Dynamic intelligent prediction and early warning method and system for stope roof

Abstract

The invention discloses a dynamic intelligent prediction and early warning method and system for a stope roof. The early warning method comprises the steps that pressure data of a hydraulic support in a plurality of historical weighting periods are obtained; processing the pressure data in the historical weighting period to obtain an effective data set; according to the effective data set, calculation processing is carried out through a mathematical statistics method, and a first prediction result of the next weighting propulsion degree is obtained; acquiring real-time pressure data of the hydraulic support; the real-time pressure data of the hydraulic support are input into a prediction model library, pressure data in a certain time in the future are obtained, the prediction model library is constructed through a deep learning network and obtained through training of a training set, and the training set is pressure data in a plurality of historical weighting periods; and obtaining a second prediction result of the next weighting propulsion degree based on the pressure data and weighting criteria of a certain time in the future, and sending out an early warning signal. The method can further meet the requirements of coal mine safety production, and the prediction precision and reliability are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of safe mining of coal resources and intelligent mining thereof, and in particular to a method and system for dynamic intelligent prediction and early warning of a mining site roof. Background Art

[0002] In recent years, coal-related industries have continued to develop and improve, and mine pressure monitoring technology and equipment have gradually matured. At the same time, the depth of coal mining in my country continues to increase, the geology and mining conditions of the working face are complex, and the damage and movement of the overburden are difficult to monitor accurately. Dynamic monitoring and prediction of the roof has become a major technical problem for safe mining of coal mining working faces. During the mining process of the fully mechanized working face, the breaking and movement of the roof rock layer will lead to periodic roof pressure, thereby exerting additional pressure on the hydraulic support and underground equipment, and even affecting the safety of underground workers. However, the current prediction system still has problems such as insufficient prediction accuracy, lack of real-time and rapid response capabilities of the prediction system, poor usability and maintainability of the prediction system, and lack of comprehensiveness and depth of prediction results. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a dynamic intelligent prediction and early warning method and system for the mine roof, which can further meet the needs of coal mine safety production, improve the accuracy and reliability of prediction, thereby realizing comprehensive monitoring, law analysis and advance warning of the coal mining working face, providing mines with accurate and timely mine pressure information and decision-making support, and further improving the level of safe, efficient and intelligent mining in mines.

[0004] The present invention discloses a method for dynamic intelligent prediction and early warning of a stope roof, comprising:

[0005] Obtain the pressure data of the hydraulic support in several historical pressure cycles;

[0006] Processing the pressure data within the historical pressure cycle to obtain a valid data set;

[0007] According to the effective data set, a mathematical statistics method is used to perform calculation processing to obtain a first prediction result of the next push progress;

[0008] Obtain real-time pressure data of hydraulic supports;

[0009] Input the real-time pressure data of the hydraulic support into the prediction model library to obtain the pressure data of a certain time in the future, wherein the prediction model library is constructed by using a deep learning network, and the prediction model library is obtained by training with a training set, and the training set is the pressure data within several historical pressure cycles;

[0010] Based on the pressure data and pressure judgment criteria for a certain period of time in the future, a second prediction result of the progress of the next pressure is obtained, and an early warning signal is issued;

[0011] The prediction results are self-evaluated at different levels, and the prediction model library is optimized through feedback to improve the prediction accuracy, wherein the prediction results are the first prediction results and the second prediction results.

[0012] Optionally, processing the pressure data in the historical pressure cycle to obtain a valid data set includes:

[0013] Processing the pressure data in the historical pressure cycle to remove pressure values ​​in the pressure data that are less than the initial support force, wherein the initial support force is a preset pressure value;

[0014] The final resistance of the hydraulic support in each working cycle is selected from the processed pressure data to form a valid data set.

[0015] Optionally, obtaining a first prediction result of the next push progress includes:

[0016] Calculate the sum of the mean value of the end-of-cycle resistance and its mean square error in the valid data set as a roof pressure determination index, and determine the historical pressure step distance of each bracket based on the index;

[0017] Eliminate abnormal values ​​in the historical pressure step distances, wherein the abnormal values ​​in the pressure step distances refer to pressure step distances that differ by more than 50% from the average value of several recent pressure step distances;

[0018] According to the historical pressure step lengths after removing the outliers, the average value and mean square error of the historical pressure step lengths after removing the outliers are calculated;

[0019] Determine the most recent pressure advancement speed, add and subtract the mean square error of the pressure step distance from the average value of the pressure step distance to obtain the fluctuation range of the next pressure step distance, and superimpose the fluctuation range of the next pressure step distance on the most recent pressure advancement speed to obtain the next pressure advancement speed.

[0020] Optionally, the method for predicting the next pressure advancement speed according to the average pressure step distance and pressure position is:

[0021]

[0022] Among them, M i is the i-th pressing step, W z is the most recent pressure position, ΔZ is the geological information correction parameter, W l The predicted value for the next cycle, R C is the compressive strength of the basic roof rock, h is the basic roof thickness, V is the average advancement speed of the working face, H is the coal seam mining depth, α is the coal seam inclination, d is the inclined length of the working face, γ is the basic roof rock density, σ τ It is the basic top rock tensile strength.

[0023] Optionally, the training set is obtained as:

[0024] The historical incoming pressure data transmitted by the pressure sensor is collected, the historical incoming pressure data is screened and processed, abnormal values ​​obtained during the monitoring process are removed, missing data caused by sensor insensitivity or operational errors are supplemented, and then the processed data is smoothed and denoised to obtain the training set.

[0025] Optionally, inputting the real-time pressure data of the hydraulic support into a prediction model library to obtain pressure data for a certain period of time in the future includes:

[0026] For each support on the working surface, the pressure data of the past several hours is used as the input of the prediction model library. The pressure data of the support in the next several hours is calculated and output according to the prediction model library. A cyclic continuous prediction method is adopted, and the pressure data is output once at a certain interval to obtain the pressure data for a certain period of time in the future.

[0027] Optionally, obtaining a second prediction result of the next pressure advancement progress based on the pressure data and the pressure judgment criterion at a certain time in the future includes:

[0028] Based on the pressure data of a certain period of time in the future, combined with the pressure judgment criteria, it is judged whether the working face is under pressure, and combined with the pressure arch pressure decision, it is judged whether the working face is under pressure continuously;

[0029] According to the continuous pressure results, a second prediction result of the next pressure advancement speed is obtained;

[0030] Wherein, judging whether the working surface is under pressure according to the pressure signal includes:

[0031] Get the mean square error of the final resistance:

[0032]

[0033] Among them, σ P is the mean square error of the average resistance at the end of the cycle, n is the number of measured cycles, P t is the measured end-of-cycle resistance of each cycle, is the average value of resistance at the end of cycle;

[0034] According to the mean square error of the final resistance, the top plate pressure criterion is obtained:

[0035]

[0036] Among them, P' t The top plate is used to press the judgment threshold. is the average value of the measured resistance at the end of the cycle;

[0037] According to the top plate pressure criterion, it is determined whether the working surface is under pressure.

[0038] Optionally, combined with the pressure arch pressure decision, judging whether the working face is continuously pressurized includes:

[0039] In the same time period, the pressure of n consecutive brackets at the previous and next moments is greater than the top plate pressure judgment threshold P' t , that is, the conditions are met:

[0040]

[0041] Among them, P j,t1 is the pressure of the jth bracket at time t1, P' t is the top plate pressure judgment threshold, n is the number of consecutive brackets, t1 and t2 represent the time points before and after respectively;

[0042] Based on the conditions, determine whether continuous pressure is occurring on the working face.

[0043] Optionally, a graded self-evaluation of the prediction results may include:

[0044] According to the error between the predicted pressure advancement speed and the actual pressure advancement speed, the prediction errors are graded and marked;

[0045] For prediction models with large prediction errors, remove them from the training set and prediction model library;

[0046] When multiple prediction models can be matched during the prediction process, the prediction model with the smallest matching error is used for prediction.

[0047] The present invention also provides a dynamic intelligent prediction and early warning system for the stope roof, comprising: a data acquisition module, a data processing module and an early warning module;

[0048] The data acquisition module is used to collect data on the support pressure in real time;

[0049] The data processing module is used to process the collected data and predict the next push progress and prediction results;

[0050] The early warning module is used to issue an early warning according to the next push progress and the prediction result. Compared with the prior art, the present invention has the following advantages and technical effects:

[0051] 1. The present invention inverts the roof movement state by identifying the support bearing and support group load distribution characteristics, thereby realizing comprehensive monitoring, law analysis and advance warning of the coal mining working face, providing accurate and timely mine pressure information and decision support for the mine, and further improving the mine's safety, efficiency and intelligent mining level.

[0052] 2. In terms of reducing production costs, the system significantly reduces manual inspection and data processing costs through automated and intelligent monitoring. The real-time monitoring and early warning mechanism reduces direct economic losses, compensation costs and production suspension losses caused by mine pressure accidents, further improving the economic benefits of the enterprise.

[0053] 3. In terms of improving production efficiency, the intelligent mine pressure monitoring and early warning system can accurately analyze mine pressure data, provide scientific guidance for mining, optimize mining plans, and improve resource utilization and production efficiency. This not only increases the output of enterprises, but also improves product quality and enhances market competitiveness.

[0054] 4. In terms of risk management, the present invention can identify and warn potential mine pressure risks in advance through powerful data analysis and prediction capabilities, helping enterprises to take timely measures to reduce the probability of risk occurrence and the degree of loss. This improvement in risk management capabilities can bring more stable income and broader development space to enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0056] Figure 1 This is a flow chart of a method for dynamic intelligent prediction and early warning of a stope roof according to an embodiment of the present invention;

[0057] Figure 2 is a flow chart of a mathematical statistics prediction method according to an embodiment of the present invention;

[0058] Figure 3 is a flowchart of a deep learning prediction method according to an embodiment of the present invention;

[0059] Figure 4 It is an architecture diagram of a mine roof dynamic intelligent prediction and early warning system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0061] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] The present invention proposes a method for dynamic intelligent prediction and early warning of the stope roof, such as Figure 1As shown, the specific steps include:

[0063] Obtain real-time pressure data of hydraulic supports;

[0064] Input the real-time pressure data of the hydraulic support into the prediction model library to obtain the pressure data for a certain period of time in the future, wherein the prediction model library is constructed using a deep learning network, and the prediction model library is obtained by training with a training set, which is the pressure data within several historical pressure cycles;

[0065] Based on the pressure data and pressure judgment criteria for a certain period of time in the future, a second prediction result of the progress of the next pressure is obtained, and an early warning signal is issued.

[0066] The prediction results are graded and self-evaluated, and feedback is provided to optimize the prediction model library and improve the prediction accuracy, wherein the prediction results are the first prediction results and the second prediction results.

[0067] Specifically, the main contents of the dynamic intelligent prediction and early warning method of the mine roof include mathematical statistics prediction method and deep learning prediction method.

[0068] Among them, the mathematical and statistical prediction method refers to the statistical analysis of several recent historical pressure data that have been monitored, calculating the average and mean square error of key indicators such as pressure step distance, pressure time, and pressure intensity, and then predicting the relevant information and fluctuation range of the next pressure. Its characteristic is that the prediction results can be obtained early, and the next pressure result can generally be predicted one cycle in advance, but the accuracy of the prediction results is greatly affected by geological conditions. For working faces with simple geological conditions and relatively fixed pressure laws, the prediction results are more accurate; while for working faces with complex geological conditions and relatively different pressure laws, the prediction results are usually in a larger range. Therefore, the mathematical and statistical prediction method can be used to make an early and rough prediction of the next pressure situation.

[0069] The deep learning prediction method refers to the construction and training of a deep learning-based roof pressure prediction model based on the analysis of historical roof pressure data and its rules. The prediction model can be used to determine the mine pressure trend within a certain period of time in the future, and then determine the next pressure time and location, and issue a pressure warning when the prediction result meets the set pressure warning signal. The characteristic of this method is that the prediction result is relatively accurate, but the lead time is limited, and the closer to the pressure time, the more accurate the prediction result. Therefore, the deep learning prediction method can be used to make real-time and accurate predictions of the next pressure situation. Combining the above two methods can achieve complementary advantages, that is, to ensure that the pressure prediction results have a certain lead time, and can be timely and accurate.

[0070] Furthermore, the pressure data in the historical pressure cycle is processed to obtain a valid data set including:

[0071] Process the pressure data in the historical pressure cycle and remove the pressure values ​​in the pressure data that are less than the initial support force, wherein the initial support force is a preset pressure value;

[0072] The final resistance of the hydraulic support in each working cycle is selected from the processed pressure data to form a valid data set.

[0073] Furthermore, obtaining the first prediction result of the next push progress includes:

[0074] The sum of the mean value of the end-of-cycle resistance and its mean square error in the valid data set is calculated as the top plate pressure determination index, and the historical pressure step distance of each support is determined based on the index;

[0075] The outliers in the historical pressure step distances are eliminated, wherein the outliers in the pressure step distances refer to the pressure step distances that differ by more than 50% from the average value of the pressure step distances in recent times;

[0076] According to the historical pressure step lengths after removing outliers, calculate the average value and mean square error of the historical pressure step lengths after removing outliers;

[0077] Determine the most recent pressure advancement speed, and add or subtract the mean square error of the future pressure step lengths to the average value of the future pressure step lengths to obtain the fluctuation range of the next pressure step length. Superimpose the fluctuation range of the next pressure step length on the most recent pressure advancement speed to obtain the next pressure advancement speed.

[0078] Specifically, the dynamic monitoring and pressure prediction mechanism of the roof proposed by this method is as follows: 1) Calculate the periodic pressure step. First, obtain the historical pressure data of the support within the range to be predicted, remove the pressure values ​​less than the initial support force, calculate the mean and standard deviation of the pressure data of each support, and filter out the pressure data with a large degree of dispersion according to the calculated mean and standard deviation to obtain the pressure data that meets the pressure conditions. Then pre-process the pressure data, determine the pressure interval and the start time, and calculate the advancement corresponding to each pressure time. Finally, calculate the pressure step of each support, remove the pressure step data that is too large or too small, and take the average. 2) Calculate the most recent pressure advancement. Record the last pressure advancement of all supports, and then perform statistical analysis, calculate the mean, variance and standard deviation, and analyze the concentrated distribution data of the advancement. Use the mean as the estimated value, calculate the MSE and MAE to determine the prediction range. 3) Calculate the next pressure advancement, that is, the most recent pressure advancement + periodic pressure step.

[0079] Specifically, mathematical and statistical prediction methods such as Figure 2As shown in Figure 1, historical pressure data and advancement data are used as data sources to predict the pressure location and the support resistance during pressure. That is, mathematical statistics methods are used to roughly predict the roof pressure. The prediction of working face pressure based on mathematical statistics methods mainly includes predicting the resistance of the pressure support and predicting the location of the pressure.

[0080] Prediction of pressure resistance: The prediction of the resistance of the working face support is mainly based on the statistics of the pressure values ​​that have been monitored, and on this basis, the pressure values ​​are averaged, and then the mean square error of the pressure values ​​is calculated. The addition and subtraction of the mean square error of the pressure value is the fluctuation range of the next pressure value. The calculation process formula of the mean pressure value and the pressure prediction value is:

[0081]

[0082] Where P c is the average pressure of the bracket, MPa.

[0083]

[0084] Where P i is the pressure value for the ith time, MPa; P l To predict the incoming pressure value, MPa.

[0085] Pressure position prediction: The pressure position prediction of the working face support mainly counts the monitored pressure step distances, performs average processing on this basis, and then calculates the mean square error of the pressure step distance. The mean square error of the average pressure step distance plus or minus the average pressure step distance is the predicted pressure step distance fluctuation range. Superimposing it on the most recent pressure position, the next pressure position prediction value can be obtained. Pressure step distance prediction judgment formula:

[0086]

[0087] In the formula, M i is the i-th pressing step, m; W z is the most recent pressing position, m; W l is the predicted value of incoming pressure, m.

[0088] S11. Determine the pressure according to the roof periodic pressure determination index:

[0089] According to existing research, the sum of the average cycle end resistance of the support and its mean square deviation is used as the main indicator for judging the periodic pressure of the roof. At the same time, the pressure step distance can be determined according to the corresponding mining position of the pressure value. The calculation process formula of the mean square deviation of the end resistance is:

[0090]

[0091] σ Pis the mean square error of the average resistance at the end of the cycle; n is the number of measured cycles; P t The measured end-of-cycle resistance of each cycle, MPa; Average value of resistance at the end of cycle, MPa.

[0092] The measured average value of the end-of-cycle resistance plus the mean square error of the end-of-cycle resistance is the top plate pressure judgment value. The top plate pressure judgment criterion is as follows:

[0093]

[0094] In the formula, P' t Criterion for top plate pressure, MPa.

[0095] S12, predicting incoming pressure according to the incoming pressure criterion and the processed incoming pressure step distance:

[0096] The prediction of the pressure position of the working face support mainly counts the monitored pressure step distances, performs mean processing on this basis, and then calculates the mean square error of the pressure step distance. The mean square error of the average pressure step distance plus or minus the average pressure step distance is the predicted pressure step distance fluctuation range. Superimposing it on the most recent pressure position can get the predicted value of the next pressure position.

[0097] The formula for predicting the incoming pressure step distance is as follows:

[0098]

[0099] In the formula, M i is the i-th pressing step, m; W z is the most recent pressure position, m; ΔZ is the geological information correction parameter, W l The predicted value for the next cycle is m.

[0100] In the geological information correction parameter term ΔZ, R C is the compressive strength of the basic top rock, MPa; h is the basic top thickness, m; V is the average advancement speed of the working face, m / d; H is the mining depth of the coal seam, m; α is the inclination angle of the coal seam, d is the inclination length of the working face, m; γ is the basic top rock density, KN / m3; σ τ is the basic top rock tensile strength, MPa.

[0101] S13. Algorithm implementation steps based on mathematical statistics method:

[0102] First, preprocess the data to obtain the historical pressure data of all supports within the current prediction range, remove the pressure values ​​less than the initial support force, calculate the mean and standard deviation of each support pressure data, and filter out the pressure data with a large degree of dispersion based on the mean and standard deviation to obtain the pressure data that meets the pressure conditions. Then process the obtained pressure data and determine the starting time of each pressure for each support according to the configured pressure interval (the default is 6 hours). According to the calibrated advancement and coal mining machine position statistical data, calculate the advancement corresponding to each pressure time. Then calculate the pressure step distance of each support, remove data that is too large or too small, and then calculate the average pressure step distance within the target mining range. It is also necessary to calculate the advancement of the most recent pressure, and calculate the mean, variance and standard deviation of the advancement. According to the normal distribution law of the data, analyze the concentrated distribution data of the advancement.

[0103] Finally, a predictive calculation is performed, and the calculated mean value is used as an estimate of the most recent advancement of the working face, and the advancement of the next advancement is predicted.

[0104] Furthermore, the training set is obtained as:

[0105] Collect historical incoming pressure data transmitted by the pressure sensor, filter and process the historical incoming pressure data, remove abnormal values ​​obtained during the monitoring process, complete the missing data caused by sensor insensitivity or operational errors, and then smooth and reduce noise on the processed data to obtain a training set.

[0106] Furthermore, the mine pressure data to be predicted is input into the prediction model to obtain the prediction results, including:

[0107] According to the predicted mine pressure data, the pressure signal is identified, and whether the working face is under pressure is determined according to the pressure signal. Combined with the pressure arch pressure decision, it is determined whether the working face is under continuous pressure.

[0108] Get the prediction results based on the continuous pressure results.

[0109] Specifically, S2 predicts incoming pressure by analyzing historical pressure trends based on deep learning, that is, accurately predicts the incoming pressure on the roof of the comprehensive mining working face.

[0110] The deep learning-based periodic pressure prediction algorithm is a model that uses historical data to predict future values. It predicts future values ​​by learning patterns and trends in historical data based on the characteristics and rules of time series data. After training the deep learning model with a large amount of roof pressure data, the model can automatically, in real time and continuously predict roof changes in the future. The prediction results can represent important information such as the predicted pressure value in the future.

[0111] The logical process of the incoming pressure prediction method based on deep learning is as follows Figure 3As shown, it mainly includes: 1) Data preprocessing: Collect historical mine pressure data of each support on the working face in the past two months and filter out outliers. Store high-quality data in the database for deep learning model training. 2) Train LSTM model: Based on historical mine pressure data, train the deep learning model, and through iterative training optimization, learn the inherent laws and patterns in the data, and then obtain features that can reflect periodic changes, trend changes and potential influencing factors. 3) Predict the periodic pressure of the working face: For each support on the working face, use the past pressure data as the input of the deep learning model, and the model predicts and outputs the future pressure change trend of the support. Using a cyclic continuous prediction method, the model prediction results are optimized and updated in real time every half an hour to ensure the accuracy of the prediction.

[0112] S21. Learn the historical pressure patterns. By learning the patterns and internal laws in the historical data, the model is trained to facilitate the prediction of future values. First, the historical pressure data transmitted by the pressure sensor is collected. By screening and processing these data, the abnormal values ​​obtained during the monitoring process are removed, and the missing data caused by sensor insensitivity or operational errors are supplemented. Then the data is smoothed and denoised, stored in the InfluxDb database, and the deep learning model is trained. Through multiple iterations of training and optimization, the characteristics that can reflect the periodic changes, trend changes and potential influencing factors of the pressure are obtained.

[0113] S22. Predict the changing trend of roof pressure by inputting the real-time feedback data from the sensor into the system, matching it with the established function model, and outputting the future pressure trend. That is, for each support on the working face, the pressure data of the past 48 hours is used as the input of the deep learning model, and the model predicts and outputs the pressure change trend of the support in the next 6 hours. The cyclic continuous prediction method is adopted to optimize and update the model prediction results in real time every half an hour to ensure the accuracy of the prediction.

[0114] S23, identify the pressure signal, and determine whether the working face is under pressure according to the pressure signal. The pressure judgment criterion of a single support is the sum of the average cycle end resistance of the support and its mean square deviation as the main indicator for judging the periodic pressure of the roof. At the same time, the pressure step distance can be determined according to the corresponding mining position of the pressure value. The calculation process formula of the mean square deviation of the end resistance is:

[0115]

[0116] σ P is the mean square error of the average resistance at the end of the cycle; n is the number of measured cycles; P t The measured end-of-cycle resistance of each cycle, MPa; Average value of resistance at the end of cycle, MPa.

[0117] The measured average value of the end-of-cycle resistance plus the mean square error of the end-of-cycle resistance is the top plate pressure judgment value. The top plate pressure judgment criterion is as follows:

[0118]

[0119] In the formula, P' t is the top plate pressure criterion threshold, MPa.

[0120] The "pressure arch" is used as a criterion for continuous pressure. During the entire cycle of pressure, when the pressure of each support is on the rise, the pressure of the entire working surface is low on both sides and high in the middle, and the pressure of the entire working surface is distributed in an arch shape, that is, a "pressure arch" appears. This phenomenon is used as a criterion for continuous pressure. Once the "pressure arch" phenomenon appears and reaches the pressure judgment threshold range, it is judged that pressure is about to occur.

[0121] Define the variables as follows:

[0122] P i,t : The pressure of the i-th bracket at time t. P' t : Top plate pressure judgment threshold n: Number of consecutive brackets t1 and t2: Represent the time points of the first and second cuts respectively;

[0123] In the same time period (i.e. for a certain starting bracket index i), the pressure of n consecutive brackets at the front and rear cuts (t1 and t2) is greater than the top plate pressure judgment threshold P' t , that is, satisfying the conditions:

[0124] and

[0125] Determine if there is pressure on the working surface and issue an early warning.

[0126] S24, according to the identified pressure signal, after the working face has pressure, a pressure warning signal is issued in time. According to the monitored pressure step distance, specific location, specific time and duration of the pressure, after the pressure warning decision is made, a pressure warning is issued.

[0127] By comprehensively judging multiple indicators and then issuing an early warning, it is possible to effectively avoid false pressure alarms caused by abnormalities in individual supports on the working surface.

[0128] S3. Comprehensively evaluate the prediction results and optimize them. According to theoretical conditions, technical difficulty, actual needs and operability, the evaluation is divided into four levels: Level I: <0.5m, Level II: 0.5-1.0m, Level III: 1.0-1.6m, Level IV: >1.6m. Then conduct a graded self-evaluation of the prediction results, and provide feedback to optimize the prediction model and improve the prediction accuracy.

[0129] S31. Classify and count the prediction errors of each bracket, comprehensively classify the prediction errors of each bracket, and systematically count, analyze and record them for subsequent improvement and optimization.

[0130] S32. Quantitatively evaluate the accuracy of each function model, because the pressure forecasting of the fully mechanized mining face requires the establishment of multiple function models, and the prediction accuracy is improved through the combined action of multiple function models. However, the established models are inevitably not accurate enough, so it is necessary to quantitatively evaluate the accuracy of the models and select the model with high accuracy for pressure forecasting.

[0131] S33. Eliminate and correct models with accuracy below level IV. Models with errors > 1.6m can be defined as inaccurate and must be eliminated.

[0132] S34. Give priority to selecting function models with high accuracy for pressure prediction, obtain the movement law of the rock layer on the roof of the fully mechanized mining face, give mine pressure warning signals and production guidance suggestions, ensure the system's usability and scalability, and provide strong technical support for safe production in coal mines.

[0133] The present invention also discloses a dynamic intelligent prediction and early warning system for the roof of a mining area, comprising: a data acquisition module, a data processing module and an early warning module;

[0134] Data acquisition module, used to collect data on the support pressure in real time;

[0135] The data processing module is used to process the collected data and predict the next push progress and prediction results;

[0136] The early warning module is used to issue early warnings based on the next push progress and prediction results.

[0137] Specifically, the architecture of the mine roof dynamic intelligent prediction and early warning system is as follows: Figure 4 As shown in the figure, the software platform adopts a distributed architecture design, which mainly includes the data acquisition end, the server end and the user end. The data acquisition end is installed on the underground hydraulic support, responsible for real-time data collection on the support pressure, and ensures real-time upload to the server. The server end integrates the data collected by the data acquisition end, stores it, analyzes the data, runs the prediction algorithm, establishes a database, and takes into account the transportation function. The user end interacts with the server through the Web interface to obtain real-time mine pressure monitoring information and prediction results.

[0138] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for dynamic intelligent prediction and early warning of a stope roof, characterized in that: include: Obtain the pressure data of the hydraulic support in several historical pressure cycles; Processing the pressure data in the historical pressure cycle to obtain a valid data set; According to the effective data set, a mathematical statistics method is used to perform calculation processing to obtain a first prediction result of the next push progress; Obtain real-time pressure data of hydraulic supports; Input the real-time pressure data of the hydraulic support into the prediction model library to obtain the pressure data of a certain time in the future, wherein the prediction model library is constructed by using a deep learning network, and the prediction model library is obtained by training with a training set, and the training set is the pressure data within several historical pressure cycles; Based on the pressure data and pressure judgment criteria for a certain period of time in the future, a second prediction result of the progress of the next pressure is obtained, and an early warning signal is issued; The prediction results are self-evaluated at different levels, and the prediction model library is optimized through feedback to improve the prediction accuracy, wherein the prediction results are the first prediction results and the second prediction results.

2. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1 is characterized in that: The pressure data in the historical pressure cycle is processed to obtain a valid data set including: Processing the pressure data in the historical pressure cycle to remove pressure values ​​in the pressure data that are less than the initial support force, wherein the initial support force is a preset pressure value; The final resistance of the hydraulic support in each working cycle is selected from the processed pressure data to form a valid data set.

3. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1 is characterized in that: The first prediction results of the next push progress include: Calculate the sum of the mean value of the end-of-cycle resistance and its mean square error in the valid data set as a roof pressure determination index, and determine the historical pressure step distance of each bracket based on the index; Eliminate abnormal values ​​in the historical pressure step distances, wherein the abnormal values ​​in the pressure step distances refer to pressure step distances that differ by more than 50% from the average value of several recent pressure step distances; According to the historical pressure step lengths after removing the outliers, the average value and mean square error of the historical pressure step lengths after removing the outliers are calculated; Determine the most recent pressure advancement speed, add and subtract the mean square error of the pressure step distance from the average value of the pressure step distance to obtain the fluctuation range of the next pressure step distance, and superimpose the fluctuation range of the next pressure step distance on the most recent pressure advancement speed to obtain the next pressure advancement speed.

4. The method for dynamic intelligent prediction and early warning of stope roof according to claim 3 is characterized in that: Based on the average pressure step length and pressure position, the method for predicting the next pressure advancement speed is: Among them, M i is the i-th pressing step, W z is the most recent pressure position, ΔZ is the geological information correction parameter, W l The predicted value for the next cycle, R C is the compressive strength of the basic roof rock, h is the basic roof thickness, V is the average advancement speed of the working face, H is the coal seam mining depth, α is the coal seam inclination, d is the inclined length of the working face, γ is the basic roof rock density, σ τ It is the basic top rock tensile strength.

5. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1, characterized in that: The training set is obtained as: The historical incoming pressure data transmitted by the pressure sensor is collected, the historical incoming pressure data is screened and processed, abnormal values ​​obtained during the monitoring process are removed, missing data caused by sensor insensitivity or operational errors are supplemented, and then the processed data is smoothed and denoised to obtain the training set.

6. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1, characterized in that: Inputting the real-time pressure data of the hydraulic support into the prediction model library to obtain the pressure data for a certain period of time in the future includes: For each support on the working surface, the pressure data of the past several hours is used as the input of the prediction model library. The pressure data of the support in the next several hours is calculated and output according to the prediction model library. A cyclic continuous prediction method is adopted, and the pressure data is output once at a certain interval to obtain the pressure data for a certain period of time in the future.

7. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1, characterized in that: Based on the pressure data and the pressure judgment criterion at a certain time in the future, obtaining a second prediction result of the progress of the next pressure increase includes: Based on the pressure data of a certain time in the future, combined with the roof pressure judgment criteria, it is judged whether the working face is under pressure, and combined with the pressure arch pressure decision, it is judged whether the working face is under pressure continuously; According to the continuous pressure results, a second prediction result of the next pressure advancement speed is obtained; Among them, judging whether the working surface is under pressure includes: Get the mean square error of the final resistance: Among them, σ P is the mean square error of the average resistance at the end of the cycle, n is the number of measured cycles, P t is the measured end-of-cycle resistance of each cycle, is the average value of resistance at the end of the cycle; Get the top plate pressure criterion: Among them, P' t The top plate is used to press the judgment threshold. It is the average value of the resistance at the end of the cycle measured.

8. The method for dynamic intelligent prediction and early warning of stope roof according to claim 7 is characterized in that: Combined with the pressure arch pressure decision, judging whether the working face is continuously pressurized includes: In the same time period, the pressure of n consecutive brackets at the previous and next moments is greater than the top plate pressure judgment threshold P' t , that is, the conditions are met: and Among them, P j,t1 is the pressure of the jth bracket at time t1, P' t is the top plate pressure judgment threshold, n is the number of consecutive brackets, t1 and t2 represent the time points before and after respectively; Based on the conditions, determine whether continuous pressure is occurring on the working face.

9. The method for dynamic intelligent prediction and early warning of stope roof according to claim 1, characterized in that: The graded self-evaluation of the forecast results includes: According to the error between the predicted pressure advancement speed and the actual pressure advancement speed, the prediction errors are graded and marked; For prediction models with large prediction errors, remove them from the training set and prediction model library; When multiple prediction models can be matched during the prediction process, the prediction model with the smallest matching error is used for prediction.

10. The mine roof dynamic intelligent prediction and early warning system is characterized by: include: Data collection module, data processing module and early warning module; The data acquisition module is used to collect data on the support pressure in real time; The data processing module is used to process the collected data and predict the next push progress and prediction results; The early warning module is used to issue an early warning based on the next push progress and prediction results.

Images