LSTM-GPC-based targeted regulation and control method for air passing quantity of mine louvered air window

Abstract

The invention relates to an LSTM-GPC-based mine louvered air window air passing quantity targeted regulation and control method, and belongs to the technical field of mine ventilation automatic control. In order to solve the problems that traditional air window manual adjustment is low in efficiency and poor in precision, and PID control is difficult to adapt to a nonlinear system, a solution integrating LSTM dynamic modeling and nonlinear generalized predictive control is provided, a predictive model is constructed by identifying an air window angle-air volume nonlinear relation on line, and the optimal angle adjustment amount is generated in combination with a generalized predictive algorithm; dynamic compensation is introduced to eliminate environmental interference; and a closed-loop feedback mechanism is adopted to correct model parameters in real time until the regulated air volume meets the error threshold requirement. According to the method, the angle prediction time of the louvered air window can be greatly shortened, the anti-interference capability is improved, the air passing amount of the louvered air window is accurately and rapidly regulated and controlled, and the mine ventilation safety and stability are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of mine ventilation automation control, and relates to a method for targeted regulation of the air volume passing through a louvered air window in a mine based on LSTM-GPC. Background Technique

[0002] The mine ventilation system is a key link to ensure underground safety production. Its core task is to ensure sufficient oxygen, controllable harmful gas concentration, and maintain a stable climate environment in the roadway by reasonably distributing the air volume in each area. As a key regulating device in the mine ventilation network, the opening of the air window directly affects the air resistance of the roadway where it is located and the air volume distribution of the entire network. The traditional louvered air window adjusts the air passing area by adjusting the blade angle, thereby realizing air volume regulation, and is the main actuator for dynamic air volume balance underground.

[0003] Currently, most mines in China still rely on manual adjustment of the blade angle of the air window. Operators need to estimate the target opening based on experience and gradually approach the target air volume through multiple cycles of "adjustment - measurement - feedback". This process has significant defects:

[0004] (1) Each time the air volume is adjusted, it takes dozens of minutes to meet the requirements through multiple measurements and repeated adjustments, which is difficult to meet the real-time demand for air volume during the production process change of the mining and excavation working face.

[0005] (2) Manual experience has subjective biases. Especially in complex ventilation networks, the non-linear relationship between the blade angle and the air volume results in no clear proportional relationship between the opening adjustment amount and the air volume change, and it is easy to have repeated oscillating adjustments.

[0006] (3) Improper adjustment of the blade angle of the air window is likely to cause air flow disorder, resulting in accidents such as insufficient air volume and gas overrun in the working face, seriously affecting the safety production of the mine.

[0007] In recent years, scholars have proposed a mutual feedback iterative regulation method based on "air window regulation + air volume monitoring" or "air window regulation + differential pressure back-calculation of air volume", which automatically adjusts the blade angle by real-time feedback of the air volume error through sensors. However, such methods still have the following technical bottlenecks:

[0008] Existing methods only rely on the current error direction (such as increasing or decreasing the opening) for fixed-step adjustment (such as ±1° each time), lacking the modeling of the system's dynamic characteristics, resulting in a slow convergence speed. Experiments show that in the scenario of sudden change of the target air volume, traditional methods require more than 10 iterations to converge, which cannot meet the emergency regulation requirements.

[0009] Some research attempts to introduce the PID control algorithm, but its linear control logic is difficult to adapt to the strong non-linear characteristics of the windshield system. For example, when the windshield opening is close to the limit position, the sensitivity of the air volume to the angle change drops sharply. Fixed PID parameters will lead to integral saturation or overshoot, and the measured data shows that the steady-state error is as high as 8% - 12%.

[0010] The air pressure fluctuation in the mine, the roadway deformation, and the change of air resistance caused by the start and stop of equipment will disturb the air volume of the windshield. The traditional method lacks an online compensation mechanism for time-varying disturbances, and the regulation stability is poor.

[0011] With the advancement of the intelligent mine construction, the mine ventilation system is developing towards the integration of "perception - prediction - decision - regulation". Although existing research has tried to introduce intelligent algorithms such as neural networks and fuzzy control into the air volume prediction, there are still key challenges:

[0012] The prediction error of a single LSTM or ARIMA model increases sharply under working conditions outside the training data range (such as extreme opening and high wind speed), and it is not deeply integrated with the control algorithm, so closed-loop optimization cannot be achieved.

[0013] High-precision non-linear models (such as deep reinforcement learning) require a large amount of computing resources and are difficult to be deployed in mine embedded controllers, resulting in regulation delay.

[0014] In summary, the existing windshield regulation methods have significant defects in terms of response speed, control accuracy, and anti-interference ability. There is an urgent need for an intelligent regulation method that can online identify the non-linear dynamic characteristics of the windshield, combine multi-step prediction with adaptive compensation, so as to improve the reliability and regulation efficiency of the mine ventilation system. Summary of the Invention

[0015] In view of this, the purpose of the present invention is to provide a targeted regulation method for the air volume passing through the mine shutter windshield based on LSTM-GPC. This method uses an intelligent modeling and identification method to identify the non-linear relationship between the air volume passing through the windshield and the shutter angle, and uses the data of air volume - shutter angle at historical moments to establish a dynamic prediction model of the air volume passing through the windshield - shutter angle, accurately capturing the dynamic characteristics of the air volume change, so as to realize the real-time dynamic regulation of the air volume; by integrating the non-linear generalized predictive control algorithm, the system can adaptively adjust the control strategy according to the prediction result and the actual air volume deviation, generate the optimal adjustment amount of the blade angle, and then improve the control accuracy and robustness.

[0016] To achieve the above purpose, the present invention provides the following technical solutions:

[0017] A targeted regulation method for the air volume passing through the mine shutter windshield based on LSTM-GPC, the method comprising the following steps:

[0018] S1: Set the target air volume of the windshield to Q0 and the air volume error threshold to δ0. Generally, the value of the air volume error threshold δ0 should be less than 5%.

[0019] S2: Obtain the initial data of the blade angle and the air volume passing through at the initial moments k = 1, 2, 3, namely [θ1, Q1], [θ2, Q2], [θ3, Q3]. The specific steps are as follows:

[0020] S21: Record the windshield blade angle and the air volume passing through at the initial moment k = 1 as [θ1, Q1], and calculate the relative error δ1 between the actual air volume and the target air volume at k = 1. When δ1 ≤ δ0, there is no need to adjust the blade angle, and the air volume regulation ends, where:

[0021] S22: When δ1 > δ0, adjust the windshield blade angle at k = 2 to θ2, record the current windshield blade angle and the air volume passing through as [θ2, Q2], and calculate the relative error δ2 between the actual air volume and the target air volume at k = 2. When δ2 ≤ δ0, there is no need to adjust the blade angle, and the air volume regulation ends; where: θ2 = θ1 + Δθ2, Δθ2 = sign(Q0 - Q1) * 1°,

[0022] S23: When δ2 > δ0, adjust the windshield blade angle at k = 3 to θ3, record the current windshield blade angle and the air volume passing through as [θ3, Q3], and calculate the relative error δ3 between the actual air volume and the target air volume at k = 3. When δ3 ≤ δ0, there is no need to adjust the blade angle, and the air volume regulation ends; otherwise, perform S3. Where: θ3 = θ2 + Δθ3, Δθ3 = sign(Q0 - Q2) * 1°,

[0023] S3: Based on the initial data, establish a dynamic prediction model of the air volume passing through the windshield - blade angle Q k = f(Q k-1 , Q k-2 , θ k , θ k-1 , θ k-2 ), k ≥ 3;

[0024] S4: Combine the nonlinear generalized predictive control algorithm to generate the optimal adjustment amount θ k of the blade angle at k moment, adjust the blade angle to θ k , measure the air volume Q k corresponding to the angle θ k at k moment;

[0025] S5: When When the air volume regulation ends; otherwise, repeat S4 until the air volume regulation ends. The prediction model of the air window blade angle and the air volume passing through is expressed as:

[0026] Q k+1 = f(Q k ,Q k-1 ,θ k+1 ,θ k ,θ k-1 )

[0027] In the formula: k+1, k,... are sampling times; Q k+1 is the predicted air volume at the (k+1)-th moment; θ k+1 is the blade angle at the (k+1)-th moment; Q k ,Q k-1 ,θ k ,θ k-1 are the air volume and blade angle at historical moments; f(·) ∈ R is an unknown non-linear function. Expand the model near the operating point into the form of the sum of a low-order linear model and a non-linear term:

[0028]

[0029] In the formula: α(k) = [a1, a2, b0, b1, b2], a i (i = 1, 2) and b j (j = 0, 1, 2) are the first-order Taylor coefficients of the air window air volume regulation system at the operating point; is the non-linear term and is bounded.

[0030] The establishment process of the prediction model of the air window blade angle and the air volume passing through is as follows:

[0031] 1) Use the least squares method with forgetting factor (FFRLS) to identify and obtain the estimated value of the linear parameter α(k) which can be recursively obtained through the following formula:

[0032]

[0033] Among them: λ is the forgetting factor, 0 < λ ≤ 1; K(k) is the gain matrix of the prediction error; P(k) is the covariance matrix; I is the identity matrix of the same dimension as P.

[0034] 2) Calculate the estimated output of the linear part from , calculate including the linear modeling error and the non-linear part, and let x k = [Q k ,Q k-1 ,θ k+1 ,θ k ,θk-1 T As the input of the LSTM, is the output, and the estimated value of the non - linear part is trained and generated accordingly.

[0035] Compare with When (where ε is the minimum error), repeat steps 1) and 2) until the error meets the requirements, and the prediction model is established.

[0036] The process of generating the optimal blade angle adjustment amount θ(k) at time k by the non - linear generalized predictive control algorithm is as follows:

[0037] 1) Represent the prediction model obtained by intelligent modeling with a polynomial containing the time - delay operator z -1 , that is: where: A(z -1 ) = 1 + a1z -1 + a2z -2 , B(z -1 ) = b0 + b1z -1 + b2z -2 , A(z -1 ) and B(z -1 ) are the coefficient polynomials of the low - order linear model, including unknown non - linear terms such as modeling errors and external disturbances;

[0038] 2) Design a generalized predictive controller based on the low - order linear model to generate the blade angle adjustment amount θ1(k), and define the performance index: Q0 is the target air volume, Np is the prediction time - domain length, Nu is the control time - domain length. By minimizing the air volume tracking error and the blade angle fluctuation within the prediction time - domain, let we can obtain the linear generalized predictive controller H c (z -1 )Δθ1(k) = P T Q0 - F c (z -1 )Q(k), where, H and G are obtained through the Diophantine equations 1 = E k (z -1 )ΔA(z -1 )+ z -k F k (z -1 ) and E k (z -1 )B(z -1 ) = G k (z -1 )+ z -k H​k (z -1 )Solve.

[0039] 3) According to the tracking error of the linear generalized predictive control closed-loop system \(e(k + 1)=Q_0(k + 1)-Q(k + 1)\), design the dynamic compensation angle value \(\theta_2(k)\) to eliminate the influence of the non-linear term on the louvered air window air regulation system, introduce \(J_2 = e(k + 1)\) 2 , let The non-linear generalized predictive control law equation based on LSTM is obtained as: where: M c (z -1 ) is solved through the Diophantine equation \((1 - z -1 )A(z -1 )H c (z -1 )+z -1 B(z -1 )F c (z -1 )+z -1 M c (z -1 ) = 1.

[0040] 4) Add the above linear generalized predictive control quantity \(\theta_1(k)\) and the dynamic compensation angle value \(\theta_2(k)\) to generate the optimal blade angle adjustment quantity \(\theta(k)=\theta_1(k)+\theta_2(k)\) at time \(k\).

[0041] The beneficial effects of the present invention are as follows:

[0042] (1) By integrating the LSTM network and the dynamic parameter identification technology, a non-linear prediction model of the air volume passing through the air window and the blade angle is constructed, which can online capture the dynamic influence of complex underground working conditions (such as air pressure fluctuations, roadway deformation) on the air volume-angle relationship, and solve the limitation of the traditional method relying on a fixed parameter model. Experiments show that the model prediction error can be controlled within 3%, which is significantly lower than the steady-state error (8% - 12%) of the traditional PID control.

[0043] Based on the generalized predictive control (GPC) algorithm, combined with the air volume tracking error and angle adjustment within the multi-step prediction time domain, an optimal control sequence is generated to achieve the leap from "trial-and-error approximation" to "one-step prediction precise regulation". In practical applications, the number of convergence iterations in the scenario of sudden change of the target air volume is reduced from more than 10 times of the traditional method to 3 - 5 times, and the regulation efficiency is increased by more than 50%.

[0044] (2) Adopt a two - layer control architecture that combines linear GPC control and dynamic compensation. By decomposing the system model into low - order linear terms and non - linear perturbation terms, the main controller and compensator are designed respectively, effectively overcoming the mismatch problem of the traditional PID algorithm under the strong non - linear characteristics of the windshield. For example, when the windshield opening is close to the limit position (sensitivity - decreasing area), the dynamic compensator can adaptively adjust the angle increment to avoid overshoot or oscillation caused by integral saturation.

[0045] Introduce an online parameter identification with forgetting factor (FFRLS) and an LSTM real - time compensation mechanism, which can quickly respond to underground environmental disturbances (such as the start - stop of local ventilators and sudden changes in air resistance caused by gas outbursts). The control error can be restored within the threshold within 2 - 3 sampling periods after the disturbance occurs, and the anti - interference ability of the system is improved by more than 40%.

[0046] (3) Through the error feedback of real - time collection of the over - air volume data and the model prediction value, form a closed - loop control loop of "perception - prediction - decision - regulation", dynamically correct the model parameters and control strategies, and avoid the long - term cumulative error caused by model drift in traditional open - loop control. Long - term operation tests show that the system can still maintain the stability of the error threshold (<5%) after continuous operation for 72 hours.

[0047] Combined with an intelligent judgment mechanism of the air volume error threshold (δ0), automatically terminate the adjustment after reaching the target accuracy, avoid energy waste and equipment wear caused by over - adjustment, and at the same time support manual setting of the threshold to meet the requirements of different working conditions (such as δ0 can be relaxed to 8% during emergency ventilation).

[0048] (4) The introduction of the LSTM network enables the model to have the ability to deeply extract features from historical data, adapt to the ventilation network characteristics of different mines (such as roadway topology, fan configuration), without relying on a large number of prior parameter calibrations, and the deployment period is shortened by 60%.

[0049] The system supports modular expansion. By adding sensor data (such as temperature, humidity, gas concentration), the prediction model can be further optimized, providing a technical basis for the multi - objective collaborative control (such as energy - efficiency optimization, disaster warning) of future intelligent ventilation systems.

[0050] Other advantages, objectives and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. Brief Description of the Drawings

[0051] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0052] Figure 1 The control structure diagram of the louvered air window proposed by the present invention;

[0053] Figure 2 The establishment process of the air volume-louver angle prediction model for the air window proposed by the present invention;

[0054] Figure 3 The specific example diagram of the adaptive control of the air volume passing through the air window provided by the present invention. Detailed implementation manners

[0055] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0056] Among them, the drawings are only used for exemplary illustration, showing only schematic diagrams, not physical diagrams, and cannot be understood as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the drawings will be omitted, enlarged or reduced, and do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0057] In the drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the drawings are only used for exemplary illustration and cannot be understood as a limitation to the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0058] Please refer to Figures 1 to 3 , a method for targeted regulation of the air volume passing through the louvered air window in a mine based on LSTM-GPC, characterized in that: the implementation of this method includes the following steps:

[0059] S1: Set the target air volume of the air window as Q0 = 665m 3 / min, and the air volume error threshold as δ0 = 2%;

[0060] S2: Determine the initial values [θ1, Q1], [θ2, Q2], [θ3, Q3] of the prediction model for the windshield blade angle and the air flow rate. The specific steps are as follows:

[0061] S21: Record the windshield blade angle and the air flow rate at the initial moment k = 1 as [θ1, Q1] = [20°, 238.02m 3 / min]; Calculate

[0062] S22: Since δ1 > δ0, adjust the windshield blade angle at the moment k = 2 to θ2 = 21°, and record the current windshield blade angle and the air flow rate as [θ2, Q2] = [21°, 243.52m 3 / min], and calculate the relative error between the actual air flow rate and the target air flow rate at the moment k = 2 where: θ2 = θ1 + Δθ2, Δθ2 = 1°;

[0063] S22: Since δ2 > δ0, adjust the windshield blade angle at the moment k = 3 to θ3 = 22°, and record the current windshield blade angle and the air flow rate as [θ3, Q3] = [22°, 249.04m 3 / min], and calculate the relative error between the actual air flow rate and the target air flow rate at the moment k = 3 where: θ3 = θ2 + Δθ3, Δθ3 = 1°;

[0064] S3: According to the initial values [θ1, Q1], [θ2, Q2], [θ3, Q3] of the prediction model for the windshield blade angle and the air flow rate determined in step S2, as Figure 2 shown: Use the intelligent modeling method to online identify and establish the prediction model for the louvered windshield blade angle and the air flow rate Q3 = 0.5171Q2 + 0.5054Q1 + 0.0467θ3 + 0.0446θ2 + 0.0425θ1 + 0.0065;

[0065] S4: Let k = k + 1. From S3, the optimal adjustment amount θ of the blade angle at the moment k can be obtained by combining the prediction model for the windshield blade angle and the air flow rate with the nonlinear generalized predictive control algorithm k = 31°, adjust the blade angle to θ k , measure the air flow rate Q corresponding to the angle at the moment k k = 302.49m 3 / min;

[0066] S5: Repeat step S4 until the air flow rate regulation ends.

[0067] As Figure 3As shown: The air volume is adjusted 7 times in total to meet the requirement that δ7 = 1.2% < δ0, and the specific regulation process is shown in Table 1.

[0068] Table 1

[0069] Moment k Angle (°) <![CDATA[Air volume (m 3 / min)]]> Air volume error 1 20 238.02 64.3％ 2 21 243.52 63.4％ 3 22 249.05 62.6％ 4 31 302.49 54.6％ 5 73 564.77 15.2％ 6 74 571.58 14.2％ 7 89 673.89 1.2％

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for targeted regulation of the air volume passing through the shutter-type air window in a mine based on LSTM-GPC, characterized in that: Including the following Steps: S1: Set the target air volume Q0 and the air volume error threshold δ0, where δ0 is less than 5%; S2: Obtain the initial data of blade angles and air flow rates at the initial moments \(k = 1, 2, 3\), i.e., \([\theta_1, Q_1]\), \([\theta_2, Q_2]\), \([\theta_3, Q_3]\). If the actual air flow rate error \(\delta\) at any moment k \(\leq\delta_0\), then end the regulation; otherwise, execute S3; S3: Based on the initial data, establish a dynamic prediction model Q of the air volume passing through the windshield - blade angle using an intelligent modeling method k = f(Q k-1 , Q k-2 , θ k , θ k-1 , θ k-2 ), where k ≥ 3; S4: Combine the non - linear generalized predictive control algorithm to generate the optimal blade angle adjustment amount θ at the current moment k , and measure the actual air flow rate Q k ; S5: If the current air volume error δ k ≤ δ0, then end the regulation; otherwise, repeat S4 until the error requirement is met.

2. The method for targeted regulation of air volume passing through a mine louvered air window based on LSTM-GPC according to claim 1, characterized in that: The adjustment rule of the initial blade angle in S2 is: θ2 = θ1 + sign(Q0 - Q1)×1°, where sign(Q0 - Q1) represents the air volume deviation direction; θ3 = θ2 + sign(Q0 - Q2)×1°.

3. The method for targeted regulation of the air volume passing through the mine louvered air window based on LSTM-GPC according to claim 1, characterized in that: The establishment of the air volume - blade angle dynamic prediction model of the air window includes: Decompose the windshield air intake - blade angle dynamic prediction model into low-order linear terms and non-linear terms where α(k) is a linear parameter; Online identification of the linear parameter α(k) is carried out using the least squares method with forgetting factor FFRLS, and the non-linear term is trained through the LSTM network 4. The method for targeted regulation of the air volume passing through the mine louvered air window based on LSTM-GPC according to claim 3, characterized in that: The parameter recurrence formula of the FFRLS is: Among them, λ is the forgetting factor, 0 < λ ≤ 1.

5. The method for targeted regulation of the air volume passing through the mine louvered air window based on LSTM-GPC according to claim 3, characterized in that: The input of the LSTM network is x k = [Q k , Q k-1 , θ k+1 , θ k , θ k-1 , and the output is the estimated value of the non - linear term And through iterative training, make 6. The method for targeted regulation of air volume passing through mine louvered air windows based on LSTM-GPC according to claim 1, wherein: The non - linear generalized predictive control algorithm includes: Design a generalized predictive controller based on a low - order linear model to generate a linear control quantity θ1(k); Design a dynamic compensation quantity θ2(k) according to the tracking error e((k + 1) = Q0(k + 1) - Q((k + 1) to offset the influence of non - linear terms; Superimpose θ1(k) and θ2(k) to generate the optimal adjustment quantity θ((k) = θ1(k) + θ2((k).

7. The method for targeted regulation of the air volume passing through the mine louvered air window based on LSTM-GPC according to claim 5, characterized in that: The linear control quantity θ1(k) is obtained by minimizing the performance index J1: where, N p is the prediction horizon, N u is the control horizon, and λ1 is the weight coefficient.

8. The method for targeted regulation of the air volume passing through the mine louvered air window based on LSTM-GPC according to claim 5, characterized in that: The dynamic compensation quantity θ2(k) is obtained by solving the equation: Among them, M c (z -1 ) is solved by a Diophantine equation.

9. The method for targeted regulation of air volume passing through a mine louvered air window based on LSTM-GPC according to claim 1, wherein: The air volume error is calculated as: And δ is updated in real time during the regulation process k until δ k ≤ δ0 is satisfied.

10. The method for targeted regulation of air volume passing through the mine louvered air window based on LSTM-GPC according to claim 1, characterized in that: The method dynamically optimizes the blade angle through a closed - loop control mechanism, specifically including: Real-time collection of the over-air volume data Q k And compare it with the predicted value; Adjust the prediction model parameters and control quantities according to the error feedback until the actual air volume converges to the target value.

Images