Intelligent prediction method for coal slime flotation control parameters based on large model

Abstract

The invention relates to the technical field of industrial control systems, in particular to an intelligent prediction method for coal slime flotation control parameters based on a large model, which comprises the following steps of: based on coal slime characteristic data and ore pulp flow data input in real time, performing feature extraction and structured arrangement for large model input and training the large model; the method comprises the following steps of: deducing and calculating through a large model to obtain dosage set value prediction, aeration set value prediction and ore pulp liquid level height prediction, carrying out disturbance analysis on predicted values to quantify a fluctuation range, and establishing a control parameter scheme. According to the method, coal slime characteristic data and ore pulp flow data are collected in real time to perform deep structured feature extraction, and a deep neural network model is trained on the basis to predict the dosage, the aeration volume and the ore pulp liquid level height in the flotation process, so that a dynamic control parameter scheme is established, and a process adjustment prediction value is obtained; meanwhile, disturbance analysis is added to define a parameter fluctuation range, and robustness and stability of parameter control are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial control systems, and in particular to an intelligent prediction method for coal slime flotation control parameters based on a large model. BACKGROUND

[0002] The technical field of industrial control systems is a comprehensive technical system that integrates automation control technology, sensor technology, communication technology, and computer technology, and is widely used in production and manufacturing, energy management, chemical process control, mining automation, transportation control, and power system management. The core goal is to achieve automatic monitoring, coordination, and optimized control of equipment and processes in industrial production processes.

[0003] The prior art cannot timely perceive and capture the complex dynamic changes of coal slime characteristics and slurry flow, and also fails to quantify and control the adjustment range and fluctuation range of process parameters, resulting in frequent lag and slow response problems of flotation control parameters. In actual process operation, the fluctuations of clean coal yield and tail coal ash often exceed the allowed range, and manual intervention by experienced operators is required, increasing production costs and the risk of human error in process adjustment. Therefore, improvements are needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide an intelligent prediction method for coal slime flotation control parameters based on a large model.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solution: an intelligent prediction method for coal slime flotation control parameters based on a large model, comprising the following steps: Based on the real-time input of coal slime characteristic data and slurry flow data, feature extraction and structured arrangement are performed for large model input and training of the large model. The dosage set value prediction and the aeration amount set value prediction and the slurry liquid level height prediction are obtained through the large model deduction calculation, and the disturbance analysis quantization fluctuation range of the prediction value is performed to establish a control parameter scheme; Based on the control parameter scheme, the dosage set value prediction and the aeration amount set value prediction and the slurry liquid level height prediction are extracted to generate a to-be-executed flotation control instruction. Based on the to-be-executed flotation control instruction, the configuration is applied, and the clean coal yield monitoring value and the tail coal ash monitoring value are continuously monitored within a set time window to collect key performance data of the working condition; Based on the key performance data of the working condition, the clean coal yield monitoring value and the tail coal ash monitoring value are extracted to obtain production effect feedback results. Based on the production effect feedback results, the prediction confidence interval of each parameter in the control parameter scheme is compared and analyzed, the deviation degree of the calculation effect and the prediction interval boundary is calculated, and the control parameter adjustment difference is determined; Based on the control parameter adjustment difference, combined with the initial reagent dosage setting value prediction and the aeration amount setting value prediction and the slurry liquid level prediction in the control parameter scheme, parameter adjustment operation is performed to generate an adjusted parameter instruction set, and based on the adjusted parameter instruction set, an optimized setting is issued to the slime flotation control unit as the next production batch, forming an iterative optimization control parameter set.

[0006] Preferably, the control parameter scheme acquisition step is: Based on the real-time input of the slime characteristic data and the slurry flow data, the ash content, particle size distribution and moisture content in the slime characteristic data and the flow fluctuation rate and flow stability characteristics in the slurry flow data are extracted, and the structured data feature set is obtained in a predetermined order. Based on the structured data feature set, the structured data feature set is input to the large model in a standardized numerical format, and the node parameters and weights of the model are adjusted to make the large model learn the change rule of the data features, forming a trained large model, and the large model is a deep neural network model. Based on the trained large model, the structured data feature set is called as input data for forward inference calculation to obtain reagent dosage setting value prediction, aeration amount setting value prediction and slurry liquid level prediction under corresponding conditions, and random disturbance analysis is performed on the prediction results to obtain the fluctuation range of each prediction result. After combining the prediction results and the corresponding fluctuation range, the control parameter scheme is obtained.

[0007] Preferably, the acquisition step of the to-be-executed flotation control instruction is: Based on the control parameter scheme, the reagent dosage setting value prediction, the aeration amount setting value prediction and the slurry liquid level prediction in the control parameter scheme are analyzed, and the reagent dosage setting value prediction, the aeration amount setting value prediction and the slurry liquid level prediction are extracted as independent control parameter items to generate an independent control parameter item set. Based on the independent control parameter item set, the reagent dosage setting value prediction, the aeration amount setting value prediction and the slurry liquid level prediction in the independent control parameter item set are sequentially data encoded according to the execution instruction format specification of the flotation control unit to form a data instruction template conforming to the input standard of the flotation control unit, and a to-be-executed flotation control instruction is generated.

[0008] Preferably, the working condition key performance data acquisition step is: Based on the to-be-executed flotation control instruction, the reagent dosage setting value prediction, the aeration amount setting value prediction, and the pulp liquid level prediction in the to-be-executed flotation control instruction are analyzed, and the reagent dosage setting value prediction, the aeration amount setting value prediction, and the pulp liquid level prediction are sequentially issued to the reagent feeding system, the aeration system, and the liquid level control system, respectively, to apply and configure the control parameter setting, and to generate configured control parameters; Based on the configured control parameters, the start time and the end time of a time window are set, the clean coal yield monitoring value and the tail coal ash content monitoring value are recorded continuously within the set time window, the clean coal yield monitoring value and the tail coal ash content monitoring value are sequentially archived according to the time stamp, and the working condition key performance data is generated.

[0009] Preferably, the obtaining step of the production effect feedback result is: Based on the working condition key performance data, all clean coal yield monitoring values and tail coal ash content monitoring values in the working condition key performance data are sequentially extracted, the clean coal yield monitoring values and the tail coal ash content monitoring values are one-to-one corresponding matched according to the time stamp, and the data records with missing or abnormal data are eliminated, and the valid monitoring data pair is generated. Based on the valid monitoring data pair, the change amplitude, the fluctuation interval, and the extreme value are respectively counted according to the change trend of the clean coal yield monitoring value and the change trend of the tail coal ash content monitoring value, the performance of the valid monitoring data pair within the set time window is evaluated, and the monitoring trend evaluation result is generated. Based on the monitoring trend evaluation result, it is determined whether the flotation effect of the production process meets the expected index, and the production effect feedback result is generated.

[0010] Preferably, the obtaining step of the control parameter adjustment difference is: Based on the production effect feedback result, the clean coal yield monitoring value, the tail coal ash content monitoring value, and the monitoring value of the pulp liquid level prediction in the production effect feedback result are extracted, the prediction confidence interval boundary of the reagent dosage setting value prediction, the aeration amount setting value prediction, and the pulp liquid level prediction in the control parameter scheme is extracted, and the daily average absolute fluctuation value of each control parameter in the last 30 days and the average value in the last 30 days are recorded respectively, and the parameter comparison set is formed; Based on the parameter comparison set, the comprehensive deviation degree is calculated. Based on the comprehensive deviation degree, first, the relative positional relationship between the production effect feedback value of each control parameter and the prediction confidence interval is judged, if the production effect feedback value exceeds the upper limit of the confidence interval, the control parameter is adjusted in the decreasing direction, if it is lower than the lower limit of the confidence interval, the control parameter is adjusted in the increasing direction, the adjustment amplitude is determined combined with the comprehensive deviation degree, and the control parameter adjustment difference is generated.

[0011] Preferably, the obtaining step of the adjusted parameter instruction set is: Based on the control parameter adjustment difference, extract the reagent addition amount adjustment value, the aeration amount adjustment value and the slurry liquid level height adjustment value in the control parameter adjustment difference, match the reagent addition amount set value prediction, the aeration amount set value prediction and the slurry liquid level height prediction in the control parameter scheme respectively, and extract the maximum allowable variation range of the control parameter scheme in the last 30 days to form a parameter adjustment pairing set; Based on the parameter adjustment pairing set, calculate the adjusted set value; Based on the adjusted set value of each control parameter, convert the reagent addition amount adjusted set value, the aeration amount adjusted set value and the slurry liquid level height adjusted set value into a data format conforming to the execution standard of the flotation control unit in sequence to generate an adjusted parameter instruction set.

[0012] Preferably, the step of obtaining the iterative optimization control parameter set is: Based on the adjusted parameter instruction set, analyze the reagent addition amount adjusted set value, the aeration amount adjusted set value and the slurry liquid level height adjusted set value in the adjusted parameter instruction set, and convert them into standard data instructions according to the communication protocol of the coal slime flotation control unit in sequence to form a standard data instruction set; Based on the standard data instruction set, send the reagent addition amount adjusted set value standard data instruction in the set to the reagent addition system control interface, send the aeration amount adjusted set value standard data instruction to the aeration system control interface, and send the slurry liquid level height adjusted set value standard data instruction to the liquid level control system interface to complete the parameter configuration issuing action and form the iterative optimization control parameter set.

[0013] Compared with the prior art, the advantages and positive effects of the present application are: The present application extracts deep structured features by collecting coal slime characteristic data and slurry flow data in real time, trains a deep neural network model based thereon to predict the reagent addition amount, the aeration amount and the slurry liquid level height in the flotation process, establishes a dynamic control parameter scheme to obtain process adjustment prediction values, adds disturbance analysis to determine the parameter fluctuation range, enhances the robustness and stability of parameter control, iteratively calculates the control parameter adjustment difference by using real-time monitoring feedback data of clean coal yield and tail coal ash content, corrects process deviation in time and improves the timeliness and accuracy of process parameter response, and optimizes the flotation process control by using a dynamic adjustment parameter instruction set to realize automatic iterative adjustment of control parameters, improve the overall adaptability of the coal slime flotation process, improve the control accuracy and production efficiency, and achieve the long-term stable clean coal yield improvement and tail coal ash content reduction target. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The present application is a step schematic diagram. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0016] Please refer to Figure 1 The present application provides a technical scheme, an intelligent prediction method for coal slime flotation control parameters based on a large model, comprising the following steps: Based on the real-time input of coal slime characteristic data and slurry flow data, feature extraction and structured arrangement are performed for large model input and large model training, and through large model deduction calculation, the reagent dosage set value prediction, the aeration amount set value prediction and the slurry liquid level height prediction are obtained, and the prediction value is subjected to disturbance analysis and quantization of fluctuation range, and a control parameter scheme is established; Based on the control parameter scheme, the reagent dosage set value prediction and the aeration amount set value prediction and the slurry liquid level height prediction are extracted, the to-be-executed flotation control instruction is generated, based on the to-be-executed flotation control instruction, the configuration is applied, and the clean coal yield monitoring value and the tail coal ash content monitoring value are continuously monitored within the set time window, and are collected as working condition key performance data; Based on the working condition key performance data, the clean coal yield monitoring value and the tail coal ash content monitoring value are extracted, the production effect feedback result is obtained, based on the production effect feedback result, the prediction confidence interval of each parameter in the control parameter scheme is compared and analyzed, the deviation degree of the calculation effect and the prediction interval boundary is calculated, and the control parameter adjustment difference is determined; Based on the control parameter adjustment difference, combined with the initial reagent dosage set value prediction and the aeration amount set value prediction and the slurry liquid level height prediction in the control parameter scheme, parameter adjustment operation is performed, the adjusted parameter instruction set is generated, based on the adjusted parameter instruction set, as the optimized setting of the next production batch, it is issued to the coal slime flotation control unit to form an iterative optimization control parameter set.

[0017] The acquisition step of the control parameter scheme is: Based on the real-time input of coal slime characteristic data and slurry flow data, the ash content, particle size distribution and moisture content in the coal slime characteristic data and the flow fluctuation rate and flow stability characteristics in the slurry flow data are extracted, and the structured data feature set is formed in a predetermined order in sequence, and the structured data feature set is obtained; Based on the structured data feature set, the structured data feature set is input to the large model in a standardized numerical format item by item, and the node parameters and weights of the model are adjusted, so that the large model learns the change rule of the data features, and a trained large model is formed, and the large model is a deep neural network model; Based on the trained large model, the structured data feature set is called as input data, forward inference calculation is carried out, and the setting value prediction of the reagent amount, the setting value prediction of the aeration amount and the prediction of the ore pulp liquid level height under the corresponding conditions are deduced. Random disturbance analysis is carried out on the prediction results to obtain the fluctuation range of each prediction result. The prediction results and the corresponding fluctuation range are combined to obtain the control parameter scheme.

[0018] Specifically, based on the real-time input of the coal slime characteristic data and the ore pulp flow data, first, the key indicators are extracted from the coal slime characteristic data, specifically including the ash content collected by the online ash analyzer (such as X-ray fluorescence analyzer) every 60 seconds, and the particle size distribution data obtained every 300 seconds, which is divided into three particle size intervals, such as the mass percentage of particles less than 74 microns, 74 to 150 microns, and greater than 150 microns, and the moisture content measured by the microwave moisture meter every 60 seconds, and the flow dynamic characteristics are extracted from the ore pulp flow data, specifically including the flow fluctuation rate and the flow stability characteristics, wherein the calculation method of the flow fluctuation rate is: in the past 10 minutes, the instantaneous flow value of the ore pulp is collected every second, and the average value of the 600 flow values is calculated and the standard deviation , then the flow fluctuation rate , for example, if the average flow in 10 minutes is 120 cubic meters / hour, and the standard deviation is 6 cubic meters / hour, then the flow fluctuation rate is calculated as , and the flow stability characteristics are represented by the coefficient of variation of the ore pulp flow data in the past 30 minutes , the calculation method is the ratio of the standard deviation of the flow data in the time window to the average value, for example, if the average flow in 30 minutes is 118 cubic meters / hour, and the standard deviation is 7.08 cubic meters / hour, then the flow stability characteristic value is , then the extracted ash content, moisture content, particle percentage less than 74 microns, particle percentage between 74 and 150 microns, particle percentage greater than 150 microns, flow fluctuation rate, and flow stability characteristics are arranged in this predetermined order to form a vectorized data record, and the predetermined order is determined according to the analysis of the slime flotation process mechanism and the importance of historical data, and the coal quality characteristics that have a significant impact on flotation are arranged first, followed by the hydrodynamic characteristics, to form a structured data feature set.

[0019] Based on the structured data feature set, first, each feature value in the structured data feature set is standardized, specifically using the Z-score standardization method, that is, for any feature value , the standardized value , wherein is the current batch the original value of the i-th feature, and are the average and standard deviation of the i-th feature in the past three months of historical production data, respectively, for example, if the three-month average of the ash content is 25% and the standard deviation is 3%, and the current real-time ash content is 28%, then the normalized feature value of the ash content is After all the features are normalized, the vector obtained is input into a pre-constructed and initially parameter-set deep neural network large model, the specific structure of which is as follows: the input layer contains 7 neurons, corresponding to the seven normalized features in the structured data feature set, followed by three fully connected hidden layers, the first hidden layer contains 128 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 128 neurons, all of which use the ReLU activation function, i.e. The output layer contains 3 neurons, corresponding to the three outputs of the reagent dosage setting value prediction, the aeration amount setting value prediction, and the slurry liquid level height prediction, respectively, and the output layer uses the linear activation function, i.e. The model training process is as follows: prepare a training set containing at least one year of historical production data, each data containing the structured data feature set at that time and the accurate record of the actual reagent dosage, aeration amount, and slurry liquid level height in the verified efficient production state, divide the data set into 70% training set, 15% validation set, and 15% test set, use mean squared error (MSE) as the loss function during training, which is the average of the square of the difference between the predicted value and the actual value, adjust the node parameters (i.e. weights and biases) of the model through the Adam optimizer, the initial learning rate of the Adam optimizer is set to 0.001, which is determined by testing a series of learning rates (such as 0.01, 0.005, 0.001, 0.0005) on the validation set and selecting the learning rate with the fastest convergence and the lowest validation loss, for example, in the test, when the learning rate is 0.001, the model's validation loss decreases by 30% after 20 cycles and the trend is stable, while the loss starts to fluctuate when the learning rate is 0.005, so 0.001 is selected, the training uses the mini-batch gradient descent method, the batch size is set to 64, the training iteration is performed for 150 cycles, or the training is terminated early when the loss on the validation set does not decrease significantly (less than 0.0001) for 15 consecutive cycles, the gradients are calculated and all weights and biases in the network are updated through the backpropagation algorithm according to the loss function, so that the large model can learn and capture the complex nonlinear variation between the input data features and the output control parameters, forming the trained large model.

[0020] ​Based on the trained large model, first, the current real-time generated and standardized structured data feature set by the same method as training is called as input data, the standardized structured data feature set is input into the trained large model for forward inference calculation, that is, data is calculated layer by layer from the input layer to the output layer, and the initial reagent dosage setting value prediction, the initial aeration amount setting value prediction and the initial slurry liquid level prediction under the current coal slime characteristics and slurry flow conditions are directly deduced according to the learned weight and bias parameters of the model, then, random disturbance analysis is performed on the three initial prediction results to quantify their fluctuation range, the specific disturbance analysis method is: for each input feature, a random Gaussian noise with a mean of 0 and a standard deviation of 1% of the historical data standard deviation of the feature is superimposed on the basis of its standardized value (for example, if the historical standard deviation of a standardized feature is 1, the superimposed noise standard deviation is 0.01), repeat this process to generate 100 sets of slightly disturbed input feature sets, input these 100 sets of disturbed input feature sets into the large model for forward inference, and get 100 sets of corresponding reagent dosage, aeration amount and slurry liquid level prediction values, for each prediction parameter (such as reagent dosage), calculate the average value and standard deviation of the 100 prediction values , and the average value is taken as the final setting value prediction, and the fluctuation range is determined according to the distribution of the 100 disturbance prediction results, for example, the 2.5th percentile of the 100 prediction values is taken as the lower boundary of the fluctuation range, and the 97.5th percentile is taken as the upper boundary of the fluctuation range, thus forming a 95% confidence interval, for example, for the reagent dosage setting value prediction, the average value is 1250g / t, among the 100 prediction values obtained by the above disturbance analysis, the 2.5th percentile is 1220g / t, and the 97.5th percentile is 1280g / t, then the reagent dosage setting value prediction is 1250g / t, and the fluctuation range is [1220, 1280]g / t, the same processing is performed for the aeration amount setting value prediction and the slurry liquid level prediction, finally, the setting value prediction and the corresponding fluctuation range (i.e. the upper and lower boundaries of the confidence interval) of each parameter are combined to form a set containing the reagent dosage setting value prediction and the fluctuation range, the aeration amount setting value prediction and the fluctuation range, and the slurry liquid level prediction and the fluctuation range, and the control parameter scheme is obtained.

[0021] The acquisition step of the to-be-executed flotation control instruction is: Based on the control parameter scheme, the reagent dosage setting value prediction, the aeration amount setting value prediction and the slurry liquid level prediction in the control parameter scheme are analyzed, and the reagent dosage setting value prediction, the aeration amount setting value prediction and the slurry liquid level prediction are extracted as independent control parameter items to generate an independent control parameter item set. Based on the independent control parameter item set, the reagent dosage setting value prediction, the aeration amount setting value prediction and the pulp liquid level height prediction in the independent control parameter item set are sequentially data encoded according to the execution instruction format specification of the flotation control unit to form a data instruction template conforming to the input standard of the flotation control unit, and a flotation control instruction to be executed is generated.

[0022] Specifically, based on the control parameter scheme, the scheme contains the reagent dosage setting value prediction and its fluctuation range, the aeration amount setting value prediction and its fluctuation range, and the pulp liquid level height prediction and its fluctuation range, first, the control parameter scheme is parsed, and the specific operation is to extract the core prediction setting value from the scheme item by item, that is, the numerical value of the reagent dosage setting value prediction is extracted, for example, if the reagent dosage part in the scheme is (1250 g / t, [1220, 1280] g / t), the extracted reagent dosage setting value prediction is 1250 g / t, and the corresponding fluctuation range [1220, 1280] g / t is ignored, similarly, the numerical value of the aeration amount setting value prediction is extracted from the control parameter scheme, for example, if the aeration amount part in the scheme is (2.5 m3 / min, [2.3, 2.7] m3 / min), the extracted aeration amount setting value prediction is 2.5 m3 / min, finally, the numerical value of the pulp liquid level height prediction is extracted from the control parameter scheme, for example, if the pulp liquid level height part in the scheme is (2.8 m, [2.75, 2.85] m), the extracted pulp liquid level height prediction is 2.8 m, then, these extracted reagent dosage setting value prediction, aeration amount setting value prediction and pulp liquid level height prediction are respectively created as independent control parameter items, each item contains parameter name identification, parameter prediction value and parameter unit, for example, the reagent dosage setting value prediction item is {Parameter name: "Reagent dosage setting value prediction", Parameter value: 1250, Unit: "g / t"}, the aeration amount setting value prediction item is {Parameter name: "Aeration amount setting value prediction", Parameter value: 2.5, Unit: "m3 / min"}, and the pulp liquid level height prediction item is {Parameter name: "Pulp liquid level height prediction", Parameter value: 2.8, Unit: "m"}; the three independent control parameter items are collected to generate an independent control parameter item set.

[0023] Based on the independent control parameter entry set, which contains each independent reagent dosage setting value prediction entry, aeration amount setting value prediction entry and ore pulp liquid level height prediction entry, the prediction values in these entries are strictly data encoded according to the pre-defined "floatation control unit execution instruction format specification", which is formulated according to the communication protocol and data interface requirements of the floatation control system (such as Siemens PCS7 or similar DCS system) actually adopted by the coal preparation plant, and is fixed as part of the device technical file, which specifies in detail the target address (such as register address) of each control parameter in the control system, data type, value range and necessary scaling and offset, the data encoding process is as follows: first, for the reagent dosage setting value prediction in the independent control parameter entry set, for example, its value is 1250 g / t, according to the specification, its target register address is MW100 (Modbus address example), the required data type is 16-bit unsigned integer, the unit is directly g / t, no scaling is required, then the encoded value is integer 1250, second, for the aeration amount setting value prediction, for example, its value is 2.5 m3 / min, the specification specifies that its target register address is MW102, the data type is 16-bit unsigned integer, but the control unit expects the unit to be 0.01 m3 / min, therefore scaling is required, the scaling factor is 100, this factor 100 is set according to the requirement of preserving two decimal places of accuracy (i.e. accuracy of 0.01), the calculation method is = , i.e. , then the encoded value is integer 250, third, for the ore pulp liquid level height prediction, for example, its value is 2.8 m, the specification specifies that its target register address is MW104, the data type is 16-bit unsigned integer, the control unit expects the unit to be mm, therefore the scaling factor is 1000, this factor 1000 is also based on the unit conversion requirement (1 m = 1000 mm), the calculation method is , then the encoded value is integer 2800, organize the parameter values (1250, 250, 2800) after the above encoding process according to the order and structure defined by the specification, for example, construct a sequence containing target address and corresponding encoded value: , this sequence is a specific example of the data instruction template that meets the floatation control unit input standard, generating the floatation control instruction to be executed.

[0024] The working condition key performance data acquisition step is: Based on the to-be-executed flotation control instruction, the reagent dosage setting value prediction, the aeration amount setting value prediction, and the pulp liquid level prediction in the to-be-executed flotation control instruction are analyzed, and the reagent dosage setting value prediction, the aeration amount setting value prediction, and the pulp liquid level prediction are sequentially issued to the reagent feeding system, the aeration system, and the liquid level control system, respectively, to complete the control parameter setting by configuration, and to generate configured control parameters. Based on the configured control parameters, the start time and the end time of a time window are set, the clean coal yield monitoring value and the tail coal ash content monitoring value are recorded continuously in the set time window, the clean coal yield monitoring value and the tail coal ash content monitoring value are sequentially archived according to the time stamp, and the working condition key performance data is generated.

[0025] Specifically, based on the to-be-executed flotation control instruction, the instruction is a sequence containing the target address of each control parameter and the corresponding encoded set value, for example, an instruction list such as The instruction list is first analyzed, and the specific operation is to read the instructions one by one, identify the control parameter types contained in the instructions, such as finding MW100 corresponding to reagent dosage control, MW102 corresponding to aeration amount control, and MW104 corresponding to liquid level control through a pre-set address mapping table, and extracting the encoded set values of each, that is, parsing the set value 1250 targeted at the reagent feeding system from the instruction (MW100, 1250), the set value 250 targeted at the aeration system from the instruction (MW102, 250), and the set value 2800 targeted at the liquid level control system from the instruction (MW104, 2800), and then, according to the instruction sequence, the encoded reagent dosage set value 1250 is written into the predetermined address MW100 in the reagent control PLC through the industrial Ethernet interface using, for example, the ModbusTCP protocol Write Single Register function code, then the encoded aeration amount set value 250 is also written into the predetermined address MW102 in the aeration control PLC through the above communication mode, and then the encoded pulp liquid level height set value 2800 is written into the predetermined address MW104 in the liquid level control PLC, after each time the instruction is issued, the system will wait for the target PLC to return the confirmation message of successful operation, such as the normal response message in the Modbus protocol, this waiting has a timeout time, for example, 5 seconds, the setting of 5 seconds is based on the statistical analysis of the network communication delay in the field and the time required for the PLC to execute the instruction, which is 1.5 times the average response time plus a fixed margin of 2 seconds (for example, the average response is 1 second, 1*1.5+2≈3.5 seconds, rounded up to 5 seconds), if the successful confirmation is received within the timeout time, it is recorded that the parameter has been successfully configured, if the timeout or error code is received, it is recorded that the configuration fails and an alarm is triggered, until all instructions are successfully issued and confirmed, or in the case that part of the instructions fail but the process allows to continue, the final configuration status and the actual issued value of each parameter are summarized to generate the configured control parameters.

[0026] After confirming that the new control parameter settings have taken effect in the reagent feeding system, the aeration system, and the liquid level control system based on the configured control parameters, a time window for monitoring production effectiveness is immediately set. The starting time of the time window is defined as the exact moment when the last control parameter is confirmed as successfully configured in the PLC. The duration of the time window, for example, is set to 2 hours. This 2-hour monitoring duration is determined based on historical experience and process characteristics, mainly considering the time required for the coal slime flotation process to reach a new quasi-steady state (usually between 30 minutes and 1 hour) and the need to collect enough representative data points for subsequent analysis. For example, if the parameter is configured to be completed at 09:15:00 on a certain day, the starting time of the time window is 09:15:00 and the ending time is 11:15:00. Within this 2-hour set time window, the system automatically records key production performance indicators at a preset frequency. Specifically, it includes the clean coal yield monitoring value calculated by collecting data every 1 minute from the electronic belt scale installed on the clean coal product output belt and the corresponding coal flow detection equipment. The value is recorded in percentage form, for example, 78.5%. At the same time, the tailings ash content monitoring value obtained by automatically sampling and analyzing every 5 minutes through the online ash content analyzer (such as a gamma-ray ash content analyzer) installed on the tailings flow pipeline is also recorded in percentage form, for example, 42.3%. Each clean coal yield monitoring value and tailings ash content monitoring value obtained is attached with a system time stamp at the time of collection, for example, (“2024-05-29 09:16:00”, clean coal yield: 78.5%), (“2024-05-29 09:20:00”, tailings ash content: 42.3%), and these time-stamped monitoring data are stored and archived in chronological order to form the working condition key performance data.

[0027] The production effectiveness feedback result acquisition step is: Based on the working condition key performance data, all clean coal yield monitoring values and tailings ash content monitoring values in the working condition key performance data are extracted in sequence. The clean coal yield monitoring values and the tailings ash content monitoring values are matched one by one according to the time stamp, and the data records with missing or abnormal data are excluded to generate a valid monitoring data pair. Based on the valid monitoring data pair, the change amplitude, fluctuation interval, and extreme value of the clean coal yield monitoring value and the tailings ash content monitoring value are respectively counted according to the change trend of the clean coal yield monitoring value and the change trend of the tailings ash content monitoring value. The performance of the valid monitoring data pair in the set time window is evaluated to generate a monitoring trend evaluation result. Based on the monitoring trend evaluation result, it is determined whether the flotation effect of the production process meets the expected indicators to generate a production effectiveness feedback result.

[0028] Specifically, based on the working condition key performance data, which contains the clean coal yield monitoring value sequence and the tail coal ash content monitoring value sequence with time stamp in a set time window, first, all the clean coal yield monitoring values and tail coal ash content monitoring values are extracted from these sequences respectively, then, in order to conduct comprehensive analysis, the two monitoring value sequences with different sampling frequencies are aligned and matched according to the time stamp, and the specific matching method is as follows: taking the time stamp of the tail coal ash content monitoring value as the benchmark, for each tail coal ash content monitoring value data point (for example, time stamp , value ), find the clean coal yield monitoring value data point (for example, time stamp , value ) closest to its time stamp in the clean coal yield monitoring value sequence, and the absolute value of the time difference between and should not exceed a preset maximum allowable time difference, for example, 60 seconds, this 60-second allowable time difference is set according to the clean coal yield data collection frequency (for example, every minute) and the process response characteristics, if the clean coal yield monitoring value meeting the condition can be found, a preliminary matching data point is formed, if not, the tail coal ash content data point cannot be matched temporarily, then, the validity of all preliminary matching data points and unmatching but single existing monitoring values is checked, and the data records with missing values (for example, the reading of a certain time stamp is empty) or obvious abnormalities are removed, and the judgment criteria of abnormal data records include: first, exceeding the preset reasonable value range, for example, the reasonable range of clean coal yield monitoring value is set to 30% to 95% according to historical operation data and process design, any reading exceeding this interval, such as 25% or 98%, is considered abnormal, and the reasonable range of tail coal ash content monitoring value is set to 10% to 80%, and the reading exceeding this range is also considered abnormal, these range values are determined by the coal preparation plant process engineers according to long-term production practice and coal quality characteristics analysis, second, the instantaneous change rate is too large, for example, if the absolute value of the change of clean coal yield monitoring value between adjacent two sampling points (interval 1 minute) exceeds 15 percentage points (this 15% threshold is set based on the analysis of the maximum adjustment rate of the equipment and the normal working condition fluctuation), it is considered that the latter data point may be abnormal, after the above matching and removal processing, the remaining data records generate the effective monitoring data pair.

[0029] Based on the effective monitoring data pair, which is the paired sequence of clean coal yield monitoring value and tail coal ash content monitoring value after time stamp alignment and abnormal data elimination, statistical analysis is performed on the dynamic performance of the clean coal yield monitoring value and the tail coal ash content monitoring value in the entire set time window (for example, 2 hours). First, for the clean coal yield monitoring value sequence, the maximum value and the minimum value are calculated, and the difference between the two is the change range of the clean coal yield, and the fluctuation interval is the closed interval formed by the maximum value and the minimum value. At the same time, the average value and the standard deviation of the sequence are calculated to evaluate the central tendency and dispersion degree, for example, if the minimum value of the clean coal yield monitoring value in 2 hours is 72.5%, the maximum value is 78.3%, the change range is 5.8%, the fluctuation interval is [72.5%, 78.3%], the average value is 75.6%, and the standard deviation is 1.2%. Secondly, trend analysis is performed on the clean coal yield monitoring value sequence, and a linear trend line is fitted by using the least squares method, and the slope is calculated, for example, the slope of +0.2 indicates that the clean coal yield increases by 0.2 percentage points per hour on average. For the tail coal ash content monitoring value sequence, the same statistical calculation is performed, including determining the change range, fluctuation interval, extreme value (maximum value and minimum value), average value and standard deviation, and calculating the slope of the linear trend line, for example, if the minimum value of the tail coal ash content monitoring value is 45.1%, the maximum value is 49.5%, the change range is 4.4%, the fluctuation interval is [45.1%, 49.5%], the average value is 47.2%, and the standard deviation is 0.8%, the trend line slope is -0.1, indicating that the tail coal ash content decreases by 0.1 percentage points per hour on average. The statistical indicators calculated for the clean coal yield monitoring value and the tail coal ash content monitoring value respectively, including their respective change range, fluctuation interval, extreme value, average value, standard deviation and trend line slope, are summarized and generated into monitoring trend evaluation results.

[0030] Based on the monitoring trend evaluation results, which include the statistical characteristics (such as mean, standard deviation, trend slope, etc.) of the clean coal yield monitoring value and the tail coal ash content monitoring value within a set time window, these statistical characteristics are compared with a set of "expected indicators" to determine the flotation effect of the current production process. The "expected indicators" are formulated by the production technology department of the coal preparation plant based on the washing characteristics of the current coal type, market requirements for product quality, and best practice experience of equipment operation, and are regularly reviewed and updated. For example, for the current batch of slime, the expected indicators may specify that the average value of the clean coal yield monitoring value should not be less than 75.0%, the average value of the tail coal ash content monitoring value should not exceed 50.0%, the standard deviation of the clean coal yield monitoring value should not be greater than 1.5 percentage points (a smaller value indicates greater stability, and this value is set based on the fluctuation level of the 75% percentile in historical data), and the trend line slope of the clean coal yield monitoring value should not be negative (i.e., greater than or equal to 0), and the trend line slope of the tail coal ash content monitoring value should not be positive (i.e., less than or equal to 0). The determination process is as follows: compare the actual average clean coal yield (e.g., 75.6%) with the expected indicator of 75.0%, the actual average tail coal ash content (e.g., 47.2%) with the expected indicator of 50.0%, the actual standard deviation of the clean coal yield (e.g., 1.2%) with the expected indicator of 1.5%, and the respective trend slopes (e.g., +0.2 for clean coal yield, -0.1 for tail coal ash content) with the expected direction. If all statistical characteristics meet or exceed the corresponding expected indicators, such as 75.6% 75.0%, 47.2% 50.0%, 1.2% 1.5%, +0.2 0, -0.1 0, then the flotation effect is determined to be "as expected". If any one or more do not meet the expected indicators, it is determined to be "not as expected" or "partially as expected", and the specific items that do not meet the indicators and the degree of deviation are recorded. These determination conclusions are summarized to generate production effect feedback results.

[0031] The control parameter adjustment difference acquisition step is: Based on the production effect feedback results, the clean coal yield monitoring value, tail coal ash content monitoring value, and slurry liquid level height prediction monitoring value are extracted from the production effect feedback results, and the respective prediction confidence interval boundaries of the reagent addition amount set value prediction, air volume set value prediction, and slurry liquid level height prediction in the control parameter scheme are extracted. The daily average absolute fluctuation value of each control parameter in the last 30 days and the average value in the last 30 days are recorded to form a parameter comparison set; Based on the parameter comparison set, the comprehensive deviation degree is calculated, and the calculation formula is: ; wherein, is the comprehensive deviation degree of the th control parameter, is the actual production effect feedback value of the th control parameter, is the lower boundary value of the prediction confidence interval of the th control parameter and to avoid division by zero, is the lower boundary value of the prediction confidence interval of the th control parameter, is the upper boundary value of the prediction confidence interval of the th control parameter and to avoid division by zero, is the upper boundary value of the prediction confidence interval of the th control parameter; is the relative daily fluctuation range of the th control parameter in the last 30 days, calculated as wherein, is the standard deviation of the th control parameter in the last 30 days, is the average value of the th control parameter in the last 30 days; Based on the comprehensive deviation degree, first determine the relative position relationship between the production effect feedback value and the prediction confidence interval of each control parameter. If the production effect feedback value exceeds the upper limit of the confidence interval, the control parameter is adjusted in the decreasing direction. If it is lower than the lower limit of the confidence interval, the control parameter is adjusted in the increasing direction. The adjustment range is determined in combination with the comprehensive deviation degree to generate the control parameter adjustment difference.

[0032] Specifically, based on the production effect feedback result, which contains a comprehensive evaluation of the current production batch flotation effect, such as “as expected” and specific performance data such as the average value of the clean coal yield monitoring value is 75.6%, the average value of the tail coal ash monitoring value is 47.2%, first extract the average value of the above clean coal yield monitoring value 75.6% and the average value of the tail coal ash monitoring value 47.2% from the production effect feedback result. At the same time, obtain the average monitoring value of the actual operation of the ore pulp liquid level in the same production evaluation period, for example, the average ore pulp liquid level height of 2.82 meters calculated by calling the liquid level sensor data of the corresponding time window in the historical database (SCADA system record). These three (75.6%, 47.2%, 2.82 meters) will be used as the actual production effect feedback value of each control target parameter (clean coal yield, tail coal ash, ore pulp liquid level) , then from the control parameter scheme generated at the early stage of the flow, the confidence interval boundary of the previously large model predicted reagent dosage set value prediction (for example, the predicted reagent dosage 1250 grams / ton, the lower boundary of the confidence interval is 1220 grams / ton, the upper boundary is 1280 grams / ton), the confidence interval boundary of the predicted aeration amount set value prediction (for example, the predicted aeration amount 2.5 cubic meters / minute, the lower boundary of the confidence interval is 2.3 cubic meters / minute, the upper boundary is 2.7 cubic meters / minute), and the confidence interval boundary of the predicted pulp liquid level height prediction (for example, the predicted pulp liquid level 2.8 meters, the lower boundary of the confidence interval is 2.75 meters, the upper boundary is 2.85 meters), and then, for each actual applied control parameter, i.e., reagent dosage, aeration amount, and pulp liquid level, the entire set value record of the last 30 days is queried and counted from the historical operation database, and the respective set value average value and set value standard deviation of the last 30 days are calculated, for example, for the reagent dosage, the average value of all set values of the last 30 days is 1240 grams / ton, and the standard deviation of these set values is 30 grams / ton, and the same statistical calculation is performed for the aeration amount and the pulp liquid level. The extracted and calculated actual production effect feedback value, the predicted confidence interval boundary, and the average value and the standard deviation of the last 30 days are collected to form a parameter comparison set.

[0033] Based on the parameter comparison set, the comprehensive deviation degree is calculated, the formula is: , the beneficial part of the formula is that the degree to which the actual production effect deviates from its predicted confidence interval can be quantified, and the relative volatility of the historical set value of the control parameter itself is also considered , the numerator part is designed as a quadratic term, when exactly falls on the boundary of the confidence interval, the deviation degree is zero, and when is farther away from the center of the confidence interval or exceeds the interval range, the numerator term rapidly increases, and the in the denominator plays a regulating role, if the historical set value of a parameter itself often fluctuates greatly (i.e., is large), then the current deviation will be considered relatively not so “abnormal”, and thus the overall deviation degree is appropriately reduced , on the contrary, for a parameter with stable historical set values, even a small deviation will lead to a significant This design makes deviation assessment more intelligent and adaptive, enabling it to more accurately identify deviations in parameters that truly require attention and adjustment. The steps to obtain it are as follows: Representing the The feedback value of the actual production effect of each control parameter within the current evaluation period. Here, "control parameter" specifically refers to the average performance of the setpoints directly issued to the control system (such as dosage, aeration rate, and slurry level) in actual operation, or the actual average value of key production indicators (such as clean coal yield and tailings ash content) generated under the action of these control parameters. To calculate the dosage control parameter... ,but This is the average actual dosage of the dosing system during the current evaluation period. This value is obtained by collecting feedback signals from the dosing actuators (such as the actual speed of the dosing pump or the flow meter reading) through the SCADA system and calculating their average value over a set time window. For example, for the dosage (index) By collecting feedback on the actual stroke frequency of the dosing pump during the evaluation period (e.g., 2 hours), this frequency has a calibration relationship with the actual dosage. For example, if data is recorded once per minute, totaling 120 data points, the average calibrated dosage is 1265 grams per ton. grams per ton.

[0034] The steps to obtain it are as follows: Representing the The lower boundary value of the prediction confidence interval for each control parameter, which originates from the generation of the "control parameter scheme" in an early step of the application process. This value is obtained by extrapolating and calculating specific coal slime characteristics and slurry flow data using a large model and performing random disturbance analysis. For example, for the dosage (index) The lower boundary of the confidence interval for the predicted dosage setpoint obtained from the aforementioned "control parameter scheme" is 1220 g / ton. Therefore... grams per ton.

[0035] The steps to obtain it are as follows: Representing the The upper boundary value of the prediction confidence interval for each control parameter is obtained in the same way as... The same principle applies, both stemming from the "control parameter scheme," for example, regarding the dosage (index). The upper boundary of the confidence interval for the predicted dosage setpoint obtained from the "control parameter scheme" is 1280 g / ton. Therefore... grams per ton.

[0036] The acquisition step of is: is a small correction introduced to avoid the occurrence of division by zero or small value in the calculation, resulting in numerical instability.

[0037] The acquisition step of is: is the lower boundary value of the predicted confidence interval after correction of to ensure the stability of the calculation, and its calculation formula is , substituting the previously obtained drug addition parameters grams / ton and , then grams / ton.

[0038] The acquisition step of is: is the upper boundary value of the predicted confidence interval after correction of to ensure the stability of the calculation, and its calculation formula is , substituting the previously obtained drug addition parameters grams / ton and , then grams / ton.

[0039] The acquisition step of is: represents the average value of the set value of the th control parameter in the last 30 days, which is obtained by querying the historical database of the control system, extracting all set value records of the control parameter in the past 30 days (for example, recording the average set value once a day, or recording once an hour, and then taking the average), and then calculating the average value of these records, for example, for the drug addition (index ), querying the historical records to obtain the average drug addition set value every day in the past 30 days, if the sum of the 30 daily average set values is 37200 grams / ton, then grams / ton.

[0040] The acquisition step of is: represents the standard deviation of the set value of the th control parameter in the last 30 days, reflecting the fluctuation degree or instability of the parameter set value in the past month, when calculating, first obtain all control parameters in the past 30 days as when calculating the set value data sequence, and then based on this data sequence, using the statistical calculation formula of standard deviation, that is, the square root of the average of the square sum of the difference between each data point and the mean value, for example, for the reagent dosage (index ), based on the past 30-day daily average reagent dosage set value sequence and the calculated mean value , the standard deviation of these set values is 30 g / t, so .

[0041] The acquisition step is: represent the relative daily fluctuation amplitude of the first control parameter in the last 30 days, also known as the coefficient of variation, which is a dimensionless quantity used to compare the fluctuation degree of parameters of different orders of magnitude or different units, and its calculation formula is , here we need to ensure that is not zero, if is close to zero, then may be meaningless or extremely large, but for coal preparation plant control parameters, this situation usually does not occur, substituting the reagent dosage parameters g / t and g / t obtained in the previous step, we get .

[0042] Calculation process: Take the reagent dosage control parameter (index ) as an example, substitute the obtained parameter values for calculation: g / t; g / t; g / t; g / t; g / t; ; Calculate the product part in the numerator: First term: ; Second term: ; Multiply the two terms: ; Take the absolute value: ; Calculate the denominator part: ; Calculate the comprehensive deviation : ; The results indicate that, for the control parameter of dosage, the current actual production effect feedback value... g / ton is within the predicted confidence interval Within grams per ton, and relatively close to the center of the confidence interval (closer to) Some), the calculated comprehensive deviation This is a very small value, much less than 1, which usually indicates that the actual value matches the expected range of the prediction model well, especially considering its historical volatility. Such deviation is considered slight, close to 0. A value indicates that the parameter is running stably and in line with the prediction; the larger the value, the more serious the deviation, and the more significant the adjustment is required.

[0043] Based on the overall deviation, such as the overall deviation of the dosage calculated above. In addition to the comprehensive deviation calculated for other control parameters (such as aeration volume and slurry level), the actual production performance feedback value within the current evaluation period is first calculated for each control parameter. (e.g., dosage) (g / ton) and the prediction confidence interval of this parameter given in the "Control Parameter Scheme" (e.g., dosage). Compare them (grams / ton) to determine their relative positional relationship. The specific judgment logic is as follows: if... Greater than the upper boundary of the prediction confidence interval If so, it is initially determined that the setpoint of the control parameter needs to be adjusted to a lower value. Less than the lower boundary of the prediction confidence interval If so, it is initially determined that the setpoint of the control parameter needs to be adjusted in the direction of increase. lie in and Between (including boundaries), such as the dosage in this example. If the parameter operates within the expected range, the decision to fine-tune it and the degree of fineness depends primarily on the magnitude of the overall deviation. Then, the overall deviation of the parameter is considered. To determine the specific adjustment range, a tiered or continuous strategy can be adopted, for example, by setting a base adjustment step size. This step size is preset based on the physical meaning of each parameter and the control sensitivity, such as the basic adjustment step size for the dosage. The basic adjustment step for inflation volume can be set to 5 grams per ton. The basic adjustment step for the slurry level is set at 0.05 cubic meters per minute. Set to 0.02 meters, these basic step size set reference to the minimum effective adjustment amount recommended by the device manual and senior engineers based on long-term operation experience summed up the safe adjustment range, and then the actual adjustment amount Can be calculated as Where the "direction coefficient" is determined according to the aforementioned position relationship (for example, -1 if it needs to be reduced, +1 if it needs to be raised, and if it needs to be fine-tuned within the interval, it is determined according to The direction of deviation from the midpoint of the interval), Is an adjustable sensitivity coefficient, for example, set to 10, The value is selected by observing the response speed and stability of the system adjustment under different Values through offline simulation testing, and selecting a value that can converge in fewer iteration times and is not easy to cause system oscillation, for parameters within the confidence interval, if it Deviation from the midpoint of the interval Exceeds a certain proportion (for example, 10% of the interval half-width) and Greater than a small threshold value (for example, 0.001, this threshold value is used to avoid adjusting for small, insignificant deviations), fine-tuning can also be performed, and the adjustment value of each control parameter is calculated according to the above logic to form the control parameter adjustment difference.

[0044] The acquisition step of the adjusted parameter instruction set is: Based on the control parameter adjustment difference, extract the reagent addition amount adjustment value, the aeration amount adjustment value and the pulp liquid level adjustment value in the control parameter adjustment difference, match the reagent addition amount set value prediction, the aeration amount set value prediction and the pulp liquid level prediction in the control parameter scheme respectively, and extract the maximum allowable variation amplitude of the control parameter scheme in the last 30 days to form a parameter adjustment pairing set; Based on the parameter adjustment pairing set, calculate the adjusted set value, the calculation formula is: ; Where, Represents the adjusted set value of the th control parameter, Represents the initial prediction value of the th control parameter, Represents the control parameter adjustment difference of the th control parameter, Represents the maximum allowable variation amplitude of the th control parameter in the last 30 days, and the Denominator ensures non-linear smoothing processing and avoids numerical mutations, and The function definition must ensure ; Based on the adjusted set value of each control parameter, the adjusted set value of the reagent dosage, the adjusted set value of the aeration amount and the adjusted set value of the pulp liquid level are converted into a data format conforming to the execution standard of the flotation control unit in sequence to generate an adjusted parameter instruction set.

[0045] Specifically, based on the control parameter adjustment amount, the adjustment amount is the recommended adjustment value for each control parameter (reagent dosage, aeration amount, pulp liquid level) calculated by the previous step according to the production effect feedback and the comprehensive deviation, for example, the obtained reagent dosage adjustment value g / t, the aeration amount adjustment value m3 / min, and the pulp liquid level adjustment value m, first, the three specific adjustment values are extracted from the control parameter adjustment amount, and then the adjustment values are matched with the initial predicted set value corresponding to each adjustment value, and the initial predicted set value is derived from the "control parameter scheme" generated in the early stage of the process of the present application, for example, the reagent dosage set value prediction g / t, the aeration amount set value prediction m3 / min, and the pulp liquid level prediction m, and for each control parameter, the maximum allowable variation range within the last 30 days is extracted The maximum allowable variation range is a limit value preset according to the safe operation rules of the coal preparation plant equipment and the process stability requirements, and is not dynamically extracted from the "control parameter scheme", but is stored in the system configuration as a fixed engineering parameter, for example, the maximum allowable variation range of the reagent dosage is set to 20 g / t, which is to prevent excessive impact on the reagent concentration caused by a single adjustment, based on historical data analysis, adjustments exceeding this range in the short term have caused "flotation tank turning" or reagent waste, the maximum allowable variation range of the aeration amount is set to 0.3 m3 / min, and the maximum allowable variation range of the pulp liquid level is set to 0.1 m, these two values are also determined based on the equipment capacity (such as the fan adjustment rate, the valve response time) and the process stability window (to avoid pulp overflow or air suction), the extracted adjustment value of each control parameter, the corresponding initial predicted value , and the preset maximum allowable variation range are combined to form a parameter adjustment pairing set.

[0046] Based on the parameter adjustment pairing set, the adjusted set value is calculated, the formula is: The benefit of the formula is that the formula provides a nonlinear way to calculate the final adjusted set value , not just simply adding the adjustment amount to the initial prediction value Instead, the adjustment result is smoothed and dynamically scaled by introducing a maximum allowed variation amplitude and a logarithmic term of the ratio of the initial prediction value to this amplitude so that when the initial prediction value is large or small (i.e. far from the case), the denominator increases, thus inhibiting the numerator more strongly, which helps to make the new set value tend towards a more conservative or more restrained value in the case where it was already in a relatively extreme (too high or too low, with respect to the reference) situation and possibly prevent excessive adjustment, stabilizing the control parameter, especially when is much greater than , the formula reduces the output value , preventing the set value from entering an undesirable area; The acquisition step of the initial prediction value of the i-th control parameter is: The initial prediction value of the i-th control parameter, which is calculated by the large model based on the real-time input of the slime property data and the slurry flow data, is recorded in the "control parameter scheme". This value represents the optimal control parameter set reference under the current working condition, for example, for the reagent addition control parameter (let ), its initial prediction value is 200 g / t, which is within the common reagent addition reference value range (usually tens to hundreds of g / t) for a specific coal type and process condition in the coal preparation industry, so g / t.

[0047] The acquisition step of the control parameter adjustment difference of the i-th control parameter is: The control parameter adjustment difference of the i-th control parameter, which is determined based on the comparison and analysis of the production effect feedback result and the prediction confidence interval in the previous step and the comprehensive deviation calculation, indicates the correction size and direction of the initial prediction value based on the current actual production effect, for example, for the reagent addition control parameter (let ), g / t.

[0048] The acquisition step of the control parameter adjustment difference of the i-th control parameter is: The control parameter adjustment difference of the i-th control parameter, which is determined based on the comparison and analysis of the production effect feedback result and the prediction confidence interval in the previous step and the comprehensive deviation calculation, indicates the correction size and direction of the initial prediction value ​​​The maximum allowable variation of a control parameter in a single adjustment is not a dynamically calculated value, but rather an engineering parameter pre-set and fixed in the system by process engineers or equipment experts based on specific equipment performance limitations, process stability requirements, and safe operating procedures. It defines the maximum absolute amount of change allowed for the control parameter in a single adjustment iteration to prevent production fluctuations or equipment damage due to excessively rapid or large adjustments. For example, for the dosage control parameter (let...),... Considering the adjustment precision of the dosing pump, the time required for the reagents to mix evenly in the slurry, and the need to avoid drastic impacts on the flotation cell environment, the maximum permissible variation in a single operation is set at 20 grams per ton. This value is based on empirical observations; for example, under similar operating conditions, a single adjustment exceeding 20 grams per ton has led to drastic fluctuations in the clean coal recovery rate in the short term. grams per ton.

[0049] Calculation process: Dosage control parameters (index) Taking (e.g.,) as an example, substitute the obtained parameter values ​​for calculation: grams per ton; grams per ton; grams per ton; First calculate : grams per ton; Then calculate the product in the numerator. : (grams per ton) ; Next, calculate the parameter of the logarithmic term in the denominator. : (dimensionless); because ,satisfy The domain of a function is required.

[0050] calculate : ; ; Calculate the denominator : ; Calculate the fraction within the square root: (grams per ton) ; Take the absolute value (in this example, if the numerical value is positive, the absolute value remains unchanged): (k / t) ; Finally, the square root is calculated to get 35.66.

[0051] Based on the adjusted set value of each control parameter, such as the adjusted set value of reagent dosage calculated above (k / t), and the adjusted set value of aeration rate (e.g. (m3 / min) and the adjusted set value of slurry level (e.g. (m)), these decimal values need to be converted into the instruction format that meets the specific execution standard of the slime flotation control unit. This conversion process strictly refers to the "flotation control unit execution instruction format specification" defined at the initial stage of system design. This specification details the data type (such as 16-bit integer, 32-bit floating point) required for each parameter when communicating with the underlying programmable logic controller (PLC) or distributed control system (DCS), the value range, the necessary scaling factor, and the specific register address or communication protocol tag. For example, for the adjusted set value of reagent dosage 35.66 k / t, if the control unit's corresponding register (e.g. Modbus address MW200) receives unsigned 16-bit integer data, and the dosage unit is directly in k / t, then 35.66 needs to be rounded, for example, rounded to 36. For the adjusted set value of aeration rate 1.85 m3 / min, if its corresponding register (e.g. MW202) expects the unit to be 0.01 m3 / min, then 1.85 needs to be multiplied by 100 to get 185. This scaling factor 100 is preset based on the precision requirement of 0.01 (i.e. ), for the adjusted set value of slurry level 2.72 m, if its corresponding register (e.g. MW204) expects the unit to be mm, then 2.72 needs to be multiplied by 1000 to get 2720. This scaling factor 1000 is determined based on the unit conversion from m to mm (1 m = 1000 mm). After completing this data type conversion, unit conversion, and value scaling for all adjusted set values of control parameters, the encoded values (e.g. 36, 185, 2720) are combined with their corresponding control unit target addresses (e.g. MW200, MW202, MW204) to form a series of standardized instruction entries. These entries are combined to generate the adjusted parameter instruction set.

[0052] The iterative optimization control parameter set acquisition step is: Based on the adjusted parameter instruction set, the adjusted setting value of the reagent addition amount, the adjusted setting value of the aeration amount and the adjusted setting value of the slurry liquid level height in the adjusted parameter instruction set are parsed, and are converted into standard data instructions in sequence according to the communication protocol of the slime flotation control unit, to form a standard data instruction set; Based on the standard data instruction set, the adjusted setting value standard data instruction of the reagent addition amount in the set is sent to the reagent addition system control interface, the adjusted setting value standard data instruction of the aeration amount is sent to the aeration system control interface, and the adjusted setting value standard data instruction of the slurry liquid level height is sent to the liquid level control system interface, to complete the parameter configuration issuing action and form an iterative optimization control parameter set.

[0053] Specifically, based on the adjusted parameter instruction set, the instruction set contains the target address of each control parameter (such as the reagent addition amount, the aeration amount and the slurry liquid level height) and the adjusted setting value after data type conversion and scaling. For example, the instruction of the adjusted parameter instruction set for the reagent addition amount is (target address MW200, adjusted setting value 36), the instruction for the aeration amount is (target address MW202, adjusted setting value 185), and the instruction for the slurry liquid level height is (target address MW204, adjusted setting value 2720). First, the adjusted parameter instruction set is parsed, and each instruction is read and identified. The specific control parameter name (for example, the reagent addition amount is identified by looking up the mapping table through the address MW200), the target address and the encoded setting value corresponding to each instruction are identified. Then, according to the communication protocol specification actually used by the coal preparation plant flotation control unit, for example, the Modbus TCP / IP protocol, each parsed instruction information (target address and encoded setting value) is encapsulated into a complete write operation request message in the format of a data frame conforming to the protocol, which is provided by the device supplier and fixed in the system interface document during the integration of the control system. It defines the header structure, function code, data area byte sequence, etc. of the message in detail. Taking the reagent addition amount instruction (MW200, 36) as an example, if it is converted into a Modbus TCP write single register (function code 0x06) request message, a transaction processing identifier (such as a dynamically generated sequence number), a protocol identifier (0x0000), a length field (indicating the number of subsequent bytes, for example, 6 bytes), a unit identifier (PLC Modbus slave address, for example, 0x01), a function code (0x06), a register address (for example, the specific address 199 corresponding to MW200, if the MW0 start address is 0), and the data value to be written (0x0024, which is decimal 36) need to be constructed. The same operation is performed on the adjusted setting value instructions of the aeration amount and the slurry liquid level height to generate corresponding Modbus TCP write request messages. These encapsulated single write operation request messages conforming to the specific communication protocol standard form a standard data instruction set.

[0054] Based on a standard data instruction set, which contains independent write operation request messages generated for each control parameter such as dosage, aeration rate, and slurry level, conforming to a specific communication protocol (e.g., Modbus TCP / IP), the system processes each standard data instruction sequentially according to a predetermined order, usually based on the order of the parameters' impact on the process or their importance, or according to the natural order of the parameters in the instruction set. Specifically, the system first retrieves the standard data instruction for the adjusted dosage setpoint (e.g., the previously generated instruction containing the setpoint 36 and the target...). The Modbus TCP write request message (labeled with address MW200) is sent via the factory's industrial Ethernet to the IP address and designated port (e.g., IP address 192.168.1.10, port 502) of the configured dosing system control PLC. After sending, the system starts a response waiting timer, for example, set to a timeout of 3 seconds. This 3-second setting is based on an assessment of historical network communication latency (e.g., average round-trip time of 0.5 seconds) and typical PLC instruction processing time (e.g., 0.2 seconds), with an added safety margin (e.g., ...). If a Modbus TCP response message indicating successful writing is received from the dosing system control PLC within 3 seconds, the dosing quantity parameter configuration is considered successfully sent. If no response is received within the timeout period or an error response message is received, a failure is recorded and an alarm can be triggered. A retry operation is then performed according to a preset strategy, for example, a maximum of 2 retries, each with a 5-second interval. These two retry parameters (2 retries, 5 seconds) were determined during the system debugging phase based on network stability and acceptable latency. Then, the standard data command for the adjusted inflation quantity is processed in the same way and sent to the inflation system control PLC. The corresponding interface of the PLC is configured and confirmation is awaited. Finally, the standard data command for adjusting the slurry level height is processed and sent to the corresponding interface of the PLC of the level control system and confirmation is awaited. When the standard data commands for all parameters are successfully sent and the correct response is received from the control system, the configuration and sending of all control parameters is completed. These new control parameter settings that have been successfully sent and applied to the actual production process (e.g., dosage of 36 grams / ton, aeration rate of 1.85 cubic meters / minute, slurry level of 2.72 meters) together form the iterative optimization control parameter set for the next production cycle.

[0055] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An intelligent prediction method for slime flotation control parameters based on a large model, characterized in that, The method comprises the following steps: Based on the real-time input of the slime characteristic data and the slurry flow data, feature extraction and structured arrangement are performed for large model input and large model training, and the large model is used for deducing calculation to obtain the reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction, and the prediction values are subjected to disturbance analysis and quantification of fluctuation range to establish a control parameter scheme; Based on the control parameter scheme, the reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction are extracted to generate a to-be-executed flotation control instruction, and based on the to-be-executed flotation control instruction, configuration is applied and the clean coal yield monitoring value and the tail coal ash content monitoring value are continuously monitored within a set time window to collect working condition key performance data; Based on the working condition key performance data, the clean coal yield monitoring value and the tail coal ash content monitoring value are extracted to obtain production effect feedback results, and based on the production effect feedback results, comparison and analysis are performed with the prediction confidence interval of each parameter in the control parameter scheme, the deviation degree of the effect from the prediction interval boundary is calculated, and the control parameter adjustment difference is determined; Based on the control parameter adjustment difference, combined with the initial reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction in the control parameter scheme, parameter adjustment operation is performed to generate an adjusted parameter instruction set, and based on the adjusted parameter instruction set, the optimized setting is issued to the slime flotation control unit as the next production batch to form an iterative optimization control parameter set.

2. The intelligent prediction method of slime flotation control parameters based on large models according to claim 1, characterized in that, The control parameter scheme acquisition step is: Based on the real-time input of the slime characteristic data and the slurry flow data, the ash content, particle size distribution and moisture content in the slime characteristic data and the flow fluctuation rate and flow stability characteristics in the slurry flow data are extracted, and the structured data feature set is formed in a predetermined order to obtain the structured data feature set; Based on the structured data feature set, the structured data feature set is input to the large model in a standardized numerical format, and the node parameters and weights of the model are adjusted to make the large model learn the change rule of the data features, form a trained large model, and the large model is a deep neural network model; Based on the trained large model, the structured data feature set is called as input data for forward reasoning calculation to deduce the reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction under corresponding conditions, the prediction results are subjected to random disturbance analysis to obtain the fluctuation range of each prediction result, and the prediction results and the corresponding fluctuation range are combined to obtain the control parameter scheme.

3. The intelligent prediction method of slime flotation control parameters based on large models according to claim 1, characterized in that, The to-be-executed flotation control instruction acquisition step is: Based on the control parameter scheme, the reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction in the control parameter scheme are analyzed, the reagent dosage setting value prediction, the air charge setting value prediction and the slurry liquid level height prediction are extracted as independent control parameter items, and an independent control parameter item set is generated; Based on the independent control parameter entry set, the reagent dosage setting value prediction, the aeration amount setting value prediction and the pulp liquid level prediction in the independent control parameter entry set are sequentially data encoded according to the execution instruction format specification of the flotation control unit to form a data instruction template conforming to the input standard of the flotation control unit, and a to-be-executed flotation control instruction is generated.

4. The intelligent prediction method of slime flotation control parameters based on large models according to claim 1, characterized in that, The working condition key performance data acquisition step is: Based on the to-be-executed flotation control instruction, the reagent dosage setting value prediction, the aeration amount setting value prediction and the pulp liquid level prediction in the to-be-executed flotation control instruction are parsed, and the reagent dosage setting value prediction, the aeration amount setting value prediction and the pulp liquid level prediction are sequentially issued to the reagent feeding system, the aeration system and the liquid level control system, respectively, to complete the control parameter setting by configuration, and configured control parameters are generated. Based on the configured control parameters, the start time and the end time of a time window are set, and the clean coal yield monitoring value and the tail coal ash content monitoring value are continuously recorded in the set time window, and the clean coal yield monitoring value and the tail coal ash content monitoring value are sequentially archived according to the time stamp to generate working condition key performance data.

5. The intelligent prediction method of slime flotation control parameters based on large models according to claim 1, characterized in that, The production effect feedback result acquisition step is: Based on the working condition key performance data, all clean coal yield monitoring values and tail coal ash content monitoring values in the working condition key performance data are sequentially extracted, the clean coal yield monitoring values and the tail coal ash content monitoring values are one-to-one corresponding matched according to the time stamp, and the data records with missing or abnormal data are excluded to generate valid monitoring data pairs. Based on the valid monitoring data pairs, the change amplitude, the fluctuation interval and the extreme value are respectively counted according to the change trend of the clean coal yield monitoring value and the change trend of the tail coal ash content monitoring value, the performance of the valid monitoring data pairs in the set time window is evaluated, and a monitoring trend evaluation result is generated. Based on the monitoring trend evaluation result, it is determined whether the flotation effect of the production process meets the expected index, and a production effect feedback result is generated.

6. The method of claim 1, wherein the method is characterized by, The control parameter adjustment difference acquisition step is: Based on the production effect feedback result, the clean coal yield monitoring value, the tail coal ash content monitoring value and the pulp liquid level monitoring value in the production effect feedback result are extracted, the reagent dosage setting value prediction, the aeration amount setting value prediction and the pulp liquid level prediction in the control parameter scheme are extracted, and the respective prediction confidence interval boundaries are recorded, and the daily average absolute fluctuation value of each control parameter in the last 30 days and the average value in the last 30 days are recorded to form a parameter comparison set; Based on the parameter comparison set, a comprehensive deviation degree is calculated; Based on the comprehensive deviation degree, first, the relative positional relationship between the production effect feedback value of each control parameter and the prediction confidence interval is determined, if the production effect feedback value exceeds the upper limit of the confidence interval, the control parameter is adjusted in the decreasing direction, if it is lower than the lower limit of the confidence interval, the control parameter is adjusted in the increasing direction, the adjustment amplitude is determined combined with the comprehensive deviation degree, and a control parameter adjustment difference is generated.

7. The method of claim 1, wherein the method is characterized by, The adjusted parameter instruction set acquisition step is: Based on the control parameter adjustment difference, extract the reagent addition amount adjustment value, the aeration amount adjustment value and the slurry liquid level height adjustment value in the control parameter adjustment difference, match the reagent addition amount set value prediction, the aeration amount set value prediction and the slurry liquid level height prediction in the control parameter scheme respectively, and extract the maximum allowed variation range of the control parameter scheme in the last 30 days to form a parameter adjustment pairing set; Based on the parameter adjustment pairing set, calculate the adjusted set value; Based on the adjusted set value of each control parameter, convert the reagent addition amount adjusted set value, the aeration amount adjusted set value and the slurry liquid level height adjusted set value into a data format conforming to the execution standard of the flotation control unit in sequence to generate an adjusted parameter instruction set.

8. The method of claim 1, wherein the method is characterized by, The obtaining step of the iterative optimization control parameter set is: Based on the adjusted parameter instruction set, analyze the reagent addition amount adjusted set value, the aeration amount adjusted set value and the slurry liquid level height adjusted set value in the adjusted parameter instruction set, and convert them into standard data instructions according to the communication protocol of the coal slime flotation control unit in sequence to form a standard data instruction set; Based on the standard data instruction set, send the reagent addition amount adjusted set value standard data instruction to the reagent addition system control interface, the aeration amount adjusted set value standard data instruction to the aeration system control interface, and the slurry liquid level height adjusted set value standard data instruction to the liquid level control system interface in sequence to complete the parameter configuration issuing action and form an iterative optimization control parameter set.

Images