Coal washing process production scheduling optimization system based on dynamic modeling

Abstract

The present application relates to the technical field of coal washing process production scheduling, and specifically discloses a coal washing process production scheduling optimization system based on dynamic modeling, comprising a production data real-time acquisition module, a dynamic raw coal property analysis module, a production state prediction module, a scheduling strategy generation module, a constraint condition verification module and an optimization decision module. The present application acquires raw coal properties, equipment states and quality index data in real time, obtains dynamic characterization of raw coal selectability curves through sliding window sampling, density composition analysis and adaptive filtering processing, deduces a production state prediction sequence using a Markov decision process, adjusts a process standard flow to generate an adaptive scheduling strategy scheme, screens a feasible strategy set after quality constraint verification, finally solves a multi-objective optimization problem with the goal of maximizing coal washing efficiency and minimizing energy consumption, and outputs optimal production scheduling instructions; the present application tracks raw coal property fluctuations and predicts production state evolution trends in real time, and ensures that clean coal product quality meets standards.

Description

Technical Field

[0001] This invention relates to the field of coal washing process production scheduling technology, specifically to a coal washing process production scheduling optimization system based on dynamic modeling. Background Technology

[0002] Coal washing production involves removing impurities such as gangue and sulfur from raw coal through processes like heavy media separation, jigging, and flotation to obtain clean coal products of varying quality grades. In actual production, the settings of process parameters such as heavy media separation density, hydrocyclone inlet pressure, and media addition amount directly determine the ash content, yield, and energy consumption level of the clean coal product. Traditional coal washing production scheduling typically employs two methods: one is a timed adjustment method based on a fixed process standard flow, where operators set parameters according to predetermined time nodes and fixed parameter values; the other is a lagging adjustment method based on the results of on-duty testing, where subsequent process parameters are corrected after obtaining the quick ash test results for the clean coal.

[0003] The properties of raw coal in coal washing production fluctuate frequently, resulting in dynamic changes in the selectivity curve of raw coal entering the washing process. Traditional scheduling methods rely on lagging test results and manual experience to set fixed process parameters, which cannot track the continuous evolution trend of raw coal properties in real time. It is also difficult to achieve dynamic and coordinated adjustment among multiple coupled process parameters such as heavy medium separation density, hydrocyclone pressure, and medium addition amount. This leads to large fluctuations in the quality of clean coal products, decreased coal washing efficiency, and increased medium and power consumption. Existing technologies lack a solution that can predict future production status based on the dynamic changes in raw coal properties and automatically generate adaptive scheduling instructions that take into account both coal washing efficiency and energy consumption while meeting ash content and yield quality constraints. Summary of the Invention

[0004] The purpose of this invention is to provide a production scheduling optimization system for the coal washing process based on dynamic modeling, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A coal washing process production scheduling optimization system based on dynamic modeling includes:

[0007] The real-time production data acquisition module obtains raw coal property parameters, equipment operating status data, and current coal washing product quality index data from multiple monitoring nodes on the coal washing production line.

[0008] The dynamic raw coal property analysis module processes historical raw coal data based on raw coal property parameters and time series analysis algorithms, and outputs a dynamic representation of raw coal selectivity curve that matches the current production period.

[0009] The production status prediction module, based on the dynamic characterization of the raw coal washability curve and combined with equipment operating status data, uses Markov decision process to perform state transition deduction of the coal washing process and generates a production status prediction sequence containing multiple time nodes.

[0010] The scheduling strategy generation module uses the production status prediction sequence to dynamically adjust the pre-stored standard process flow of coal washing production and generate an adaptive scheduling strategy scheme that matches the current production conditions.

[0011] The constraint verification module performs constraint matching verification between the adaptive scheduling strategy scheme and the quality indicator data, and selects a set of feasible scheduling strategies that meet the preset quality threshold.

[0012] The optimization decision module, based on the set of feasible scheduling strategies, solves a multi-objective optimization problem with the objective functions of maximizing coal washing efficiency and minimizing energy consumption, determines the optimal production scheduling instruction, and outputs it to the coal washing production line control system.

[0013] As a further aspect of the present invention: the dynamic characterization of the raw coal washability curve that matches the current production period specifically includes:

[0014] Sliding window sampling is performed on historical raw coal data to generate a sequence of raw coal property parameters for multiple consecutive time windows;

[0015] Density composition analysis was performed on the sequence of property parameters of each raw coal to obtain the yield and ash content time series data of each density level;

[0016] Adaptive exponential weighted filtering was applied to the time series yield data at each density level to obtain smoothed yield trend values.

[0017] Based on the smoothed yield trend value, a raw coal washout curve is constructed using cubic spline interpolation, and the raw coal washout curve is used as a dynamic representation that matches the current production period.

[0018] As a further aspect of the present invention: the density composition analysis of the sequence of raw coal property parameters to obtain the yield and ash content time series data of each density level specifically includes:

[0019] Based on the ash content and particle size distribution in the raw coal property parameter sequence, the raw coal is divided into a preset number of virtual density intervals, and an initial yield allocation coefficient is assigned to each virtual density interval.

[0020] The yield allocation coefficient of each virtual density interval is weighted and summed with the ash characteristic value of the corresponding density interval to obtain the calculated ash value.

[0021] With the goal of approximating the measured ash content in the raw coal property parameter sequence by calculating the ash content value, the yield allocation coefficient is adjusted iteratively until the difference between the two is less than a preset threshold.

[0022] The yield allocation coefficients after iterative convergence are used as the yields of each density level at the current time point, and combined with the preset ash content feature values, the yields and ash content data of each density level at the corresponding time point are output.

[0023] As a further aspect of the present invention: the generation of a production status prediction sequence containing multiple time points specifically includes:

[0024] The real-time change gradient of the yield of each density level is extracted from the dynamic characterization of the raw coal washability curve, and the density of the heavy medium suspension and the inlet pressure of the hydrocyclone are extracted from the equipment operation status data and combined into the production status vector at the current moment.

[0025] The production state vector is matched with the preset state space template and assigned to the corresponding discrete state category. Based on the historical state transition frequency statistics, a state transition probability table from the corresponding category to subsequent categories is constructed.

[0026] Starting from the current state, the most likely state at each subsequent time node is determined sequentially using a step-by-step recursive method based on the state transition probability table, and these most likely states are arranged in chronological order to generate a production state prediction sequence.

[0027] As a further aspect of the present invention: the step-by-step recursive method for determining the most probable state at each subsequent time point specifically includes:

[0028] Starting from the production status at the current time point, obtain all candidate states for the next time point and their corresponding transition probability values ​​according to the state transition probability table;

[0029] Calculate the cumulative probability for each candidate state, where the cumulative probability of the current time node is assigned a value of one, and the cumulative probability of subsequent nodes is the cumulative probability of the preceding node multiplied by the probability value of the corresponding transition.

[0030] The candidate state with the highest cumulative probability value is selected as the most likely state at the next time node, and the cumulative probability value of the most likely state is used as the benchmark value for the next round of recursion.

[0031] Until all preset time nodes are recursively calculated, the most likely states selected in each round are arranged in chronological order to form a production state prediction sequence.

[0032] As a further aspect of the present invention: the generation of an adaptive scheduling strategy scheme that matches the current production conditions specifically includes:

[0033] The standard process flow of coal washing production was analyzed, and several key process nodes and their default parameter values, including heavy medium separation density, hydrocyclone pressure setpoint and medium addition amount, were extracted.

[0034] The production status prediction sequence is matched one-to-one with each key process node in chronological order to determine the predicted raw coal washability status and equipment operating status corresponding to the execution time of each node.

[0035] Based on the predicted state corresponding to each node, query the pre-stored parameter-state response relationship table for the process parameter adjustment value that matches the corresponding predicted state;

[0036] The default parameter values ​​of each key process node are replaced with the queried process parameter adjustment values, and combined to generate an adaptive scheduling strategy scheme that matches the current production conditions.

[0037] As a further aspect of the present invention: the step of selecting a set of feasible scheduling strategies that meet a preset quality threshold specifically includes:

[0038] The heavy medium separation density and hydrocyclone pressure setpoints of each key process node are extracted from the adaptive scheduling strategy and combined into a parameter combination to be verified.

[0039] The parameter combination to be verified is associated with the dynamic characterization of the raw coal washability curve in the current quality index data. The pre-stored product quality response surface is input, and the corresponding predicted clean coal ash content and predicted clean coal yield are obtained by interpolation.

[0040] The predicted clean coal ash content is compared with the preset ash content threshold, and the predicted clean coal yield is compared with the preset yield threshold to determine whether both meet the threshold range requirements simultaneously.

[0041] The adaptive scheduling strategy schemes corresponding to the parameter combinations to be verified that simultaneously meet the requirements of ash content threshold and yield threshold are selected and included in the set of feasible scheduling strategies.

[0042] As a further aspect of the present invention: the step of determining the optimal production scheduling instruction and outputting it to the coal washing production line control system specifically includes:

[0043] Iterate through each strategy scheme in the set of feasible scheduling strategies, extract the heavy medium separation density sequence and hydrocyclone pressure setting sequence, combine them with the production status prediction sequence, and calculate the expected coal washing efficiency value corresponding to each strategy through the preset efficiency response function.

[0044] Based on the equipment start-up and shutdown sequence and media addition amount in each strategy scheme, the instantaneous power and media consumption of each production node are calculated and accumulated to obtain the expected energy consumption per ton of coal corresponding to the strategy.

[0045] The expected coal washing efficiency and expected energy consumption per ton of coal for each strategy are mapped to a two-dimensional decision space with efficiency as the vertical axis and energy consumption as the horizontal axis, and the non-dominated strategy located on the upper right convex envelope of the space is identified.

[0046] From the non-dominated strategy options, select the option with the largest ratio of expected coal washing efficiency to expected energy consumption per ton of coal, encapsulate its corresponding process parameter sequence according to time nodes to generate the optimal production scheduling instruction, and output it to the coal washing production line control system.

[0047] The beneficial effects of this invention are:

[0048] (1) This invention uses a dynamic raw coal property analysis module to perform sliding window sampling, density composition analysis, and adaptive exponential weighted filtering on historical raw coal data. This allows for real-time tracking of the dynamic changes in the raw coal washability curve, eliminating interference from instantaneous fluctuations, and making the output raw coal washability curve more accurately reflect the true physical characteristics of the raw coal currently being washed. Combined with the production status prediction module, which uses a Markov decision process for state transition deduction, the evolution trend of the production status can be predicted in advance, providing accurate input basis for the generation of subsequent scheduling strategies and significantly improving the response speed and adaptability of the coal washing process to fluctuations in raw coal properties.

[0049] (2) This invention optimizes the decision-making module to perform multi-objective optimization with the goals of maximizing coal washing efficiency and minimizing energy consumption, under the premise of meeting the constraints of clean coal ash content and yield quality thresholds. It identifies non-dominated strategy schemes from the set of feasible scheduling strategies and selects the scheme with the largest efficiency-energy consumption ratio as the optimal scheduling instruction. This scheme can reduce the power consumption and media consumption of the main equipment such as the heavy medium cyclone and medium pump while ensuring product quality, improve clean coal yield, maximize the economic benefits of the coal washing production process, and provide executable adaptive scheduling instructions for the production line control system. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1As shown, this invention is a production scheduling optimization system for the coal washing process based on dynamic modeling, comprising:

[0054] The real-time production data acquisition module obtains raw coal property parameters, equipment operating status data, and current coal washing product quality index data from multiple monitoring nodes on the coal washing production line.

[0055] The dynamic raw coal property analysis module processes historical raw coal data based on raw coal property parameters and time series analysis algorithms, and outputs a dynamic representation of raw coal selectivity curve that matches the current production period.

[0056] The production status prediction module, based on the dynamic characterization of the raw coal washability curve and combined with equipment operating status data, uses Markov decision process to perform state transition deduction of the coal washing process and generates a production status prediction sequence containing multiple time nodes.

[0057] The scheduling strategy generation module uses the production status prediction sequence to dynamically adjust the pre-stored standard process flow of coal washing production and generate an adaptive scheduling strategy scheme that matches the current production conditions.

[0058] The constraint verification module performs constraint matching verification between the adaptive scheduling strategy scheme and the quality indicator data, and selects a set of feasible scheduling strategies that meet the preset quality threshold.

[0059] The optimization decision module, based on the set of feasible scheduling strategies, solves a multi-objective optimization problem with the objective functions of maximizing coal washing efficiency and minimizing energy consumption, determines the optimal production scheduling instruction, and outputs it to the coal washing production line control system.

[0060] The real-time production data acquisition module obtains raw coal property parameters, equipment operating status data, and current coal washing product quality index data from multiple monitoring nodes on the coal washing production line, specifically including:

[0061] The real-time production data acquisition module is connected to the data interfaces of multiple online monitoring instruments and control systems deployed on the coal washing production line, and synchronously acquires three types of production data according to a preset sampling period.

[0062] The first category consists of raw coal property parameters. Online ash and moisture analyzers are installed on the feed conveyor of the coal washing production line, and a particle size analyzer is installed at the feed chute to monitor the ash percentage, moisture percentage, and particle size distribution percentage of the raw coal entering the production line in real time. Simultaneously, the cumulative weight data of raw coal for the current shift is collected at the belt scale in the raw coal storage yard. These data are aggregated to a data acquisition unit via a fieldbus, forming a time-stamped sequence of raw coal property parameters.

[0063] The second category is equipment operating status data. Real-time data is obtained from the distributed control system of the heavy medium separation system, including the inlet pressure of the heavy medium hydrocyclone, the liquid level in the combined medium tank, the liquid level in the dilute medium tank, and the suspension density value fed back by the media density meter. Vibration amplitude and bearing temperature values ​​of the desliming screen and centrifuge are read from equipment vibration monitoring sensors. Instantaneous current, active power, and cumulative power consumption of each device are collected from the smart meters in the power distribution cabinet. All of this data is uploaded to the real-time database server via a programmable logic controller (PLC).

[0064] The third category consists of quality indicator data for the current coal washing products. Online ash and moisture analyzers are installed on the clean coal discharge conveyor to monitor the ash and moisture percentages of the final clean coal product in real time. The quick ash analysis results of the clean coal samples taken during the current shift, as well as the ash content data of the flotation tailings, are automatically retrieved from the coal washing plant's laboratory information management system. These quality indicator data, along with the real-time data collected on the production line, are synchronously stored in the industrial real-time database, providing a benchmark for subsequent constraint verification modules.

[0065] In the dynamic raw coal property analysis module, based on raw coal property parameters, historical raw coal data is processed according to a time-series analysis algorithm to output a dynamic representation of raw coal selectivity curves that match the current production period. Specifically, this includes:

[0066] The first step is to perform sliding window sampling on historical raw coal data. Extract all time-stamped raw coal ash content, moisture content, particle size distribution, and cumulative weight of washed raw coal from the industrial real-time database over the past 48 hours. Set the sliding window time span to 120 minutes and the window sliding step size to 15 minutes. Starting from the current moment and moving backward, extract a 120-minute time window every 15 minutes. Arrange the raw coal property parameters of all sampling moments within this window in chronological order to generate multiple consecutive time window sequences of raw coal property parameters.

[0067] The second step involves density composition analysis of the raw coal property parameter sequences. For each time window, based on the average ash content and average particle size distribution data within that window, the raw coal is divided into six pre-defined virtual density intervals. The specific interval boundaries are set as less than 1.3 g / cm³, 1.3–1.4 g / cm³, 1.4–1.5 g / cm³, 1.5–1.6 g / cm³, 1.6–1.8 g / cm³, and greater than 1.8 g / cm³. An initial yield allocation coefficient is assigned to each virtual density interval, with the sum of all initial coefficients being 100%. The yield allocation coefficient for each interval is then weighted and summed with the pre-defined ash content characteristic value for that interval to obtain the calculated ash content value for that time window. The ash content characteristic values ​​for each density interval are statistically derived from historical buoyancy test data. For example, the ash content characteristic value for the interval less than 1.3 g / cm³ is set to 8%, and the ash content characteristic value for the interval greater than 1.8 g / cm³ is set to 75%. The calculated ash content value is compared with the measured average ash content value in the raw coal property parameter sequence for that time window, and the difference between the two is calculated. With the goal of minimizing this difference, the yield allocation coefficients for each density interval are iteratively adjusted using the least squares method. In each iteration, the coefficients for each interval are increased or decreased by a preset step size according to the size of the difference, until the absolute value of the difference between the calculated ash content value and the measured ash content value is less than 0.1 percentage points. The yield allocation coefficients for the six density intervals obtained after the iteration convergence are used as the yield data for each density level in that time window. Combined with the preset ash content characteristic values ​​for each density interval, the yield and ash content data for each density level in that time window are output. The above process is repeated for all sliding windows to obtain the time-series data of yield and ash content for each density level arranged in chronological order.

[0068] The third step involves adaptive exponential weighted filtering of the yield time series data for each density level. For each density level, the yield sequence over time is processed separately. The initial smoothing factor is set to 0.3. For the first data point in the sequence, the smoothed yield trend value is directly taken from the original yield value at that point. For the yield data at each subsequent time point, the difference between the original yield value at the current time point and the smoothed yield trend value at the previous time point is calculated. Based on the direction of this difference, the smoothing factor is adaptively adjusted: if the difference is positive and its absolute value is greater than 5 percentage points, the smoothing factor is increased to 0.6 to accelerate the response to new trends; if the difference is negative and its absolute value is greater than 5 percentage points, the smoothing factor is also increased to 0.6; if the absolute value of the difference is between 2 and 5 percentage points, the smoothing factor remains at 0.3; if the absolute value of the difference is less than 2 percentage points, the smoothing factor is decreased to 0.1 to enhance filtering stability. The adjusted smoothing factor is then weighted with the current original yield value and the previous smoothed value to obtain the smoothed yield trend value at the current time point. By processing all time points of all density levels sequentially, a smoothed yield trend value sequence for each density level is obtained.

[0069] The fourth step involves constructing a raw coal washability curve using cubic spline interpolation. For the current production time, the latest yield trend value for each density level, after filtering and smoothing, is taken as the representative yield data for the current period. The median of the density intervals corresponding to each density level is used as the x-axis, specifically 1.25 g / cm³, 1.35 g / cm³, 1.45 g / cm³, 1.55 g / cm³, 1.70 g / cm³, and 1.90 g / cm³. The cumulative yield for each density level is used as the y-axis, obtained by progressively accumulating the yield from the low-density interval to the high-density interval. Between two adjacent density median points, a cubic polynomial is used for interpolation. The polynomial coefficients are determined by the y-axis values ​​at both ends of the interval and the continuity condition of the derivative at those ends. Through interpolation calculations, the cumulative yield value is obtained for every 0.01 g / cm³ within the density range from 1.2 g / cm³ to 2.0 g / cm³. Connecting these discrete points forms a smooth curve, which serves as a dynamic representation corresponding to the raw coal washability curve, and is output to the production status prediction module.

[0070] In the production status prediction module, based on the dynamic characterization of the raw coal washability curve and combined with equipment operating status data, a Markov decision process is used to perform state transition deduction of the coal washing process, generating a production status prediction sequence containing multiple time points, specifically including:

[0071] The first step is to construct the production state vector for the current moment. From the dynamic raw coal washability curve, the real-time change gradient of the yield at each density level is extracted. Specifically, for each density level, the smoothed yield trend value at the current moment is compared with the yield trend value of the same density level at the previous moment. The difference between the two is calculated, and this difference is divided by the sampling time interval of 15 minutes to obtain the yield change rate for that density level, expressed as percentage points per minute. Simultaneously, the measured density of the heavy medium suspension at the current moment, expressed as grams per cubic centimeter, and the measured pressure at the hydrocyclone inlet, expressed as kilopascals, are extracted from the equipment operating status data. All of the above data are combined into a multi-dimensional vector, serving as the production state vector for the current moment. This vector contains the yield change gradients for 6 density levels, 1 heavy medium suspension density value, and 1 hydrocyclone inlet pressure value, for a total of 8 dimensions.

[0072] The second step is to assign the production state vectors to discrete state categories. A state space template is pre-constructed offline. Specifically, the construction method involves collecting production state vectors from all times over the past 90 days and using K-means clustering to divide them into 20 pre-defined state categories. Each category generates a cluster center vector and a corresponding cluster radius. The cluster radius is the maximum Euclidean distance from all samples within that category to the center vector. For the current production state vector, its Euclidean distance to each cluster center vector is calculated sequentially. If this distance is less than the cluster radius of the corresponding category, the vector is assigned to that category; if it falls into multiple categories, it is assigned to the category with the smallest distance; if it does not fall into any category, it is assigned to the closest category. The assigned category number is the discrete state category at the current time.

[0073] The third step is to construct a state transition probability table. This involves statistically analyzing historical data to determine the frequency of transitions from each state category to other state categories. The table iterates through the state sequences of the past 90 days in chronological order. For each pair of adjacent states, if the state at the previous time step was of category [missing information], then [missing information]. The state at the next moment is the category. Then in the counting matrix, Increase by 1. After the statistics are completed, for each state category... Calculate its orientation to all state categories The transition probability is calculated using the following formula: ;

[0074] in, Indicates from state category Transfer to state category The probability, For historical statistics from categories Transfer to Category The number of times, the denominator is from the category The total number of transfers to all possible categories. If the denominator is zero, it means there has been no transfer from any category in history. The departure transfer is then set. All are equal to 1 / 20. Store all probability values ​​in a 20x20 transition probability table, with each row... Each column corresponds to the source state category. The corresponding target state category, and the sum of the probabilities of each row is 1.

[0075] The fourth step involves a step-by-step recursive prediction starting from the current state. The prediction step size is set to the next 8 time nodes, with each node spaced 15 minutes apart, meaning the production state is predicted for the next 2 hours. First, the current state category is determined, denoted as S0, and its cumulative probability is assigned a value of 1. For the first future time node, all 20 candidate target state categories and their transition probability values ​​corresponding to row S0 are read from the transition probability table. For each candidate state category... Calculate its cumulative probability value, which is equal to the cumulative probability of the current node multiplied by the transition probability. The candidate state category with the highest cumulative probability value is selected as the most probable state of the first future node, denoted as S1, and its cumulative probability value is used as the baseline value for the next round of recursion. Then, starting from S1, the above process is repeated: the transition probabilities corresponding to row S1 are read from the transition probability table, the cumulative probability of each candidate state category is calculated (equal to the baseline value multiplied by the corresponding transition probability), the largest one is selected as the most probable state S2 of the second future node, and the baseline value is updated to the cumulative probability of that state. This process is repeated until all eight time nodes are predicted, resulting in the state category sequence S1, S2, ..., S8. This sequence is then arranged in chronological order to generate the production state prediction sequence.

[0076] In the scheduling strategy generation module, the pre-stored standard process flow for coal washing is dynamically adjusted using the production status prediction sequence to generate an adaptive scheduling strategy scheme that matches the current production conditions. Specifically, this includes:

[0077] The first step is to analyze the standard coal washing production process flow. A pre-stored standard coal washing production process flow file is read from the database. This file records the complete operation flow from raw coal input to product output in the form of a process timeline. Several process nodes that have a critical impact on the final product quality and yield are extracted from this flow, specifically including: the heavy medium separation density setting node, the hydrocyclone inlet pressure setting node, and the medium addition amount adjustment node. Each process node is associated with a default parameter value; for example, the default value for heavy medium separation density is 1.45 g / cm³, the default value for hydrocyclone inlet pressure is 220 kPa, and the default value for medium addition amount is 1.5 m³ / h. These default parameter values ​​are applicable to normal production conditions when the raw coal properties are stable.

[0078] The second step is to match the production status prediction sequence with key process nodes. The production status prediction sequence contains predicted statuses for eight future time nodes, each 15 minutes apart, covering the production conditions for the next two hours. Key process nodes in the standard process flow are ordered according to their execution sequence on the timeline; for example, the 15-minute mark is the first adjustment node for heavy media separation density, the 30-minute mark is the hydrocyclone pressure adjustment node, and the 45-minute mark is the media addition adjustment node. The execution time of each process node is matched one-to-one with the predicted status at the corresponding time point in the production status prediction sequence. Specifically, if the execution time of a process node is the 30-minute mark in the future, the predicted raw coal washability status and equipment operating status corresponding to the 30-minute mark in the production status prediction sequence are selected as the associated status data for that node.

[0079] The third step is to query the process parameter adjustment values ​​in the parameter-state response relationship table. The parameter-state response relationship table is pre-constructed offline by collecting production data before and after each process parameter adjustment over the past 180 days. This includes the shape of the raw coal washability curve, the density of the heavy medium suspension, the hydrocyclone inlet pressure at the time of adjustment, and the clean coal ash content and yield achieved after the adjustment. Each set of data is categorized according to the shape of the raw coal washability curve into three types: easy-to-wash coal, medium-washable coal, and difficult-to-wash coal. For each type, the average clean coal yield and ash content corresponding to different heavy medium separation density settings are statistically analyzed. The density value corresponding to the highest clean coal yield and ash content within the standard is selected as the optimal density setting for that type of raw coal. Similarly, the impact of different hydrocyclone pressure settings on the fine-particle separation effect is statistically analyzed, and the optimal pressure value is selected. These optimal settings are then compiled into a table, forming a parameter-state response relationship table with the raw coal washability category and the current equipment status as query conditions. For the predicted state corresponding to the current process node, first determine its sorting category based on the characteristics of the raw coal sorting curve in the predicted state, and then, in combination with the equipment operating status in the predicted state, query the matching heavy medium separation density adjustment value, hydrocyclone pressure adjustment value, and medium addition amount adjustment value in the response relationship table.

[0080] The fourth step is to generate an adaptive scheduling strategy. The default parameter values ​​of each key process node in the standard process flow are replaced one by one with the adjusted process parameter values ​​obtained in the third step. For example, if the default value for the heavy medium sorting density at the 15th minute is 1.45 g / cm³, and the obtained adjusted value is 1.42 g / cm³, then 1.42 is used as the new setting value for that node. All the replaced process nodes and their parameter values ​​are then recombined in their original chronological order to form a complete scheduling strategy. This strategy includes the specific time point of each process adjustment within the next two hours, the name of the adjusted parameter, and the adjusted target value, matching the current predicted production status and serving as input for subsequent constraint verification.

[0081] In the constraint verification module, the adaptive scheduling strategy scheme is matched and verified with the quality index data to select a set of feasible scheduling strategies that meet the preset quality threshold, specifically including:

[0082] The first step is to extract the parameter combinations to be verified. For each adaptive scheduling strategy, the adjustment instructions for multiple key process nodes within the next two hours are parsed. From these instructions, all nodes involving heavy medium sorting density settings are extracted, and all density settings are arranged in chronological order to form a density setting sequence. Similarly, all nodes involving hydrocyclone inlet pressure settings are extracted and arranged in chronological order to form a pressure setting sequence. These two sequences are combined as the parameter combinations to be verified for this strategy. Each parameter combination to be verified corresponds to a unique scheduling strategy identifier.

[0083] The second step involves constructing and querying the product quality response surface. The product quality response surface is constructed offline beforehand by collecting data on clean coal ash content and clean coal yield achieved under different raw coal washability conditions, different heavy media separation densities, and different hydrocyclone pressure settings over the past three years. The cumulative yield of floating matter at a density of 1.4 g / cm³ in the raw coal washability curve is used as the raw coal washability characteristic value. This characteristic value, the heavy media separation density setting, and the hydrocyclone pressure setting are used as input variables, and clean coal ash content and clean coal yield are used as output variables. Kriging interpolation is used to spatially interpolate the discrete historical data points, constructing two three-dimensional response surfaces, corresponding to the clean coal ash content response surface and the clean coal yield response surface, respectively. These two surfaces use the raw coal washability characteristic value, heavy media separation density, and hydrocyclone pressure as coordinate axes. Each point on the surface represents the predicted clean coal ash content or predicted clean coal yield achievable under that coordinate combination.

[0084] For the current parameter combination to be verified, the cumulative yield of floating matter at a density of 1.4 g / cm³ is first extracted from the dynamic characterization of the raw coal washability curve at the current moment, serving as the current raw coal washability characteristic value. This characteristic value is then paired one-to-one with each heavy medium separation density setpoint and hydrocyclone pressure setpoint in the parameter combination to be verified, and input into the clean coal ash content response surface and clean coal yield response surface, respectively. Through spatial interpolation calculation of the response surfaces, the predicted clean coal ash content and predicted clean coal yield corresponding to each set of setpoints are obtained. Since a scheduling strategy scheme contains setpoints for multiple time nodes, it is necessary to calculate the predicted value for each node separately, and take the average value over the entire time period as the final predicted clean coal ash content and predicted clean coal yield of the scheme.

[0085] The third step involves comparing the ash content with preset quality thresholds. The preset ash content thresholds are set based on the grade requirements of the clean coal product; for example, for Grade 1 clean coal, the ash content threshold is set to be less than or equal to 9.5%. The preset yield thresholds are set based on the economic benefits requirements of the coal washing plant; for example, the clean coal yield threshold is set to be greater than or equal to 75%. The predicted clean coal ash content calculated in the second step is compared with 9.5% to determine if it is less than or equal to 9.5%. The predicted clean coal yield is compared with 75% to determine if it is greater than or equal to 75%. If both conditions are met simultaneously, the scheduling strategy passes the quality constraint verification.

[0086] The fourth step involves filtering and incorporating feasible scheduling strategies into the set. For scheduling strategies that pass the verification in step three, their corresponding parameter combinations and scheme identifiers are saved to the set of feasible scheduling strategies. Schemes that fail verification, i.e., those predicting a clean coal ash content greater than 9.5% or a predicting clean coal yield less than 75%, are excluded from subsequent optimization processes. The final set of feasible scheduling strategies includes all candidate schemes that simultaneously meet the ash content and yield threshold requirements. Each scheme is accompanied by its complete sequence of process parameter settings, serving as input for the next optimization decision.

[0087] In the optimization decision module, based on the set of feasible scheduling strategies, a multi-objective optimization problem is solved with the objective functions of maximizing coal washing efficiency and minimizing energy consumption. The optimal production scheduling instruction is determined and output to the coal washing production line control system, specifically including:

[0088] The first step is to calculate the expected coal washing efficiency for each strategy. All candidate strategies in the feasible scheduling strategy set are traversed. For each strategy, the heavy medium separation density setpoint sequence and hydrocyclone pressure setpoint sequence are extracted, along with the corresponding production state prediction sequence. The heavy medium separation density setpoint, hydrocyclone pressure setpoint, and the predicted raw coal washability curve characteristic value for each time point are substituted into a pre-constructed efficiency response function. This efficiency response function is constructed as follows: actual clean coal yield data under different process parameter combinations over the past two years are collected, and a multiple regression method is used to fit the functional relationship between clean coal yield and heavy medium separation density, hydrocyclone pressure, and raw coal washability characteristic value. For each time point, the three input parameters are substituted into the function to calculate the instantaneous clean coal yield for that time point. The instantaneous clean coal yields of all time points are arithmetically averaged to obtain the expected coal washing efficiency value of the strategy, expressed as a percentage.

[0089] The second step is to calculate the expected energy consumption per ton of coal for each strategy. For each strategy, based on the start-up and shutdown sequence of the included equipment, the operating time of the main equipment such as the heavy medium cyclone separator, desliming screen, centrifuge, and media pump is calculated for the next two hours. The rated power of each piece of equipment is read from the real-time database, and the operating time of each piece of equipment is multiplied by the rated power to obtain the expected power consumption of a single piece of equipment. The expected power consumption of all equipment is summed to obtain the total expected power consumption of the strategy. At the same time, the media addition setpoints for each time node are extracted from the strategy, and the media addition amount at each node is summed over time to obtain the total media consumption of the strategy. The total expected power consumption is added to the total media consumption, and then divided by the expected total amount of raw coal to be washed in the corresponding time period of the strategy to obtain the expected energy consumption per ton of coal for the strategy, converted to kilowatt-hours per ton.

[0090] The third step is to identify non-dominated strategy options. The expected coal washing efficiency and expected energy consumption per ton of coal for each strategy option are mapped onto a two-dimensional decision space with expected coal washing efficiency as the vertical axis and expected energy consumption per ton of coal as the horizontal axis. In this space, all strategy options are compared one by one: for any two options A and B, if the expected coal washing efficiency of option A is greater than that of option B, and the expected energy consumption per ton of coal of option A is less than that of option B, then option A dominates option B. If the expected coal washing efficiency of option A is greater than that of option B, but the expected energy consumption per ton of coal of option A is also greater than that of option B, then the two options do not dominate each other. This process is repeated for all options to identify all options that are not dominated by any other option. These options constitute the set of non-dominated strategy options located on the upper right convex envelope of the space. Each option in this set is a candidate option whose merits cannot be easily compared in terms of both efficiency and energy consumption objectives.

[0091] The fourth step is to select the optimal solution and generate scheduling instructions. For all non-dominated strategy solutions identified in the third step, the ratio of the expected coal washing efficiency to the expected energy consumption per ton of coal for each solution is calculated. The solution with the largest ratio is selected as the final optimal scheduling strategy. The complete sequence of process parameters is extracted from this solution, including the heavy medium separation density setpoint, hydrocyclone pressure setpoint, medium addition amount setpoint, and equipment start / stop instructions for each time point within the next two hours. These parameters are encapsulated into a standard format production scheduling instruction in chronological order and transmitted to the distributed control system of the coal washing production line via industrial Ethernet. The control system then automatically adjusts the corresponding parameters.

[0092] The working principle of this invention is as follows: A real-time production data acquisition module acquires raw coal property parameters, equipment operating status data, and current coal washing product quality index data from multiple monitoring nodes on the coal washing production line. A dynamic raw coal property analysis module, based on the raw coal property parameters, uses a time-series analysis algorithm to perform sliding window sampling, density composition analysis, adaptive exponential weighted filtering, and cubic spline interpolation on historical raw coal data, outputting a dynamic representation of the raw coal washability curve that matches the current production period. A production status prediction module, based on the dynamic representation of the raw coal washability curve and combined with equipment operating status data, extracts the real-time yield change gradient of each density level, the density of the heavy medium suspension, and the hydrocyclone inlet... The inlet pressure constitutes a production state vector, which is categorized into preset discrete state categories to construct a state transition probability table. Starting from the current state, a cumulative probability recursive method is used to determine the most probable state for each subsequent time node, generating a production state prediction sequence containing multiple time nodes. The scheduling strategy generation module uses the production state prediction sequence to analyze the pre-stored standard coal washing process flow, extracting multiple key process nodes and their default parameter values, such as heavy medium separation density, hydrocyclone pressure setpoint, and medium addition amount. After matching the production state prediction sequence with each key process node according to time, the module queries and predicts the state in the pre-stored parameter-state response relationship table. Matching process parameter adjustment values ​​are used to replace default parameter values, generating an adaptive scheduling strategy scheme that matches the current production conditions. The constraint verification module extracts the heavy medium separation density sequence and hydrocyclone pressure setting sequence from the adaptive scheduling strategy scheme to form a parameter combination to be verified. This combination is then correlated with the feature values ​​in the dynamic characterization of the raw coal washability curve and input into a pre-stored product quality response surface for interpolation calculation to obtain the predicted clean coal ash content and predicted clean coal yield. The predicted clean coal ash content is compared with the preset ash content threshold, and the predicted clean coal yield is compared with the preset yield threshold to select a set of feasible scheduling strategies that simultaneously meet the threshold requirements. The optimization decision module traverses the feasible scheduling options. For each strategy scheme in the strategy set, the heavy medium separation density sequence and the hydrocyclone pressure setting sequence are extracted. Combined with the production status prediction sequence, the expected coal washing efficiency is calculated through a preset efficiency response function. The expected energy consumption per ton of coal is obtained by summing the instantaneous power and medium consumption of each production node according to the equipment start-up and shutdown sequence and the amount of medium added. The expected coal washing efficiency and expected energy consumption per ton of coal are mapped to a two-dimensional decision space to identify the non-dominated strategy scheme located on the upper right convex envelope. The scheme with the largest ratio of expected coal washing efficiency to expected energy consumption per ton of coal is selected. Its corresponding process parameter sequence is encapsulated according to time nodes to generate the optimal production scheduling instruction and output to the coal washing production line control system.

[0093] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A coal washing process production scheduling optimization system based on dynamic modeling, characterized in that, include: The real-time production data acquisition module obtains raw coal property parameters, equipment operating status data, and current coal washing product quality index data from multiple monitoring nodes on the coal washing production line. The dynamic raw coal property analysis module, based on raw coal property parameters, processes historical raw coal data using a time-series analysis algorithm, and outputs a dynamic representation of raw coal washability curves that match the current production period. Specifically, it includes: Sliding window sampling is performed on historical raw coal data to generate a sequence of raw coal property parameters for multiple consecutive time windows; Density composition analysis was performed on the sequence of property parameters of each raw coal to obtain the yield and ash content time series data of each density level; Adaptive exponential weighted filtering was applied to the time series yield data at each density level to obtain smoothed yield trend values. Based on the smoothed yield trend value, a raw coal washout curve is constructed using cubic spline interpolation, and the raw coal washout curve is used as a dynamic representation that matches the current production period. The production status prediction module, based on the dynamic characterization of the raw coal washability curve and combined with equipment operating status data, uses Markov decision process to perform state transition deduction of the coal washing process and generates a production status prediction sequence containing multiple time nodes. The scheduling strategy generation module uses the production status prediction sequence to dynamically adjust the pre-stored standard process flow of coal washing production and generate an adaptive scheduling strategy scheme that matches the current production conditions. The constraint verification module performs constraint matching verification between the adaptive scheduling strategy scheme and the quality indicator data, and selects a set of feasible scheduling strategies that meet the preset quality threshold. The optimization decision module, based on the set of feasible scheduling strategies, solves a multi-objective optimization problem with the objective functions of maximizing coal washing efficiency and minimizing energy consumption, determines the optimal production scheduling instruction, and outputs it to the coal washing production line control system.

2. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 1, characterized in that, The density composition analysis of the sequence of raw coal property parameters to obtain the yield and ash content time series data for each density level specifically includes: Based on the ash content and particle size distribution in the raw coal property parameter sequence, the raw coal is divided into a preset number of virtual density intervals, and an initial yield allocation coefficient is assigned to each virtual density interval. The yield allocation coefficient of each virtual density interval is weighted and summed with the ash characteristic value of the corresponding density interval to obtain the calculated ash value. With the goal of approximating the measured ash content in the raw coal property parameter sequence by calculating the ash content value, the yield allocation coefficient is adjusted iteratively until the difference between the two is less than a preset threshold. The yield allocation coefficients after iterative convergence are used as the yields of each density level at the current time point, and combined with the preset ash content feature values, the yields and ash content data of each density level at the corresponding time point are output.

3. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 1, characterized in that, The generation of the production status prediction sequence containing multiple time points specifically includes: The real-time change gradient of the yield of each density level is extracted from the dynamic characterization of the raw coal washability curve, and the density of the heavy medium suspension and the inlet pressure of the hydrocyclone are extracted from the equipment operation status data and combined into the production status vector at the current moment. The production state vector is matched with the preset state space template and assigned to the corresponding discrete state category. Based on the historical state transition frequency statistics, a state transition probability table from the corresponding category to subsequent categories is constructed. Starting from the current state, the most likely state at each subsequent time node is determined sequentially using a step-by-step recursive method based on the state transition probability table, and these most likely states are arranged in chronological order to generate a production state prediction sequence.

4. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 3, characterized in that, The method of determining the most probable state for each subsequent time point using a step-by-step recursive approach specifically includes: Starting from the production status at the current time point, obtain all candidate states for the next time point and their corresponding transition probability values ​​according to the state transition probability table; Calculate the cumulative probability for each candidate state, where the cumulative probability of the current time node is assigned a value of one, and the cumulative probability of subsequent nodes is the cumulative probability of the preceding node multiplied by the probability value of the corresponding transition. The candidate state with the highest cumulative probability value is selected as the most likely state at the next time node, and the cumulative probability value of the most likely state is used as the benchmark value for the next round of recursion. Until all preset time nodes are recursively calculated, the most likely states selected in each round are arranged in chronological order to form a production state prediction sequence.

5. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 1, characterized in that, The generation of an adaptive scheduling strategy scheme that matches the current production conditions specifically includes: The standard process flow of coal washing production was analyzed, and several key process nodes and their default parameter values, including heavy medium separation density, hydrocyclone pressure setpoint and medium addition amount, were extracted. The production status prediction sequence is matched one-to-one with each key process node in chronological order to determine the predicted raw coal washability status and equipment operating status corresponding to the execution time of each node. Based on the predicted state corresponding to each node, query the pre-stored parameter-state response relationship table for the process parameter adjustment value that matches the corresponding predicted state; The default parameter values ​​of each key process node are replaced with the queried process parameter adjustment values, and combined to generate an adaptive scheduling strategy scheme that matches the current production conditions.

6. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 1, characterized in that, The selection of a set of feasible scheduling strategies that meet the preset quality threshold specifically includes: The heavy medium separation density and hydrocyclone pressure setpoints of each key process node are extracted from the adaptive scheduling strategy and combined into a parameter combination to be verified. The parameter combination to be verified is associated with the dynamic characterization of the raw coal washability curve in the current quality index data. The pre-stored product quality response surface is input, and the corresponding predicted clean coal ash content and predicted clean coal yield are obtained by interpolation. The predicted clean coal ash content is compared with the preset ash content threshold, and the predicted clean coal yield is compared with the preset yield threshold to determine whether both meet the threshold range requirements simultaneously. The adaptive scheduling strategy schemes corresponding to the parameter combinations to be verified that simultaneously meet the requirements of ash content threshold and yield threshold are selected and included in the set of feasible scheduling strategies.

7. The coal washing process production scheduling optimization system based on dynamic modeling according to claim 1, characterized in that, The process of determining the optimal production scheduling instruction and outputting it to the coal washing production line control system specifically includes: Iterate through each strategy scheme in the set of feasible scheduling strategies, extract the heavy medium separation density sequence and hydrocyclone pressure setting sequence, combine them with the production status prediction sequence, and calculate the expected coal washing efficiency value corresponding to each strategy through the preset efficiency response function. Based on the equipment start-up and shutdown sequence and media addition amount in each strategy scheme, the instantaneous power and media consumption of each production node are calculated and accumulated to obtain the expected energy consumption per ton of coal corresponding to the strategy. The expected coal washing efficiency and expected energy consumption per ton of coal for each strategy are mapped to a two-dimensional decision space with efficiency as the vertical axis and energy consumption as the horizontal axis, and the non-dominated strategy located on the upper right convex envelope of the space is identified. From the non-dominated strategy options, select the option with the largest ratio of expected coal washing efficiency to expected energy consumption per ton of coal, encapsulate its corresponding process parameter sequence according to time nodes to generate the optimal production scheduling instruction, and output it to the coal washing production line control system.

Images