阅读说明：本技术 一种基于小样本暂态数据的汽轮机高调门流量特性优化方法 (Steam turbine high throttle flow characteristic optimization method based on small sample transient data ) 是由 霍红岩 于海存 辛晓钢 张谦 秦成果 张国斌 殷建华 范志强 周磊 郭瑞君 杜荣 于 2019-11-19 设计创作，主要内容包括：本发明公开了一种基于小样本暂态数据的汽轮机高调门流量特性优化方法,通过采样汽轮机变负荷过程中的暂态数据,再对所采集的数据进行清洗和聚类分析,得到各个负荷工况下的样本数据集；根据汽轮机调节级压力和进汽量之间的正比关系,对样本数据集中的汽轮机进汽量进行修正,得到相同边界条件下汽轮机进汽量；最后采用最小二乘拟合算法,对进汽流量与总阀门指令进行线性化拟合,并结合原有流量与阀门指令之间的关系,对汽轮机高调门流量特性曲线进行修正,得到线性化后的汽轮机各高调门配汽曲线。(The invention discloses a steam turbine high throttle flow characteristic optimization method based on small sample transient data, which comprises the steps of sampling transient data in a variable load process of a steam turbine, cleaning and clustering the acquired data to obtain a sample data set under each load working condition; correcting the steam inlet quantity of the steam turbine with concentrated sample data according to the proportional relation between the pressure of the regulating stage of the steam turbine and the steam inlet quantity to obtain the steam inlet quantity of the steam turbine under the same boundary condition; and finally, performing linear fitting on the steam inlet flow and the total valve instruction by adopting a least square fitting algorithm, and correcting the flow characteristic curve of the high-speed regulating valve of the steam turbine by combining the relation between the original flow and the valve instruction to obtain the linear steam distribution curve of each high-speed regulating valve of the steam turbine.)

1. A steam turbine high throttle flow characteristic optimization method based on small sample transient data is characterized by comprising the following steps:

the method comprises the following steps: acquiring small sample transient data, namely acquiring the small sample transient data of a unit under the continuously variable load working condition of 50% -100% load in a second-level sampling period based on historical data stored in a DCS;

step two: cleaning historical data, namely, adopting an isolated forest algorithm to check and remove abnormal values in the small sample transient data acquired in the step one, and adopting a linear difference method to supplement;

step three: clustering historical data, namely, taking the load of the unit as a dividing basis of a clustering center of mass, selecting initial clustering center points at intervals of 1MW from the historical data cleaned in the second step according to the variation amplitude of the load, and obtaining a target data set of steam parameters and valve position instructions of the unit under different load working conditions by adopting a PAM (pulse amplitude modulation) clustering algorithm;

step four: correcting the steam inlet quantity of main steam of the steam turbine, selecting the main steam pressure and the regulating stage corresponding to the target data set obtained in the step three under the highest load working condition as reference values, and correcting the steam inlet quantity of the steam turbine in the target data set to the steam inlet quantity of the steam turbine corresponding to the reference values according to the direct ratio between the pressure and the steam flow of the regulating stage of the unit;

step five: optimizing a high governing valve steam distribution curve, and performing linear fitting on the steam turbine steam admission quantity corrected in the step four and the total valve position instruction by adopting a least square algorithm to obtain an expected linear relation curve between the steam turbine steam admission quantity and the total valve position instruction;

step six: and (4) correcting the high governing valve steam distribution curve, and correcting and optimizing the command function of distributing the total valve position command to each single valve by utilizing the linear relation between the steam inlet quantity of the steam turbine and the corresponding total valve position command obtained by the fitting in the step five and combining the original high governing valve flow characteristic curve of the steam turbine to obtain a new steam distribution function of the steam turbine valve.

2. The method for optimizing the high governing valve flow characteristic of the steam turbine based on the small sample transient data according to claim 1, wherein the sampling period in the first step is 1 second, and the acquired parameters comprise: the system comprises a unit power, a main steam flow, a main steam pressure, a main steam temperature, a regulation stage pressure, a high-pressure cylinder exhaust temperature, a total valve position opening, a GV1 valve position opening, a GV2 valve position opening, a GV3 valve position opening and a GV4 valve position opening.

3. The steam turbine high governing valve flow characteristic optimization method based on small sample transient data according to claim 1, wherein the isolated forest algorithm in the second step is composed of iTrees, binary search trees are constructed in a multi-iteration mode, the binary search trees are composed into a forest, and the construction process comprises the following steps:

(1) randomly extracting m sub-samples from n training data, and putting the m sub-samples into a root node of an isolated tree;

(2) randomly appointing a dimension, and randomly generating a cutting point p in the range of the current node data, wherein the cutting point is generated between the maximum value and the minimum value of the appointed dimension in the current node data;

(3) the selection of the cutting point generates a hyperplane, and the data space of the current node is divided into 2 subspaces: placing points smaller than p in the currently selected dimension on the left branch of the current node, and placing points larger than or equal to p on the right branch of the current node;

(4) recursion steps 2 and 3 are carried out on the left branch node and the right branch node of the node, new leaf nodes are continuously constructed until only one piece of data (the cutting can not be continued) is arranged on the leaf nodes or the tree grows to the set height;

(5) recording the number of edges passing from the root node to the leaf node of each data, namely the path length h (x), so as to calculate the abnormal index S (x, n) of the data:

c(n)＝2H(n-1)-(2(n-1)/n)

here, h (n) ═ ln (n) + ξ ═ 0.5772156649 is an euler constant, and S (x, n) closer to 1 indicates a higher probability that the point is an abnormal point, and closer to 0 indicates a higher probability that the point is a normal point.

4. The steam turbine high governing valve flow characteristic optimization method based on small sample transient data according to claim 1, wherein 50% -100% of continuous variable load data segments are selected in the third step, and the number of clustering centroids is determined by taking 1MW interval as a centroid division basis.

5. The method for optimizing the high governing valve flow characteristic of the steam turbine based on the small sample transient data according to claim 1, wherein the calculation method for correcting the steam inlet amount of the steam turbine in the target data set to the corresponding steam inlet amount of the steam turbine under the reference value in the fourth step is as follows:

wherein, ReaValue represents the steam inlet quantity of the steam turbine corresponding to the correction of the steam inlet quantity of the steam turbine in the target data set to the reference value;

G. p respectively represents the steam inlet quantity t/h and the pressure MPa of the steam turbine; g_ARepresenting the steam inlet amount of the reference working condition; g_BRepresenting the steam inlet amount of the actual working condition; subscript G₀Represents a primary steam parameter; p_1BIndicating the regulated steam parameter, P, under actual conditions_0AThe representation represents the main steam pressure, P, under the reference condition_0BThe representation represents the main steam pressure, P, under actual conditions_1AThe representation represents the steam pressure of the regulating stage, P, under the reference operating conditions_1BThe representation represents the regulator stage steam pressure at actual conditions.

Technical Field

The invention belongs to the technical field of steam turbine operation optimization, and relates to a steam turbine high throttle flow characteristic optimization technology based on small sample transient data. The transient data of the small sample of the unit is used as a basis, and the linearity of the high-speed regulating valve of the steam turbine is optimized through technologies such as data cleaning, cluster analysis and steam flow correction, so that the dynamic regulation performance of the steam turbine can be effectively improved, and the load control precision and the primary frequency modulation control quality of the unit are improved.

Background

With the large access of new energy electric power such as wind energy, solar energy and the like, in order to ensure the stability of the load and frequency of a power grid and stabilize the influence of a random fluctuation power supply on the safety of the power grid, a conventional coal-fired power generating unit needs to frequently participate in peak regulation and frequency regulation of the power grid, the control requirement of the power grid on a thermal power generating unit is continuously improved, the unit needs to rapidly cross out of a regulation dead zone, and the rapid and high-precision load regulation is realized.

The load control of the unit is usually completed by a digital electro-hydraulic control system (DEH system) of the steam turbine, the linearity deviation of a valve flow characteristic curve of the DEH system is overlarge, the response time of the unit to a load can be increased, frequent oscillation of a high-speed governor is easily caused in the rapid adjustment process, the adjustment rate and the adjustment precision performance of the unit load are deteriorated, main steam parameters fluctuate, the fatigue damage of a valve actuating mechanism is increased, the examination indexes of two detailed rules of a power grid are influenced, and the safety and the economic operation of the unit are not facilitated.

In the actual operation process of the unit, the unit frequently participates in load adjustment every day, a large amount of unit transient state data exist in a Distributed Control System (DCS), and the data can objectively and comprehensively reflect the current operation characteristics and performance of the unit. How to analyze the flow characteristic of the turbine high-pressure regulating valve from the transient data of the small sample to make the turbine high-pressure regulating valve better serve for the actual operation is also a subject in front of engineering technicians.

Disclosure of Invention

In order to solve the problems, the invention provides a turbine high-speed governing valve flow characteristic optimization technology based on small sample transient data. The invention aims to provide a steam turbine high governing valve flow characteristic optimization technology based on small sample transient data, which identifies and optimizes a linear relation between a steam turbine main valve position instruction and main steam flow through the unit transient data of the small sample, so that a steam distribution function of the steam turbine high governing valve is more reasonable, and the variable load dynamic performance of a unit is improved.

In order to achieve the purpose, the invention adopts the technical scheme that:

the first step is as follows: and collecting transient data of the small samples. Selecting a continuous variable load data section of the unit with the load within a 50% -100% interval by using historical data stored in the DCS;

the second step is that: and (6) data cleaning. Adopting an isolated deep forest algorithm to check and eliminate abnormal values in the small sample data, and adopting a linear difference method to supplement;

the third step: and (6) clustering data. Taking the load of the unit as the basis of the division of the mass center, taking the change of 1MW as one mass center, and carrying out clustering calculation on the cleaned data by adopting a PAM (pulse amplitude modulation) clustering algorithm to obtain a target data set representing the operating characteristics of the steam turbine;

the fourth step: and (5) correcting the main steam flow. Selecting main steam pressure and regulating stage pressure corresponding to the highest load working condition in the target data set as reference values, and correcting the steam inlet quantity of the steam turbine in the target data set to the steam inlet quantity of the steam turbine for use under the reference values according to the proportional relation between the regulating stage pressure and the steam flow of the unit;

the fifth step: and (4) linearly fitting the flow characteristic of the valve. Performing linear fitting by adopting a minimum quadratic algorithm according to the corrected steam turbine steam admission amount and the total valve position instruction to obtain an expected linear relation between the steam turbine steam admission amount and the total valve position instruction;

and a sixth step: and (5) correcting a steam distribution curve of the high-speed regulating valve. And according to the linear curve between the steam inlet quantity of the steam turbine and the position in front of the main valve obtained by fitting, combining the original steam distribution function of the high valve of the steam turbine, and distributing the main valve position instruction to each single valve instruction function for correction to obtain a new steam distribution function of the steam turbine valve.

Drawings

Fig. 1 is a graph showing an actual flow characteristic curve and a corrected flow characteristic curve according to an embodiment of the present invention.

FIG. 2 is an exemplary plot of the optimized total valve position command versus the valve position command of GV1 according to embodiments of the present invention.

Detailed Description

The invention discloses a steam turbine high governing valve flow characteristic optimization technology based on small sample transient data, which comprises the following steps:

the first step is as follows: and collecting transient data of the small samples. Selecting a continuous variable load data section of the unit with the load within a 50% -100% interval by using historical data stored in the DCS; as shown in table 1, the patent acquisition parameters of the present invention include: the system comprises a unit power, a main steam flow, a main steam pressure, a main steam temperature, a regulation stage pressure, a high-pressure cylinder exhaust temperature, a total valve position opening, a GV1 valve position opening, a GV2 valve position opening, a GV3 valve position opening and a GV4 valve position opening.

The adoption period is as follows: for 1 second.

The second step is that: and (3) adopting an isolated deep forest algorithm to check and eliminate abnormal values in the small sample data, and adopting a linear difference method to supplement.

The isolated forest is composed of iTrees, binary search trees are constructed in a multi-iteration mode, and the binary trees form the forest. The construction process is as follows:

(1) randomly extracting m sub-samples from n training data, and putting the m sub-samples into a root node of an isolated tree;

(2) randomly appointing a dimension, and randomly generating a cutting point p in the range of the current node data, wherein the cutting point is generated between the maximum value and the minimum value of the appointed dimension in the current node data;

(3) the selection of the cutting point generates a hyperplane, and the data space of the current node is divided into 2 subspaces: placing points smaller than p in the currently selected dimension on the left branch of the current node, and placing points larger than or equal to p on the right branch of the current node;

(4) and (3) recursing the left branch node and the right branch node of the node, and continuously constructing new leaf nodes until only one datum (the cutting can not be continued) is on the leaf node or the tree grows to the set height.

(5) Recording the number of edges passing from the root node to the leaf node of each data, namely the path length h (x), so as to calculate the abnormal index S (x, n) of the data:

c(n)＝2H(n-1)-(2(n-1)/n)

here, h (n) ═ ln (n) + ξ ═ 0.5772156649 is an euler constant, S (x, n) closer to 1 indicates a higher possibility that the point is an abnormal point, and closer to 0 indicates a higher possibility that the point is a normal point.

The third step: and (6) clustering data. And taking the load of the unit as a basis for dividing the mass center, randomly extracting an initial clustering center when 1MW changes into one mass center, and performing clustering calculation on the cleaned data by adopting a PAM (pulse amplitude modulation) clustering algorithm to obtain a target data set representing the operating characteristics of the steam turbine. Taking the unit load variation range [150MW,300MW ] as an example, taking 1MW interval as a centroid division basis, determining 150 clustering centroids, and acquiring a target data set by adopting a PAM clustering algorithm;

the fourth step: and correcting the steam inlet flow of the steam turbine. Selecting the main steam pressure and the regulating pressure corresponding to the highest unit load condition in the target data set as reference values, correcting the steam flow in the target data set to be in the same steam pressure level, and adopting the following calculation method:

wherein G, p respectively represents the steam inlet quantity t/h and pressure MPa of the steam turbine; subscript a represents a reference condition; the small scale B represents the actual working condition; subscript 0 represents the main steam parameter; the subscript 1 indicates the conditioning stage steam parameters.

The fifth step: and (4) performing linear fitting on the high-modulation gate flow. And performing linear fitting by adopting a minimum quadratic algorithm according to the corrected main steam flow characteristic value RealValue and the total valve position instruction to obtain the expected linear relation between the steam inlet quantity of the steam turbine and the total valve position instruction, as shown in FIG. 2.

And a sixth step: and (5) correcting a steam distribution curve of the high-speed regulating valve. And according to the linear curve between the steam turbine steam inlet flow and the position before the main valve position obtained by fitting, combining the original steam turbine high governing valve steam distribution function, and correcting the distribution of the main valve position instruction to each single valve instruction function to obtain a new steam turbine high governing valve steam distribution function. FIG. 2 shows a graph of the optimized total valve position command versus the valve position command of GV 1.