CN116628880A - Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network - Google Patents
Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network Download PDFInfo
- Publication number
- CN116628880A CN116628880A CN202310586835.9A CN202310586835A CN116628880A CN 116628880 A CN116628880 A CN 116628880A CN 202310586835 A CN202310586835 A CN 202310586835A CN 116628880 A CN116628880 A CN 116628880A
- Authority
- CN
- China
- Prior art keywords
- data
- hydraulic support
- neural network
- dynamic programming
- travel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 42
- 239000003245 coal Substances 0.000 claims abstract description 55
- 238000005065 mining Methods 0.000 claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 42
- 238000004140 cleaning Methods 0.000 claims description 17
- 230000008878 coupling Effects 0.000 claims description 10
- 238000010168 coupling process Methods 0.000 claims description 10
- 238000005859 coupling reaction Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 238000012216 screening Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000013499 data model Methods 0.000 claims description 4
- 238000006073 displacement reaction Methods 0.000 claims description 4
- 238000007405 data analysis Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 description 10
- 238000004458 analytical method Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000000052 comparative effect Effects 0.000 description 5
- 238000012937 correction Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005520 cutting process Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Pure & Applied Mathematics (AREA)
- Biophysics (AREA)
- Mathematical Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computational Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Mathematical Optimization (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a hydraulic support straightening method based on a dynamic programming algorithm fused BP neural network, and relates to the technical field of underground fully-mechanized mining face automation of coal mines. The hydraulic support straightening method based on the dynamic programming algorithm fused with the BP neural network comprises the following steps: step one: collecting coal mine underground discrete data; step two: preprocessing data, and searching data conforming to the travel of the hydraulic support; step three: the problems of loss of processing data, insufficient accuracy and the like of a dynamic programming algorithm are utilized, so that the hydraulic support stroke data is straightened; step four: and establishing an optimal model by establishing a BP neural network, and realizing the straightening function of the hydraulic support. The method solves the problem of loss of underground acquisition data of the coal mine by combining a dynamic planning algorithm, realizes the alignment of travel data of the hydraulic support acquired by an upper computer of the underground coal mine, establishes an optimal model by utilizing a neural network, and realizes the real-time alignment of the hydraulic support.
Description
Technical Field
The invention relates to the technical field of automation of underground fully mechanized mining working faces of coal mines, in particular to a hydraulic support straightening method based on a dynamic programming algorithm fused BP neural network.
Background
Coal is a basic energy source and an important industrial raw material in China, more than 960 hundred million tons of raw coal are accumulated since the country is built, reliable energy source guarantee is provided for the development of national economy and other technologies in China, and important research on mineral resource safety and efficient development and utilization technology is clearly pointed out in the national institute of State long-term science and technology development planning class (2006-2020). The current coal mining technology is generally a traditional mechanized mining technology, a fully-mechanized mining working face is formed by a coal mining machine, a hydraulic support and a scraper conveyor, and the scraper conveyor is supported by the hydraulic support, so that the operation of the coal mining machine is pushed, and the reciprocating coal cutting action of the coal mining machine is realized. The straightness of the hydraulic support is a key device for ensuring that the coal mining machine can run linearly and improving the coal mining efficiency, however, due to the large inclination angle of the stratum and a series of complex underground working condition conditions with complex geological conditions of a thick coal seam and an extra thick coal seam, difficulties are brought to the stability and straightness adjustment of the hydraulic support, so that the hydraulic support can obtain dangerous postures such as rolling and torsion, and serious accidents such as equipment damage and casualties are caused.
At present, the traditional hydraulic support positioning method is usually carried out by manually pulling ropes on a coal face or adjusting by infrared beams, and the method can not meet the requirement of fully mechanized coal face automation, can not ensure the precision, and greatly influences the coal mining efficiency and the safety of a coal mine. In recent years, with the rise of machine learning and deep learning technologies, various fields perform neural network training by utilizing a large amount of discrete data generated under actual working conditions, only input and output data of a controlled object are utilized to analyze the characteristics of the controlled object, and various optimizing algorithms are utilized, so that an optimized model applicable to the actual working conditions is obtained, and the utilization rate of the existing data is greatly improved.
The Chinese patent application CN113587927A discloses a working face straightening method based on BP neural network and inertial navigation, which comprises the steps of establishing an inertial navigation coordinate system and setting the BP neural network through a positioning curve, a pushing curve and an ideal straight line on an upper computer, so that the prediction of the pushing curve of the whole hydraulic support is carried out next time, and real-time straightening is realized; however, the positioning curve obtained by the inertial navigation system has the conditions of data loss and poor data accuracy, so that the BP neural network has low prediction accuracy, and a larger gap from the actual working condition exists, and the coal mining efficiency of the coal mining machine cannot be greatly improved.
Therefore, a prediction control method capable of simultaneously processing data loss, realizing the advanced alignment and improving the accuracy of a hydraulic support travel data prediction control model is needed, so as to realize the automatic alignment of the hydraulic support of the underground fully-mechanized coal face of the coal mine and improve the coal mining efficiency.
Disclosure of Invention
The invention aims to provide a hydraulic support straightening method based on a dynamic programming algorithm fused BP neural network, which solves the technical problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions: the hydraulic support straightening method based on the dynamic programming algorithm fused with the BP neural network comprises the following steps:
step one: collecting coal mine underground discrete data: classifying, screening and cleaning the hydraulic support work discrete data collected underground coal mine, and screening travel data representing the real work state of the hydraulic support;
step two: data preprocessing, namely searching data conforming to the travel of the hydraulic support: the method comprises the steps that the stroke data of the hydraulic support obtained after cleaning is subjected to global searching for an optimal interval through a dynamic programming algorithm, the algorithm is utilized to realize the alignment of the discrete data of the stroke of the hydraulic support, and the discrete data of the stroke of the hydraulic support is determined to be the real stroke data of the hydraulic support in an ideal acquisition state;
step three: the problems of loss of processing data, insufficient accuracy and the like of a dynamic programming algorithm are utilized, and the hydraulic support stroke data are straightened: considering the coupling relation of a plurality of hydraulic support groups and a group, wherein the factor that error ranges exist between adjacent supports in the group, dividing the data of a certain complete hydraulic support advancing process according to the hydraulic support advancing time sequence, and determining various parameters of the neural network;
step four: establishing a BP time sequence neural network model, and establishing an optimal hydraulic support travel data model by multiple tuning parameters: building a BP neural network in an upper computer, and obtaining an optimal model through model training of multi-trip data time sequence prediction;
step five: based on an optimal model, the travel data acquired in real time are imported, and the hydraulic support is straightened in real time: when the coal mining machine starts to work, inputting stroke data corresponding to a hydraulic support where the current coal mining machine is located into a trained optimal model, obtaining a next group of strokes which are required to be actually pushed of the hydraulic support where the current coal mining machine is located, and sending a pushing instruction to a support controller of the next hydraulic support according to the strokes; realizing real-time alignment.
Further, the method for preprocessing the collected underground discrete data of the coal mine, namely the discrete data collected by the fully mechanized mining face comprises the following steps:
step one: travel data screening, timestamp format conversion and bracket state abnormal data processing: after acquiring the acquired discrete data of the actual working condition, data preprocessing is performed firstly, and positioning and searching of travel data and conversion of various related data formats are realized by utilizing a data analysis method;
step two: processing long-time motionless data and processing repeated data of a single bracket according to a time relation: cleaning again based on the data obtained in the step one, and cleaning the collected data in the same second and the data with unchanged time-increasing data value;
step three: dividing travel data of the single-travel hydraulic support: dividing the working condition based on the data obtained in the second step, finding out each group of data values which accord with single advancing of the whole hydraulic support group, and dividing the data values according to time;
step four: maximum and minimum difference positioning bracket travel: searching a maximum value and a minimum value based on each group of data values obtained in the third step, and taking the difference between the maximum value and the minimum value as the single-time travelling distance of the hydraulic support according to the actual working condition;
step five: based on the single travelling distance of the hydraulic support determined in the step four, optimizing based on a dynamic programming algorithm, determining an optimal interval, traversing data to find a start position and an end position of the optimal interval, taking the start position and the end position as data replacement standards, and cleaning data by combining with an adjacent frame error coupling relation; and obtaining the stroke data value of the hydraulic support in an ideal acquisition state under the required actual working condition, and realizing the straightening function of the stroke data.
Further, the dynamic programming algorithm comprises the following steps:
step one: defining an array through initial hydraulic support travel data to store variables calculated in each interval;
step two: calculating the maximum value of a single interval, adjacent frame difference values and other variables according to working conditions;
step three: judging whether the acquired data accords with ideal acquired data, if not, returning to the execution step II; if yes, executing the fourth step;
step four: a plurality of high-quality intervals are compared and found out to find out an optimal interval;
step five: and correcting other data outside the interval according to the adjacent frame relation.
Further, the generation process based on BP neural network time series prediction comprises the following steps:
step one: searching data of a global optimal solution based on a dynamic programming algorithm, dividing single-group travel data of the hydraulic support according to the group follow-up relation of the hydraulic support and the adjacent frame coupling relation of the hydraulic support under actual working conditions, setting an output layer, an input layer and determining the number of hidden layer nodes according to continuous debugging after dividing;
step two: setting a training set and a testing set by using single-set travel data of the hydraulic support, wherein 70% of data are used as training data, 30% of data are used as testing data, and training is performed in Matlab by using a written BP neural network;
step three: adjusting data to normalize the data to a (0, 1) interval, setting parameters such as training algebra, training function, learning rate, minimum error and the like, determining each optimal value through continuous debugging, further obtaining a result and deviation of the whole BP time sequence prediction, and continuously debugging to obtain an optimal model;
step four: training the acquired real-time hydraulic support data based on the optimal model obtained by training in the step four to obtain the next group of strokes of the hydraulic support where the current coal mining machine is located, which are required to be actually pushed, and sending pushing instructions to a support controller of the next hydraulic support according to the strokes to realize real-time alignment.
Further, the hydraulic support real-time alignment comprises the following steps:
step one: the upper computer is put into the algorithm and the neural network model of the invention;
step two: acquiring hydraulic support travel data at the previous moment;
step three: training to obtain an optimal model and storing;
step four: predicting the displacement of the hydraulic stroke at the next moment based on the model;
step five: and the bracket controller receives the adjustment instruction to complete the real-time straightening of the bracket.
Compared with the related art, the hydraulic support straightening method based on the dynamic programming algorithm fused BP neural network has the following beneficial effects:
the hydraulic support straightening method based on the dynamic programming algorithm fused BP neural network, provided by the invention, solves the problem of loss of underground coal mine collected data by combining the dynamic programming algorithm, realizes the straightening of the hydraulic support travel data collected by an underground coal mine upper computer, and then establishes an optimal model by using the neural network, thereby realizing the real-time straightening of the hydraulic support.
Drawings
Fig. 1 is a schematic flow chart of a hydraulic support alignment method based on a dynamic programming algorithm fused BP neural network;
FIG. 2 is a block diagram of a method for preprocessing discrete data collected on a fully-mechanized coal mining face;
FIG. 3 is a diagram of a dynamic programming algorithm provided by the present invention;
FIG. 4 is a schematic diagram of a process for generating a BP neural network time series prediction according to the present invention;
FIG. 5 is a schematic diagram of a real-time alignment process of a hydraulic support provided by the invention;
FIG. 6 is a graph I of the result of a comparative experiment after correction and alignment based on a dynamic programming algorithm;
FIG. 7 is a graph II of the result of a comparative experiment after correction and alignment based on a dynamic programming algorithm;
FIG. 8 is a third comparative experiment result graph after correction and alignment based on a dynamic programming algorithm;
FIG. 9 is a graph IV of the result of a comparative experiment after correction and alignment based on a dynamic programming algorithm;
FIG. 10 is a comparative experiment of trip data without alignment using a dynamic programming algorithm;
FIG. 11 is a graph showing the average relative error analysis of FIG. 10;
FIG. 12 is an analysis result of training set 1;
FIG. 13 is an average relative error analysis result of training set 1;
FIG. 14 is an analysis result of training set 2;
FIG. 15 is an average relative error analysis result for training set 2;
FIG. 16 is an analysis result of training set 3;
FIG. 17 is an average relative error analysis result of training set 3;
FIG. 18 is an analysis result of training set 4;
fig. 19 shows the average relative error analysis result of training set 4.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a flow chart of a hydraulic support alignment method based on a dynamic programming algorithm fused with a BP neural network is shown. The invention provides a technical scheme that: the hydraulic support straightening method based on the dynamic programming algorithm fused with the BP neural network comprises the following steps:
step one: collecting coal mine underground discrete data: classifying, screening and cleaning the hydraulic support work discrete data collected underground coal mine, and screening travel data representing the real work state of the hydraulic support;
step two: data preprocessing, namely searching data conforming to the travel of the hydraulic support: the method comprises the steps that the stroke data of the hydraulic support obtained after cleaning is subjected to global searching for an optimal interval through a dynamic programming algorithm, the algorithm is utilized to realize the alignment of the discrete data of the stroke of the hydraulic support, and the discrete data of the stroke of the hydraulic support is determined to be the real stroke data of the hydraulic support in an ideal acquisition state;
step three: the problems of loss of processing data, insufficient accuracy and the like of a dynamic programming algorithm are utilized, and the hydraulic support stroke data are straightened: considering the coupling relation of a plurality of hydraulic support groups and a group, wherein the factor that error ranges exist between adjacent supports in the group, dividing the data of a certain complete hydraulic support advancing process according to the hydraulic support advancing time sequence, and determining various parameters of the neural network;
step four: establishing a BP time sequence neural network model, and establishing an optimal hydraulic support travel data model by multiple tuning parameters: building a BP neural network in an upper computer, and obtaining an optimal model through model training of multi-trip data time sequence prediction;
step five: based on an optimal model, the travel data acquired in real time are imported, and the hydraulic support is straightened in real time: when the coal mining machine starts to work, inputting stroke data corresponding to a hydraulic support where the current coal mining machine is located into a trained optimal model, obtaining a next group of strokes which are required to be actually pushed of the hydraulic support where the current coal mining machine is located, and sending a pushing instruction to a support controller of the next hydraulic support according to the strokes; realizing real-time alignment.
As shown in fig. 1, firstly, various data of a hydraulic support of a real underground fully-mechanized coal mining face of a coal mine are collected by an upper computer; secondly, dividing data by using Python, and cleaning the data to obtain a hydraulic support stroke data value optimized by using an algorithm subsequently; furthermore, optimization is performed based on a dynamic programming algorithm, the best acquisition effect in the acquired hydraulic support travel data is set, a section of interval with adjacent frame errors according with working conditions is taken as an optimal interval, data correction is performed based on the whole hydraulic support travel data, the influence of missing data and precision data difference is removed, the hydraulic support travel data in an ideal acquisition state is obtained, and straightening of the hydraulic support travel data is realized; establishing a BP neural network time sequence prediction model, and establishing an optimal model by multiple tuning; and finally, training the acquired real-time hydraulic support data based on an optimal model to obtain the next group of strokes of the hydraulic support where the current coal mining machine is located, which are required to be actually pushed, and sending a pushing command to a support controller of the next hydraulic support according to the strokes to realize real-time straightening.
Referring to fig. 2, a block diagram of a discrete data preprocessing method for fully mechanized mining face acquisition is shown. The method for preprocessing the discrete data acquired by the underground coal mine, namely the fully-mechanized mining face comprises the following steps:
step one: travel data screening, timestamp format conversion and bracket state abnormal data processing: after acquiring the acquired discrete data of the actual working condition, data preprocessing is performed firstly, and positioning and searching of travel data and conversion of various related data formats are realized by utilizing a data analysis method;
step two: processing long-time motionless data and processing repeated data of a single bracket according to a time relation: cleaning again based on the data obtained in the step one, and cleaning the collected data in the same second and the data with unchanged time-increasing data value;
step three: dividing travel data of the single-travel hydraulic support: dividing the working condition based on the data obtained in the second step, finding out each group of data values which accord with single advancing of the whole hydraulic support group, and dividing the data values according to time;
step four: maximum and minimum difference positioning bracket travel: searching a maximum value and a minimum value based on each group of data values obtained in the third step, and taking the difference between the maximum value and the minimum value as the single-time travelling distance of the hydraulic support according to the actual working condition;
step five: based on the single travelling distance of the hydraulic support determined in the step four, optimizing based on a dynamic programming algorithm, determining an optimal interval, traversing data to find a start position and an end position of the optimal interval, taking the start position and the end position as data replacement standards, and cleaning data by combining with an adjacent frame error coupling relation; and obtaining the stroke data value of the hydraulic support in an ideal acquisition state under the required actual working condition, and realizing the straightening function of the stroke data.
As shown in fig. 2, the upper computer cannot independently take hydraulic bracket travel data, so that data cleaning is required in advance; for example: the data types of various hydraulic supports are converted into data such as push rod travel, support pressure and the like, a time stamp is converted into Beijing time and the like, the problem of abnormal support data is solved, the same data of the same second in the data collected by an upper computer and the data which are not moved for a long time along with the time are removed, the cleaned data are further divided into single travel according to time, and the difference between the maximum value and the minimum value is used as the travel data of the initial hydraulic support which is actually collected.
Referring to fig. 3, a dynamic programming algorithm structure is shown. The dynamic programming algorithm comprises the following steps:
step one: defining an array through initial hydraulic support travel data to store variables calculated in each interval;
step two: calculating the maximum value of a single interval, adjacent frame difference values and other variables according to working conditions;
step three: judging whether the acquired data accords with ideal acquired data, if not, returning to the execution step II; if yes, executing the fourth step;
step four: a plurality of high-quality intervals are compared and found out to find out an optimal interval;
step five: and correcting other data outside the interval according to the adjacent frame relation.
As shown in fig. 3, a bottom-up dynamic programming algorithm is adopted to search an optimal interval of hydraulic support stroke data, dynamic programming is generally performed from bottom to top, an optimal solution (i.e., an optimal sub-structure) of each sub-problem is obtained from bottom to top through a state transition equation, the optimal sub-structure is actually an optimal solution obtained by exhausting all cases, and after the optimal solution of each sub-problem is obtained, namely, each optimal solution is actually a global optimal solution of the sub-problem. Firstly defining an array to store an optimal interval sum with each data point as an end, and initializing one data in the array as a value of a first data point; secondly, traversing the whole initial travel data from the second data, setting judging conditions according to adjacent frames of the bracket and an error threshold (the adjacent frame error which is set to be in accordance with the working condition is 100 mm), and screening; further selecting a plurality of high-quality intervals to fill into the array defined at the beginning, and selecting the interval with the smallest accumulated error as the optimal interval; and finally, correcting the stroke data of other brackets based on the optimal interval and the actual working condition, and realizing the alignment of the stroke data of the hydraulic bracket based on a dynamic programming algorithm.
Referring to fig. 4, a generation process of the BP neural network time-series prediction is shown. The generation process based on BP neural network time series prediction comprises the following steps:
step one: searching data of a global optimal solution based on a dynamic programming algorithm, dividing single-group travel data of the hydraulic support according to the group follow-up relation of the hydraulic support and the adjacent frame coupling relation of the hydraulic support under actual working conditions, setting an output layer, an input layer and determining the number of hidden layer nodes according to continuous debugging after dividing;
step two: setting a training set and a testing set by using single-set travel data of the hydraulic support, wherein 70% of data are used as training data, 30% of data are used as testing data, and training is performed in Matlab by using a written BP neural network;
step three: adjusting data to normalize the data to a (0, 1) interval, setting parameters such as training algebra, training function, learning rate, minimum error and the like, determining each optimal value through continuous debugging, further obtaining a result and deviation of the whole BP time sequence prediction, and continuously debugging to obtain an optimal model;
step four: training the acquired real-time hydraulic support data based on the optimal model obtained by training in the step four to obtain the next group of strokes of the hydraulic support where the current coal mining machine is located, which are required to be actually pushed, and sending pushing instructions to a support controller of the next hydraulic support according to the strokes to realize real-time alignment.
As shown in fig. 4, firstly, according to the adjacent frame relation of the hydraulic frames, setting each 7 frames as a group, adopting BP time sequence prediction, adopting the first 7 hydraulic frames to predict the frame 8 data, taking the data as the input layer and the output layer of the neural network, and selecting the optimal node number of the hidden layer according to an empirical formula; then determining the most suitable optimizing excitation function according to the actual model condition, selecting the most suitable training function, taking 70% as a training set and 30% as a testing set according to the previous description, and setting the maximum algebra of training and the minimum error gradient; and finally, calculating the relative error of the model and the fitting goodness of the whole model to judge, if the fitting goodness of the two indexes is bad, re-optimizing the model, and continuously debugging to obtain an optimal hydraulic support stroke data model.
Referring to fig. 5, a schematic diagram of a hydraulic support real-time alignment process is shown. The hydraulic support real-time alignment comprises the following steps:
step one: the upper computer is put into the algorithm and the neural network model of the invention;
step two: acquiring hydraulic support travel data at the previous moment;
step three: training to obtain an optimal model and storing;
step four: predicting the displacement of the hydraulic stroke at the next moment based on the model;
step five: and the bracket controller receives the adjustment instruction to complete the real-time straightening of the bracket.
As shown in fig. 5, firstly, a dynamic programming algorithm and a BP time sequence prediction model which are written in advance are put into an upper computer, and after the hydraulic support advances one or more times, training is started to obtain an optimal model and the optimal model is stored in the upper computer; secondly, collecting real-time data when the hydraulic support works next time, uploading the real-time data to an upper computer, and predicting by the upper computer based on the real-time data and combining with a trained optimal model to obtain the stroke displacement of the hydraulic support next time; and finally, the bracket controller receives an adjustment instruction from the upper computer to complete the real-time straightening function of the hydraulic bracket.
Description of the Experimental results
Referring to fig. 6, 7, 8, 9: the method is a comparison experiment result description of hydraulic support travel data collected underground coal mines and corrected and straightened based on a dynamic programming algorithm.
The experimental data were 3 times 157 single pushing strokes of the stent.
In the experiment, the end headstock has the reciprocating coal cutting condition in the actual working condition, so the end headstock is manually corrected based on the adjacent frame coupling principle before algorithm alignment for removing the influence. It can be seen that the acquired data has the best acquisition effect of 30-60 (different strokes) due to the problems of complex working conditions, poor sensitivity of the sensor, poor data acquisition effect of the upper computer and the like, so that the dynamic programming algorithm is adopted to correct the whole data section based on the section, so that the whole acquired stroke data is closest to the real hydraulic support track, and the hydraulic support stroke data is aligned.
Referring to fig. 10-19: and establishing a BP neural network prediction model effect experimental result based on the acquired data.
The experiment was performed using the four sets of data in fig. 6, 7, 8, and 9; FIG. 11 is a graph of the raw data of FIG. 8, which is used as a comparison experiment of the trip data not found using the dynamic programming algorithm.
In the experiment, the fact that the end head frame can have reciprocating coal cutting conditions in actual working conditions is considered, so that the end head frame (6 frames) is removed for establishing a BP neural network model for better realizing an optimal model of a bracket except the end head frame and realizing real-time alignment.
It can be seen that in fig. 11, the underground working condition is complicated, the data acquisition precision of the sensor is not high enough, and the situation of data loss exists when the upper computer acquires the data, if the data is directly put into the neural network for training, the obtained model has large relative error and is generally positioned above 20%; and the model fitting goodness is negative due to too poor data, so that the model fitting goodness cannot be displayed correctly, and an ideal model which can be directly used for realizing hydraulic support alignment cannot be directly obtained.
After the hydraulic support stroke data realized based on the dynamic programming algorithm is straightened, the neural network model is constructed, and the accuracy and the relative error of the model are greatly improved, wherein the relative error of the model is within 12%, and the maximum error is only 80.131mm (the fifth support of the third group of data). The model fitting goodness is more than 80%, wherein the lowest model fitting goodness is 82.6934%, and the highest model fitting goodness is 93.794%, and the model fitting goodness is greatly improved.
Claims (5)
1. The hydraulic support straightening method based on the dynamic programming algorithm fused with the BP neural network is characterized by comprising the following steps of:
step one: collecting coal mine underground discrete data: classifying, screening and cleaning the hydraulic support work discrete data collected underground coal mine, and screening travel data representing the real work state of the hydraulic support;
step two: data preprocessing, namely searching data conforming to the travel of the hydraulic support: the method comprises the steps that the stroke data of the hydraulic support obtained after cleaning is subjected to global searching for an optimal interval through a dynamic programming algorithm, the algorithm is utilized to realize the alignment of the discrete data of the stroke of the hydraulic support, and the discrete data of the stroke of the hydraulic support is determined to be the real stroke data of the hydraulic support in an ideal acquisition state;
step three: the problems of loss of processing data, insufficient accuracy and the like of a dynamic programming algorithm are utilized, and the hydraulic support stroke data are straightened: considering the coupling relation of a plurality of hydraulic support groups and a group, wherein the factor that error ranges exist between adjacent supports in the group, dividing the data of a certain complete hydraulic support advancing process according to the hydraulic support advancing time sequence, and determining various parameters of the neural network;
step four: establishing a BP time sequence neural network model, and establishing an optimal hydraulic support travel data model by multiple tuning parameters: building a BP neural network in an upper computer, and obtaining an optimal model through model training of multi-trip data time sequence prediction;
step five: based on an optimal model, the travel data acquired in real time are imported, and the hydraulic support is straightened in real time: when the coal mining machine starts to work, inputting stroke data corresponding to a hydraulic support where the current coal mining machine is located into a trained optimal model, obtaining a next group of strokes which are required to be actually pushed of the hydraulic support where the current coal mining machine is located, and sending a pushing instruction to a support controller of the next hydraulic support according to the strokes; realizing real-time alignment.
2. The hydraulic support alignment method based on the dynamic programming algorithm fused BP neural network according to claim 1, which is characterized in that: the method for preprocessing the discrete data collected by the underground coal mine, namely the fully-mechanized mining face comprises the following steps:
step one: travel data screening, timestamp format conversion and bracket state abnormal data processing: after acquiring the acquired discrete data of the actual working condition, data preprocessing is performed firstly, and positioning and searching of travel data and conversion of various related data formats are realized by utilizing a data analysis method;
step two: processing long-time motionless data and processing repeated data of a single bracket according to a time relation: cleaning again based on the data obtained in the step one, and cleaning the collected data in the same second and the data with unchanged time-increasing data value;
step three: dividing travel data of the single-travel hydraulic support: dividing the working condition based on the data obtained in the second step, finding out each group of data values which accord with single advancing of the whole hydraulic support group, and dividing the data values according to time;
step four: maximum and minimum difference positioning bracket travel: searching a maximum value and a minimum value based on each group of data values obtained in the third step, and taking the difference between the maximum value and the minimum value as the single-time travelling distance of the hydraulic support according to the actual working condition;
step five: based on the single travelling distance of the hydraulic support determined in the step four, optimizing based on a dynamic programming algorithm, determining an optimal interval, traversing data to find a start position and an end position of the optimal interval, taking the start position and the end position as data replacement standards, and cleaning data by combining with an adjacent frame error coupling relation; and obtaining the stroke data value of the hydraulic support in an ideal acquisition state under the required actual working condition, and realizing the straightening function of the stroke data.
3. The hydraulic support alignment method based on the dynamic programming algorithm fused BP neural network according to claim 1, which is characterized in that: the dynamic programming algorithm comprises the following steps:
step one: defining an array through initial hydraulic support travel data to store variables calculated in each interval;
step two: calculating the maximum value of a single interval, adjacent frame difference values and other variables according to working conditions;
step three: judging whether the acquired data accords with ideal acquired data, if not, returning to the execution step II; if yes, executing the fourth step;
step four: a plurality of high-quality intervals are compared and found out to find out an optimal interval;
step five: and correcting other data outside the interval according to the adjacent frame relation.
4. The hydraulic support alignment method based on the dynamic programming algorithm fused BP neural network according to claim 1, which is characterized in that: the generation process based on BP neural network time sequence prediction comprises the following steps:
step one: searching data of a global optimal solution based on a dynamic programming algorithm, dividing single-group travel data of the hydraulic support according to the group follow-up relation of the hydraulic support and the adjacent frame coupling relation of the hydraulic support under actual working conditions, setting an output layer, an input layer and determining the number of hidden layer nodes according to continuous debugging after dividing;
step two: setting a training set and a testing set by using single-set travel data of the hydraulic support, wherein 70% of data are used as training data, 30% of data are used as testing data, and training is performed in Matlab by using a written BP neural network;
step three: adjusting data to normalize the data to a (0, 1) interval, setting parameters such as training algebra, training function, learning rate, minimum error and the like, determining each optimal value through continuous debugging, further obtaining a result and deviation of the whole BP time sequence prediction, and continuously debugging to obtain an optimal model;
step four: training the acquired real-time hydraulic support data based on the optimal model obtained by training in the step four to obtain the next group of strokes of the hydraulic support where the current coal mining machine is located, which are required to be actually pushed, and sending pushing instructions to a support controller of the next hydraulic support according to the strokes to realize real-time alignment.
5. The hydraulic support alignment method based on the dynamic programming algorithm fused BP neural network according to claim 1, which is characterized in that: the hydraulic support real-time alignment comprises the following steps:
step one: the upper computer is put into the algorithm and the neural network model of the invention;
step two: acquiring hydraulic support travel data at the previous moment;
step three: training to obtain an optimal model and storing;
step four: predicting the displacement of the hydraulic stroke at the next moment based on the model;
step five: and the bracket controller receives the adjustment instruction to complete the real-time straightening of the bracket.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586835.9A CN116628880A (en) | 2023-05-23 | 2023-05-23 | Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586835.9A CN116628880A (en) | 2023-05-23 | 2023-05-23 | Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116628880A true CN116628880A (en) | 2023-08-22 |
Family
ID=87591455
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310586835.9A Pending CN116628880A (en) | 2023-05-23 | 2023-05-23 | Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116628880A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117473313A (en) * | 2023-10-30 | 2024-01-30 | 长江师范学院 | Fewer-sensing estimation system and method for pushing and shifting gestures of hydraulic support group |
-
2023
- 2023-05-23 CN CN202310586835.9A patent/CN116628880A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117473313A (en) * | 2023-10-30 | 2024-01-30 | 长江师范学院 | Fewer-sensing estimation system and method for pushing and shifting gestures of hydraulic support group |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114329810B (en) | Real-time prediction method for working posture of shield tunneling machine based on big data | |
| LU501938B1 (en) | Method and system for intelligent analysis of big data on unmanned mining in mine | |
| CN106194181B (en) | Intelligent working face coal-rock interface recognition method based on geological data | |
| CN116628880A (en) | Hydraulic support straightening method based on dynamic programming algorithm fused BP neural network | |
| CN109903383B (en) | Method for accurately positioning coal mining machine in three-dimensional model of working face coal seam | |
| CN111365015A (en) | Shield tunneling parameter feature extraction and attitude deviation prediction method based on XGboost | |
| CN106295873B (en) | A kind of Grey Markov Chain trajectory predictions method towards coal cutter memorized cutting | |
| CN111723448A (en) | Digital twin intelligent fully mechanized coal mining face hydraulic support straightness monitoring method | |
| CN110929384A (en) | Mine pressure big data real-time analysis system and method based on fully mechanized coal mining face | |
| CN110728003B (en) | Intelligent prediction method for hydraulic support pressure of working face of underground coal mine | |
| CN112668109B (en) | Method for establishing cutting route model of fully-mechanized coal mining face | |
| CN110107295B (en) | Unmanned coal mining path planning method based on working face high-precision navigation model | |
| CN113128109A (en) | Test and evaluation method for intelligent fully-mechanized mining robot production system | |
| CN110847968A (en) | Periodic pressure-bearing dividing method based on maximum working resistance curve of support cycle | |
| CN113689032A (en) | Multi-sensor fusion gas concentration multi-step prediction method based on deep learning | |
| CN116108339A (en) | Actual measurement data outlier detection and correction method for tunnel boring machine | |
| CN106256991A (en) | A kind of coal cutter memorized cutting and memory location combination learning method | |
| CN106194180B (en) | Method for predicting heightening track of coal cutter based on sliding window and gray Markov chain | |
| CN113821917B (en) | Method for estimating shape of top and bottom plates of mined area in operation process of fully mechanized mining equipment | |
| CN103439964A (en) | On-line updating system and method for OCSVM monitoring model | |
| CN112036609B (en) | Dynamic prediction method for water inflow of coal mine working face based on multi-order power system model | |
| CN112149912A (en) | Coal mine working face mine pressure data prediction method based on reverse propagation neural network | |
| CN117348500B (en) | Automatic control method and system for fully-mechanized coal mining face | |
| CN112003887A (en) | Cloud-edge collaborative deep learning device for industrial internet time sequence data prediction | |
| CN114194745A (en) | Scraper conveyor straightness control algorithm based on inertial navigation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |