Semi-autogenous grinding process intelligent control method based on multi-stage working condition recognition and prediction

Abstract

The invention relates to the technical field of mineral separation automation, in particular to a semi-autogenous grinding process intelligent control method based on multi-stage working condition recognition and prediction, which comprises the following steps: acquiring sample data of ore grinding equipment in a preset time range, and judging the load state of a mill by replacing a fuzzy set of multiple variables with a multi-working condition recognition result, in addition, working condition trend and working condition transfer state prediction results serve as feed-forward information of working condition changes to be provided for a fuzzy controller, fuzzy control is carried out based on the information, and the control quantity is adjusted in advance before the working conditions change, so that the mill load is kept stable, the ore grinding efficiency is improved, and the ore grinding efficiency is improved. The ore grinding process state is stabilized.

Description

Technical Field

[0001] This invention relates to the field of technology, specifically to an intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction. Background Technology

[0002] Grinding is a crucial step in mineral processing, primarily responsible for crushing and classifying ore to provide suitable particle sizes for subsequent beneficiation processes. Semi-autogenous grinding mills are commonly used grinding equipment, offering advantages such as simultaneous crushing and coarse grinding, a simple process flow, fewer production stages, and fewer pieces of equipment. Intelligent optimization control of the semi-autogenous grinding process is a research hotspot in the mineral processing field. However, due to the highly nonlinear and dynamic nature of the grinding process, significant uncertainties exist in the system's behavior. Furthermore, the grinding process involves numerous variables, complex control objectives, and frequent changes in operating conditions, making most optimization control algorithms difficult to implement. Fuzzy control methods, through relatively simple rule design, can effectively address the challenges of nonlinearity, complexity, uncertainty, and real-time requirements in the grinding process, thereby achieving automated control and energy conservation in the grinding process.

[0003] Fuzzy control, as a simple and efficient optimization control method, has high application value. However, traditional fuzzy control methods need to comprehensively consider the influence of multiple variables on the grinding process state. Excessive complexity of variables may lead to rule explosion. Furthermore, the variables in the grinding process are highly volatile, and changes in operating conditions affect grinding efficiency, energy consumption, and product quality. Traditional operating condition identification methods cannot describe the changing trends and state transitions of the grinding process state. When operating conditions change, they only manifest as abrupt changes between adjacent operating conditions. Control methods based on traditional operating condition identification cannot provide timely control outputs to maintain stable operating conditions in response to changing trends, resulting in control lag. Therefore, it is necessary to conduct more detailed analysis of the operating conditions and refine the control schemes under different operating conditions so that the fuzzy controller can respond promptly and effectively when the operating conditions may or are about to change, thereby ensuring the stable and efficient operation of the grinding process. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent control method for semi-autogenous grinding processes based on multi-level operating condition identification and prediction, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A semi-autogenous grinding process intelligent control method based on multi-level working condition identification and prediction includes the following steps:

[0007] S1, Obtain sample data of the grinding process within a preset time range;

[0008] S2, Based on the multi-level working condition identification and prediction method, the working condition identification and prediction of the working condition transition state are performed according to the sample data;

[0009] S3, acquire the data required for controlling the grinding process at the current moment, and perform fuzzification processing on the data to obtain the fuzzification result, and denote the current moment as t1;

[0010] S4, trigger the corresponding rule in the fuzzy rule base based on the fuzzification result, multi-condition recognition result, condition trend and condition transition state prediction result;

[0011] S5, based on the triggered rules, performs fuzzy reasoning to obtain the ore feed rate setpoint, water feed rate setpoint, and rotation speed setpoint;

[0012] S6: Send the set values ​​of feed rate, water flow rate and rotation speed to the grinding process, and record the current time as t2; S7: Calculate the relationship between Δt = t2 - t1 and T, where T is the control period. If Δt > T, return to S6; otherwise, return to S1.

[0013] As a preferred embodiment of the present invention, the sample data of S1 includes industrial sensor data, production data, and electromechanical equipment operation data that can be used to evaluate on-site working conditions within a preset time range.

[0014] As a preferred embodiment of the present invention, step S2 employs a multi-level working condition identification and prediction method to provide the fuzzy controller with working condition identification and prediction results, and the steps are as follows:

[0015] S21. The obtained mill power (P), return sand amount (R), feed rate (F), and large ore block size (D) distribution are relatively normalized to ensure dimensional uniformity. The normalization formula is as follows: Where Y represents the variable that needs to be normalized, Y opt Y represents the optimal value of this variable. max Y represents the maximum value of the variable within that time period. * This represents the result after normalization of variable Y;

[0016] S22, based on the normalized mill power P * and return sand amount R * Calculate the operating condition score S at the current moment, identify the operating condition based on the actual value of the operating condition score S, and calculate the operating condition identification result S. t ;

[0017] S23, based on the normalized ore feed rate F * and large ore blocks of size D * Multi-timescale trend analysis of mill load dynamics was used to identify operating condition trends and obtain the operating condition trends of mill feed rate and large ore size distribution at time t.

[0018] S24, based on the multi-condition identification result S t Operating condition boundary S tL S tU and operating condition trend identification results Identify the working condition transition state C t ;

[0019] S25 uses an improved Transformer model based on multi-scale feature extraction to perform single-step classification prediction of multi-level working conditions, and obtains the prediction results of working condition transition states.

[0020] As a preferred embodiment of the present invention, the weighted scoring formula for the working condition in S22 is: S=αR * +βP * Where S is the operating condition score, and α and β are R... * P * The corresponding weighting factor, and α+β=1, α≤β, are used to divide the working conditions into N categories based on the actual value of S and the actual production situation, and a number is used to represent a category of working conditions.

[0021] As a preferred embodiment of the present invention, the specific method for identifying the working condition trend in S23 is as follows:

[0022] S231. The grinding process is characterized by nonlinearity, strong noise, and slow time variation. The short-term trend mainly reflects the time-varying characteristics, the long-term trend reflects the drift of the operating conditions, and the medium-term trend combines the advantages of both. In order to accurately judge the changing trend of the mill operating conditions, it is necessary to use the multi-scale time trend analysis method for operating condition identification.

[0023] S232, using the first-order difference method to calculate the short-term trend of mill load ST t It identifies rapid changes in operating conditions over a short period of time. (ST) t The short-term trend of variable y at time t is represented by the formula: Δy t =y t -y t-1 , where Δy t y represents the magnitude of change in data at adjacent time points. t y t-1 Here, t and t-1 are the data values ​​at time points t and t-1, respectively, and θ1 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent a downward trend, a steady state, and an upward trend, respectively.

[0024] S233, then the mid-term trend of mill load MT is calculated using the linear regression slope. tIt identifies the stable changing trend of operating conditions within a certain time window. MT t The formula for calculating the intermediate trend of variable y at time t is: Among them MT t y is the regression slope, n is the number of data points, t is the time index, and y is the regression slope. t θ2 is the intermediate trend threshold, and the calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0025] S234, then the Z-score method is used to calculate the long-term trend of mill load LT. t Identify long-term trends in operating conditions, LT t The formula for calculating the intermediate trend of variable y at time t is: Z t y represents the degree of deviation of the data at any given time. t Let μ be the data value at time t. t Let σ be the mean of the historical data window up to time t. t θ3 is the standard deviation of the historical data window before time t, and θ3 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0026] S235 employs a multi-scale trend voting method, using weighted voting to calculate short-term, medium-term, and long-term trends, collectively forming the final operating condition trend result. The calculation formula for the individual variable x-trend is FT. t =ω s ST t +ω m MT t +ω l LT t FT t Let ω be the operating condition trend score of variable x at time t. s ST t ω m MT t ω l LT t These are the weights of the short-term, medium-term, and long-term trends of variable x, respectively, and ω s +ω m +ω l =1.

[0027] S236, Since the feed rate (F) and the distribution of large ore blocks (D) directly affect the mill load and grinding efficiency, the feed rate is obtained through the multi-scale trend calculation formula in S135. and large ore blocks The final operating condition trend calculation expression for the distribution is as follows: The operating condition trend is divided into 3 categories, and the calculation results 0, 1 and 2 represent a downward trend, steady state and upward trend, respectively.

[0028] As a preferred embodiment of the present invention, the working condition boundary in S24 needs to be defined separately for each working condition according to the actual situation of the industrial site, with its upper boundary S defined separately. tU With lower boundary S tL .

[0029] As a preferred embodiment of the present invention, the specific steps for identifying the working condition transition state in S24 are as follows:

[0030] S241 performs boundary detection on the operating conditions, assuming that only when the operating condition score approaches the boundary can the changing trend of the operating conditions potentially cause the mill operating conditions to cross the boundary and trigger a condition transition. The detection formula is as follows: Where δ L δ U S represents the distance from the current operating condition to the lower and upper boundaries, respectively. tL S tU These are the lower and upper boundaries of the current operating condition, respectively.

[0031] S242, identify the working condition transition state, if the condition is satisfied... or And the operating condition is sufficiently close to the lower boundary of the current operating condition, i.e., δ L <ε, where ε is the boundary judgment threshold, and the current working condition S is calculated through S12. t If the value is not 0, the operating condition is considered to be in a descent transition state. or And the operating condition is sufficiently close to the upper boundary of the current operating condition, i.e., δ U If the condition is less than ε and the current operating condition St is not operating condition 4, then the operating condition is considered to be in an upward transition state. If neither of the above two conditions is met, then the operating condition is determined not to have transitioned and is defined as a non-transition state, and the operating condition transition state category is C. t The formula is defined as follows Where 0, 1, and 2 represent the descending transition state, the non-transition state, and the ascending transition state, respectively, and ε is the boundary judgment threshold.

[0032] As a preferred embodiment of the present invention, the improved Transformer model in S25 includes a UnetTSF feature extraction module, a Transformer encoder module, and an output fully connected module connected in sequence. The UnetTSF feature extraction module is used to extract multi-scale features of the data, the Transformer encoder module is used for global temporal modeling, and the output fully connected module is used to classify and predict mill load conditions (static class), large ore proportion trend (dynamic1 class), mill feed trend (dynamic2 class), and mill condition transition state (transfer class).

[0033]

[0034] Where X∈R B×T×C The input is the original time series data, where B is the number of sample batches, T is the sequence length, C is the feature dimension at each time step, and H is the number of time series data. UnetTSF H is the multi-scale feature extracted by UnetTSF. Trans The global temporal features extracted for the Transformer are taken as the final feature input to the classifier, since the Transformer's output is a time series feature. W is the feature extracted from the last time step of the Transformer's output. static For static class tasks, the weight matrix, W dynamicl The weight matrix W for dynamic1 type tasks dynamic2 The weight matrix W for dynamic2 type tasks transfer Let b be the weight matrix for transfer-type tasks. static For the bias term corresponding to the static class task, b dynamicl For the bias term corresponding to the dynamic1 class task, b dynamic2 For the bias term corresponding to the dynamic2 type task, b transfer Y is the bias term corresponding to transfer-type tasks. static For the prediction results of static class tasks, Y dynamicl For the prediction results of dynamic1 type tasks, Y dynamic2 For the prediction results of dynamic2 type tasks, Y transfer This is the prediction result for transfer-type tasks.

[0035] As a preferred embodiment of the present invention, in step S3, fuzzy set partitioning is performed on all variables, and fuzzification results are output. Based on actual production experience, the variables are divided into 3-fuzzy subset variables and 5-fuzzy subset variables. The 3-fuzzy subset is represented as {L,M,H} = {"low", "normal", "high"}, and the 5-fuzzy subset is represented as {LL,L,M,H,HH} = {"very low", "low", "normal", "high", "very high"}.

[0036] As a preferred embodiment of the present invention, the specific rule triggering process of the multi-module-based fuzzy control rule base design method in step S4 is as follows: The specific rule triggering process of the multi-module-based fuzzy control rule base design method in step S4 is as follows: Determine the input variables and output variables, trigger the corresponding rules in the fuzzy rule base according to the fuzzification results, multi-condition identification results, and condition transition state prediction results, and determine whether the current input variable is in an extreme case or the condition is in an extreme condition; if so, trigger the rules of the anomaly judgment module; if not, determine the current mill load state according to the multi-condition identification results and trigger the rules in the instantaneous load control module, then trigger the rules in the trend response control module according to the condition trend and condition transition state prediction results, and finally send the triggering rules to the fuzzy controller for fuzzy inference.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] 1. This invention proposes a multi-level operating condition identification method that integrates static features and changing trends. First, it identifies multiple operating conditions of the grinding process based on key features of process variables. Then, it identifies operating condition trends based on the changing trend features of input variables. Finally, based on operating condition boundary information and operating condition trends, it determines whether the current operating condition is in a transitional state between different operating conditions, and identifies the operating condition transition state accordingly. Based on this, an improved Transformer model based on multi-scale feature extraction is proposed to predict multi-level operating conditions, providing feedforward information for the control method. Addressing the problems of redundant variables leading to rule explosion and the inability to respond promptly to rapidly changing operating conditions in traditional fuzzy control methods, a semi-autogenous grinding process intelligent control method based on multi-level operating condition identification and prediction is proposed. This method uses the multi-operating condition identification results, rather than the state variables of the grinding process, as the fuzzy control input, reducing the number of variables involved in the fuzzy rules and simplifying the rules. Furthermore, the operating condition trends and the predicted results of the operating condition transition state are provided as feedforward information to the fuzzy controller to adjust the control quantity in advance before the operating conditions change, maintaining stable operating conditions. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the overall steps of an embodiment of the present invention;

[0040] Figure 2 This is a flowchart of a multi-level working condition identification and prediction method based on the fusion of static features and changing trends according to an embodiment of the present invention.

[0041] Figure 3 This is a structural diagram of the improved Transformer model based on multi-scale feature extraction according to an embodiment of the present invention;

[0042] Figure 4 This is a structural diagram of the fuzzy controller according to an embodiment of the present invention;

[0043] Figure 5 This is a flowchart illustrating the fuzzy rule base triggering process according to an embodiment of the present invention.

[0044] Figure 6 This is the multi-level operating condition prediction confusion matrix in an embodiment of the present invention;

[0045] Figure 7 This invention provides a comparison of the control effects of intelligent control and manual control under non-abnormal operating conditions in simulation experiments.

[0046] Figure 8 This invention presents a comparison of the effects of intelligent control and manual control under abnormal operating conditions in simulation experiments. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] To facilitate understanding of the present invention, a more comprehensive description of the invention will be given below with reference to the accompanying drawings, and several embodiments of the invention will be provided. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.

[0049] For examples, please refer to Figure 1-5 The present invention provides a technical solution:

[0050] A semi-autogenous grinding process intelligent control method based on multi-level working condition identification and prediction includes the following steps:

[0051] S1, Obtain sample data of the grinding process within a preset time range;

[0052] S2, Based on the multi-level working condition identification and prediction method, the working condition identification and prediction of the working condition transition state are performed according to the sample data;

[0053] S3, acquire the data required for controlling the grinding process at the current moment, and perform fuzzification processing on the data to obtain the fuzzification result, and denote the current moment as t1;

[0054] S4, trigger the corresponding rule in the fuzzy rule base based on the fuzzification result, multi-condition recognition result, condition trend and condition transition state prediction result;

[0055] S5, based on the triggered rules, performs fuzzy reasoning to obtain the ore feed rate setpoint, water feed rate setpoint, and rotation speed setpoint;

[0056] S6, send the set values ​​of feed rate, water flow rate and rotation speed to the grinding process, and record the current time as t2;

[0057] S7, calculate the relationship between Δt = t2 - t1 and T, where T is the control period. If Δt > T, return to S6; otherwise, return to S1.

[0058] Furthermore, the sample data of S1 includes industrial sensor data, production data, and electromechanical equipment operation data that can be used to evaluate on-site working conditions within a preset time range.

[0059] Furthermore, in step S2, a multi-level working condition identification and prediction method is used to provide the fuzzy controller with working condition identification and prediction results. The steps are as follows:

[0060] S21. The obtained mill power (P), return sand amount (R), feed rate (F), and large ore block size (D) distribution are relatively normalized to ensure dimensional uniformity. The normalization formula is as follows: Where Y represents the variable that needs to be normalized, Y opt Y represents the optimal value of this variable. max Y represents the maximum value of the variable within that time period. * This represents the result after normalization of variable Y;

[0061] S22, based on the normalized mill power P * and return sand amount R * Calculate the operating condition score S at the current moment, identify the operating condition based on the actual value of the operating condition score S, and calculate the operating condition identification result S. t ;

[0062] S23, based on the normalized ore feed rate F * and large ore blocks of size D * Multi-timescale trend analysis of mill load dynamics was used to identify operating condition trends and obtain the operating condition trends of mill feed rate and large ore size distribution at time t.

[0063] S24, based on the multi-condition identification result S t Operating condition boundary S tL S tU and operating condition trend identification results Identify the working condition transition state C t ;

[0064] S25 uses an improved Transformer model based on multi-scale feature extraction to perform single-step classification prediction of multi-level working conditions, and obtains the prediction results of working condition transition states.

[0065] Furthermore, the weighted scoring formula for the working condition in S22 is: S=αR * +βP * Where S is the working condition score, and α and β are R... * P * The corresponding weighting factor, and α+β=1, α≤β, are used to divide the working conditions into N categories based on the actual value of S and the actual production situation, and a number is used to represent a category of working conditions.

[0066] Furthermore, the specific method for identifying operating condition trends in S23 is as follows:

[0067] S231. The grinding process is characterized by nonlinearity, strong noise, and slow time variation. The short-term trend mainly reflects the time-varying characteristics, the long-term trend reflects the drift of the operating conditions, and the medium-term trend combines the advantages of both. In order to accurately judge the changing trend of the mill operating conditions, it is necessary to use the multi-scale time trend analysis method for operating condition identification.

[0068] S232, using the first-order difference method to calculate the short-term trend of mill load ST t It identifies rapid changes in operating conditions over a short period of time. (ST) t The short-term trend of variable y at time t is represented by the formula: Δy t =y t -y t-1 , where Δy t y represents the magnitude of change in data at adjacent time points. t y t-1 Here, t and t-1 are the data values ​​at time points t and t-1, respectively, and θ1 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent a downward trend, a steady state, and an upward trend, respectively.

[0069] S233, then the mid-term trend of mill load MT is calculated using the linear regression slope. t It identifies the stable changing trend of operating conditions within a certain time window. MT t The formula for calculating the intermediate trend of variable y at time t is: Among them MT t y is the regression slope, n is the number of data points, t is the time index, and y is the regression slope. t θ2 is the intermediate trend threshold, and the calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0070] S234, then the Z-score method is used to calculate the long-term trend of mill load LT. t Identify long-term trends in operating conditions, LT t The formula for calculating the intermediate trend of variable y at time t is: Z t y represents the degree of deviation of the data at any given time. t Let μ be the data value at time t. t Let σ be the mean of the historical data window up to time t. t θ3 is the standard deviation of the historical data window before time t, and θ3 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0071] S235 employs a multi-scale trend voting method, using weighted voting to calculate short-term, medium-term, and long-term trends, collectively forming the final operating condition trend result. The calculation formula for the individual variable x-trend is FT. t =ω s ST t +ω m MT t +ω l LT t FT t Let ω be the operating condition trend score of variable x at time t. s ST t ω m MT t ω l LT t These are the weights of the short-term, medium-term, and long-term trends of variable x, respectively, and ω s +ω m +ω l =1.

[0072] S236, Since the feed rate (F) and the distribution of large ore blocks (D) directly affect the mill load and grinding efficiency, the feed rate is obtained through the multi-scale trend calculation formula in S135. and large ore blocks The final operating condition trend calculation expression for the distribution is as follows: The operating condition trend is divided into 3 categories, and the calculation results 0, 1 and 2 represent a downward trend, steady state and upward trend, respectively.

[0073] Furthermore, the working condition boundaries in S24 need to be defined separately for each working condition based on the actual conditions of the industrial site, defining its upper boundary S. tU With lower boundary S tL .

[0074] Furthermore, the specific steps for identifying the working condition transition state in S24 are as follows:

[0075] S241 performs boundary detection on the operating conditions, assuming that only when the operating condition score approaches the boundary can the changing trend of the operating conditions potentially cause the mill operating conditions to cross the boundary and trigger a condition transition. The detection formula is as follows: Where δ L δ U S represents the distance from the current operating condition to the lower and upper boundaries, respectively. tL S tU These are the lower and upper boundaries of the current operating condition, respectively.

[0076] S242, identify the working condition transition state, if the condition is satisfied... or And the operating condition is sufficiently close to the lower boundary of the current operating condition, i.e., δ L <ε, where ε is the boundary judgment threshold, and the current working condition S is calculated through S12. t If the value is not 0, the operating condition is considered to be in a descent transition state. or And the operating condition is sufficiently close to the upper boundary of the current operating condition, i.e., δ U If the condition is less than ε and the current operating condition St is not operating condition 4, then the operating condition is considered to be in an upward transition state. If neither of the above two conditions is met, then the operating condition is determined not to have transitioned and is defined as a non-transition state, and the operating condition transition state category is C. t The formula is defined as follows Where 0, 1, and 2 represent the descending transition state, the non-transition state, and the ascending transition state, respectively, and ε is the boundary judgment threshold.

[0077] Furthermore, the improved Transformer model in S25 includes a UnetTSF feature extraction module, a Transformer encoder module, and an output fully connected module connected in sequence. The UnetTSF feature extraction module is used to extract multi-scale features of the data, the Transformer encoder module is used for global temporal modeling, and the output fully connected module is used to classify and predict mill load conditions (static class), large ore proportion trends (dynamic1 class), mill feed trends (dynamic2 class), and mill condition transition states (transfer class).

[0078]

[0079] Where X∈RB×T×C The input is the original time series data, where B is the number of sample batches, T is the sequence length, C is the feature dimension at each time step, and H is the number of time series data. UnetTSF H is the multi-scale feature extracted by UnetTSF. Trans The global temporal features extracted for the Transformer are taken as the final feature input to the classifier, since the Transformer's output is a time series feature. W is the feature extracted from the last time step of the Transformer's output. static For static class tasks, the weight matrix W dynamicl The weight matrix W for dynamic1 type tasks dynamic2 The weight matrix W for dynamic2 type tasks transfer Let b be the weight matrix for transfer-type tasks. static For the bias term corresponding to the static class task, b dynamicl For the bias term corresponding to the dynamic1 class task, b dynamic2 For the bias term corresponding to the dynamic2 type task, b transfer Y is the bias term corresponding to transfer-type tasks. static For the prediction results of static class tasks, Y dynamicl For the prediction results of the dynamic1 type task, Y dynamic2 For the prediction results of dynamic2 type tasks, Y transfer This is the prediction result for transfer-type tasks.

[0080] Furthermore, in step S3, fuzzy set partitioning is performed on all variables, and fuzzification results are output. Based on actual production experience, the variables are divided into 3-fuzzy subset variables and 5-fuzzy subset variables. The 3-fuzzy subset is represented as {L,M,H} = {"low", "normal", "high"}, and the 5-fuzzy subset is represented as {LL,L,M,H,HH} = {"very low", "low", "normal", "high", "very high"}.

[0081] Furthermore, the specific rule triggering process of the multi-module fuzzy control rule base design method in step S4 is as follows: The input variables are determined to be ore feed rate, water feed rate, rotational speed, and large ore block size distribution; the output variables are ore feed adjustment amount, water feed adjustment amount, and rotational speed adjustment amount. Based on the fuzzification result, multi-condition identification result, and condition transition state prediction result, the corresponding rules in the fuzzy rule base are triggered. It is determined whether the current input variable is in an extreme situation or the operating condition is under extreme conditions. If so, the anomaly judgment module rule is triggered; if not, the current mill load state is determined based on the multi-condition identification result, and the rules in the instantaneous load control module are triggered. Then, based on the operating condition trend and condition transition state prediction result, the rules in the trend response control module are triggered. Finally, the triggering rules are sent to the fuzzy controller for fuzzy inference.

[0082] Specific Implementation Cases

[0083] Example 1

[0084] The sample data includes H*N past mill power, return sand amount, feed amount, and large ore block size distribution data, including the current time t, where H is the sampling frequency and N is the time window width.

[0085] Data on mill power, return sand volume, feed rate, and large ore block size distribution over a period of time are obtained and then normalized to ensure dimensional consistency. The normalization formula is as follows:

[0086]

[0087] Where P, R, F, and D are the mill power, return sand amount, feed rate, and large ore block size distribution, respectively. * R * F * D * These represent the mill power, return sand quantity, feed rate, and large ore block size distribution after relative normalization, respectively. opt R opt F opt D opt These represent the optimal operating values ​​for mill power, return sand quantity, feed rate, and large ore size distribution, respectively. max R max F max D max These represent the maximum values ​​of mill power, return sand amount, feed rate, and large ore block size distribution, respectively.

[0088] P is obtained after normalization. * and R *The system calculates a condition score for the current moment to identify multiple operating conditions, representing the current load state of the mill. Through in-depth communication with on-site operators, the boundaries of each operating condition are determined. Based on the condition score results, the operating conditions are divided into 5 categories. The operating condition S is then assigned to a specific category based on the condition score S. t The identification process was performed, and the identified content is shown in the table below.

[0089]

[0090] In this case, assuming that the mill is operating under non-extreme conditions, the overload and underload conditions in the operating condition category are not mill overload or idling in the traditional sense, but rather the load is too high or too low relative to the standard operating condition.

[0091] According to F * and D * Multi-timescale trend analysis of mill load dynamics to identify operating condition trends;

[0092] The first-order finite difference method is used to calculate the short-term trend of mill load and identify the rapid changes in operating conditions over a short period of time. The calculation formula is Δy. t =y t -y t-1 , where Δy t y represents the magnitude of change in data at adjacent time points. t y t-1 The data values ​​at time points t and t-1 are respectively, and Δy is adjusted using a threshold θ1. t The classification is performed using the following formula: where ST t The short-term trend at time t is represented by 0.1, where 0, 1, and 2 represent a downward trend, a steady state, and an upward trend, respectively.

[0093] Next, the medium-term trend (MT) of the mill load was calculated using linear regression slope. t It identifies the stable changing trend of operating conditions within a certain time window. MT t The formula for calculating the intermediate trend of variable y at time t is: Among them MT t y is the regression slope, n is the number of data points, t is the time index, and y is the regression slope. t θ2 is the intermediate trend threshold, set to 0.05. The calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0094] Next, the Z-score method is used to calculate the long-term trend of mill load LT. t Identify long-term trends in operating conditions, LT t The formula for calculating the intermediate trend of variable y at time t is: Z ty represents the degree of deviation of the data at any given time. t Let μ be the data value at time t. t Let σ be the mean of the historical data window up to time t. t θ is the standard deviation of the historical data window before time t, θ3 is the intermediate trend threshold, set to 1.96, and the calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively.

[0095] A multi-scale trend voting method is adopted, which uses weighted voting to calculate short-term, medium-term, and long-term trends, collectively forming the final operating condition trend result. The calculation formula for the individual variable x trend is FT. t =ω s ST t +ω m MT t +ω l LT t FT t Let ω be the operating condition trend score of variable x at time t. s ST t ω m MT t ω l LT t These are the weights of the short-term, medium-term, and long-term trends of variable x, respectively, and ω s +ω m +ω l =1, take ω s =0.3, ω m =0.5, ω l =0.2, the expression for the working condition trend identification result is: in The operating trends of the semi-autogenous mill feed rate and ore particle size distribution at time t are respectively. The operating trends are divided into 3 categories, and the calculation results 0, 1 and 2 represent the downward trend, steady state and upward trend, respectively.

[0096] Identify the working condition transition state based on the multi-working condition identification results, working condition boundaries, and working condition trend identification results;

[0097] Boundary detection is performed on the operating conditions. It is assumed that only when the operating condition score approaches the boundary will the changing trend of the operating condition cause the mill operating condition to cross the boundary and trigger an operating condition transition. The detection formula is as follows: Where δ L δ U S represents the distance from the current operating condition to the lower and upper boundaries, respectively. tL S tU These are the lower and upper boundaries of the current operating condition, respectively.

[0098] Identify the operating condition transition state; if the conditions are met... or And the operating condition is sufficiently close to the lower boundary of the current operating condition, i.e., δ L <ε, where ε is the boundary judgment threshold, and the current working condition S is calculated through S12. t If the value is not 0, the operating condition is considered to be in a descent transition state. or And the operating condition is sufficiently close to the upper boundary of the current operating condition, i.e., δ U If the condition is less than ε and the current operating condition St is not operating condition 4, then the operating condition is considered to be in an upward transition state. If neither of the above two conditions is met, then the operating condition is determined not to have transitioned and is defined as a non-transition state, and the operating condition transition state category is C. t The formula is defined as follows Where 0, 1, and 2 represent the descending transition state, the non-transition state, and the ascending transition state, respectively, and ε is the boundary judgment threshold, set to 0.1;

[0099] 8. An improved Transformer model based on multi-scale feature extraction is used to perform single-step classification and prediction of multi-level operating conditions. Specifically, the improved Transformer model includes a UnetTSF feature extraction module, a Transformer encoder module, and an output fully connected module connected in sequence. The UnetTSF feature extraction module is used to extract multi-scale features of the data, the Transformer encoder module is used for global time series modeling, and the output fully connected module is used to classify and predict mill load conditions (static class), large ore proportion trend (dynamic1 class), mill feed trend (dynamic2 class), and mill operating condition transition state (transfer class).

[0100] 9.

[0101] Where X∈R B×T×C The input is the original time series data, where B is the number of sample batches, T is the sequence length, and C is the feature dimension at each time step. After multiple experiments, B = 256, T = 15, C = 12, and H... UnetTSF H is the multi-scale feature extracted by UnetTSF. Trans The global temporal features extracted for the Transformer are taken as the final feature input to the classifier, since the Transformer's output is a time series feature. W is the feature extracted from the last time step of the Transformer's output. static For static class tasks, the weight matrix W dynamicl The weight matrix W for dynamic1 type tasks dynamic2 The weight matrix W for dynamic2 type tasks transfer Let b be the weight matrix for transfer-type tasks.static For the bias term corresponding to the static class task, b dynamicl For the bias term corresponding to the dynamic1 class task, b dynamic2 For the bias term corresponding to the dynamic2 type task, b transfer Y is the bias term corresponding to transfer-type tasks. static For the prediction results of static class tasks, Y dynamicl For the prediction results of dynamic1 type tasks, Y dynamic2 For the prediction results of dynamic2 type tasks, Y transfer This is the prediction result for transfer-type tasks.

[0102] 10. Obtain the mill power, return sand amount, feed rate, large ore size distribution, water supply, and rotation speed data at the current time t, and perform fuzzification processing. Record the current time as t1.

[0103] The input variables are defined as ore feed rate, water feed rate, rotation speed, and large ore size distribution; the output variables are ore feed adjustment, water feed adjustment, and rotation speed adjustment; and the target variables are power and return sand volume, expressed in the form of operating conditions. Furthermore, fuzzy set partitioning is performed on the input and output variables. Based on actual production experience, the variables are divided into 3-fuzzy subset variables and 5-fuzzy subset variables. The large ore size distribution, rotation speed, and rotation speed adjustment are 3-fuzzy subset variables, while the ore feed rate, water feed rate, ore feed adjustment, and water feed adjustment are 5-fuzzy subset variables. The 3-fuzzy subset is represented as {L,M,H} = {"low", "normal", "high"}, and the 5-fuzzy subset is represented as {LL,L,M,H,HH} = {"very low", "low", "normal", "high", "very high"}.

[0104] Based on the actual situation on site, the maximum and minimum ranges of the variables required for control will be determined. The five levels will be represented by LL, L, M, H, and HH, and the three levels will be represented by L, M, and H. The feed rate and water supply will be divided into five levels, the rotation speed and the distribution of large ore pieces into three levels, the feed rate and water supply output by the controller into five levels, and the rotation speed into three levels.

[0105] The corresponding rules in the fuzzy rule base are triggered based on the fuzzification results, multi-condition recognition results, and the prediction results of condition trends and condition transition states.

[0106] A multi-module fuzzy control rule base design method is adopted, including an anomaly detection module, an instantaneous load control module, and a trend response control module. The specific rule triggering process is shown in the appendix. Figure 5 As shown, it includes the following steps:

[0107] Determine whether the current input variable is in an extreme situation or the operating condition is in an extreme condition; if so, trigger the exception judgment module rules and proceed to the final step; otherwise, proceed to the next step.

[0108] The current mill load status is determined based on the multi-condition identification results, and the rules within the real-time load control module are triggered.

[0109] Rules within the trend response control module are triggered based on the predicted results of operating condition trends and operating condition transition states.

[0110] Determine whether the rules within the real-time load control module or trend response control module have been triggered.

[0111] The triggering rules are sent to the fuzzy controller for fuzzy inference;

[0112] The anomaly detection module adjusts the input variable significantly by determining whether the current input variable is in an extreme state and whether the current operating condition is in an extreme state. This prevents the equipment from exceeding its safe operating range and improves system safety. When a rule in the anomaly detection module is triggered, rules in subsequent modules will not be triggered, preventing rules from other modules from affecting the control results. If the anomaly detection module has no executable rules, it will proceed to subsequent modules to search for executable rules. Specific rules are as follows:

[0113] IF(semi-autogenous mill feed rate is LL) Then(feed rate adjustment is HH)

[0114] IF (semi-autogenous mill feedwater is LL) Then (feedwater adjustment is HH)

[0115] IF (semi-autogenous mill speed is LL) Then (speed adjustment amount is HH)

[0116] IF(semi-autogenous mill feed rate is HH) Then(feed rate adjustment is LL)

[0117] IF (semi-autogenous mill feedwater is HH) Then (feedwater adjustment is LL)

[0118] The real-time load control module identifies multiple operating conditions based on the mill load characteristics, determines the current mill load level, and sets independent fuzzy control rules according to the control requirements of different load levels to avoid control conflicts. When S... t =1, meaning a relatively low load, the rule follows the principle of first adjusting the feed rate, then the water feed rate, and finally the rotational speed. When S t =3 is similar. Some rules are as follows:

[0119] When S t =1:

[0120] IF (feed rate is L) and (rotation speed is M or H) Then (feed rate adjustment is H)

[0121] IF (Feed rate is M) and (Water feed rate is M or H) and (Rotation speed is M or H) Then (Water feed rate adjustment is H)

[0122] IF (Feed rate is M) and (Water feed rate is L) and (Rotation speed is M or H) Then (Rotation speed adjustment is L)

[0123] The trend response control module sets rules based on multi-level operating condition prediction results. First, it divides the rules into an upward transition rule module and a downward transition rule module based on the predicted operating condition transition state. Then, within each module, the cause of the operating condition transition state is determined based on the predicted operating condition trend, thereby further refining the rules. Some rules are as follows:

[0124] When C t+1 =2, S t =2、 hour:

[0125] IF (Feed rate is M or H) and (Water feed rate is L or M) Then (Water feed rate adjustment is H)

[0126] IF (Feed rate is M or H) and (Water feed rate is H) and (Rotation speed is L or M) Then (Rotation speed adjustment is H)

[0127] Based on the triggering rules, fuzzy reasoning is performed, and the set values ​​of ore feed rate, water feed rate and rotation speed are output after defuzzification;

[0128] Specifically, the Mamdani fuzzy inference method is used to calculate the final fuzzy output set, and the centroid method is used as the defuzzification method to obtain the final adjustment amount.

[0129] The set values ​​for feed rate, water flow rate, and rotation speed are sent to the grinding process.

[0130] Determine if the current time t is less than T, where T is the control period; if yes, return to the previous step; otherwise, return to the first step.

[0131] The intelligent control method for the semi-autogenous grinding process proposed in this invention was simulated using DCS data from the grinding circuit of a mineral processing plant. A power model of the semi-autogenous grinding mill was used as the verification model. Data sampling was conducted from October 10th to November 1st, 2024. 120 minutes of actual production data under both non-abnormal and abnormal operating conditions were selected. Using the power of the semi-autogenous grinding mill as an indicator, the power under fuzzy control was compared with the expected power value and the actual power value under manual control during the current time period. The power reflects the control effect. The comparison effect under non-abnormal operating conditions is shown in the figure below. Figure 7 As shown in the table below, under manual control, the power increase rate of the semi-autogenous mill is relatively slow and highly volatile. Under intelligent control, however, the power increase rate is faster, reaching the desired power value in approximately 850s, with an overshoot of about 15kW at 1500s. Finally, the power stabilizes around 4500s, with a steady-state error of approximately 2.25kW. This indicates that the intelligent control method can quickly adjust the mill power and maintain stable mill load when the power is relatively low, i.e., when the mill load is low. Furthermore, the overshoot and steady-state error are within acceptable ranges, therefore, the intelligent control method is considered to have good control performance. Specific simulation comparison results are shown in the table below.

[0132]

[0133] Comparison diagrams under abnormal operating conditions are shown below. Figure 8 As shown in the table, an abnormal power output was observed in the semi-autogenous mill during the current time period. Based on actual production conditions, the main cause of this power fluctuation was the rapid increase in the size distribution of large ore blocks, leading to a rapid increase in mill load within a short period. To prevent overloading, the operator significantly reduced the feed rate, resulting in a rapid drop in semi-autogenous mill power output around 3000s and 4000s. Under intelligent control, although the power output experienced a brief increase due to changes in block size distribution, it gradually stabilized after approximately 1000s. Finally, the power output reached the expected level for the current time period around 1300s and remained stable over a long period. This demonstrates that the intelligent control method achieves good control performance even under abnormal operating conditions. Specific simulation comparison results are shown in the table below.

[0134]

[0135] Based on the above analysis, it can be seen that the intelligent control method for semi-autogenous grinding processes based on multi-level working condition identification and prediction proposed in this invention can ensure the stability of the grinding process, increase ore throughput, and improve grinding efficiency.

[0136] All standard parts used in this application can be purchased from the market. The specific connection methods of each part adopt conventional methods such as bolts, rivets, and welding that are mature in the prior art. The machinery, parts and equipment adopt conventional models in the prior art and are also general components, which are common knowledge in this field.

[0137] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A semi-autogenous grinding process intelligent control method based on multi-level working condition identification and prediction, characterized in that, Includes the following steps: S1, Obtain sample data of the grinding process within a preset time range; S2, Based on the multi-level working condition identification and prediction method, the working condition identification and prediction of the working condition transition state are performed according to the sample data; S3, acquire the data required for controlling the grinding process at the current moment, and perform fuzzification processing on the data to obtain the fuzzification result, and denote the current moment as t1; S4, trigger the corresponding rule in the fuzzy rule base based on the fuzzification result, multi-condition recognition result, condition trend and condition transition state prediction result; S5, based on the triggered rules, performs fuzzy reasoning to obtain the ore feed rate setpoint, water feed rate setpoint, and rotation speed setpoint; S6, send the set values ​​of feed rate, water flow rate and rotation speed to the grinding process, and record the current time as t2; S7, calculate the relationship between Δt = t2 - t1 and T, where T is the control period. If Δt > T, return to S6; otherwise, return to S1.

2. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 1, characterized in that: The sample data of S1 includes industrial sensor data, production data, and electromechanical equipment operation data that can be used to evaluate on-site working conditions within a preset time range.

3. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 1, characterized in that: In step S2, a multi-level operating condition identification and prediction method is used to provide the fuzzy controller with operating condition identification and prediction results. The steps are as follows: S21. The obtained mill power (P), return sand amount (R), feed rate (F), and large ore block size (D) distribution are relatively normalized to ensure dimensional uniformity. The normalization formula is as follows: Where Y represents the variable that needs to be normalized, Y opt Y represents the optimal value of this variable. max Y represents the maximum value of the variable within that time period. * This represents the result after normalization of variable Y; S22, based on the normalized mill power P * and return sand amount R * Calculate the current working condition score S to perform multi-working condition identification, and obtain the current working condition identification result S. t ; S23, based on the normalized ore feed rate F * and large ore blocks of size D * Multi-timescale trend analysis of mill load dynamics was used to identify operating condition trends and obtain the operating condition trends of mill feed rate and large ore size distribution at time t. S24, based on the multi-condition identification result S t Operating condition boundary S tL S tU and operating condition trend identification results Identify the working condition transition state C t ; S25 uses an improved Transformer model based on multi-scale feature extraction to perform single-step classification prediction of multi-level working conditions, and obtains the prediction results of working condition transition states.

4. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 3, characterized in that: The weighted scoring formula for the working condition in S22 is: S=αR * +βP * Where S is the operating condition score, and α and β are R... * P * The corresponding weighting factor, and α+β=1, α≤β, are used to divide the working conditions into N categories based on the actual value of S and the actual production situation, and a number is used to represent a category of working conditions.

5. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 3, characterized in that: The specific method for identifying operating condition trends in S23 is as follows: S231. The grinding process is characterized by nonlinearity, strong noise, and slow time variation. The short-term trend mainly reflects the time-varying characteristics, the long-term trend reflects the drift of the operating conditions, and the medium-term trend combines the advantages of both. In order to accurately judge the changing trend of the mill operating conditions, the multi-scale time trend analysis method is used for operating condition identification. S232, using the first-order difference method to calculate the short-term trend of mill load ST t It identifies rapid changes in operating conditions within a short period of time. ST t The short-term trend of variable y at time t is represented by the formula: Δy t =y t -y t-1 , where Δy t y represents the magnitude of change in data at adjacent time points. t y t-1 Here, t and t-1 are the data values ​​at time t and t-1, respectively, and θ1 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively. S233, then the mid-term trend of mill load MT is calculated using the linear regression slope. t It identifies the stable changing trend of operating conditions within a certain time window. MT t The formula for calculating the intermediate trend of variable y at time t is: Among them MT t y is the regression slope, n is the number of data points, t is the time index, and y is the regression slope. t θ2 is the intermediate trend threshold, and the calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively. S234, then the Z-score method is used to calculate the long-term trend of mill load LT. t Identify long-term trends in operating conditions, LT t The long-term trend of variable y at time t is calculated using the following formula: Z t y represents the degree of deviation of the data at time t. t Let μ be the data value at time t. t Let σ be the mean of the historical data window up to time t. t θ3 is the standard deviation of the historical data window before time t, and θ3 is the intermediate trend threshold. The calculation results 0, 1, and 2 represent the downward trend, steady state, and upward trend, respectively. S235 employs a multi-scale trend voting method, using weighted voting to calculate short-term, medium-term, and long-term trends, collectively forming the final operating condition trend result. The calculation formula for the individual variable x-trend is FT. t =ω s ST t +ω m MT t +ω l LT t FT t Let ω be the operating condition trend score of variable x at time t. s ST t ω m MT t ω l LT t These are the weights of the short-term, medium-term, and long-term trends of variable x, respectively, and ω s +ω m +ω l =1. S236. Since the feed rate (F) and the distribution of large ore blocks (D) directly affect the mill load and grinding efficiency, the feed rate is obtained through the multi-scale trend calculation formula in S235. and large ore blocks The final operating condition trend calculation expression for the distribution is as follows: The operating conditions are divided into three categories, with the calculation results of 0, 1, and 2 representing a downward trend, a steady state, and an upward trend, respectively.

6. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 3, characterized in that: The operating condition boundaries in S24 need to be defined separately for each operating condition based on the actual conditions of the industrial site. tU With lower boundary S tL .

7. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 3, characterized in that: The specific steps for identifying the operating condition transition state in S24 are as follows: S241 performs boundary checks on the operating conditions, assuming that changes in the operating conditions are only likely to cause the mill operating conditions to cross the boundary and trigger a condition transition when the operating condition score is close to the boundary. The detection formula is as follows: Where δ L δ U S represents the distance from the current operating condition to the lower and upper boundaries, respectively. tL S tU These are the lower and upper boundaries of the current operating condition, respectively. S242, identify the working condition transition state, if the condition is satisfied... or And the operating condition is sufficiently close to the lower boundary of the current operating condition, i.e., δ L <ε, where ε is the boundary judgment threshold, and the current working condition S is calculated through S22. t If the value is not 0, the operating condition is considered to be in a descent transition state. or And the operating condition is sufficiently close to the upper boundary of the current operating condition, i.e., δ U If the condition is less than ε and the current operating condition St is not operating condition 4, then the operating condition is considered to be in an upward transition state. If neither of the above two conditions is met, then the operating condition is determined not to have transitioned and is defined as a non-transition state, and the operating condition transition state category is C. t The formula is defined as follows Where 0, 1, and 2 represent the descending transition state, the non-transition state, and the ascending transition state, respectively, and ε is the boundary judgment threshold.

8. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 2, characterized in that: The improved Transformer model in S25 includes a UnetTSF feature extraction module, a Transformer encoder module, and an output fully connected module connected in sequence. The UnetTSF feature extraction module is used to extract multi-scale features of the data, the Transformer encoder module is used for global temporal modeling, and the output fully connected module is used to classify and predict mill load conditions (static class), large ore proportion trends (dynamic1 class), mill feed trends (dynamic2 class), and mill condition transition states (transfer class). Where X∈R B×T×C The input is the original time series data, where B is the number of sample batches, T is the sequence length, C is the feature dimension at each time step, and H is the number of time series data. UnetTSF H is the multi-scale feature extracted by UnetTSF. Trans The global temporal features extracted for the Transformer are taken as the final feature input to the classifier, since the Transformer's output is a time series feature. W is the feature extracted from the last time step of the Transformer's output. static For static class tasks, the weight matrix W dynamicl The weight matrix W for dynamic1 type tasks dynamic2 The weight matrix W for dynamic2 type tasks transfer Let b be the weight matrix for transfer-type tasks. static For the bias term corresponding to the static class task, b dynamicl For the bias term corresponding to the dynamic1 class task, b dynamic2 For the bias term corresponding to the dynamic2 type task, b transfer Y is the bias term corresponding to transfer-type tasks. static For the prediction results of static class tasks, Y dynamicl For the prediction results of the dynamic1 type task, Y dynamic2 For the prediction results of dynamic2 type tasks, Y transfer This is the prediction result for transfer-type tasks.

9. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 1, characterized in that: In step S3, fuzzy sets are partitioned for all variables, and the fuzzification result is output. Based on actual production experience, the variables are divided into 3-fuzzy subset variables and 5-fuzzy subset variables. The 3-fuzzy subset is represented as {L,M,H} = {"low", "normal", "high"}, and the 5-fuzzy subset is represented as {LL,L,M,H,HH} = {"very low", "low", "normal", "high", "very high"}.

10. The intelligent control method for a semi-autogenous grinding process based on multi-level working condition identification and prediction according to claim 1, characterized in that: The specific rule triggering process of the multi-module fuzzy control rule base design method in step S4 is as follows: Determine the input and output variables, trigger the corresponding rules in the fuzzy rule base according to the fuzzification results, multi-condition identification results, and condition transition state prediction results, and determine whether the current input variable is in an extreme situation or the condition is under extreme conditions; if so, trigger the rules of the anomaly judgment module; if not, determine the current mill load state according to the multi-condition identification results and trigger the rules in the instantaneous load control module. Then, trigger the rules in the trend response control module according to the condition trend and condition transition state prediction results. Finally, send the triggering rules to the fuzzy controller for fuzzy inference.

Images