CN102799151A - Statistical-classification-based method for real-time balance adjustment of metallurgical gas system - Google Patents

Abstract

A real-time balance adjustment method for metallurgical gas systems based on statistical classification, which is characterized by firstly applying a Gaussian process classifier to separate the time corresponding to the adjustment unit data into adjustment time and non-adjustment time, and separating the generation unit data and consumption unit data corresponding to the adjustment time The data and the adjusted unit are used as samples of the fuzzy rule base to establish an adjustment sample base; then use the fuzzy C-means clustering algorithm to convert each sample in the sample base into an If-Then fuzzy rule to establish a complete fuzzy rule base; when monitoring When the gas system is about to run unbalanced at a certain moment, the gas generation unit data and consumption unit data at that moment are converted into If-Then fuzzy rules using the fuzzy C-means clustering algorithm, and compared with the established fuzzy rule base, Determine the adjustable unit at this moment; then use the difference calculation method to obtain the adjusted total amount of the gas system; finally, according to the priority of the gas system adjustment unit and the maximum load capacity of each adjustment unit, distribute the adjusted total amount to different adjustment units , to realize the real-time balance adjustment of the metallurgical gas system.

Description

A real-time balance adjustment method for metallurgical gas system based on statistical classification

technical field

The invention belongs to the field of information technology, relates to a Gaussian process, a fuzzy rule base and difference calculation, and is a real-time balance adjustment method for a metallurgical gas system based on statistical classification. The present invention utilizes a large amount of existing data at the metallurgical enterprise site, first uses a Gaussian process classifier to separate the time corresponding to the adjustment unit data into an adjustment time and a non-adjustment time, and separates the generation unit data, consumption unit data and adjusted unit corresponding to the adjustment time As a sample of the fuzzy rule base, establish an adjustment sample base; then use the fuzzy C-means clustering algorithm to convert each sample in the adjustment sample base into an If-Then fuzzy rule, and establish a complete fuzzy rule base; A moment is about to run out of balance, the gas generation unit data and consumption unit data at this moment are converted into If-Then fuzzy rules using the fuzzy C-means clustering algorithm, compared with the established fuzzy rule base, and the moment is determined Adjustable units; then use the difference calculation method to obtain the adjusted total amount of the gas system; finally, according to the priority of the gas system adjustment unit and the maximum load capacity of each adjustment unit, the adjusted total amount is distributed to different adjustment units to realize the metallurgical gas system. Real-time balance adjustment of the system.

Background technique

Energy is the most important resource other than human resources in the production process of metallurgical enterprises. Whether the operation of the energy system is stable, economical and reliable will directly affect the quality of products and the economic benefits of enterprises. Therefore, how to efficiently utilize the by-product gas produced in the production process to make the enterprise low consumption, low emission, low cost, and low pollution operation has become the goal pursued by metallurgical enterprises (Jun Zhao, Quanli Liu, Wei Wang, WitoldPedrycz, and Liqun Cong, Hybrid Neural Prediction and Optimized Adjustment for Coke Oven Gas System in Steel Industry [J]. IEEE trans. on neural networks and learning systems., vol.23, no.3, pp.439-450, Mar.2012).

The production process of metallurgical enterprises will produce a large amount of by-product gas, and such by-products are also important secondary energy sources that can be used in coking, heating furnaces, power plants, heat treatment and other links. Their effective and reasonable use will directly affect the energy consumption standards of metallurgical enterprises and output costs. By-product gas includes coke oven gas, blast furnace gas and converter gas. Due to the irregularity of iron and steel production, the gas system often has an unbalanced situation of production and consumption. Too much or too little production and consumption margin can easily cause the counter to exceed the upper or lower limit, which affects system safety. Therefore, the gas must be monitored. Adjust effectively. Under normal circumstances, when the gas cabinet is abnormal, according to the change trend of the production and consumption units of the gas system, increasing or reducing the gas consumption of the adjustable unit in the pipeline network can balance the gas cabinet, and the most important thing is to ensure normal production Under the premise of safe operation of gas cabinets and gas cabinets, the utilization rate of gas can be maximized, the emission of by-product gas can be reduced, and production costs can be reduced.

At present, in the actual production process, dispatchers often rely on manual experience to achieve the balance adjustment of the entire gas system. Referring to the actual situation on site, dispatchers generally estimate whether adjustments are needed based on the actual operation of the counters at the current moment and the changes in the production and consumption of each unit in the system, and roughly calculate the total amount of adjustment, and then adjust the total amount based on manual experience. If the adjusted operating state of the gas system meets the expected requirements, the adjustment will stop, otherwise it will continue to adjust until the requirements are met. The disadvantage of this method is that the dispatcher needs to complete a huge workload. Although the total amount of adjustment can be given, the specific adjustment plan cannot be given directly and accurately. This adjustment method relying on manual experience needs to be constantly adjusted. It will cause a lag in the response of the adjustment means, and the balance adjustment of the gas system cannot be completed in a timely and effective manner. It may also lead to abnormal production of the enterprise and affect the production efficiency of the enterprise.

Contents of the invention

The technical problem to be solved by the invention is the real-time balance adjustment problem of the metallurgical gas system. In order to solve the above problems, the Gaussian process classifier is first used to separate the adjustment unit data corresponding to the adjustment time and the adjustment unit data corresponding to the non-adjustment time, and the generation unit data, consumption unit data and adjusted unit corresponding to the adjustment time are used as fuzzy rules The sample of the library is used to establish the adjustment sample library, and the fuzzy rule library is established based on the adjustment sample library to determine the adjustment unit; then the total amount of gas adjustment is obtained through the difference calculation method; finally, the priority of the adjustment unit and each adjustment unit are adjusted according to the on-site gas system. The maximum load capacity of the unit, and distribute the adjustment amount to different adjustment units. By using the invention, a better adjustment scheme can be provided in a short time for the reference of dispatchers, and the balance adjustment of the metallurgical gas system can be realized.

The overall implementation process of the technical solution of the present invention is shown in Figure 1, and the specific steps are as follows:

1. On-site data reading: read the required gas system adjustment unit data, adjusted unit, generation unit data, consumption unit data, and gas cabinet position data from the on-site real-time database of the metallurgical gas system;

2. Gaussian process binary classification model: use Gaussian process classifier to separate the time corresponding to the adjustment unit data obtained in the first step into adjustment time and non-adjustment time, and record the adjustment unit data corresponding to the adjustment time;

3. Establish the fuzzy rule base: use the data of the generation unit and the consumption unit corresponding to the adjustment time obtained in the second step as the input sample of the fuzzy rule base, and use the adjusted unit corresponding to the adjustment time as the output sample of the fuzzy rule base, Establish an adjusted sample library, use the fuzzy C-means clustering algorithm to convert each sample in the adjusted sample library into If-Then fuzzy rules, and establish a relatively complete fuzzy rule library;

4. Real-time online determination of the adjustment unit: monitor the moment when the gas system is unbalanced, use the fuzzy C-means clustering algorithm to convert the corresponding generation unit data and consumption unit data into If-Then fuzzy rules, and step 3 Compare the established fuzzy rule base to find the most similar fuzzy rule in the fuzzy rule base, and its output is the adjustable unit at the current moment;

5. Calculation of the total amount of adjustment: use the difference calculation method to obtain the total amount of gas that needs to be adjusted;

6. Allocate the adjustment amount of each adjustment unit: according to the adjustment unit obtained in step 4, according to the priority of the on-site adjustment unit and the maximum load capacity of each adjustment unit, distribute the adjustment amount obtained in step 5 to different adjustment units .

Effect and benefit of the present invention are:

Considering the blindness of on-site dispatchers to determine the adjustment unit by manual experience, the present invention adopts a method based on the combination of Gaussian process classifier and fuzzy rule base to determine the adjustment unit on-line in real time, which can effectively avoid the dependence of the system on manual experience, and effectively The speed of determining the adjustment unit is greatly improved, the problem of adjustment lag is solved, and the rational utilization and distribution of gas is realized, thereby realizing the automation and intelligent operation of industrial production;

The present invention makes full use of the existing on-site data of metallurgical enterprises to accurately determine the adjustment unit in real time, and after obtaining the total adjustment amount by using the differential calculation method, according to the priority of the adjustment unit and the maximum load capacity of each adjustment unit, the adjustment total The amount is distributed to different adjustment units to meet the real-time and stability requirements of the site, thus providing a complete and feasible adjustment plan for the dispatchers of the gas system.

Description of drawings

Figure 1 is a flow chart of the overall implementation of the technical solution.

Figure 2 is a structural diagram of the gas system pipe network in metallurgical enterprises.

Figure 3 is a schematic diagram of the calculation of the total adjustment of the gas system.

Detailed ways

In order to better understand the technical solution of the present invention, the embodiment of the present invention will be described in detail below in conjunction with accompanying drawing 2, and accompanying drawing 2 is the pipe network structural diagram of the gas system of a certain metallurgical enterprise, blast furnace, coke oven and converter are gas system Generating unit, the blast furnace gas, coke oven gas and converter gas produced by it are pressurized by the pressurization station, and then supplied to consumption units such as lime kiln, sintering plant, continuous casting, cold rolling, hot rolling, steel pipe plant, and blooming. The gas will be supplied to low-pressure boilers and power plant boilers to generate steam and electricity. Low-pressure boilers and power plant boilers are adjustable units of the gas system. They are important adjustment means to ensure system balance in the gas system. The gas cabinet connected to the pipe network is the storage device of the gas system. Under normal circumstances, the gas system will operate in a state of balance between production and consumption. Sometimes, due to changes in industrial production or industrial failures, the gas system will be unbalanced. When the supply exceeds demand, the number of gas cabinets may exceed its operating limit. In this case, it is necessary to open the release tower to release the excess gas. When the supply exceeds the demand, it may cause the stagnation of industrial production. Therefore, when it is detected that an abnormal situation is about to occur, it is necessary to adjust the gas system in time to make the gas system reach a new balance. At present, the dispatchers of the on-site gas system monitor the operation status of the gas system in real time to judge whether the gas system needs to be adjusted in the future. This not only has a large workload, but also depends on the experience of dispatchers, which can easily lead to a lag in adjustment. Therefore, the present invention proposes a real-time balance adjustment method for metallurgical gas systems based on statistical classification to realize automatic analysis, automatic control and adjustment of metallurgical gas systems. auto-adjust. According to the method flow shown in Fig. 1, the specific implementation steps of the present invention are as follows:

Step 1: Reading of Field Data

Read the required gas system adjustment unit data, adjusted unit type, generation unit data, consumption unit data, and gas cabinet position data from the on-site real-time database of the metallurgical enterprise.

For the convenience of description, the meaning of the adjustment time is first given: within a few minutes, if the gas cabinet position is always at a higher or lower state, the amount of any adjustment unit suddenly changes, indicating that the adjustment unit performs a balance adjustment on the gas system , then the moment corresponding to this mutation point is called the adjustment moment. Regarding the characteristics of the adjustment time, the applicant of the present invention communicated with the on-site personnel of Shanghai Baosteel Energy Center and concluded that the adjustment time of the gas system has the following characteristics:

(1) The gas cabinet has been in a high or low state for a long time. If it is not adjusted, the gas cabinet is likely to exceed the limit for a period of time in the future, which will affect the safety of the system.

(2) If the consumption of one or several adjustment units suddenly changes from a steady state at this time, balance the gas system to make the gas cabinet position gradually tend to normal.

In general, when the gas cabinet is in a high state, the amount of the adjustment unit will be suddenly increased, and when the gas cabinet is in a low state, the amount of the adjustment unit will be suddenly reduced to balance and adjust the gas system to ensure The system is functioning normally.

Step 2: Build a Gaussian process binary classification model

Read the known adjustment unit data to form a sample set, calculate the slope of the points with the same interval in each sample in the sample set, and use all the obtained slope values as the input samples of the Gaussian process binary classification model, and the output samples are each The category labels corresponding to input samples, the data of input samples and output samples

The sample set is expressed as D={(xi _, y _i )| _i=1,...,N }, where the input sample x _i ∈ R ^d , the output sample y _i ∈ {-1, 1}, y _i = 1 indicates that the time corresponding to the sample is an adjustment time, y _i =-1 indicates that the time corresponding to this sample is a non-adjustment time, and N is the number of input samples.

In the case of a given input sample x _i , in order to calculate the output sample y _i , introduce an implicit function f, let f _i =f(xi ₎ , then record all implicit function values of the implicit function as f=[f ₁ , f ₂ ,..., f _N ] ^T . It is known that there is the following dependency p(y _i |f _i )=Φ(y _i f _i ) between the output sample y _i and each implicit function value f _i , and each output sample y _i is independent of each other, then the output sample’s The joint likelihood function can be described as:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; \&Phi;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Given the hyperparameter θ, according to the Bayesian criterion, the posterior distribution of the implicit function value f is expressed as:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; \&Phi;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Among them, X is the input sample set, recorded as X=[x ₁ ,...,x _N ]; assuming that the prior distribution of each implicit function value f _i is a zero-mean Gaussian distribution, then given the input sample set X, the implicit function The joint distribution of values should obey the multivariate Gaussian distribution, that is, p(f|X, θ)=N(f|0, K); K is the covariance matrix of the implicit function value f, and each element in K is defined as K _ij =K(x _i , x _j , θ), K is a positive definite covariance function. Given a detection sample x _* , and wanting the posterior probability distribution of its implicit function value f _* , the posterior distribution of the implicit function value f in formula (2) can be marginalized, then the posterior distribution of f _* can be expressed as :

p(f _* |D, θ, x _* )=∫p(f _* |f, X, θ, x _* )p(f|D, θ)df (3)

Substituting the posterior distribution of the implicit function value f in formula (2) into formula (3), the approximate posterior probability is obtained

$\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

The mean and variance are:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; K</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; K</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; AK</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Here k _* is the covariance of detection sample x _* and input sample set X, k _* = [K(x ₁ , x _* ),..., K(x _N , x _* )] ^T ; m and A are implicit The function value f obeys the mean and variance of the Gaussian distribution. Use q(f|D, θ) to represent the approximate distribution of the implicit function value f when the data sample set D, satisfying q(f|D, θ)=N(f| m, A). According to formula (4) and the expected calculation formula, given the detection sample x _* , the approximate value of the probability that the output sample y _* belongs to category 1 can be expressed as:

$\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&Integral;</span>\mathrm{<span\; class="notranslate">\; \&Phi;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}_{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&Phi;</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{<span\; class="notranslate">\; *</span>}}{\sqrt{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{<span\; class="notranslate">\; *</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

θ is the hyperparameter of the model, which has a great influence on the probability estimation. Therefore, when the model is constructed, the hyperparameter θ needs to be determined in advance. The present invention adopts the maximum likelihood function estimation method to obtain the hyperparameter θ, and the maximum likelihood estimation method The optimal hyperparameter is found by solving the maximum value of the likelihood function of the hyperparameter, such as formula (7)

p(D|θ)＝∫p(y|f)p(f|X，θ)df (7)

For the detection sample x _* of the binary classification, it can be set that when the probability q(y _* = 1 | D, θ, x _* ) > 0.5 of x _* belonging to the positive class, it is classified as the positive class, that is, at this moment adjustment moment; otherwise, it is classified as a negative class, that is, the moment is a non-adjustment moment. The classification accuracy comparison of different methods is shown in Table 1:

Table 1 Comparison of classification accuracy among Laplace method, expectation propagation method and support vector machine method

Step 3: Build a fuzzy rule base

Establish a fuzzy rule base based on If-Then fuzzy rules, the specific process is as follows:

① Use the generation unit data and consumption unit data corresponding to the adjustment time obtained in step 2 as the input samples of the fuzzy rule base, and the corresponding output samples are the adjusted unit types corresponding to the adjustment time;

② Divide the input sample space and output sample space into fuzzy domains, that is, through the analysis of input samples and output samples, the optimal clustering number of gas system generation unit data and consumption unit data can be obtained, so that it can fully reflect the input samples Characteristics;

③ Use the fuzzy C-means clustering algorithm to cluster the input samples, and record the category to which each segment of data belongs, and generate an initial fuzzy rule base based on If-Then fuzzy rules;

③④Record the membership degree of each category to which each piece of data belongs;

⑤ Simplify the fuzzy rule base, eliminate the fuzzy rules of the same input sample and the same output sample, and merge the fuzzy rules of the same input sample and different output samples into one fuzzy rule, indicating that there are multiple adjustment units at the current moment to adjust the balance of the gas system.

Step 4: Determine the adjustment unit online in real time

Monitor the moment when the gas system is unbalanced, use the fuzzy C-means clustering algorithm to convert the corresponding generation unit data and consumption unit data into If-Then fuzzy rules, and compare them with the fuzzy rule base established in step 3. Find the closest fuzzy rule to the fuzzy rule base, and its output sample is the adjustable unit at the current moment.

Take out some data points that need to be adjusted from the field database for verification experiments. The error statistics of the experimental data are shown in Table 2:

Table 2 Error statistics table

The adjusted units of the gas system in the present invention are represented by letters, which are specifically expressed as: A-the No. 1 boiler of the power plant, B-the No. 2 boiler of the power plant, C-the No. 4 generator, and D-the low-pressure boiler.

Step 5: Calculation of adjusted totals

The adjusted total amount of the gas system is obtained by using the differential calculation method, and the specific process is as follows in conjunction with Figure 3:

During the time period when the gas cabinet exceeds the limit, three points t ₁ , t ₂ and t ₃ are selected;

The value of the gas tank level at the instant is gh _i , then the value of the gas tank level at time t ₁ can be described as:

${\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="gh"\; filenames="HDA00001853496100031.png"\; state="\{\{state\}\}">gh</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; gh</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="t"\; filenames="HDA00001853496100031.png"\; state="\{\{state\}\}">t</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; dflow</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

Among them, dflow ₁ (t) is the flow difference between the generation unit data and the consumption unit data of the gas system at time t, and gh ₂ and gh ₃ can also be obtained in the same way; if the gas cabinet is adjusted to the normal level, the target value gh _o , define the flow difference between the generated unit data and the consumed unit data of the adjusted gas system at time t as dflow _o (t), then gh _o can be expressed as:

${\mathrm{<span\; class="notranslate">\; gh</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; gh</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="t"\; filenames="HDA00001853496100031.png"\; state="\{\{state\}\}">t</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; dflow</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

Subtract formula (8) and (9) to get formula (10), further, formula (10) can be abbreviated as formula (11);

${\mathrm{<span\; class="notranslate">\; gh</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; gh</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; dflow</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; dflow</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

Δgh ₁ =t ₁ ·Δdflow ₁ (11)

In this way, when the over-limit points of the three cabinets are adjusted to the target value, the change of flow difference can be obtained: Δdflow ₁ = Δgh ₁ /t ₁ , Δdflow ₂ = Δgh ₂ /t ₂ , Δdflow ₃ = Δgh ₃ / _t3 .

Further obtain the adjusted total amount according to formula (12).

Δdflow=max{Δdflow ₁ , Δdflow ₂ , Δdflow ₃ } (12)

Step 6: Distribute the adjustment amount of each adjustment unit

According to the adjustment units obtained in step 4, according to the priority of the on-site adjustment units and the maximum load capacity of each adjustment unit, the total amount of adjustment obtained in step 5 is allocated to different adjustment units.

Claims (1)

1. real-time balance adjustment method of metallurgical coal gas system based on statistical classification, its characteristic comprises the steps:

(1) from metallurgical coal gas system for field real-time data base read required adjustment unit data, by adjustment unit classification, generating unit data, consumable unit data, gas-holder location data;

(2) set up Gaussian process two disaggregated models

${\mathrm{\&mu;}}_{*}=k_{*}^T{K}^{-1}m,$ ${\mathrm{\&sigma;}}_{*}^2=K(x_{*},x_{*})-k_{*}^T(K^{-1}-K^{-1}{\mathrm{AK}}^{-1})k_{*}---\left(5\right)$

According to formula (4) and expectation computing formula, given detection sample x _*, output sample y then _*The approximate table that belongs to the probability of classification 1 is shown:

$q(y_{*}=1|D,\mathrm{\&theta;},x_{*})=\&Integral;\mathrm{\&Phi;}\left(f_{*}\right)N\left(f_{*}\right|{\mathrm{\&mu;}}_{*},{\mathrm{\&sigma;}}_{*}^2){\mathrm{df}}_{*}=\mathrm{\&Phi;}\left(\frac{{\mathrm{\&mu;}}_{*}}{\sqrt{1+{\mathrm{\&sigma;}}_{*}^2}}\right)---\left(6\right)$

Adopt the maximum likelihood function estimation technique to ask for ultra parameter θ, i.e. the maximal value of the likelihood function through finding the solution ultra parameter, and then the ultra parameter of seeking is optimumly seen formula (7)

p(D|θ)＝∫p(y|f)p(f|X，θ)df (7)

Detection sample x for two classification _*, set and work as x _*Belong to positive type probability q (y _*=1|D, θ, x _*)＞0.5 o'clock then is divided into positive type with it, promptly should be constantly for adjustment constantly; Otherwise, it is divided into negative type, promptly should constantly be the non-adjustment moment;

(3) set up fuzzy rule base

1. the corresponding constantly generating unit data and the input sample of consumable unit data of adjustment that step (2) classification obtained as fuzzy rule base, corresponding output sample for adjustment constantly correspondence by the adjustment unit classification;

2. will import sample space and output sample spatial division is fuzzy field, promptly through to input sample, output sample analysis, draws the best cluster number of coal gas system generating unit data and consumable unit data, makes it can complete reflection import the characteristic of sample;

3. use the fuzzy C-means clustering algorithm to the input sample section of carrying out cluster, and write down the classification under every segment data, produce initial fuzzy rule base based on the If-Then fuzzy rule;

4. write down the degree of membership of affiliated each classification of every segment data;

5. simplify fuzzy rule base, reject identical input sample, the fuzzy rule of identical output sample, with identical input sample, the fuzzy rule of different output samples is merged into a fuzzy rule, explains that current time has a plurality of adjustment units that the coal gas system balance is adjusted;

(4) real-time online is confirmed adjustment unit

The monitoring coal gas system moves the unbalanced moment; Generating unit data and consumable unit data that this is corresponding constantly; Use the fuzzy C-means clustering algorithm to turn to the If-Then fuzzy rule; Compare with the fuzzy rule base that step (3) is set up; Find out with fuzzy rule base in the most close fuzzy rule, but its output sample is exactly the adjustment unit of current time;

(5) calculating of adjustment total amount adopts the acquisition of Difference Calculation method to need the coal gas total amount of adjustment, and idiographic flow is following:

In the time period of transfiniting in the gas chamber position, select three cabinet positions t that transfinites ₁, t ₂, t ₃Suppose that the gas chamber place value at initial time t=0 is gh _i, so at t ₁The gas chamber place value is described as constantly:

${\mathrm{gh}}_1={\mathrm{gh}}_i+\underset{t=1}{\overset{t_1}{\mathrm{\&Sigma;}}}{\mathrm{dflow}}_1\left(t\right)---\left(8\right)$

In like manner, try to achieve gh ₂And gh ₃

If the desired value gh when normal level is adjusted in the gas chamber position _o, defining adjusted coal gas system is dflow in the flow difference of t moment generating unit data and consumable unit data _o(t), gh so _oBe expressed as:

${\mathrm{gh}}_o={\mathrm{gh}}_i+\underset{t=1}{\overset{t_1}{\mathrm{\&Sigma;}}}{\mathrm{dflow}}_o\left(t\right)---\left(9\right)$

With formula (8) and (9) subtract each other formula (10), further, formula (10) is write a Chinese character in simplified form an accepted way of doing sth (11);

${\mathrm{gh}}_1-{\mathrm{gh}}_o=\underset{t=0}{\overset{t_1}{\mathrm{\&Sigma;}}}[{\mathrm{dflow}}_1\left(t\right)-{\mathrm{dflow}}_o\left(t\right)]---\left(10\right)$

Δgh ₁＝t ₁·Δdflow ₁ (11)

Trying to achieve with this transfinites three cabinet positions when a little being transferred to desired value the variable quantity of flow difference:

Δdflow ₁＝Δgh ₁/t ₁，

Δdflow ₂＝Δgh ₂/t ₂，

Δdflow ₃＝Δgh ₃/t ₃；

Further obtain the adjustment total amount according to formula (12):

Δdflow＝maX{Δdflow ₁，Δdflow ₂，Δdflow ₃} (12)

(6) distribute the adjustment amount of each adjustment unit

According to the adjustment unit that obtains in the step (4),, the adjustment total amount that step (5) obtains is distributed to different adjustment units according to the priority of on-the-spot adjustment unit and the peak load ability of each adjustment unit.

Images