Data-driven multi-rate ore grinding loop control method and system

Abstract

The invention discloses a data-driven multi-rate ore grinding loop control method and system, and the method comprises the steps: collecting the historical data of an ore grinding loop control system at different time scales, and estimating the future trajectory of the control system according to the characteristics of the historical data; the method comprises the following steps: predicting future data by taking known historical data of a set moment as a reference, rolling the set moment as the reference forwards along a time domain, constructing time-varying Hankel matrixes of historical output, historical input, future output and future input, and predicting future output data according to historical data characteristics; and a correction item is introduced for controller correction, the corrected frequency of the electric vibration feeder and the corrected opening degree of the water supplementing valve act on an ore grinding loop control system, and the ore grinding loop ore feeding amount, the water feeding amount and the set value are tracked through the predicted value. According to the control method, the control strategy is adjusted in real time in a data driving mode, and the control precision is improved so as to adapt to the time-varying characteristic and the multi-rate characteristic of the ore grinding loop.

Description

Technical Field

[0001] The present invention relates to the technical field of mineral processing and process control, and particularly relates to a data-driven multi-rate grinding circuit control method and system. Background Art

[0002] In the actual grinding process, the grinding circuit is a key link for ore crushing and classification. Traditional grinding circuit control systems usually adopt a single-rate control method, that is, the sampling and control periods of all control variables and output variables are the same. However, in the actual production process, due to factors such as equipment aging, working environment changes, and operating conditions changes, the model parameters of the grinding circuit exhibit time-varying characteristics, and this time-varying characteristic makes it difficult for the traditional single-rate control method to meet the actual production requirements.

[0003] In addition, in order to prevent equipment damage and ensure process stability, the control operations must be restricted within the safe operating range. This includes setting strict upper and lower limits for parameters such as ore feed rate, valve opening, and mill load. However, in the context of the industrial Internet, in order to meet higher standards of transient performance requirements, distributed control systems usually set the control period relatively short to respond and adjust the system state more quickly. However, the detection of ore feed rate and water flow rate in the basic circuit is limited by hardware, resulting in a relatively long output sampling period. This multi-rate characteristic makes the dynamic behavior of the system more complex. If the traditional single-rate control method continues to be used, it is not only difficult to meet the control requirements, but also may affect the stable operation of the grinding circuit, and thus have an adverse impact on the economic benefits of the concentrator.

[0004] Therefore, there is an urgent need for a grinding circuit control system that can effectively handle time-varying parameters and multi-rate characteristics to improve control accuracy and system stability, and thus enhance the economic benefits of the concentrator. Summary of the Invention

[0005] The problem to be solved by the present invention is: to provide a data-driven multi-rate grinding circuit control method and system, which can adjust the control strategy in real time through a data-driven manner to improve control accuracy and adapt to the time-varying characteristics and multi-rate characteristics of the grinding circuit.

[0006] The present invention adopts the following technical solutions: A data-driven multi-rate grinding circuit control method includes the following steps:

[0007] Step 1, data acquisition and future trajectory estimation: Based on the data acquisition device installed in the concentrator, historical data of the grinding circuit control system at different time scales are collected, including: ore feed amount, water supply amount, frequency of the electric vibration feeder, and opening degree data of the makeup water valve, and the future trajectory of the control system is estimated according to the characteristics of the historical data;

[0008] Step 2, Rolling Horizon Prediction: Use the historical data known at a set time as a benchmark to predict future data. The set time used as the benchmark rolls forward with the time domain. Construct time-varying Hankel matrices of historical output, historical input, future output, and future input, and predict future output data based on the characteristics of the historical data;

[0009] Step 3, Data-Driven Control: Based on the future output data predicted in Step 2, introduce a correction term to correct the controller, and apply the corrected frequencies of the electric-vibration feeder and the opening degrees of the water supply valves to the grinding circuit control system to perform the tracking of the ore feeding amount, water supply amount, and predicted value to the set value in the grinding circuit.

[0010] Preferably, in Step 1, estimating the future trajectory of the control system according to the characteristics of the historical data includes the following sub-steps:

[0011] Step 1.1, Collect asynchronous historical output and input data, stack them according to the frame period to form reconstructed data with the frame period as the new period;

[0012] Step 1.2, Solve the historical trajectory law. Based on the reconstructed data, define the corresponding multi-rate Hankel matrices of the historical output and input data to characterize the characteristics of the historical data;

[0013] Specifically, to solve the historical trajectory law, the methods include:

[0014] Define the law of the historical output trajectory, denoted as y p (m), and use the least squares method to solve the law of the historical output trajectory

[0015] Define the law of the historical input trajectory, denoted as u p (m), and use the least squares method to solve the law of the historical input trajectory

[0016] Step 1.3, Define the corresponding multi-rate Hankel matrices of the future output and input data, and estimate the future trajectory according to the characteristics of the historical data.

[0017] Specifically, to estimate the future trajectory, the method is as follows:

[0018] Define the future output matrix y f (m), which contains the ore feeding amount and water supply amount data from the (m + 1)T0 moment to the (m + N y )T0 + (b - 1)T a moment, arranged as column vectors, N y represents the prediction time domain of the output, and the subscript f represents future data;

[0019] Define the future output Hankel matrix H[Y f, including the ore feeding amount and water feeding amount data from NT0 + T a to (N + N y - 1 + j)T0+(b - 1)T a moments, arranged in the Hankel form of N y rows and jb columns; among them, the first column is the ore feeding amount and water feeding amount data from the moment of NT0 + T a to the moment of (N + N y )T0;

[0020] Estimate the future output matrix y f (m).

[0021] Define the future input matrix u f (m), including the data of the electromagnetic vibrator feeder frequency and the opening degree of the water replenishing valve from the moment of mT0 to (m + N u - 1)T0+(a - 1)T b moments, arranged as a column vector, and N u represents the prediction time domain of the input;

[0022] Define the Hankel matrix H[U f of the future input, including the data of the electromagnetic vibrator feeder frequency and the opening degree of the water replenishing valve from NT0 + T b to (N + N u - 1 + j)T0+(a - 1)T b moments, arranged in the Hankel form of N u rows and ja columns; among them, the first column is the data of the electromagnetic vibrator feeder frequency and the opening degree of the water replenishing valve from the moment of NT0 + T b to the moment of (N + N u )T0;

[0023] Estimate the future input matrix u f (m).

[0024] Preferably, in step 2, predicting the future output data according to the characteristics of the historical data includes the following sub - steps:

[0025] Step 2.1, Define the time - varying Hankel matrix of the historical output including the ore feeding amount and water feeding amount data from the moment of (m - N - L + 1)T0 to (m - L + j - 1)T0+(2b - 2)T a moments;

[0026] Among them, the first column is the ore feeding amount and water feeding amount data from the moment of (m - N - L + 1)T0 to (m - 1)T0+(b - 1)T a moments, and L is the prediction distance;

[0027] Solve the historical trajectory law under the time-varying Hankel matrix

[0028] Step 2.2, Define the time-varying Hankel matrix of future outputs Including the ore feeding amount and water supply amount data from the time of (m - L + 1)T0 to the time of (m - L + N y -1 + j)T+(2b - 2)T a ;

[0029] Among them, the first column is the ore feeding amount and water supply amount data from the time of (m - L + 1)T0 to the time of (m - L + N y )T0+(b - 1)T a ;

[0030] Estimate future outputs

[0031] Step 2.3, Define the time-varying Hankel matrix of historical inputs Including the electromagnetic vibrator feeder frequency and makeup water valve opening data from the time of (m - N - L)T0 to the time of ((m - L + j - 2)T0+(2a - 2)T b ) ; Among them, the first column is the electromagnetic vibrator feeder frequency and makeup water valve opening data from the time of (m - N - L)T0 to the time of (m - L - 1)T0+(a - 1)T b ;

[0032] Solve the historical trajectory law under the time-varying Hankel matrix

[0033] Step 2.4, Define the time-varying Hankel matrix of future inputs Including the electromagnetic vibrator feeder frequency and makeup water valve opening data from the time of (m - L)T0 to the time of (m - L + N y -2 + j)T0+(2a - 2)T b ;

[0034] Among them, the first column is the electromagnetic vibrator feeder frequency and makeup water valve opening data from the time of (m - L)T0 to the time of (m - L + N u )T0+(a - 1)T b ;

[0035] Estimate future inputs

[0036] Preferably, in step 3, the data-driven control includes the following sub-steps:

[0037] Step 3.1, Introduce a correction term to correct the controller and predict the input value; establish a prediction equation for the output value and an estimation equation for the input value, and perform data-driven prediction of the control system

[0038] Step 3.2: Define the performance index function, minimize the performance index function, solve the controller, and obtain the optimal correction term;

[0039] Step 3.3: Obtain the control law, apply it to the control system, and perform the tracking of the ore feeding amount, water feeding amount and predicted value to the set value in the grinding circuit.

[0040] The technical solution of the present invention also provides: A data-driven multi-rate grinding circuit control system for implementing any of the above control methods, including: a multi-rate data-driven estimation module, a rolling horizon prediction module, and a controller correction module;

[0041] The multi-rate data-driven estimation module directly collects the historical data of the grinding circuit control system at different time scales based on the data acquisition device installed in the concentrator, including: the data of the ore feeding amount, water feeding amount, vibrating feeder frequency and makeup water valve opening, and processes the data, and estimates the future trajectory of the system according to the characteristics of the historical data;

[0042] The rolling horizon prediction module is used to use the collected data to predict future data with the known data at the set time as the benchmark. The set time as the benchmark rolls forward with the time domain to enhance the correlation between data, construct the time-varying Hankel matrix of historical output, historical input, future output and future input, and predict the future output data according to the characteristics of the historical data;

[0043] The controller correction module, based on the future output data predicted by the rolling horizon prediction module, introduces a correction term to correct the controller, and applies the corrected vibrating feeder frequency and makeup water valve opening to the system to perform the tracking of the ore feeding amount, water feeding amount and predicted value to the set value in the grinding circuit.

[0044] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects:

[0045] 1. The multi-rate grinding circuit control system of the present invention adjusts the control strategy in real time in a data-driven manner, predicts future data with the known data at a certain moment as the benchmark, and at the same time, the moment as the benchmark rolls forward with the time domain to enhance the correlation between data, effectively improving the prediction accuracy to adapt to the time-varying characteristics and multi-rate characteristics of the grinding circuit.

[0046] 2. The multi-rate grinding circuit control method of the present invention, by introducing a correction term, can ensure the effective tracking of the ore feeding amount, water feeding amount and their predicted values to the set value after the corrected vibrating feeder frequency and makeup water valve opening act on the system. Description of the Drawings

[0047] Figure 1This is the flow block diagram of the data-driven multi-rate grinding circuit control method of the present invention. Detailed implementation manners

[0048] In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the application will be further elaborated in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments made by other researchers in the field based on this embodiment fall within the protection scope of the present invention. At the same time, for the step numbers in the embodiments of the present invention, they are only set for the convenience of elaboration and explanation, and no limitation is imposed on the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0049] In an embodiment of the present invention, a data-driven multi-rate grinding circuit control system, as Figure 1 shown, includes a multi-rate data-driven estimation module, a rolling horizon prediction module, and a controller correction module.

[0050] The multi-rate data-driven estimation module is used to directly collect data on the ore feed amount, water feed amount, frequency of the electric vibrator feeder, and opening of the makeup water valve at different time scales and process the data, and estimate the future trajectory of the system according to the characteristics of historical data;

[0051] The rolling horizon prediction module is responsible for using the collected data to predict future data with the data known at a certain moment as a reference. At the same time, the moment used as a reference rolls forward with the time domain, enhancing the correlation between data and improving the prediction accuracy;

[0052] The controller correction module is responsible for introducing a correction term on the basis of the above prediction module, so that after the corrected frequency of the electric vibrator feeder and the opening of the makeup water valve act on the system, it can ensure the tracking of the ore feed amount, water feed amount and their predicted values to the set values.

[0053] Specifically, in this embodiment, the specific control method of the multi-rate grinding circuit includes the following steps:

[0054] Step 1: Data acquisition and future trajectory estimation

[0055] Through the data acquisition device installed in the concentrator, collect data on the ore feed amount, water feed amount, frequency of the electric vibrator feeder, and opening of the makeup water valve respectively.

[0056] After data acquisition, there is no need to identify model parameters online. Directly use data at different time scales and stack them according to the frame period to form reconstructed data with the frame period as the new period.

[0057] Construct a multi-rate Hankel matrix to characterize the characteristics of historical data, and estimate the future trajectory of the system according to this characteristic. Specifically, it is divided into the following two steps.

[0058] 1.1. Solve the historical trajectory law

[0059] Define the law of the historical output trajectory, expressed as:

[0060] y p (m) = H[Y p g y0 (m)

[0061] where m represents the number of steps in the frame period, and y p (m) represents the historical output trajectory matrix, and Y p represents the output value within the historical fixed-step period, and H[Y p represents the Hankel matrix of the output values within the historical fixed-step period, and g y0 (m) represents the law between y p (m) and H[Y p output values.

[0062] Use the least squares method to solve the historical output trajectory law:

[0063]

[0064] where T represents the transpose.

[0065] Define the law of the historical input trajectory, expressed as:

[0066] u p (m) = H[U p g u0 (m)

[0067] where u p (m) represents the historical input trajectory matrix, and U p represents the input value within the historical fixed-step period, and H[U p represents the Hankel matrix of the input values within the historical fixed-step period, and g u0 (m) represents the law between u p (m) and H[U p input values.

[0068] Use the least squares method to solve the historical input trajectory law:

[0069]

[0070] 1.2. Estimate the future input and output matrices

[0071] Define the future output matrix y f (m), which includes the time from (m + 1)T0 to (m + N y )T0 + (b - 1)Ta The ore feeding amount and water feeding amount data at each moment are arranged in a column vector, N y represents the prediction horizon of the output. The subscript f represents future data, b represents the multiple relationship between the sampling period and the base period, and T0 and T a represent the frame period, the control period of the electro-vibrating feeder, and the control period of the water replenishing valve respectively.

[0072] Define the future output Hankel matrix H[Y f , which contains the ore feeding amount and water feeding amount data from NT0 + T a to (N + N y -1 + j)T0 + (b - 1)T a moment, and is arranged in the Hankel form of Nyb rows and jb columns. Y f represents the output value within the future prediction horizon, and N represents the length of this sequence;

[0073] Among them, the first column is the ore feeding amount and water feeding amount data from the moment of NT0 + T a to the moment of (N + N y )T0.

[0074] Estimate the future output matrix:

[0075]

[0076] Define the future input matrix u f (m), which contains the electro-vibrating feeder frequency and water replenishing valve opening data from the moment of mT0 to (m + N u -1)T0 + (a - 1)T b moment, and is arranged in a column vector, N u represents the prediction horizon of the input, a represents the multiple relationship between the control period and the base period, and T b represents the sampling period of the ore feeding amount and water feeding amount.

[0077] Define the Hankel matrix H[U f of the future input, which contains the electro-vibrating feeder frequency and water replenishing valve opening data from NT0 + T b to (N + N u -1 + j)T0 + (a - 1)T b moment, and is arranged in the Hankel form of N u a rows and ja columns. U f represents the input value within the future control horizon;

[0078] Among them, the first column is the electro-vibrating feeder frequency and water replenishing valve opening data from the moment of NT0 + T b to the moment of (N + N u )T0.

[0079] Estimated future input matrix:

[0080]

[0081] Step 2, Rolling horizon prediction

[0082] Using the known data at a certain moment as a benchmark to predict future data, and at the same time, the moment used as the benchmark rolls forward with the time domain. Construct time-varying Hankel matrices of historical output, historical input, future output, and future input, and predict future output data according to the characteristics of historical data to enhance the correlation between data and improve the prediction accuracy.

[0083] Define the time-varying Hankel matrix of historical output Containing the ore feeding amount and water feeding amount data from the moment of (m - N - L + 1)T0 to the moment of (m - L + j - 1)T0 + (2b - 2)T a moment, representing the output values within a fixed step length rolling along the time axis;

[0084] Among them, the first column is the ore feeding amount and water feeding amount data from the moment of (m - N - L + 1)T0 to the moment, which is the prediction distance.

[0085] Solve the historical trajectory law under the time-varying Hankel matrix

[0086]

[0087] Define the time-varying Hankel matrix of future output Containing the ore feeding amount and water feeding amount data from the moment of (m - L + 1)T0 to (m - L + N y - 1 + j)T + (2b - 2)T a moment, representing the output values within the future prediction time domain rolling along the time axis;

[0088] Among them, the first column is the ore feeding amount and water feeding amount data from the moment of (m - L + 1)T0 to the moment.

[0089] Estimate future output

[0090]

[0091] Define the time-varying Hankel matrix of historical input Containing the vibrating feeder frequency and makeup water valve opening data from the moment of (m - N - L)T0 to ((m - L + j - 2)T0 + (2a - 2)T b ) moment;

[0092] Among them, the first column is the data of the vibratory feeder frequency and the water replenishing valve opening from the moment of (m - N - L)T0 to the moment of (m - L - 1)T0+(a - 1)T b for the vibratory feeder and the water replenishing valve opening at the moment from (m - L - 1)T0+(a - 1)T to (m - L - 1)T0+(a - 1)T

[0093] Solve the historical trajectory law under the time-varying Hankel matrix

[0094]

[0095] Define the time-varying Hankel matrix of the future input including the data of the vibratory feeder frequency and the water replenishing valve opening from the moment of (m - L)T0 to the moment of (m - L + N y -2 + j)T0+(2a - 2)T b for the vibratory feeder and the water replenishing valve opening at the moment from (m - L)T0+(2a - 2)T to (m - L)T0+(2a - 2)T

[0096] Among them, the first column is the data of the vibratory feeder frequency and the water replenishing valve opening from the moment of (m - L)T0 to the moment of (m - L + N u )T0+(a - 1)T b for the vibratory feeder and the water replenishing valve opening at the moment from (m - L)T0+(a - 1)T to (m - L)T0+(a - 1)T

[0097] Estimate the future input

[0098]

[0099] Step 3, Data-driven control

[0100] Based on the above prediction module, introduce a correction term so that after the corrected vibratory feeder frequency and water replenishing valve opening act on the system, it can ensure the tracking of the ore feeding amount, water feeding amount and their predicted values to the set values

[0101] Specifically, it is divided into the following two steps

[0102] 3.1. Conduct a data-driven prediction process

[0103]

[0104] Among them is the historical data matrix composed of and is and arranged in column vector form to form a matrix, z p (m), z f (m) represent historical data and predicted data respectively

[0105] Add a correction term Add a correction term so that after the predicted input value acts on the system, the output value can be tracked. Then the output value and the input value prediction process after introducing the correction term can be described as:

[0106]

[0107] Predicted input value

[0108]

[0109] Among them, is N u a×N u The identity matrix of a, O u is N u a×N y The zero matrix of b, represents the correction term.

[0110] 3.2. Solve the controller

[0111] Define the performance index function:

[0112]

[0113] s.t. z f = z r

[0114] Among them, z r is the actually collected data. The purpose of the equality constraint z f = z r is to make the predicted value equal to the actually collected value. w represents the set value of the ore feeding amount and the water feeding amount data, W * (m)=[w(m)], [w(m)] represents the matrix composed of the set values, and are the weighting factors.

[0115] In this embodiment,

[0116]

[0117] Let the performance index function Take the derivative of U(m), substitute the prediction equation of the ore feeding amount and the water feeding amount value and the estimation equation of the input value into the performance index function to make the performance index function minimum. Take the derivative of Take the derivative and make it equal to 0, then the optimal correction term

[0118] Substitute the optimal correction term into the estimation equation of the input value to obtain the control law:

[0119]

[0120] Apply the first element in to the system. When the time domain scrolls to the (m + 1)-th moment, repeat the above process to achieve a rolling time domain and online optimization.

[0121] The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A data driven multi-rate grinding circuit control method, characterized in that: The steps include: Step 1: Data collection and future trajectory estimation: Based on the data collection device installed in the ore dressing plant, the historical data of the grinding circuit control system at different time scales are collected, including: ore feed rate, water feed rate, electric vibrating feeder frequency and water supply valve opening data, and the future trajectory of the control system is estimated according to the characteristics of the historical data; Step 2, rolling time domain prediction: Use the historical data known at the set time as the benchmark to predict future data. The set time as the benchmark rolls forward in the time domain to construct a time-varying Hankel matrix of historical output, historical input, future output, and future input, and predict future output data based on the characteristics of historical data; Step 3, data-driven control: Based on the future output data predicted in step 2, a correction term is introduced to calibrate the controller, and the corrected electric vibrating feeder frequency and water supply valve opening are applied to the grinding circuit control system to track the grinding circuit feed rate, water supply and predicted values ​​against the set values.

2. The data driven multi-rate grinding circuit control method according to claim 1, characterized in that: In step 1, the future trajectory of the control system is estimated based on the characteristics of historical data, including the following sub-steps: Step 1.1, collect asynchronous historical output and input data, stack them according to the frame period, and form reconstructed data with the frame period as the new period; Step 1.2, solve the historical trajectory law, define the multi-rate Hankel matrix corresponding to the historical output and input data based on the reconstructed data, and characterize the characteristics of the historical data; Step 1.3: Define the multi-rate Hankel matrices corresponding to future output and input data, and estimate future trajectories based on the characteristics of historical data.

3. The data driven multi-rate grinding circuit control method according to claim 2, characterized in that: In step 1.2, the method for solving the historical trajectory law is as follows: Define the law of historical output trajectory, expressed as: y p (m)=H[Y p ]g y0 (m) Among them, m represents the number of steps in the frame cycle, y p (m) represents the historical output trajectory matrix, Y p Represents the output value within the historical fixed-step period, H[[[Y p ]]] represents the Hankel matrix of the output value in the historical fixed-step period, g y0 (m) represents y p (m) and H[Y p ]The pattern between the output values; Use the least squares method to solve the historical output trajectory law Among them, the superscript T represents transposition; Define the law of historical input trajectory, expressed as: u p (m)=H[U p ]g u0 (m) Among them, u p (m) represents the historical input trajectory matrix, U p Indicates the input value within the historical fixed-step period, H[U p ] represents the Hankel matrix of the input value in the historical fixed-step period, g u0 (m) represents u p (m) and H[U p ]The pattern between input values; Use the least squares method to solve the historical input trajectory law 4. The data driven multi-rate grinding circuit control method according to claim 3, characterized in that: In step 1.3, the method for estimating the future trajectory is as follows: Define the future output matrix y f (m), including the time from (m+1)T0 to (m+N y )T0+(b-1)T a The ore and water feed data at each time are arranged as column vectors, N y represents the predicted time domain of the output, the subscript f represents future data, b represents the multiple relationship between the sampling period and the base period, T0, T a They represent the frame cycle, the control cycle of the electric vibrating feeder and the water supply valve respectively; Define the future output Hankel matrix H[[[Y f ]]], including from NT0+T a to (N+N y -1+j)T0+(b-1)T a The ore and water feed data at the moment are calculated according to N y Arrange in Hankel form with b rows and jb columns, Y f Represents the output value in the future prediction time domain, and N represents the length of the sequence; Among them, the first column is NT0+T a Time to (N+N y ) Data of ore and water supply at time T0; Estimate the future output matrix: Define the future input matrix u f (m), including from time mT0 to (m+N u -1)T0+(a-1)T b The frequency of the electric vibrating feeder and the opening degree of the water supply valve at the time are arranged as column vectors, N u represents the prediction time domain of the input, a represents the multiple relationship between the control period and the base period, T b Indicates the sampling period of ore and water supply; Define the Hankel matrix H[U f ], including from NT0+T b to (N+N u -1+j)T0+(a-1)T b The frequency of the electric vibrating feeder and the opening degree of the water supply valve at the moment are calculated according to N u Arrange in Hankel form with a row and ja column, U f Represents the input value in the future control time domain; Among them, the first column is NT0+T b Time to (N+N u ) The frequency of the electric vibrating feeder and the opening degree of the water supply valve at time T0; Estimate the future input matrix:

5. The data driven multi-rate grinding circuit control method according to claim 4, characterized in that: In step 2, future output data is predicted based on the characteristics of historical data, including the following sub-steps: Step 2.1: Define the time-varying Hankel matrix of the historical output Contains the time from (mN-L+1)T0 to (m-L+j-1)T0+(2b-2)T a The mineral and water supply data at each moment, Represents the output value within a fixed step length scrolling along the time axis; Among them, the first column is (mN-L+1)T0 to (m-1)T0+(b-1)T a The ore and water supply data at the time, L is the predicted distance; Solving the historical trajectory law under the time-varying Hankel matrix Step 2.2: Define the time-varying Hankel matrix of future output Contains the time from (m-L+1)T0 to (m-L+N y -1+j)T+(2b-2)T a The mineral and water supply data at each moment, Represents the output value in the future prediction domain rolling along the time axis; The first column is the time from (m-L+1)T0 to (m-L+N y )T0+(b-1)T a The mineral and water supply data at each moment. Estimating future output Step 2.3: Define the time-varying Hankel matrix of the historical input Contains the time from (mNL)T0 to ((m-L+j-2)T0+(2a-2)T b ) at the moment of the electric vibration feeder frequency and water supply valve opening data; Among them, the first column is (mNL)T0 to (mL-1)T0+(a-1)T b The frequency of the electric vibrating feeder and the opening degree of the water supply valve at the moment; Solving the historical trajectory law under the time-varying Hankel matrix Step 2.4: Define the time-varying Hankel matrix of future input Contains the time from (mL)T0 to (m-L+N y -2+j)T0+(2a-2)T b The frequency of the electric vibrating feeder and the opening degree of the water supply valve at the moment; The first column is (mL)T0 to (m-L+N u )T0+(a-1)T b The frequency of the electric vibrating feeder and the opening degree of the water supply valve at the moment; Estimating future input 6. The data driven multi-rate grinding circuit control method according to claim 4, characterized in that: In step 3, the data driven control includes the following sub-steps: Step 3.1, introduce correction terms to calibrate the controller and predict the input value; establish the prediction equation of the output value and the estimation equation of the input value to perform data-driven prediction of the control system; Step 3.2, define the performance index function, minimize the performance index function, solve the controller, and obtain the optimal correction term; Step 3.3, obtain the control law and apply it to the control system to track the grinding circuit feed rate, water supply and predicted values ​​against the set values.

7. The data driven multi-rate grinding circuit control method according to claim 6, characterized in that: In step 3.1, data-driven prediction is performed as follows: The data-driven prediction process is described as: in, Is and The historical data matrix is ​​composed of yes and The matrix is ​​arranged in the form of column vectors, z p (m), z f (m) represent historical data and forecast data respectively; Add amendment Tracking output values ​​after the predicted input values ​​act on the control system; After introducing the correction term, the prediction process of output and input values, the prediction equations of ore feed and water feed values ​​are described as: Predicting Input Values The estimation equation for the input values ​​is described as: in, N u a×N u The identity matrix of a, O u N u a×N y The zero matrix of b, Indicates a correction item.

8. The data driven multi-rate grinding circuit control method according to claim 7, characterized in that: In step 3.2, solve the controller as follows: Defining performance indicator functions s.t.z f =z r Among them, z r is the real collected data, and the equality constraint z f =z r It is used to make the predicted value equal to the actual collected value. w represents the set value of the ore feed and water feed data. * (m) = [w(m)], [w(m)] represents a matrix composed of set values, and is the weighting factor; Let the performance index function Derivative U(m), substitute the prediction equation of ore feed and water feed and the estimated equation of input value into the performance index function to minimize the performance index function. Take the derivative and set the result to 0 to obtain the optimal correction term Substitute the optimal correction term into the estimated equation of the input value to obtain the control law: Will The first element in acts on the system. When the time domain rolls to time m+1, steps 2 to 3 are repeated to perform rolling time domain online optimization.

9. A data-driven multi-rate grinding circuit control system for implementing the control method according to any one of claims 1 to 8, characterized in that: include: Multi-rate data-driven estimation module, rolling horizon prediction module, controller correction module; The multi-rate data-driven estimation module directly collects historical data of the grinding circuit control system at different time scales based on the data acquisition device installed in the concentrator, including: data on ore feed rate, water feed rate, electric vibrating feeder frequency and water supply valve opening, and processes the data, and estimates the future trajectory of the system based on the characteristics of the historical data; The rolling time domain prediction module is used to use the collected data to predict future data with the data known at the set time as the benchmark, and the set time as the benchmark rolls forward with the time domain to enhance the correlation between data, construct a time-varying Hankel matrix of historical output, historical input, future output, and future input, and predict future output data according to the characteristics of historical data; The controller correction module introduces a correction term to perform controller correction based on the future output data predicted by the rolling time domain prediction module, applies the corrected electric vibrating feeder frequency and water supply valve opening to the system, and tracks the grinding circuit feed rate, water supply and predicted values ​​to the set values.

Images