Thickener operation optimization method and device considering feeding amount fluctuation

Abstract

The invention provides a thickener operation optimization method and device considering feeding amount fluctuation, and the method comprises the steps: obtaining production data of a mineral separation site, and carrying out the preprocessing of the production data; the preprocessed production data are screened based on time sequence correlation, target data are obtained, and the time sequence correlation represents the time sequence matching degree between the production data and the thickener feeding amount; a pre-constructed prediction model is called, the target data are input into the prediction model, a predicted value of the feeding amount of the thickener in a specified time period in the future is obtained, and the prediction model is formed based on a rolling ARIMA model and a PLS model; and calling a pre-constructed thickener operation optimization model, and inputting the predicted value of the feeding amount into the thickener operation optimization model to obtain an ore drawing starting time sequence, a continuous ore drawing time sequence, a pump starting time sequence of a filter pressing pump and pump starting duration of the filter pressing pump of the thickener.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of metallurgical technology, and in particular to a method and device for optimizing thickener operation by taking feed rate fluctuation into consideration. Background Art

[0002] The thickening and dehydration process is a key step in hydrometallurgy, serving to increase slurry concentration, separate solids from liquids, and mitigate the impact of upstream production disturbances on downstream processes. A thickener is a common solid-liquid separation device, primarily used for concentration and clarification, and is widely used in mining, metallurgy, and chemical industries. Optimizing the thickening and filter pressing process can reduce energy consumption and improve production efficiency while ensuring safety. During the operation of the thickener, the feed rate of the thickener is a very important process control parameter for controlling the production of the thickener. It is affected by the upstream process and the feed rate of the thickener fluctuates. The thickening and dehydration process is mostly manually operated. Workers filter according to the number of filter press cabinets per day. Since the upstream flotation feed will fluctuate and the future changes in the ore volume in the thickener are unclear, in order to prevent the rake from pressing, workers dare not store too much ore in the thickener, resulting in too low underflow concentration, which makes the underflow pump and ore pump run longer, the thickening and dehydration process time increased, energy consumption increased, and work efficiency reduced.

[0003] The currently designed thickener optimization model primarily plans thickener operations based on the planned thickener feed rate. However, in actual production operations, the thickener feed rate is often affected by upstream production processes and fluctuates continuously. This can lead to drastic changes in the thickener feed rate within a short period of time, deviating from the planned value. However, in actual production, the thickener feed rate is affected by upstream production processes, resulting in uncertainty in the thickener feed rate. This can increase the energy consumption and safety risks of the thickening and dehydration process during subsequent optimization. Summary of the Invention

[0004] In order to solve the above technical problems, an embodiment of the present invention provides a thickener operation optimization method considering feed rate fluctuations, comprising: Obtaining production data from the mineral processing site and preprocessing the production data; The pre-processed production data is screened based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; Calling a pre-built prediction model, inputting the target data into the prediction model, and obtaining a predicted value of the feed rate of the thickener within a specified future time period, wherein the prediction model is formed based on a rolling ARIMA model and a PLS model; A pre-built thickener operation optimization model is called, and the feed rate prediction value is input into the thickener operation optimization model to obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration time series. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain.

[0005] In one embodiment, constructing the thickener feed rate prediction model includes: Constructing the rolling ARIMA model, which is an input variable prediction model for predicting the input variables and determining the prediction range of the thickener feed amount, wherein the input variables match the target data; Assisting in constructing the PLS model based on the predicted values ​​of the input variables output by the rolling ARIMA model; The prediction model is constructed based on the rolling ARIMA model and the PLS model.

[0006] In one embodiment, constructing the rolling ARIMA model includes: Preprocessing and screening the first historical production data to obtain candidate training data; Processing the candidate training data based on a difference method to obtain first training data, wherein the time series corresponding to the first training data is stationary; The initial ARIMA model is trained based on the first training data and a rolling prediction mechanism to obtain the rolling ARIMA model. The rolling prediction mechanism periodically inserts new data into the head of the current first training data and removes the old data at the tail, so as to continue training the rolling ARIMA model with the updated first training data to obtain a new rolling ARIMA model.

[0007] In one embodiment, the training of the initial ARIMA model based on the first training data and the rolling forecast mechanism includes: Calculating an autocorrelation function, a partial autocorrelation function, and a difference order based on the first training data to sequentially determine the number of autoregressive terms, the number of lags of the forecast error, and the input parameters of the initial ARIMA model; Determining a delay time between each of the input variables in the first training data and the thickener feed rate; Determining the number of prediction steps of the initial ARIMA model in combination with the preset prediction time span and the delay time; The initial ARIMA model is trained by combining the number of autoregressive terms, the number of lags of the prediction error, the input parameters, the number of prediction steps, and the rolling prediction mechanism.

[0008] In one embodiment, constructing the PLS model includes: Preprocessing and screening the second historical production data to obtain second training data, the second training data including grinding variables, flotation variables, and the thickener feed rate upstream of the thickener, wherein the grinding variables, flotation variables, and the thickener feed rate are time-series aligned; Train an initial PLS model based on a preset step size and the second training data to obtain a PLS model; Inputting the test input variables into the rolling ARIMA model to obtain predicted values ​​of the test input variables; The PLS model is tested based on the predicted values ​​of the test input variables.

[0009] In one embodiment, building a thickener operation optimization model includes: Determine the thickener underflow concentration prediction model; Determine the calculation model of energy consumption economic indicators; Determine the calculation model for the optimal ore storage volume and number of filter press cabinets; Determine the basic optimization model; The thickener operation optimization model is constructed based on the basic optimization model, the thickener underflow concentration prediction model, the energy consumption economic indicator calculation model, and the optimal ore storage capacity and filter press cabinet number calculation model.

[0010] In one embodiment, determining the thickener underflow concentration prediction model includes: Determining a predicted value of the thickener underflow concentration based on a relationship between the underflow concentration of the thickener and a change in the internal ore storage volume; The energy consumption economic indicator calculation model is determined, including: Determine the energy consumption economic indicators of the thickener underflow pump and the filter press pump based on the tiered electricity prices in the region; Determining the energy consumption economic index calculation model based on the energy consumption economic index of the thickener underflow pump and the energy consumption economic index of the filter press pump; The calculation model for determining the optimal ore storage capacity and the number of filter press cabinets includes: Under the condition of a certain feed rate, the optimal ore storage volume is achieved to minimize energy consumption and ensure safe operation of the thickener; Determining the number of ore storage cabinets based on the optimal ore storage volume; A calculation model for the optimal ore storage capacity and the number of filter press cabinets is determined based on the optimal ore storage capacity and the number of ore discharge cabinets.

[0011] In one embodiment, determining the basic optimization model includes: Construct an objective function with the goal of minimizing energy consumption; Constructing the basic optimization model based on the objective function and preset constraint conditions for ensuring normal and stable operation of the thickener; The constraints include that within one optimization cycle, the maximum underflow concentration of the thickener cannot exceed a preset upper limit, the minimum cannot be lower than a preset lower limit, and the timing constraints of the thickener discharge and the filter press start filtering, so as to ensure that the optimized discharge operation and filter press operation meet the timing requirements.

[0012] In one embodiment, the screening of the pre-processed production data based on time series correlation to obtain target data includes: Performing a delay analysis on the relationship between each input variable in the preprocessed production data and the feed rate of the thickener to obtain a time sequence relationship between the input variable and the feed rate of the thickener; determining, based on a cross-correlation function, a correlation between each of the input variables and the feed rate of the thickener at different delay times; Based on the correlation degree and the time series relationship, the pre-processed production data is screened based on the time series correlation to obtain the target data; The input variables include operating data of various devices involved in the upstream process of the thickener.

[0013] Another embodiment of the present invention also provides a thickener operation optimization device that takes into account feed rate fluctuations, including: A first acquisition module is used to obtain production data of the mineral processing site and pre-process the production data; A first screening module is configured to screen the pre-processed production data based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; a first calling module, configured to call a pre-built thickener feed rate prediction model, input the target data into the feed rate prediction model, and obtain a predicted value of the thickener feed rate within a specified future time period, wherein the feed rate prediction model is formed based on a rolling ARIMA model and a PLS model; The second calling module is used to call the pre-built thickener operation optimization model, input the feed rate prediction value into the thickener operation optimization model, and obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain. Based on the disclosure of the above embodiments, it can be known that the beneficial effects of the embodiments of the present invention include being able to more accurately and efficiently predict the feed rate of the thickener in the future period, and accurately determine the optimized operation content of the thickener in the future period based on the predicted feed rate, including the thickener's ore discharge start time sequence, continuous ore discharge time sequence, filter press pump start time sequence, and filter press pump start duration, ensuring that the overall energy consumption is minimized and the thickener operates normally and stably.

[0014] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0015] The technical solution of the present application is further described in detail below through the accompanying drawings and examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 Schematic diagram of the flow of a thickener operation optimization method considering feed rate fluctuations in an embodiment of the present invention.

[0018] Figure 2 This is a prediction flow chart of the input variable prediction model based on rolling ARIMA in this embodiment.

[0019] Figure 3 This is the partial least squares regression flowchart.

[0020] Figure 4 Flowchart for solving the optimization problem in an embodiment of the present invention.

[0021] Figure 5This is a flowchart of industrial data transmission in an embodiment of the present invention.

[0022] Figure 6 This is a structural block diagram of a thickener operation optimization device taking feed rate fluctuation into consideration in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, but are not intended to limit the present invention.

[0024] It should be understood that various modifications may be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope of the present disclosure will occur to those skilled in the art.

[0025] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.

[0026] These and other characteristics of the invention will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0027] It should also be understood that although the invention has been described with reference to certain specific examples, those skilled in the art will be able to realize many other equivalent forms of the invention that have the characteristics recited in the claims and are therefore within the scope of protection defined thereby.

[0028] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0029] Specific embodiments of the present disclosure will be described hereinafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repetitive functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant detail. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but rather serve merely as a basis and representative basis for the claims to teach those skilled in the art to variously employ the present disclosure with substantially any suitable detailed structure.

[0030] This description may use the phrases "in one embodiment," "in another embodiment," "in a further embodiment," or "in other embodiments," each of which may refer to one or more of the same or different embodiments according to the present disclosure.

[0031] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0032] like Figure 1 As shown, an embodiment of the present invention provides a thickener operation optimization method considering feed rate fluctuations, comprising: S1: Obtaining production data at the mineral processing site and preprocessing the production data; S2: Screening the pre-processed production data based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; S3: calling a pre-built prediction model, inputting the target data into the prediction model, and obtaining a predicted value of the feed rate of the thickener within a specified future time period, wherein the prediction model is formed based on a rolling ARIMA model and a PLS model; S4: Call a pre-built thickener operation optimization model, input the feed rate prediction value into the thickener operation optimization model, and obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration time series. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain.

[0033] The prediction model in this embodiment is based on a rolling ARIMA model and a PLS model. When applied, the input variable prediction model based on the rolling ARIMA is used to predict the input variables. The thickener feed rate is then predicted based on the predicted values ​​of the input variables and the thickener feed rate prediction model based on PLS regression. The thickener operation optimization model described in this embodiment is an optimization model that targets energy consumption economic indicators and is constrained by production safety conditions. It is used to generate optimized thickener operations based on the feed rate prediction value, specifically including the thickener's ore discharge start time series, ore discharge duration time series, filter press pump start time series, and filter press pump on-time duration.

[0034] Since the feed rate prediction value may have a certain error, and in order to reduce the impact of feed rate uncertainty on the optimization results in operation, in this embodiment, the uncertainty in the optimization target is transferred to the constraint, and an adaptive discretization constraint relaxation scheme is adopted to transform the original optimization problem into an approximate optimization problem. At the same time, in order to obtain the feed rate prediction error corresponding to the worst case under a specific ore-dropping sequence, an auxiliary optimization problem is also defined. The purpose of this auxiliary optimization problem is to find the feed rate prediction error when the degree of constraint violation is the largest under a certain ore-dropping time sequence. Based on this, the "worst case" of the feed rate prediction error under the current ore-dropping operation conditions can be found. Through this worst case, it is verified whether the currently obtained optimization operation is still applicable in this worst case. If so, the result of the optimization operation can be output. The constraint satisfaction rate of the operation result optimized by the interval optimization model under the condition of feed rate uncertainty proposed in this embodiment can reach 100%, which significantly improves the operational safety of the thickener and reduces energy consumption.

[0035] In one embodiment, it is first necessary to obtain the production data of the mineral processing site and pre-process the production data. The production data of the mineral processing site is mostly obtained by measuring the various devices on the production line by various measuring devices installed in the production process to obtain corresponding values. In this embodiment, they are collectively referred to as variables. The obtained variables are stored in the database for subsequent call by the system or model. Due to the complexity of the on-site production conditions and the various inevitable interferences in the measurement process, the real-time values ​​of the variables measured on site also have a certain amount of noise. Sometimes, due to sensor failure, the local value of the measured variable is too large or too small, and due to temporary suspension of production due to production failure, etc., the measured data becomes abnormal. In order to eliminate the influence of these abnormal data points, it is first necessary to clean the data. Since outliers account for a minority of the samples and their values ​​deviate significantly from the rest of the observed values ​​of the samples to which they belong, this embodiment adopts an outlier detection method based on box plots. The box plots can be used to find the values ​​of data that deviate greatly from the majority of the data. The box plot can be summarized by five numbers: the minimum sample value, the maximum sample value, the lower quartile value (at 25% of the maximum value), the upper quartile value (at 75% of the maximum value), and the median value (at 50% of the maximum value). Outliers can be displayed by the box plot. First, the interquartile range (IQR) is obtained by the following formula:

[0036] Then define the lower and upper bounds:

[0037] In the interval [ L,U] Points outside the range are outliers of the data. The process data variables are tested for outliers and the mean of the data is used to fill the outliers.

[0038] Furthermore, the collected production data includes various variables from the grinding and flotation process preceding the thickening process. However, some of these variables have little impact on the thickener feed rate, while others can interfere with the prediction model, affecting the accuracy of the thickener feed rate prediction. Therefore, before establishing the prediction model in this embodiment and subsequently using it, the model's input variables should be carefully selected to maximize prediction accuracy. Furthermore, because the location of sensors relative to the target variables varies in actual processes, data delay characteristics must also be analyzed.

[0039] The pre-processed production data is screened based on time series correlation to obtain target data, including: S201: performing delay analysis on the relationship between each input variable in the pre-processed production data and the feed rate of the thickener to obtain a time sequence relationship between the input variable and the feed rate of the thickener; S202: determining, based on a cross-correlation function, a correlation degree between each of the input variables and the feed rate of the thickener at different delay times; S203: Filtering the pre-processed production data based on the time series correlation based on the correlation degree and the time series relationship to obtain the target data; S204: The input variables include operating data of various devices involved in the upstream process of the thickener.

[0040] Specifically, in process industries, based on mechanism analysis, variables in upstream processes often affect the thickener feed rate only after a delay. Therefore, when predicting thickener feed rate, historical data corresponding to the time series relationship between each input variable and the thickener feed rate should be selected and input into the prediction model. To achieve this, a delay analysis should first be performed on the relationship between each input variable and the thickener feed rate to determine the time series relationship between the input variables and the thickener feed rate, thereby determining the correct input data for the model.

[0041] In order to obtain the delay time between each input variable and the thickener feed rate, and at the same time select those input variables with a high correlation with the thickener feed rate, it is necessary to calculate the degree of correlation between each input variable and the thickener feed rate at different delay times. In this embodiment, the cross-correlation function (CCF) is used to calculate the correlation between two variables at different delay times.

[0042] The calculation formula of the cross-correlation function (CCF) is as follows:

[0043] in x , y represents two time series, and y lag behind x , U x , U y Represents the sequence x , y The mean of S x , S y Represents the sequence x , y The variance of P xy ( k ) indicates the delay time k Time Series x and y The mutual correlation coefficient between k max Indicates the set maximum delay time.

[0044] Due to the complexity of the field environment, the data collected on-site will inevitably introduce some noise, which will have a certain impact on the analysis of the time series relationship between the variables. Since the correlation between two time series data can be approximately expressed by the changing trend of the two time series, before analyzing the correlation between the time series, it is possible to first extract the changing trend of the time series and analyze the correlation of the extracted trend data.

[0045] The trend extraction of time series adopts a method based on sliding convolution window. Practice shows that this method can better extract the changing trend of time series and effectively filter out clutter.

[0046] Assume the original time series is x , the convolution window setting value is p , define the vector:

[0047] in, v There are p elements, and satisfy .

[0048] Pick x and v Convolution as the time series after trend extraction w ,Right now w No. kThe elements are:

[0049] For each input variable, the CCF calculation formula is used to perform a correlation analysis between the trend-extracted input variable and the thickener feed data. The correlation between the input variable and the thickener feed at different delay times is obtained, and the maximum correlation coefficient is selected, as shown in Table 1 below:

[0050] The variables in the above table are the input variables described in this embodiment.

[0051] The correlation analysis results show that the 9# belt real-time value, cyclone inlet concentration, cyclone inlet ore volume, cyclone overflow concentration, cyclone overflow ore volume, south slurry pump current, south ore discharge pipe flow rate, and concentrate buffer tank level are highly correlated with the thickener feed volume. Based on the maximum correlation between each variable and the thickener feed volume, a correlation threshold of 0.16 was taken to select variables with a high correlation with the thickener feed volume. These variables are the 9# belt real-time value, cyclone inlet concentration, cyclone inlet ore volume, cyclone overflow concentration, cyclone overflow ore volume, south slurry pump current, south ore discharge pipe flow rate, and concentrate pump buffer tank level. For each selected input variable, the delay time corresponding to the maximum correlation is the time by which it leads the thickener feed volume. The delay time of the corresponding variable can be calculated. The maximum delay time in this embodiment is 50 minutes.

[0052] After obtaining the input data of the prediction model, i.e., the input variables, the prediction model can be called to predict the feed rate of the thickener. The prediction model in this embodiment needs to be constructed in advance, and constructing the prediction model includes: S5: constructing the rolling ARIMA model, which is an input variable prediction model for predicting the input variables and determining the prediction range of the thickener feed rate, wherein the input variables match the target data; S6: Assisting in constructing the PLS model based on the predicted values ​​of the input variables output by the rolling ARIMA model; S7: Construct the prediction model based on the rolling ARIMA model and the PLS model.

[0053] That is, in this embodiment, a rolling ARIMA model is first constructed to predict the input variables, and then a PLS model is constructed to predict the thickener feed rate using the predicted input variables. Therefore, the prediction model in this embodiment is a two-stage thickener feed rate prediction model based on a rolling ARIMA-PLS model, which is a combination of the above two models.

[0054] Specifically, constructing the rolling ARIMA model includes: S501: Preprocess and screen the first historical production data to obtain candidate training data; S502: Processing the candidate training data based on a difference method to obtain first training data, where the time series corresponding to the first training data is stationary; S503: Train the initial ARIMA model based on the first training data and a rolling prediction mechanism to obtain the rolling ARIMA model. The rolling prediction mechanism periodically inserts new data into the head of the current first training data and removes old data at the tail, so as to continue training the rolling ARIMA model with the updated first training data to obtain a new rolling ARIMA model.

[0055] The training of the initial ARIMA model based on the first training data and the rolling forecast mechanism includes: S504: Calculating an autocorrelation function, a partial autocorrelation function, and a difference order based on the first training data to sequentially determine the number of autoregressive terms, the number of lags of the prediction error, and input parameters of the initial ARIMA model; S505: Determine the delay time between each input variable in the first training data and the thickener feed rate; S506: Determine the number of prediction steps of the initial ARIMA model in combination with the preset prediction time span and the delay time; S507: Training the initial ARIMA model based on the number of autoregressive terms, the number of delays in the prediction error, the input parameters, the number of prediction steps, and the rolling prediction mechanism.

[0056] For example, at an actual mineral processing site, workers are generally divided into three shifts per day to manually operate the thickener's ore discharge. One shift lasts for 8 hours, and each shift is divided into two half-shifts, with each half-shift lasting 4 hours. Therefore, in order to meet actual production needs, the cycle for adjusting the ore discharge operation is set at 4 hours. Based on the actual needs of the mineral processing site and the cycle of operation adjustment, the predicted value of the thickener feed rate is also the average value of the next 4 hours. The maximum delay time between each variable in the upstream process and the thickener feed rate is 50 minutes. From the time series relationship, it can be seen that the thickener feed rate can only be predicted up to 50 minutes later by predicting the true value of the upstream process variable. Since the prediction range of the thickener feed rate is the average value of the next 240 minutes, this embodiment proposes to establish an input variable prediction model based on rolling ARIMA to predict the future values ​​of the input variables to increase the prediction range of the thickener feed rate.

[0057] ARIMA is a time series prediction model that can predict future data using historical data of a stationary time series. Before using the ARIMA model, the time series needs to be "stationary". The main stationary methods include difference method, logarithm method, etc. The formula used by the ARIMA model to predict a stationary time series is a linear formula. The parameters in the formula include the time series to be predicted. Y The number of autoregressive terms p and the number of lags in the prediction error q At the same time, the difference order used to stabilize the time series needs to be set d .

[0058] To predict time series Y exist t The value of the moment Y t First, the time series Y Pick d Order difference, get the stationary time series after difference Y (d) ,Right now:

[0059] For stationary time series Y (d) , use the following formula to predict Y (d) No. t value :

[0060] in, Indicates t -1 time prediction error, μ is the offset, , are the parameters to be fitted in the model.

[0061] When designing an ARIMA model, first take the difference of the time series in sequence until the time series is stable. The order of the difference is the input parameter of the model. d , and then determine the number of autoregressive terms in the model through the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the stationary time series p and the number of lags of prediction error q , and then fit the parameters of the model by input data , , the predicted value of the stationary time series is obtained by the above formula , then De-difference to obtain the predicted value of the original time series In order to reduce the computational complexity of the model, the selected input data are averaged every 10 minutes, and the difference order, number of autoregressive terms and number of prediction error lags of the model are determined for each input variable data after averaging.

[0062] Furthermore, the prediction method adopted in this embodiment is multi-step prediction, the basic idea of ​​which is that each time a point is predicted, the point is added to the input data sequence, and the newly generated input data is used to predict the next point until all points are predicted. The basic idea of ​​rolling prediction is that after each prediction cycle, the newly detected real data point is inserted into the head of the training data, and the data point at the end of the training data is popped out, the ARIMA model is trained again with the newly generated training data, and the newly trained model is used to make predictions. The rolling prediction method can continuously update the model parameters. According to the delay time of each variable and the thickener feed rate, the number of steps required for prediction can be determined by the following formula:

[0063] in, Indicates the i The number of prediction steps for each input variable, N p Indicates the time prediction span. In this embodiment, the predicted value of the thickener feed volume is the average value of the next 4 hours. N p =240. Indicates the i The delay time between an input variable and the thickener feed amount. T span Indicates the size of the data averaging window. Since the data is averaged every 10 minutes in this embodiment, T span =10.

[0064] Furthermore, for each input variable, such as Figure 2 As shown in the figure, the forecasting process of the input variable forecasting model based on rolling ARIMA is as follows: Step 1: Input Data X , specify the length of training data l train , current prediction point k = l train ,Pick X Before l train data as the current training data .

[0065] Step 2: Determine ARIMA model parameters p, q, d, determine the number of steps to be predicted according to the above step calculation formula N out .

[0066] Step 3: Based on the parameters p, q, d Build an ARIMA forecasting model through training data Train the ARIMA model and get the trained model .

[0067] Step 4: Pass Prediction Point The value of X midpoint Merge to get sequence ,right Take the average and get k The predicted value of the point .

[0068] Step 5: Judgement Is it established, l x For sequence X If , then X Middle k +1 point added to the sequence and remove the head The last point at the end, , go to step 1; if , the prediction is completed and the loop exits.

[0069] Furthermore, constructing the PLS model includes: S601: Preprocessing and screening the second historical production data to obtain second training data, where the second training data includes grinding variables, flotation variables, and the thickener feed rate upstream of the thickener, and the grinding variables, flotation variables, and the thickener feed rate are time-series aligned; S602: Train an initial PLS model based on a preset step size and the second training data to obtain a PLS model; S603: Input the test input variable into the rolling ARIMA model to obtain the predicted value of the test input variable; S604: Testing the PLS model based on the predicted values ​​of the test input variables.

[0070] For example, after establishing a prediction model for input variables based on rolling ARIMA, it is necessary to establish a prediction model for thickener feed rate based on PLS regression. Partial Least Squares is a widely used technique in industrial data forecasting. Application practice shows that this method can achieve good prediction results when the input data dimension is high and the correlation between input variables is large. y and m independent variables The basic idea of ​​partial least squares regression is to first extract the first component from the independent variable set. t 1 ( t 1 Yes linear combination of , and extract as much variation information as possible from the original independent variable set), and require t 1 and y The correlation between them is maximized, and then the t 1 and y If the regression accuracy reaches a satisfactory level, the algorithm terminates. Otherwise, the second component is extracted until a satisfactory accuracy is reached. The input variable data and the output variable data are averaged with a 240-minute window and a 10-minute step length. The averaged input variable and output variable data are respectively recorded as X and Y . Specify the training data length n = l train ,Will X and Y Split into training data X train 、 Y train and test data X test 、 Y test ,in:

[0071] After obtaining the training data, we need to use the data to train the PLS model. First, X train 、 Y train The data in the standardization, then the delay time between each variable and the dependent variable will be X train 、 Y train The data in are aligned, and the processed data matrices are X 0. Y 0, passed X 0. Y 0 to train the PLS regression model, such as Figure 3 As shown, the training steps are as follows: Step 1: Extract independent variable groups X The first component of 0 , so that it is consistent with the dependent variable Y The correlation is maximum at 0.

[0072] Step 2: Build Y 0 pairs t 1 and right t The return of 1, that is

[0073] in α 1 and β 1 is called the effect loading of the model.

[0074] Step 3: Use the residual matrix X 1. Y 1 instead X 0. Y 0 Repeat the above steps to get is the score vector of the second component, and so on, to get the remaining r ingredients .

[0075] Step 4: r Conduct cross-validation tests on the components to determine the optimal component ,in .

[0076] Step 5: Put Substitution , the partial least squares regression equation of the thickener feed rate is obtained as follows:

[0077] In order to match the subsequent operational adjustments, the test data used data with relatively large variations. The results showed that the test data can track the real data well. The root mean square error of the test is 0.8334 and the mean absolute percentage error is 4.05%.

[0078] The simulation results show that the mean prediction error of the thickener feed rate is , standard deviation ,Depend on Criteria, the feed rate prediction error falls within the interval The probability of is about 99.73%, that is, it is almost impossible for the situation to fall outside this range. Therefore, the value set of the feed amount prediction error is determined as follows:

[0079] Similarly, the thickener operation optimization model in this embodiment also needs to be established in advance. In this embodiment, the thickener operation optimization model is constructed, including: S8: Determine the thickener underflow concentration prediction model; S9: Determine the calculation model of energy consumption economic indicators; S10: Determine the calculation model for the optimal ore storage volume and number of filter press cabinets; S11: Determine the basic optimization model; S12: Constructing the thickener operation optimization model based on the basic optimization model, the thickener underflow concentration prediction model, the energy consumption economic index calculation model, and the optimal ore storage capacity and filter press cabinet number calculation model.

[0080] Wherein, determining the thickener underflow concentration prediction model includes: S801: Determining a predicted value of the thickener underflow concentration based on a relationship between the underflow concentration of the thickener and a change in the internal ore storage volume; The energy consumption economic indicator calculation model is determined, including: S901: Determine the energy consumption economic indicators of the thickener underflow pump and the filter press pump based on the tiered electricity prices in the region; S902: Determining the energy consumption economic indicator calculation model based on the energy consumption economic indicator of the thickener underflow pump and the energy consumption economic indicator of the filter press pump; The calculation model for determining the optimal ore storage capacity and the number of filter press cabinets includes: S101: Under the condition that the feed amount is determined, the optimal ore storage amount is determined to minimize energy consumption and ensure safe operation of the thickener; S102: Determine the number of ore storage cabinets based on the optimal ore storage amount; S103: Determine a calculation model for the optimal ore storage capacity and the number of filter press cabinets based on the optimal ore storage capacity and the number of ore draw cabinets.

[0081] For example, the change of the ore storage volume inside the thickener is the main reason affecting the change of the thickener underflow concentration. Therefore, the future t Changes in ore content in the thickener within minutes The relationship between the underflow concentration of the thickener and the change of the internal ore volume can be used to obtain the predicted value of the underflow concentration of the thickener. The internal ore volume and underflow concentration of the thickener can be soft-measured by the pressure sensor installed inside the thickener. t Soft measurement value of underflow concentration of thickener at 1 minute and t Changes in internal ore content of the thickener within minutes , the thickener can be fitted t Mineral volume changes within minutest Relationship model of underflow concentration after minutes:

[0082] in, M p is the average value of the thickener feed volume in the four hours after the prediction start time, e is the prediction error, T pre To optimize the cycle, take a fixed value of 4 hours. Q UF is the underflow rate of the thickener, which is a constant. For x The underflow concentration of the thickener at that moment. The average relative error of the prediction is 1.15%, indicating that the underflow concentration prediction model can accurately predict the underflow concentration to a certain extent.

[0083] In this embodiment, the objective function to be optimized is the energy consumption economic index of the thickening filter press process within 4 hours. The energy consumption economic index mainly consists of two parts: the energy consumption economic index of the thickener underflow pump and the energy consumption economic index of the thickener underflow pump. Energy consumption economic indicators of filter press pumps Depending on the actual situation, the unit electricity price may vary at different times of the day. The local unit electricity price is distributed in a step-like pattern throughout the day, which is called a ladder electricity price. The ladder electricity price can be expressed as follows:

[0084] in, Through the calculation formula of the above-mentioned ladder electricity price, the discontinuous ladder electricity price is converted into a time-dependent t In order to match the operation of the downstream filter press, the amount of dry ore discharged by the thickener each time is equal to the amount of dry ore filtered out by the filter press per cabinet, which is a fixed value. M PF , running time of thickener underflow pump It can be obtained by solving the following equation:

[0085] in, T Sa is the start time of discharge of the underflow pump, Q UF is the flow rate of the underflow pump, which is a fixed value. and For thickener t It should be noted that the change curve of the thickener underflow concentration is about the feed rate prediction error. e The calculated running time of the underflow pump is Also related to the feed amount prediction error e related.

[0086] Based on the underflow pump on time and the duration of the underflow pump on, the electricity cost generated each time the underflow pump is discharged can be obtained as:

[0087] in, Indicates T Always carry out the energy consumption economic indicators of the ore underflow pump, Indicates Always check the running time of the underflow pump. U UF 、 I UF 、 are the rated voltage, rated current and power factor of the underflow pump respectively, It is a tiered electricity price.

[0088] From the analysis of the data, it can be seen that the operating time of the filter press pump is approximately linearly related to the average slurry concentration in the agitator tank. The relationship model is as follows:

[0089] in, is the average concentration of the slurry in the stirring tank, 、 is the fitting parameter, For T The running time of the filter press pump starts at the moment of filter press. When fitting the running time of the filter press pump, the running time of the underflow pump is used to calculate the average slurry concentration in the mixing tank. ,and It is related to the error in the feed rate prediction, so the filter press pump running time Also related to the feed amount prediction error e related.

[0090] The energy consumption economic index of each filter press can be obtained by the start-up time and operation time of the filter press pump:

[0091] in, Indicates T Start filtration at any time, the energy consumption economic indicators generated by the filter press pump, 、 、 They respectively represent the rated voltage, rated current and power factor of the filter press pump.

[0092] Furthermore, the optimal ore storage capacity of the thickener is related to the feed rate of the next optimization cycle and the ladder electricity price. Since the ladder electricity price remains unchanged within an optimization cycle, the optimal ore storage capacity of the thickener is related to the feed rate of the next optimization cycle. Under different thickener feed rates, different initial ore storage capacities are taken, and the initial ore storage capacities are input as parameters into the optimization model with energy consumption economic indicators as the objective function and thickener safety as the constraint, to obtain the optimal value of the electricity price for different initial ore storage capacities under specific feed rate conditions. The ore storage capacity corresponding to the smallest value among these optimal values ​​is the optimal ore storage capacity under the feed rate conditions. After obtaining the optimal ore storage capacity of the thickener, the number of ore cabinets can be determined based on the optimal ore storage capacity of the thickener. In the optimization interval The number of ore drawers in a tank can be determined using the following formula:

[0093] in, Indicates The total amount of ore released by the thickener is Indicates the amount of ore pressed out by each cabinet of the filter press. m in Indicates the predicted value of thickener feed amount, Indicates that the thickener is T START The amount of ore reserves at that time, Indicates the optimal ore storage capacity of the thickener.

[0094] Based on the above content, it can be seen that the determination of the basic optimization model described in this embodiment includes: S111: Construct an objective function with the goal of minimizing energy consumption; S112: constructing the basic optimization model based on the objective function and preset constraint conditions for ensuring normal and stable operation of the thickener; S113: The constraints include that within one optimization cycle, the maximum underflow concentration of the thickener cannot exceed a preset upper limit, the minimum cannot be lower than a preset lower limit, and the timing constraints of the thickener discharge and the filter press start filtering, so as to ensure that the optimized discharge operation and filter press operation meet the timing requirements.

[0095] Specifically, based on the above, the optimization problem of this embodiment can be described as follows: Based on the predicted value of the thickener feed rate and taking into account all possible values ​​of the prediction error, plan the ore discharge time for each cabinet so that the underflow concentration of the thickener meets the constraint conditions when the prediction error takes all values, and under this condition, minimize the energy consumption economic indicators of the underflow pump and filter press pump. As mentioned above, the energy consumption economic indicators of the underflow pump and filter press pump can be expressed as follows:

[0096] in, For the i The start time of the underflow pump during the second ore drawing is used as the optimized operating variable. Indicates The energy consumption economic index of the underflow pump when the underflow pump is turned on, and the error between this value and the predicted feed rate of the thickener e Related; Assuming that the underflow pump is turned on T d After a certain time, filter pressing is started. Indicates the start time of the filter press pump. T d is a constant. Indicates The energy consumption index of the filter press pump is also related to the feed amount prediction error. e the relevant quantity; N = is the number of ore discharges within an optimization cycle. Since the change of the ladder electricity price at the ore dressing site is based on a daily cycle, the range of each optimization of the prediction model established in this embodiment is from 0:00 to 24:00, that is, the cycle is from 0:00 to 24:00. When the ore discharge time is based on minutes, there are .

[0097] The goal of optimization is to make the feed amount error e In the range of its possible values, the optimized ore-drawing operation can meet the constraints. Under this condition, the ore-drawing operation that minimizes the energy consumption economic index is selected. In order to achieve this goal, first find the optimization problem about the feed prediction error. e The worst case scenario is then considered, and the operating variables are optimized to minimize the value of the objective function under these worst cases. Based on this idea, the objective function of interval optimization of the thickening filter pressing process under uncertain conditions is defined as follows:

[0098] Based on the above formula, it can be seen that this optimization problem is a two-layer optimization problem. The operating variable of the inner optimization layer is the prediction error of the thickener feed rate. e , the purpose is to make the energy consumption economic indicators e Take the maximum value, the outer optimization operation variable is the thickener start ore discharge time sequence within an optimization cycle , used to determine the energy consumption economic indicators e When the maximum value is taken, the objective is minimized. The constraints of the optimization problem are as follows: ,have:

[0099] The first constraint and the second constraint ensure that within an optimization cycle, the maximum value of the thickener underflow concentration cannot exceed the upper limit, and the minimum value cannot be lower than the lower limit. 、 are the upper and lower limits of the thickener underflow concentration respectively. The third constraint to the last constraint is the timing constraint for the thickener to release ore and the filter press to start filtering, ensuring that the optimized operation meets the timing requirements. , .

[0100] Furthermore, in order to ensure that the optimized operation can offset the feed amount prediction error, it is necessary to e Take a collection E The constraint is satisfied when all values ​​in E is a continuous value set, so the feed amount prediction error e In the set E There are infinite values ​​in , which means that there are infinite constraints in the optimization problem. This is actually a semi-infinite optimization problem (SIP). The original optimization problem is transformed to reduce the complexity of the objective function and make it have no uncertain variables.

[0101] The optimization model can be converted into :

[0102] in, Through the above equivalent transformation, all uncertain variables are transferred to the optimization constraints. When solving the optimization problem, only the uncertainties in the constraints need to be considered.

[0103] Because the original optimization problem has an infinite number of constraints, it cannot be solved directly using a general optimization problem-solving method. Therefore, this embodiment proposes first replacing the original optimization problem with an approximate finite constraint optimization problem to obtain an approximate solution. Then, an auxiliary optimization problem is defined to obtain the value of the uncertainty variable corresponding to the maximum constraint violation. Through iterative loops, a solution that satisfies all constraints is ultimately obtained.

[0104] In this embodiment, the approximate optimization problem of the original optimization problem is obtained by a constraint relaxation technique based on adaptive discretization. The main idea of ​​this method is to find the uncertainty set E The worst-case finite subset of E k , used to replace the infinite set E , thus converting the infinite number of constraints in the semi-infinite optimization problem into a general optimization problem with finite constraints, which can be solved by a general solution algorithm. k The approximate optimization problem is as follows:

[0105]

[0106] in , E k Generated by internal program e It can be seen from the finite set of approximate optimization problems that there are finite constraints, which can be solved by general optimization methods.

[0107] The auxiliary optimization problem in this embodiment is mainly used to obtain the worst case feed rate prediction error corresponding to a specific ore drawing sequence. , and judge the feasibility of the solution. The auxiliary optimization problem is defined as follows:

[0108] in:

[0109] It should be noted that the formula Japanese style Underflow concentration and filter press pump opening and closing times and The feed rate of the thickener and the start time of the thickener underflow pump can be used to determine the feed rate of the thickener The auxiliary optimization problem is mainly to find the feed rate prediction error when the constraint degree is the largest under a certain ore drawing time series condition, so as to find the "worst case" of the feed rate prediction error under the current ore drawing operation condition. By solving the auxiliary optimization problem, it can be judged whether the current solution meets the constraint conditions of the original semi-infinite optimization problem. When When , we can prove that the solution It is feasible in the original optimization problem, otherwise, prove that the solution It is not feasible in the original optimization problem. After the approximate optimization problem of the original optimization problem and the auxiliary optimization problem are constructed, the original semi-infinite optimization problem can be solved. For the optimization problem represented by the constraints of the above-mentioned converted optimization model, such as Figure 4 As shown, the following steps can be used for optimization: Step 1: Initialization. Let the number of iterations be k = 0, select the initial finite subset of the set of feed quantity prediction errors .

[0110] Step 2: Solve the first k An approximate optimization problem, and its global optimal solution is obtained And the objective function value at this time ,in, For the k The feasible domain of an optimization problem.

[0111] Step 3: Make , solve the corresponding formula of the auxiliary optimization problem and obtain the optimal solution of the auxiliary optimization problem ,Right now:

[0112] Step 4: Judgement The positive and negative of If, the solution obtained in step 2 is In the feasible region of the original semi-infinite optimization problem, let the final solution , exit the iteration; if , explaining the solution It is not within the feasible region of the original optimization problem. , return to step 2 to continue iteration.

[0113] In the above optimization problem solving process, the solution of the approximate optimization problem needs to find the global optimal solution as much as possible. Since the objective function of the optimization problem in this embodiment is nonlinear, in order to reduce the possibility of the optimization falling into the local optimum, this embodiment uses a genetic algorithm to solve the approximate optimization problem to meet multiple constraints. The specific process can be referred to Figure 4 shown.

[0114] Through experimental verification, compared with the optimization model that does not consider the uncertainty of the feed rate prediction error, the operation optimized by the uncertainty interval optimization model proposed in the present invention has a constraint satisfaction rate increased by 98.4% to 100%, which significantly improves the operating stability of the thickener and reduces the probability of failure.

[0115] Based on the above analysis, in the actual field, this embodiment also proposes to build an intelligent optimization system for the thickening filter pressing process. The optimization of the thickener discharge operation is based on the real-time operation data collected by the field sensors. Therefore, the primary task of the system is to transmit the real-time data collected by the field sensors and transmit the thickener operation instructions to the corresponding control equipment. To this end, this embodiment establishes a data transmission mechanism of the system based on the OPC protocol, such as Figure 5 shown.

[0116] like Figure 6 As shown, another embodiment of the present invention also discloses a thickener operation optimization device taking into account feed rate fluctuations, comprising: A first acquisition module is used to obtain production data of the mineral processing site and pre-process the production data; A first screening module is configured to screen the pre-processed production data based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; a first calling module, configured to call a pre-built prediction model, input the target data into the prediction model, and obtain a predicted value of the feed rate of the thickener within a specified future time period, wherein the prediction model is formed based on a rolling ARIMA model and a PLS model; The second calling module is used to call the pre-built thickener operation optimization model, input the feed rate prediction value into the thickener operation optimization model, and obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain.

[0117] In one embodiment, constructing the thickener feed rate prediction model includes: Constructing the rolling ARIMA model, which is an input variable prediction model for predicting the input variables and determining the prediction range of the thickener feed amount, wherein the input variables match the target data; Assisting in constructing the PLS model based on the predicted values ​​of the input variables output by the rolling ARIMA model; The prediction model is constructed based on the rolling ARIMA model and the PLS model.

[0118] In one embodiment, constructing the rolling ARIMA model includes: Preprocessing and screening the first historical production data to obtain candidate training data; Processing the candidate training data based on a difference method to obtain first training data, wherein the time series corresponding to the first training data is stationary; The initial ARIMA model is trained based on the first training data and a rolling prediction mechanism to obtain the rolling ARIMA model. The rolling prediction mechanism periodically inserts new data into the head of the current first training data and removes the old data at the tail, so as to continue training the rolling ARIMA model with the updated first training data to obtain a new rolling ARIMA model.

[0119] In one embodiment, the training of the initial ARIMA model based on the first training data and the rolling forecast mechanism includes: Calculating an autocorrelation function, a partial autocorrelation function, and a difference order based on the first training data to sequentially determine the number of autoregressive terms, the number of lags of the forecast error, and the input parameters of the initial ARIMA model; determining a delay time between each of the input variables in the first training data and the thickener feed rate; Determining the number of prediction steps of the initial ARIMA model in combination with the preset prediction time span and the delay time; The initial ARIMA model is trained by combining the number of autoregressive terms, the number of lags of the prediction error, the input parameters, the number of prediction steps, and the rolling prediction mechanism.

[0120] In one embodiment, constructing the PLS model includes: Preprocessing and screening the second historical production data to obtain second training data, the second training data including grinding variables, flotation variables, and the thickener feed rate upstream of the thickener, wherein the grinding variables, flotation variables, and the thickener feed rate are time-series aligned; Train an initial PLS model based on a preset step size and the second training data to obtain a PLS model; Inputting the test input variables into the rolling ARIMA model to obtain predicted values ​​of the test input variables; The PLS model is tested based on the predicted values ​​of the test input variables.

[0121] In one embodiment, building a thickener operation optimization model includes: Determine the thickener underflow concentration prediction model; Determine the calculation model of energy consumption economic indicators; Determine the calculation model for the optimal ore storage volume and number of filter press cabinets; Determine the basic optimization model; The thickener operation optimization model is constructed based on the basic optimization model, the thickener underflow concentration prediction model, the energy consumption economic indicator calculation model, and the optimal ore storage capacity and filter press cabinet number calculation model.

[0122] In one embodiment, determining the thickener underflow concentration prediction model includes: Determining a predicted value of the thickener underflow concentration based on a relationship between the underflow concentration of the thickener and a change in the internal ore storage volume; The energy consumption economic indicator calculation model is determined, including: Determine the energy consumption economic indicators of the thickener underflow pump and the filter press pump based on the tiered electricity prices in the region; Determining the energy consumption economic index calculation model based on the energy consumption economic index of the thickener underflow pump and the energy consumption economic index of the filter press pump; The calculation model for determining the optimal ore storage capacity and the number of filter press cabinets includes: Under the condition of a certain feed rate, the optimal ore storage volume is achieved to minimize energy consumption and ensure safe operation of the thickener; Determining the number of ore storage cabinets based on the optimal ore storage volume; A calculation model for the optimal ore storage capacity and the number of filter press cabinets is determined based on the optimal ore storage capacity and the number of ore discharge cabinets.

[0123] In one embodiment, determining the basic optimization model includes: Construct an objective function with the goal of minimizing energy consumption; Constructing the basic optimization model based on the objective function and preset constraint conditions for ensuring normal and stable operation of the thickener; The constraints include that within one optimization cycle, the maximum underflow concentration of the thickener cannot exceed a preset upper limit, the minimum cannot be lower than a preset lower limit, and the timing constraints of the thickener discharge and the filter press start filtering, so as to ensure that the optimized discharge operation and filter press operation meet the timing requirements.

[0124] In one embodiment, the screening of the pre-processed production data based on time series correlation to obtain target data includes: Performing a delay analysis on the relationship between each input variable in the preprocessed production data and the feed rate of the thickener to obtain a time sequence relationship between the input variable and the feed rate of the thickener; determining, based on a cross-correlation function, a correlation between each of the input variables and the feed rate of the thickener at different delay times; Based on the correlation degree and the time series relationship, the pre-processed production data is screened based on the time series correlation to obtain the target data; The input variables include operating data of various devices involved in the upstream process of the thickener.

[0125] Furthermore, another embodiment of the present invention provides an electronic device, including: one or more processors; a memory configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the thickener operation optimization method considering feed rate fluctuation as described above.

[0126] Furthermore, another embodiment of the present invention provides a storage medium storing a computer program. When executed by a processor, the program implements the thickener operation optimization method for considering feed rate fluctuations as described above. It should be understood that each solution in this embodiment has the corresponding technical effects of the aforementioned method embodiments and will not be further described here.

[0127] Furthermore, an embodiment of the present invention also provides a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions. When the computer-executable instructions are executed, at least one processor performs a thickener operation optimization method that takes into account feed rate fluctuations, such as the one in the embodiment described above.

[0128] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the scope of the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the spirit and scope of protection of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present invention.

Claims

1. A thickener operation optimization method considering feed rate fluctuations, characterized in that: include: Obtaining production data from the mineral processing site and preprocessing the production data; The pre-processed production data is screened based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; Calling a pre-built prediction model, inputting the target data into the prediction model, and obtaining a predicted value of the feed rate of the thickener within a specified future time period, wherein the prediction model is formed based on a rolling ARIMA model and a PLS model; A pre-built thickener operation optimization model is called, and the feed rate prediction value is input into the thickener operation optimization model to obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration time series. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain.

2. The thickener operation optimization method considering feed rate fluctuation according to claim 1, characterized in that: Constructing the thickener feed rate prediction model includes: Constructing the rolling ARIMA model, which is an input variable prediction model for predicting the input variables and determining the prediction range of the thickener feed amount, wherein the input variables match the target data; Assisting in constructing the PLS model based on the predicted values ​​of the input variables output by the rolling ARIMA model; The prediction model is constructed based on the rolling ARIMA model and the PLS model.

3. The thickener operation optimization method considering feed rate fluctuation according to claim 2, characterized in that: The step of constructing the rolling ARIMA model includes: Preprocessing and screening the first historical production data to obtain candidate training data; Processing the candidate training data based on a difference method to obtain first training data, wherein the time series corresponding to the first training data is stationary; The initial ARIMA model is trained based on the first training data and a rolling prediction mechanism to obtain the rolling ARIMA model. The rolling prediction mechanism periodically inserts new data into the head of the current first training data and removes the old data at the tail, so as to continue training the rolling ARIMA model with the updated first training data to obtain a new rolling ARIMA model.

4. The thickener operation optimization method considering feed rate fluctuation according to claim 3, characterized in that: The training of the initial ARIMA model based on the first training data and the rolling forecast mechanism includes: Calculating an autocorrelation function, a partial autocorrelation function, and a difference order based on the first training data to sequentially determine the number of autoregressive terms, the number of lags of the forecast error, and the input parameters of the initial ARIMA model; Determining a delay time between each of the input variables in the first training data and the thickener feed rate; Determining the number of prediction steps of the initial ARIMA model in combination with the preset prediction time span and the delay time; The initial ARIMA model is trained by combining the number of autoregressive terms, the number of lags of the prediction error, the input parameters, the number of prediction steps, and the rolling prediction mechanism.

5. The thickener operation optimization method considering feed rate fluctuation according to claim 2, characterized in that: Constructing the PLS model includes: Preprocessing and screening the second historical production data to obtain second training data, the second training data including grinding variables, flotation variables, and the thickener feed rate upstream of the thickener, wherein the grinding variables, flotation variables, and the thickener feed rate are time-series aligned; Train an initial PLS model based on a preset step size and the second training data to obtain a PLS model; Inputting the test input variables into the rolling ARIMA model to obtain predicted values ​​of the test input variables; The PLS model is tested based on the predicted values ​​of the test input variables.

6. The thickener operation optimization method considering feed rate fluctuation according to claim 1, characterized in that: Build a thickener operation optimization model, including: Determine the thickener underflow concentration prediction model; Determine the calculation model of energy consumption economic indicators; Determine the calculation model for the optimal ore storage volume and number of filter press cabinets; Determine the basic optimization model; The thickener operation optimization model is constructed based on the basic optimization model, the thickener underflow concentration prediction model, the energy consumption economic indicator calculation model, and the optimal ore storage capacity and filter press cabinet number calculation model.

7. The thickener operation optimization method considering feed rate fluctuation according to claim 6, characterized in that: Determining the thickener underflow concentration prediction model includes: Determining a predicted value of the thickener underflow concentration based on a relationship between the underflow concentration of the thickener and a change in the internal ore storage volume; The energy consumption economic indicator calculation model is determined, including: Determine the energy consumption economic indicators of the thickener underflow pump and the filter press pump based on the tiered electricity prices in the region; Determining the energy consumption economic index calculation model based on the energy consumption economic index of the thickener underflow pump and the energy consumption economic index of the filter press pump; The calculation model for determining the optimal ore storage capacity and the number of filter press cabinets includes: Under the condition of a certain feed rate, the optimal ore storage volume is achieved to minimize energy consumption and ensure safe operation of the thickener; Determining the number of ore storage cabinets based on the optimal ore storage volume; A calculation model for the optimal ore storage capacity and the number of filter press cabinets is determined based on the optimal ore storage capacity and the number of ore discharge cabinets.

8. The thickener operation optimization method considering feed rate fluctuation according to claim 6, characterized in that: Determining the basic optimization model includes: Construct an objective function with the goal of minimizing energy consumption; Constructing the basic optimization model based on the objective function and preset constraint conditions for ensuring normal and stable operation of the thickener; The constraints include that within one optimization cycle, the maximum underflow concentration of the thickener cannot exceed a preset upper limit, the minimum cannot be lower than a preset lower limit, and the timing constraints of the thickener discharge and the filter press start filtering, so as to ensure that the optimized discharge operation and filter press operation meet the timing requirements.

9. The thickener operation optimization method considering feed rate fluctuation according to claim 4, characterized in that: The pre-processed production data is screened based on time series correlation to obtain target data, including: Performing a delay analysis on the relationship between each input variable in the preprocessed production data and the feed rate of the thickener to obtain a time sequence relationship between the input variable and the feed rate of the thickener; determining, based on a cross-correlation function, a correlation between each of the input variables and the feed rate of the thickener at different delay times; Based on the correlation degree and the time series relationship, the pre-processed production data is screened based on the time series correlation to obtain the target data; The input variables include operating data of various devices involved in the upstream process of the thickener.

10. A thickener operation optimization device taking into account feed rate fluctuations, characterized in that: include: A first acquisition module is used to obtain production data of the mineral processing site and pre-process the production data; A first screening module is configured to screen the pre-processed production data based on time series correlation to obtain target data, wherein the time series correlation represents a time series matching degree between the production data and the thickener feed rate; a first calling module, configured to call a pre-built prediction model, input the target data into the prediction model, and obtain a predicted value of the feed rate of the thickener within a specified future time period, wherein the prediction model is formed based on a rolling ARIMA model and a PLS model; The second calling module is used to call the pre-built thickener operation optimization model, input the feed rate prediction value into the thickener operation optimization model, and obtain the thickener's ore discharge start time series, continuous ore discharge time series, filter press pump start time series, and filter press pump start duration. The thickener operation optimization model is used to determine the thickener operation time series in which the underflow concentration of the thickener always meets the constraint conditions and the energy consumption of the underflow pump and filter press pump is minimized when the feed rate prediction error is uncertain.

Images