Method and device for monitoring blast furnace state based on multi-modes fusion

A method for monitoring blast furnace state based on multi-modes fusion, comprising:

• step (1) pre-calculation of sub-modes of blast furnace, comprising:
• step (1.1) acquiring blast furnace historical parameter variable data and constructing a data set;
• step (1.2) filling in missing values in the data set by a moving average method;
• step (1.3) detecting abnormal values in the data set by a boxplot method and directly eliminating the detected abnormal values;
• step (1.4) obtaining blast furnace state indication variable data and preprocessing, comprising filling the missing values and eliminating the abnormal values;
• step (1.5) calculating a grey relation degree value between each blast furnace parameter variable data and blast furnace state indication variable data, and selecting top N blast furnace parameter variables with the highest correlation as characteristic variables;
• wherein a calculation formula of the grey relation degree is as follows:
• γ
• ⁡
• (
• x
• 0
• ,
• x
• i
• )
• =
• 1
• n
• ⁢
• ∑
• p
• =
• 1
• n
• ξ
• i
• (
• t
• p
• )
• where
• ξ
• i
• (
• t
• p
• )
• =
• Δ
• min
• +
• ρ
• ⁢
• Δ
• max
• ❘
• "\[LeftBracketingBar]"
• x
• 0
• (
• t
• p
• )
• -
• x
• i
• (
• t
• p
• )
• ❘
• "\[RightBracketingBar]"
• +
• ρΔ
• max
• Δmin=(|x0(tp)−xi(tp)|)
• Δmax=(|x0(tp)−xi(tp)|)
• where x0 is the blast furnace state indication variable data of n samples, x;
• is each blast furnace parameter variable of the n samples, i=1, 2, . . . , m, m is a total number of blast furnace parameter variables, ξ is a grey relation coefficient between the two variables, tp is a p-th moment, p=1, 2, . . . , n, n is a total number of samples contained in each parameter variable, ρ is a resolution coefficient, Δmin is a two level minimum difference, Δmax is a two level maximum difference, x0(tp) is a blast furnace state indication variable value at tp and xi(tp) is a blast furnace parameter value at tp;
• step (1.6) selecting characteristic variable data calculated in step (1.5) from the blast furnace historical parameter variable data set to form a historical characteristic variable data set;
• step (1.7) applying a mean shift clustering algorithm in the historical characteristic variable data set to obtain a plurality of cluster centers;
• and
• step (1.8) calculating the Euclidean distance between the samples in the historical characteristic variable data set and each cluster center, selecting a sample point with a smallest Euclidean distance from each cluster center as the sub-modes of the blast furnace, respectively, and constructing a sub-mode characteristic variable data set by all selected sample point data, storing the blast furnace state indication variable data at a corresponding time of each sub-mode, and constructing a sub-mode indication variable data set;
• step (2) blast furnace sub-mode fusion, comprising:
• step (2.1) obtaining real-time parameter variable data of the blast furnace, and selecting the data corresponding to the N characteristic variables calculated in step (1.5) as input variables;
• step (2.2) calculating the Euclidean distance between input variable data and the characteristic variable data of each sub-mode, and constructing an Euclidean distance matrix D=(d1, d2, . . . , du, . . . , dq), where du is the Euclidean distance between the input variable data and the characteristic variable data of a u-th sub-mode, and u=1, 2, . . . , q, q is a number of types of the sub-modes obtained in step (1.8);
• step (2.3) calculating a weight on the basis of an exponential function and a sub-mode contribution rate, comprising:
• step (2.3.1) scaling the Euclidean distance calculated in step (2.2) based on the exponential function and taking a reciprocal, wherein a calculation formula is as follows:
• d
• u_t
• =
• 1
• e
• r
• ×
• d
• u
• where du_t is a transformed Euclidean distance between the input variable data and the characteristic variable data of the u-th sub-mode, and r is a scaling coefficient;
• step (2.3.2) arranging the sub-modes in an order according to corresponding du_t from big to small, and selecting first L sub-modes whose total contribution rate is greater than a set threshold;
• wherein the calculation formula of the contribution rate conu of each sub-mode is as follows:
• c
• ⁢
• o
• ⁢
• n
• u
• =
• d
• u_t
• ∑
• u
• =
• 1
• q
• ⁢
• d
• u_t
• step (2.3.3) normalizing the du_t of the selected sub-mode to obtain a corresponding weight wu, the corresponding weight of the unselected sub-mode being 0, and obtaining a weight matrix W of the sub-modes;
• step (3) monitoring of blast furnace state, comprising:
• reading the indication variable data corresponding to each sub-mode from the sub-mode indication variable data set, and performing weighted summation;
• wherein the calculation formula is as follows:
• y
• z
• ˆ
• =
• ∑
• u
• =
• 1
• q
• ⁢
• w
• u
• ×
• y
• z
• u
• where ŷz is an estimated value of a z-th blast furnace state indication variable, z=1, 2, . . . , v, v is a number of types of the blast furnace state indication variables, and yzu is a z-th blast furnace state indication variable value corresponding to the u-th sub-mode;
• and
• obtaining an estimated value of the corresponding real-time state indication variable of the blast furnace after sub-mode fusion, which realizes real-time state monitoring of the blast furnace.