}{\overset{50}{\mathrm{\&Sigma;}}}{\mathrm{de}}_l.\frac{{(x_j^l-c_{ij})}^2}{{{\mathrm{\&sigma;}}_{ij}}^3}\&CenterDot;e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}}\end{array}\right.$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, σ _ijit is degree of membership width.

Step 11: thermal power plant ball mill pulverized coal preparation system is calculated by following formula, the prediction model parameters c be finally optimized _ijand σ _ij.

$\left\{\begin{array}{c}{c}_{ij}=c_{ij}-\frac{ss\&CenterDot;T_c(i,j)}{\sqrt{\underset{q=1}{\overset{60}{\&Sigma;}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\&sigma;}}{\left(i,j\right)}^2)}}\\ {\mathrm{\&sigma;}}_{ij}={\mathrm{\&sigma;}}_{ij}-\frac{ss\&CenterDot;T_{\mathrm{\&sigma;}}(i,j)}{\sqrt{\underset{q=1}{\overset{60}{\&Sigma;}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\&sigma;}}{\left(i,j\right)}^2)}}\end{array}\right.$

In formula, ss is training pace, and its initial value is set as 0.1.

Step 12: in the training process, if RMSE changes according to certain ad hoc rules, then thermal power plant ball mill pulverized coal preparation system needs to revise the size needs of ss value.If RMSE (k) is the root-mean-square error of gained model after the training of kth step, then have:

If $\left\{\begin{array}{c}\mathrm{RMSE}\left(k\right)<\mathrm{RMSE}(k-1)\\ \mathrm{RMSE}(k-1)>\mathrm{RMSE}(k-2)\\ \mathrm{RMSE}(k-2)<\mathrm{RMSE}(k-3)\\ \mathrm{RMSE}(k-3)>\mathrm{RMSE}(k-4)\end{array}\right.,$ Then ss ← ssRd

If $\left\{\begin{array}{c}\mathrm{RMSE}\left(k\right)<\mathrm{RMSE}(k-1)\\ \mathrm{RMSE}(k-1)<\mathrm{RMSE}(k-2)\\ \mathrm{RMSE}(k-2)<\mathrm{RMSE}(k-3)\\ \mathrm{RMSE}(k-3)<\mathrm{RMSE}(k-4)\end{array}\right.,$ Then ss ← ssRi

Wherein, Rd is the attenuation rate of step-length ss, and its value is set as 0.9; Ri is the rate of growth of step-length ss, and its value is set as 1.1.

Step 13: thermal power plant ball mill pulverized coal preparation system is by iteration training step 5 ~ step 12, until RMSE reaches initial set value zero or train epochs reaches predetermined value 100.So far, thermal power plant ball mill pulverized coal preparation system finally obtains the forecast model after parameter optimization, and this forecast model is made up of 10 IF-THEN rules, specific as follows:

Rule 1:

IFBML belongs to a bunch 1ANDBMDP and belongs to a bunch 1ANDBMIP and belong to a bunch 1ANDBMOT and belong to a bunch 1ANDBMV and belong to bunch 1,

THEN

BMPC＝0.018BML-0.005BMDP-0.022BMIP+0.112BMOT+0.229BMV+0.0031

Rule 2:

IFBML belongs to a bunch 2ANDBMDP and belongs to a bunch 2ANDBMIP and belong to a bunch 2ANDBMOT and belong to a bunch 2ANDBMV and belong to bunches 2,

THEN

BMPC＝0.155BML-0.015BMDP-0.047BMIP+0.123BMOT+0.227BMV+0.0020

Rule 3:

IFBML belongs to a bunch 3ANDBMDP and belongs to a bunch 3ANDBMIP and belong to a bunch 3ANDBMOT and belong to a bunch 3ANDBMV and belong to bunches 3,

THEN

BMPC＝0.022BML+0.027BMDP+0.068BMIP+0.127BMOT+0.042BMV+0.0013

Rule 4:

IFBML belongs to a bunch 4ANDBMDP and belongs to a bunch 4ANDBMIP and belong to a bunch 4ANDBMOT and belong to a bunch 4ANDBMV and belong to bunches 4,

THEN

BMPC＝0.064BML+0.015BMDP+0.026BMIP+0.011BMOT+0.110BMV+0.0002

Rule 5:

IFBML belongs to a bunch 5ANDBMDP and belongs to a bunch 5ANDBMIP and belong to a bunch 5ANDBMOT and belong to a bunch 5ANDBMV and belong to bunches 5,

THEN

BMPC＝0.115BML+0.012BMDP+0.0002BMIP+0.060BMOT-0.007BMV+0.0010

Rule 6:

IFBML belongs to a bunch 6ANDBMDP and belongs to a bunch 6ANDBMIP and belong to a bunch 6ANDBMOT and belong to a bunch 6ANDBMV and belong to bunches 6,

THEN

BMPC＝0.267BML-0.003BMDP-0.006BMIP+0.469BMOT-0.040BMV+0.0082

Rule 7:

IFBML belongs to a bunch 7ANDBMDP and belongs to a bunch 7ANDBMIP and belong to a bunch 7ANDBMOT and belong to a bunch 1ANDBMV and belong to bunches 7,

THEN

BMPC＝0.180BML+0.018BMDP+0.006BMIP-0.014BMOT-0.054BMV+0.0009

Rule 8:

IFBML belongs to a bunch 8ANDBMDP and belongs to a bunch 8ANDBMIP and belong to a bunch 8ANDBMOT and belong to a bunch 8ANDBMV and belong to bunches 8,

THEN

BMPC＝0.411BML-0.012BMDP+0.009BMIP+0.340BMOT+0.215BMV+0.0031

Rule 9:

IFBML belongs to a bunch 9ANDBMDP and belongs to a bunch 9ANDBMIP and belong to a bunch 9ANDBMOT and belong to a bunch 9ANDBMV and belong to bunches 9,

THEN

BMPC＝-0.007BML+0.012BMDP-0.026BMIP-0.026BMOT-0.026BMV+0.0002

Rule 10:

IFBML belongs to a bunch 10ANDBMDP and belongs to a bunch 10ANDBMIP and belong to a bunch 10ANDBMOT and belong to a bunch 10ANDBMV and belong to bunches 10,

THEN

BMPC＝0.041BML-0.005BMDP-0.007BMIP+0.135BMOT+0.273BMV+0.0028

Step 14: thermal power plant ball mill pulverized coal preparation system obtains on the basis of current mill load, grinding machine gateway pressure reduction, mill entrance negative pressure, grinding machine outlet temperature and mill ventilation amount runtime value by measurement module, according to gained forecast model, realize the Accurate Prediction that pulverized coal preparation system is exerted oneself.

Described thermal power plant ball mill pulverized coal preparation system adopts the DDC system of PLC and computing machine composition, and gathers related process variable, and acquisition rate is more than 500ms.

Claims (2)

1. the thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule goes out a force prediction method, it is characterized in that: step is as follows:

Step 1: first thermal power plant ball mill pulverized coal preparation system collection signal data form on-the-spot historical data base D, this database D comprises six variablees: mill load BML, grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, mill ventilation amount BMV and pulverizer adequacy BMPC; Like this, database D is a sextuple database, wherein by mill load BML, and grinding machine gateway pressure reduction BMDP, mill entrance negative pressure BMIP, grinding machine outlet temperature BMOT, set { BML, BMDP that mill ventilation amount BMV forms, BMIP, BMOT, BMV} are the set of Takagi-Sugeno Fuzzy rule former piece, and pulverizer adequacy BMPC is Takagi-Sugeno Fuzzy consequent;

Step 2: thermal power plant ball mill pulverized coal preparation system adopts cuclear density clustering method to carry out cluster analysis to database D, specific as follows:

The object p that first extraction one is untreated from database D is as initial object, and search for its epsilon neighborhood N ε (p) according to the measuring similarity between object, if the object that N ε (p) comprises is no less than Tm, then p is a kernel object and sets up new bunch of Cv, and is included in N ε (p) a little in Cv; ε and Tm is the threshold value of setting, and the measuring similarity expression formula of object is:

$d_H(a,b)=\sqrt{2(1-K(a,b\left)\right)}$

In formula, kernel function a, b are the object in database D, and δ is the width parameter of gaussian kernel function;

Then, using a certain object p' in the epsilon neighborhood of p as EXPANDING DISPLAY AREA, if p' is kernel object, the object in the epsilon neighborhood of p' and p' is classified as bunch Cv created by p, and claims p' directly can reach from p density, otherwise only p' is classified as bunch Cv created by p; If p' directly can reach from p density, and another object p " directly can reach from p' density, then claim p " can reach from p density; All objects that can reach from p density are all classified as bunch Cv created by p; Then, from input space database, again extract an object not being classified as certain bunch as initial object, repeat process above, until all objects are all included into certain bunch in database D;

Step 3: thermal power plant ball mill pulverized coal preparation system obtains C bunch by step 2, and namely the number of fuzzy language value and Takagi-Sugeno Fuzzy rule is C, and the barycenter of definition bunch is bunch heart of this bunch; Suppose { x ₁, x ₂, x ₃, x ₄, x ₅be the former piece input variable set of bowl mill, { y} is the consequent output variable set of bowl mill; Adopt Gaussian function as membership function, then the i-th rule be subordinate to angle value μ _ican be expressed as:

${\mathrm{\&mu;}}_i=\underset{j=1}{\overset{5}{\&Pi;}}{e}^{-\frac{{(x_j-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}},i\&Element;\{1,2,...,C\},j\&Element;\{1,2,3,4,5\}$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, σ _ijdegree of membership width, degree of membership width cs _ijexpression formula be:

Step 4: thermal power plant ball mill pulverized coal preparation system is according to being subordinate to angle value μ according to step 3 gained _icalculate linear coefficient k _ijand b _i:

$\left\{\begin{array}{c}{k}_{ij}=\frac{y}{x_j}\&CenterDot;\frac{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i}{{\mathrm{\&mu;}}_i}\\ b_i=y\&CenterDot;\frac{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i}{{\mathrm{\&mu;}}_i}\end{array}\right.$

Step 5: thermal power plant ball mill pulverized coal preparation system obtains being subordinate to angle value μ required for Modling model _iand linear coefficient k _ijand b _i, and set up initial Takagi-Sugeno Fuzzy Rule-Based Modeling; Thus, the prediction of model is exerted oneself output i.e. pulverizer adequacy BMPC, can be expressed as:

$\hat{y}=\frac{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i{y}_i^{*}}{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i}$

In formula, $y_i^{*}=k_{i1}{x}_1+k_{i2}{x}_2+...+k_{i5}{x}_5+b_i;$

Step 6: thermal power plant ball mill pulverized coal preparation system is exerted oneself closer to actual pulverizer adequacy to make the prediction of model, adopts iteration optimization algorithms forecast model parameters to be carried out training optimization; If matrix P and S is as follows:

In formula, N=6C _;

Step 7: thermal power plant ball mill pulverized coal preparation system is according to one group of training data setup parameter matrix $a={\&lsqb;{\mathrm{\&mu;}}_1{x}_1^l,{\mathrm{\&mu;}}_1{x}_2^l,...,{\mathrm{\&mu;}}_1,{\mathrm{\&mu;}}_2{x}_1^l,{\mathrm{\&mu;}}_2{x}_2^l,...,{\mathrm{\&mu;}}_2,...,{\mathrm{\&mu;}}_C{x}_1^l,{\mathrm{\&mu;}}_C{x}_2^l,...,{\mathrm{\&mu;}}_C\&rsqb;}^T,$ B=y ^l, wherein l ∈ 1,2 ..., Nt}, Nt are the group numbers of training data, then matrix P and S can be optimized by following methods:

$\left\{\begin{array}{c}P\&LeftArrow;P+S\&CenterDot;\&lsqb;a\&CenterDot;\left(b^T-a^{T}P\right)\&rsqb;\\ S\&LeftArrow;S-\frac{(S\&CenterDot;a)(a^T\&CenterDot;S)}{1+a^T\&CenterDot;(S\&CenterDot;a)}\end{array}\right.$

Matrix P initial value is set as null matrix, through Nt iteration, and the linear coefficient k after being finally optimized _ijand b _i; So far, the parameter k of forecast model _ijand b _ibe optimized;

Step 8: thermal power plant ball mill pulverized coal preparation system adopts root-mean-square error RMSE as the numerical indication weighing pulverizer adequacy model accuracy, and to set RMSE initial value be Th _e; The RMSE value of gained forecast model in calculation procedure 6, if it is greater than Th _e, then the parameter c continuing optimal prediction model is needed _ijand σ _ij;

Step 9: thermal power plant ball mill pulverized coal preparation system is according to training data l ∈ 1,2 ..., Nt}, obtains intermediate variable de _l:

${\mathrm{de}}_l=(-6\&CenterDot;C)\&CenterDot;\underset{i=1}{\overset{C}{\&Sigma;}}\frac{-2{\mathrm{\&mu;}}_i{y}_i^{*}\&CenterDot;(y^l-{\hat{y}}^l)}{\underset{i=1}{\overset{C}{\&Sigma;}}{\mathrm{\&mu;}}_i}$

In formula, be input as pulverizer adequacy forecast model export;

Step 10: thermal power plant ball mill pulverized coal preparation system hypothesis initial matrix T _cand T _σbe all 6 × C matrix, and the initial value of matrix element is zero; Then there is following formula:

$\left\{\begin{array}{c}{T}_c(i,j)=T_c(i,j)+\underset{l=1}{\overset{Nt}{\mathrm{\&Sigma;}}}{\mathrm{de}}_l.\frac{(x_j^l-c_{ij})}{{{\mathrm{\&sigma;}}_{ij}}^2}\&CenterDot;e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}}\\ T_{\mathrm{\&sigma;}}(i,j)=T_{\mathrm{\&sigma;}}(i,j)+\underset{l=1}{\overset{Nt}{\mathrm{\&Sigma;}}}{\mathrm{de}}_l.\frac{{(x_j^l-c_{ij})}^2}{{{\mathrm{\&sigma;}}_{ij}}^3}\&CenterDot;e^{-\frac{{(x_j^l-c_{ij})}^2}{2{{\mathrm{\&sigma;}}_{ij}}^2}}\end{array}\right.$

In formula, c _ijthe jth dimension data value of the i-th bunch of heart, σ _ijit is degree of membership width;

Step 11: thermal power plant ball mill pulverized coal preparation system is calculated by following formula, the prediction model parameters c be finally optimized _ijand σ _ij;

$\left\{\begin{array}{c}{c}_{ij}=c_{ij}-\frac{ss\&CenterDot;T_c(i,j)}{\sqrt{\underset{q=1}{\overset{6\&CenterDot;C}{\&Sigma;}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\&sigma;}}{\left(i,j\right)}^2)}}\\ {\mathrm{\&sigma;}}_{ij}={\mathrm{\&sigma;}}_{ij}-\frac{ss\&CenterDot;T_{\mathrm{\&sigma;}}(i,j)}{\sqrt{\underset{q=1}{\overset{6\&CenterDot;C}{\&Sigma;}}(T_c{\left(i,j\right)}^2+T_{\mathrm{\&sigma;}}{\left(i,j\right)}^2)}}\end{array}\right.$

In formula, ss is training pace, and its initial value is set as 0.1;

Step 12: in the training process, if RMSE changes according to certain ad hoc rules, then thermal power plant ball mill pulverized coal preparation system needs to revise the size needs of ss value; If RMSE (k) is the root-mean-square error of gained model after the training of kth step, then have:

If $\left\{\begin{array}{c}RMSE\left(k\right)<RMSE(k-1)\\ RMSE(k-1)>RMSE(k-2)\\ RMSE(k-2)<RMSE(k-3)\\ RMSE(k-3)>RMSE(k-4)\end{array}\right.,$ Then ss ← ssRd

If $\left\{\begin{array}{c}RMSE\left(k\right)<RMSE(k-1)\\ RMSE(k-1)<RMSE(k-2)\\ RMSE(k-2)<RMSE(k-3)\\ RMSE(k-3)<RMSE(k-4)\end{array}\right.,$ Then ss ← ssRi

Wherein, Rd is the attenuation rate of step-length ss, and its value is set as 0.9; Ri is the rate of growth of step-length ss, and its value is set as 1.1;

Step 13: thermal power plant ball mill pulverized coal preparation system is by iteration training step 5 ~ step 12, until RMSE reaches initial set value Th _eor train epochs reaches predetermined value; So far, thermal power plant ball mill pulverized coal preparation system finally obtains the forecast model after parameter optimization;

Step 14: thermal power plant ball mill pulverized coal preparation system obtains on the basis of current mill load, grinding machine gateway pressure reduction, mill entrance negative pressure, grinding machine outlet temperature and mill ventilation amount runtime value by measurement module, according to gained forecast model, realize the Accurate Prediction that pulverized coal preparation system is exerted oneself.

2. the thermal power plant ball mill pulverized coal preparation system based on Takagi-Sugeno Fuzzy rule according to claim 1 goes out force prediction method, it is characterized in that: described thermal power plant ball mill pulverized coal preparation system adopts the DDC system of PLC and computing machine composition, and related process variable is gathered, acquisition rate is more than 500ms.