CN112781644A - Neural network technology-based mining sensor calibration-free method - Google Patents
Neural network technology-based mining sensor calibration-free method Download PDFInfo
- Publication number
- CN112781644A CN112781644A CN202110108809.6A CN202110108809A CN112781644A CN 112781644 A CN112781644 A CN 112781644A CN 202110108809 A CN202110108809 A CN 202110108809A CN 112781644 A CN112781644 A CN 112781644A
- Authority
- CN
- China
- Prior art keywords
- bpnn
- output
- network
- layer
- fitting
- 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 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 25
- 238000005516 engineering process Methods 0.000 title claims abstract description 9
- 238000005065 mining Methods 0.000 title claims description 24
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 11
- 230000008569 process Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 15
- 239000000523 sample Substances 0.000 claims description 7
- 230000005284 excitation Effects 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 abstract description 15
- 238000005259 measurement Methods 0.000 abstract description 13
- 238000012937 correction Methods 0.000 abstract description 11
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000000737 periodic effect Effects 0.000 abstract description 2
- 239000003245 coal Substances 0.000 description 6
- 230000001537 neural effect Effects 0.000 description 6
- 238000013459 approach Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000002068 genetic effect Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000002922 simulated annealing Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Abstract
The invention relates to a regulation-free method for a mine sensor based on a neural network technology, which comprises the following steps: training a network, constructing BPNN, training the BPNN, predicting the BPNN and processing an MCU. The BPNN is adopted to automatically obtain fitting parameters of nonlinear correction, so that the manual fitting process is reduced; the BPNN sensor based on the single chip microcomputer can realize calibration-free, and the problems of multiple or multi-point calibration, periodic calibration and the like caused by a traditional fitting mode are avoided; manual adjustment is not needed in the field application and production links, and special calibration instruments and equipment are not needed, so that the field maintenance work difficulty is reduced, and the production efficiency is improved; under-fitting or over-fitting caused by curve fitting of traditional data can be eliminated, and the measurement precision of the sensor is improved.
Description
The technical field is as follows:
the invention relates to the technical field of coal mine safety, in particular to a regulation-free method for a mine sensor based on a neural network technology.
Background art:
for coal mine safety monitoring, a mine sensor is the leading edge guarantee of coal mine safety production, and performance indexes such as reliability, stability and accuracy play an important role in system working quality and mine safety. Due to the influence of factors such as the characteristics of the sensor, the manufacturing process, the environmental change and the like, the relation between the measured parameters and the output of the sensor is generally nonlinear, and most mining sensor manufacturers adopt a traditional simple curve fitting correction method, so that the problems of complex adjustment and calibration process, frequent calibration times, low precision and the like exist in the sensor. Therefore, in the application of a coal mine field, the sensor needs to be regularly calibrated to ensure the stability of the measurement of the sensor, and is limited to the severe underground environment, which undoubtedly increases the maintenance workload of field personnel, brings certain difficulty to the calibration and maintenance work of the field personnel, and the sensor has various types, complex calibration work flow and lack of required calibration instruments, equipment or environments, such as a standard wind tunnel is required to be equipped for the calibration of the mining wind speed sensor, the calibration environment does not exist in the common coal mine field, the calibration flow is complex, and calibration personnel can easily remember the wrong calibration sequence.
Similarly, in the production link of an enterprise, the production line efficiency will be reduced by a complicated adjustment process, taking a mining wind speed sensor as an example, the sensor should be placed in a standard wind tunnel to wait for parameters such as zero adjustment, linearity adjustment and the like after the wind speed is stable, and the average production adjustment time counted by each wind speed sensor is about 30 minutes, so that the production speed is greatly reduced.
For this purpose, the non-linearity compensation and correction are performed by some hardware or software method. A few manufacturers improve the structure of hardware, and aim to provide a scheme for reducing the complexity of calibration work, such as providing a special gas storage device for a gas mining sensor or providing a gas path short-circuit device for a pressure sensor, a wind speed sensor or a flow rate sensor, and the like. Meanwhile, in the software method, the traditional fitting mode forces research and development personnel to dig data information to search for a proper calibration point, and the traditional fitting method is continuously corrected, so that double guarantees on adjustment and precision cannot be achieved.
Since the 40 th generation of the 20 th century, the self-adaptive and self-learning capabilities are gradually proposed, and the BPNN has the function of realizing any complex nonlinear mapping as a machine learning algorithm. The method is particularly suitable for solving the problem with a complex internal mechanism, reasonable solving rules can be automatically extracted by learning an example set with correct answers, and fitting parameters of nonlinear correction can be automatically acquired, namely the method has self-learning capability. The BPNN is a multi-layer feedforward neural network, and the main characteristics of the network are signal forward transmission and error backward propagation. In forward pass, the input signal is processed layer by layer from the input layer through the hidden layer to the output layer. The neuronal state of each layer only affects the neuronal state of the next layer. If the output layer can not obtain the expected output, the backward propagation is carried out, and the network weight and the broad value are adjusted according to the prediction error, so that the BPNN predicted output continuously approaches to the expected output.
In order to reduce the complexity of calibration and improve the measurement accuracy of the sensor, a BPNN model is introduced into data fitting of the mining sensor, original data obtained by the sensor is processed, analyzed and output to obtain a corresponding measurement value, the BPNN model is obtained and transplanted into a single chip microcomputer, the calibration-free performance of the mining sensor is realized, the defects of a traditional fitting mode are avoided, and the pressure of site and production is reduced. Meanwhile, the limitation caused by curve fitting of traditional data can be eliminated, and the measurement precision of the mining sensor is improved.
At present, curve fitting method is mostly adopted in searching data corresponding relation of mining sensors, a continuous curve is used for approximately describing or comparing a functional relation between coordinates represented by a group of discrete points on a plane, and the method is a method for approximating discrete data by an analytical expression, such as least square fitting, polynomial fitting method, cubic spline curve fitting and the like.
Although the curve fitting method is widely applied to mining sensors, such as the least square method is simple and convenient to implement, if the fitting mode is not properly selected, a large deviation is generated, particularly for the fitting of a complex curve, and if the mode is selected incorrectly, the fitting effect is poor. The sensor calibration is complicated, and the phenomena of over-fitting and under-fitting easily occur when the selection of the calibration points is improper.
The invention content is as follows:
in order to realize the calibration-free sensor and improve the measurement accuracy of the sensor, a mathematical model among characteristic parameters of the sensor is required to be searched, the output characteristic of the mathematical model is subjected to fitting or nonlinear compensation, and the introduction of the BPNN model is very necessary to ensure the accuracy of a measurement and control system of the sensor and realize intelligent control. The BPNN model is a machine learning algorithm, an algorithm that can learn from data, and is capable of automatically acquiring fitting parameters for non-linear correction, unlike numerical analysis and curve fitting methods.
The most important link of the whole method is the design of the BPNN model, because the BPNN model trained and finished through representative training samples can realize infinite approximation of complex relations among characteristic parameters of the sensor, and truly reflect measured values. Meanwhile, the problem of calibration failure caused by over-fitting and under-fitting brought by a curve fitting mode is solved.
The BPNN is a multi-layer feedforward neural network, and the main characteristics of the network are signal forward transmission and error backward propagation. In forward pass, the input signal is processed layer by layer from the input layer through the hidden layer to the output layer. The neuronal state of each layer only affects the neuronal state of the next layer. If the output layer can not obtain the expected output, the backward propagation is carried out, and the network weight and the broad value are adjusted according to the prediction error, so that the BPNN predicted output continuously approaches to the expected output.
The invention aims to fit original data of a mining sensor in a coal mine safety monitoring system and design nonlinear compensation, so that the sensor can realize calibration-free and improve measurement accuracy, and aims to provide a design of a BPNN (pulse coupled neural network) mining sensor calibration-free method based on a single chip microcomputer.
The invention provides a neural network technology-based adjustment-free method for a mining sensor, which comprises the following steps:
firstly, training a network: the network has associative memory and prediction capability through training;
second, BPNN construction: determining a BPNN structure according to wind speed fitting characteristics, wherein an input layer has 1 node, a hidden layer has 8 nodes, and an output layer has 1 node;
and thirdly, BPNN training: the BPNN selects the measuring points to test, obtains a plurality of groups of pressure data through different probe tests, sets measuring point labels for the pressure data, breaks up the sequence, selects part of the data to train the network, and tests the rest data for testing the fitting performance of the network;
and fourthly, predicting BPNN: outputting by using a trained network prediction function, and analyzing a prediction result;
and fifthly, MCU processing: and converting the trained network model into a C language code, writing the C language code into the singlechip, performing normalization processing on the signal when the singlechip collects the signal, transmitting the signal to the trained BPNN model for calculation, performing inverse normalization on a calculation result to form an output value with dimension, and transmitting the output value to the next stage through the TTL module.
In the first step, the training process comprises the following steps:
step l: initializing the network, determining the number n of nodes of the input layer, the number L of nodes of the hidden layer and the number of nodes of the output layer according to the system input and output sequence X, YCounting m, initializing connection weight omega among neurons of input layer, hidden layer and output layerij，ωjkInitializing a hidden layer threshold alpha and an output layer threshold b, and setting a learning rate and a neuron excitation function;
step 2: calculating the output of the hidden layer, and connecting the input layer and the hidden layer according to the input variable X and the weight omegaijAnd a hidden layer threshold alpha, calculating a hidden layer output H,
in the formula, L is the number of hidden layer nodes; f is a hidden layer excitation function having multiple expression forms, and the selected function is
And step 3: output layer output calculation, according to hidden layer output H, connecting weight omegajkAnd a threshold b, calculating the BPNN predicted output O,
and 4, step 4: error calculation, calculating a net prediction error e based on the net prediction output O and the expected output Y,
ek＝Yk-Ok k＝1,2,...,m (4)；
and 5: updating the weight value, namely updating the network connection weight value omega according to the network prediction error eij，ωjk，
wjk＝wjk+ηHjek j＝1,2,...,L；k＝1,2,...,m (6)
In the formula, η is the learning rate;
step 6: and updating the values of the network nodes, namely updating the threshold values alpha and b of the network nodes according to the network prediction error e.
bk＝bk+ek k＝1,2,...,m (8)；
And 7: and (4) judging whether the iteration of the algorithm is finished or not, and if not, returning to the step 2.
The invention has the beneficial effects that:
(1) the BPNN is adopted to automatically obtain fitting parameters of nonlinear correction, so that the manual fitting process is reduced;
(2) the BPNN sensor based on the single chip microcomputer can realize calibration-free, and the problems of multiple or multi-point calibration, periodic calibration and the like caused by a traditional fitting mode are avoided;
(3) manual adjustment is not needed in the field application and production links, and special calibration instruments and equipment are not needed, so that the field maintenance work difficulty is reduced, and the production efficiency is improved;
(4) under-fitting or over-fitting caused by curve fitting of traditional data can be eliminated, and the measurement precision of the sensor is improved.
Description of the drawings:
FIG. 1 is a BPNN topology block diagram;
FIG. 2 is a flow chart of a data fitting algorithm for BPNN-based mining wind speed sensor probe acquisition of the present invention;
FIG. 3 is a flow chart of the MCU processing of the present invention;
FIG. 4 is a graph of a BPNN network training of the present invention;
FIG. 5 is a schematic error diagram of the BPNN prediction output of the present invention.
The specific implementation mode is as follows:
the following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention more readily understood by those skilled in the art, and thus will more clearly and distinctly define the scope of the invention.
Under normal conditions, the input-output fitting relation of the mining sensor is a curve between input and output found through experience after multiple measurements. Due to variability between probes, a calibration point is required to adjust the curve to achieve the desired output. The curve fitting correction method enables the sensor to have the problems of large adjusting workload, easy calibration error and the like no matter when the sensor leaves factory for calibration or mine calibration. A scientific method for processing signals of the sensor, namely neural network fitting, is provided, a mathematical model among characteristic parameters of the sensor is sought, fitting or nonlinear compensation is carried out on the output characteristic of the mathematical model, and the accuracy of a sensor measurement and control system is ensured and intelligent control is realized.
In the following, the mining wind speed sensor is taken as an example, and when the traditional fitting mode is applied to the wind speed sensor, each measuring point of each probe needs to be continuously subjected to inverse calculation so as to find a proper curve and calibration points. The method has the following defects:
(1) sacrificing the accuracy of some measuring points, for example, using 9m/s for calibration, the conditions of good performance of low-end wind speed and poor performance of high-end wind speed can occur, and compensation needs to be performed on the conditions, as shown in table 1 below;
table 1: wind speed corresponding value under 9m/s calibration
(2) The production efficiency is low, for example, about 40 minutes is required for a common wind speed sensor to test (including zero point and linearity adjustment in production), which causes unnecessary time waste in production;
(3) the use of the mine is complicated, and when the upper computer reminds workers to calibrate the sensor, the output is abnormal due to the fact that the workers are not familiar with the use mode of the sensor, and adjustment errors can be indirectly caused.
The above problems are summarized as follows:
(1) in the traditional fitting mode, under the nonlinear condition, a compensation procedure may be required to be added;
(2) the added calibration function brings low efficiency and wrong calibration problems to production and workers.
(3) With the development of sensor intellectualization, the intelligent calculation is more and more widely applied, a BP neural network (Back Propagation) uses the most extensive model in the practical application of an artificial neural network, the outstanding advantages of the BP neural network are that the BP neural network has very strong classification and nonlinear mapping capabilities, the problem that the traditional sensor cannot realize high-precision fitting is solved, and the time of experience processing can be greatly prolonged and the use mode can be simplified when the BP neural network is used for sensor fitting.
The BPNN is a multi-layer feedforward neural network, and the main characteristics of the network are signal forward transmission and error backward propagation. In forward pass, the input signal is processed layer by layer from the input layer through the hidden layer to the output layer. The neuronal state of each layer only affects the neuronal state of the next layer. If the output layer can not obtain the expected output, the backward propagation is carried out, and the network weight and the broad value are adjusted according to the prediction error, so that the BPNN predicted output continuously approaches to the expected output. The BPNN topology is shown in fig. 1.
In fig. 1, X1, X2.., Xn is an input value of BPNN, Y1, Y2.., Ym is a predicted value of BPNN, ω isijAnd ωjkIs the BPNN weight. As can be seen from fig. 1, the BPNN may be viewed as a non-linear function with the network input values and predicted values being the independent and dependent variables of the function, respectively. When the number of input nodes is n and the number of output nodes is m, the BPNN expresses a function mapping relationship from n independent variables to m dependent variables.
The mining sensor calibration-free method based on the neural network technology comprises the processes of network training, data fitting, MCU processing and the like.
The BPNN prediction is firstly to train the network, and the network has associative memory and prediction capability through training. The BPNN training process includes the following steps:
step l: and (5) initializing the network. Determining the number n of nodes of the input layer, the number L of nodes of the hidden layer and the number m of nodes of the output layer of the network according to the input and output sequence (X, Y) of the system, and initializing the connection weight omega among neurons of the input layer, the hidden layer and the output layerij，ωjkInitializing a hidden layer threshold alpha, outputting a layer threshold b, and giving a learning rate and a neuron excitation function.
Step 2: the hidden layer outputs the computation. According to the input variable X, the connection weight omega between the input layer and the hidden layerijAnd a hidden layer threshold α, calculating a hidden layer output H.
In the formula, L is the number of hidden layer nodes; f is a hidden layer excitation function having multiple expression forms, and the selected function is
And step 3: the output layer outputs the calculation. According to the hidden layer output H, connecting the weight omegajkAnd a threshold b, calculating the BPNN predicted output O.
And 4, step 4: and (4) error calculation. And calculating the network prediction error e according to the network prediction output O and the expected output Y.
ek＝Yk-Ok k＝1,2,...,m (4)
And 5: and updating the weight value. Updating the network connection weight omega according to the network prediction error eij，ωjk。
wjk＝wjk+ηHjek j＝1,2,...,L；k＝1,2,...,m (6)
Where η is the learning rate.
Step 6: and (4) updating the aperture value. And updating the network node threshold values alpha and b according to the network prediction error e.
bk＝bk+ek k＝1,2,...,m (8)
And 7: and (4) judging whether the iteration of the algorithm is finished or not, and if not, returning to the step 2.
The BPNN-based data fitting algorithm process for the mining wind speed sensor probe acquisition can be divided into three steps, namely BPNN construction, BPNN training and BPNN prediction, as shown in FIG. 2.
The BPNN structure is determined according to wind speed fitting characteristics, and the BPNN structure is 1-8-1 due to the fact that the corresponding function of the mining wind speed sensor has one input parameter and one output parameter, namely the input layer has 1 node, the hidden layer has 8 nodes, and the output layer has 1 node.
The BPNN selects the measuring points of 0.4, 1, 2, 3, 6, 9, 12 and 15m/s for testing, 200 groups of pressure data are obtained through different probe tests, measuring point labels are arranged for the two hundred groups of data, and the sequence is disordered. 150 groups were selected for network training and 50 groups were tested for testing the fit performance of the network.
The BPNN prediction is output by using a trained network prediction function, and the prediction result is analyzed.
The trained network model is converted into a C language code and written into the single chip microcomputer, when the single chip microcomputer collects signals, the signals are subjected to normalization processing and are transmitted to the trained BPNN model for calculation, the calculation result is subjected to inverse normalization to form an output value with dimension, the output value is transmitted to the next stage through the TTL module, and the processing flow is shown in figure 3.
In Matlab, 150 sets of samples were trained on BP neural networks, respectively. The BP neural network converges after 7 times of training, and the training curve is as shown in fig. 4 (when training 10 times, the curves are Validation, Test, Train from top to bottom in sequence).
In the wind speed detection range, 0.4, 1, 2, 3, 6, 9, 12 and 15m/s are selected as detection points, the fitting result of the BP neural network is shown in a table 2, and comparison shows that if the zero setting measures of the low wind speed section are not good enough, the error is possibly large. Generally speaking, the fitting error range is small, as shown in fig. 5, the fitting error range of the BP neural network is +/-0.15m/s (the error unit is m/s), the fitting effect is good, the problem that the traditional fitting high-end effect is not good is solved, and the output signal can reach the measurement error range without calibration.
Table 2: fitting output
The result shows that compared with the traditional fitting mode, the algorithm does not need calibration, has high approximation precision, is simple to operate and has less workload. In general, the neural network idea is applied to the wind speed sensor, so that the application range of a neural network model is widened, a new method and idea are provided for nonlinear correction of the wind speed sensor, and reference significance is provided for curve fitting and error correction of other mining sensors.
The invention can be based on BPNN and carry out optimization algorithm, RBFNN can approach to any nonlinear function under any precision, the convergence speed is fast, and the problem of local minimum is not existed, thus becoming another forward neural network for replacing BPNN; the FRBFNN is used as a mixed network, combines the characteristics of RBFNN and fuzzy inference, has high algorithm speed, high stability and high precision, and can effectively perform nonlinear compensation on the sensor; the GA-NN can realize the nonlinear correction of the self and external environmental factors, and the complexity of an algorithm program is difficult to meet the realization of hardware, and the real-time performance is poor; the GSA-GA utilizes a simulated annealing algorithm to control variation and cross operation, thereby effectively preventing premature convergence of a basic genetic algorithm, improving the genetic algorithm to carry out nonlinear correction on the sensor, and having fitting accuracy far higher than that of the least square method which is applied most widely in the prior art and stronger robustness; the optimal value of the parameters of the LSSVM model can be found by utilizing the improved PSO algorithm, the nonlinear compensation effect of the LSSVM model on the sensor is obviously improved, and the measurement precision is improved. Other optimization methods also include genetic algorithms, particle swarm optimization algorithms, ant colony algorithms, simulated annealing algorithms, and the like. The optimization algorithms can realize fitting and nonlinear compensation of sensor data, but the algorithms are slow in convergence and large in calculation amount, and are not suitable for being transplanted to a single-chip microcomputer end for use.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (3)
1. A mining sensor calibration-free method based on a neural network technology is characterized by comprising the following steps:
firstly, training a network: the network has associative memory and prediction capability through training;
second, BPNN construction: determining a BPNN structure according to wind speed fitting characteristics, wherein an input layer has 1 node, a hidden layer has 8 nodes, and an output layer has 1 node;
and thirdly, BPNN training: the BPNN selects the measuring points to test, obtains a plurality of groups of pressure data through different probe tests, sets measuring point labels for the pressure data, breaks up the sequence, selects part of the data to train the network, and tests the rest data for testing the fitting performance of the network;
and fourthly, predicting BPNN: outputting by using a trained network prediction function, and analyzing a prediction result;
and fifthly, MCU processing: and converting the trained network model into a C language code, writing the C language code into the singlechip, performing normalization processing on the signal when the singlechip collects the signal, transmitting the signal to the trained BPNN model for calculation, performing inverse normalization on a calculation result to form an output value with dimension, and transmitting the output value to the next stage through the TTL module.
2. The neural network technology-based mining sensor calibration-free method according to claim 1, wherein in the first step, the training process comprises the following steps:
step l: network initialization, namely determining the number n of nodes of an input layer, the number L of nodes of a hidden layer and the number m of nodes of an output layer of the network according to a system input and output sequence X, Y, and initializing a connection weight omega among neurons of the input layer, the hidden layer and the output layerij，ωjkInitializing a hidden layer threshold alpha and an output layer threshold b, and setting a learning rate and a neuron excitation function;
step 2: calculating the output of the hidden layer, and connecting the input layer and the hidden layer according to the input variable X and the weight omegaijAnd a hidden layer threshold alpha, calculating a hidden layer output H,
in the formula, L is the number of hidden layer nodes; f is a hidden layer excitation function having multiple expression forms, and the selected function is
And step 3: output layer output calculation, according to hidden layer output H, connecting weight omegajkAnd a threshold b, calculating the BPNN predicted output O,
and 4, step 4: error calculation, calculating a net prediction error e based on the net prediction output O and the expected output Y,
ek＝Yk-Ok k＝1,2,...,m (4)；
and 5: updating the weight value, namely updating the network connection weight value omega according to the network prediction error eij，ωjk，
wjk＝wjk+ηHjek j＝1,2,...,L；k＝1,2,...,m (6)
In the formula, η is the learning rate;
step 6: and updating the values of the network nodes, namely updating the threshold values alpha and b of the network nodes according to the network prediction error e.
bk＝bk+ek k＝1,2,...,m (8)；
And 7: and (4) judging whether the iteration of the algorithm is finished or not, and if not, returning to the step 2.
3. The neural network technology-based mining sensor tuning-free method according to claim 1, wherein in the third step, the BPNN selects 0.4, 1, 2, 3, 6, 9, 12, 15m/s measuring points to test, so as to obtain 200 groups of pressure data, selects 150 groups of data to perform network training, and selects 50 groups of data to perform testing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110108809.6A CN112781644A (en) | 2021-01-27 | 2021-01-27 | Neural network technology-based mining sensor calibration-free method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110108809.6A CN112781644A (en) | 2021-01-27 | 2021-01-27 | Neural network technology-based mining sensor calibration-free method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112781644A true CN112781644A (en) | 2021-05-11 |
Family
ID=75758161
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110108809.6A Pending CN112781644A (en) | 2021-01-27 | 2021-01-27 | Neural network technology-based mining sensor calibration-free method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112781644A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107063509A (en) * | 2017-04-26 | 2017-08-18 | 深圳市相位科技有限公司 | A kind of thermosensitive thermometer calibration method based on neutral net |
| CN108332837A (en) * | 2017-12-06 | 2018-07-27 | 芯海科技（深圳）股份有限公司 | It is a kind of applied to electronic scale based on BP neural network compensation temperature gain method |
| CN109709352A (en) * | 2018-12-20 | 2019-05-03 | 中煤科工集团重庆研究院有限公司 | A kind of underground coal mine air velocity transducer monitoring accuracy sentences knowledge and adjusting process in situ |
| DE102019113539A1 (en) * | 2019-05-21 | 2020-11-26 | Tdk Electronics Ag | Method and device for operating a multi-gas sensor |
-
2021
- 2021-01-27 CN CN202110108809.6A patent/CN112781644A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107063509A (en) * | 2017-04-26 | 2017-08-18 | 深圳市相位科技有限公司 | A kind of thermosensitive thermometer calibration method based on neutral net |
| CN108332837A (en) * | 2017-12-06 | 2018-07-27 | 芯海科技（深圳）股份有限公司 | It is a kind of applied to electronic scale based on BP neural network compensation temperature gain method |
| CN109709352A (en) * | 2018-12-20 | 2019-05-03 | 中煤科工集团重庆研究院有限公司 | A kind of underground coal mine air velocity transducer monitoring accuracy sentences knowledge and adjusting process in situ |
| DE102019113539A1 (en) * | 2019-05-21 | 2020-11-26 | Tdk Electronics Ag | Method and device for operating a multi-gas sensor |
Non-Patent Citations (1)
| Title |
|---|
| 梁杰等: "基于L-M算法的反向传播网络的湿度传感器输出误差补偿研究", 《计算机测量与控制》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107688871B (en) | Water quality prediction method and device | |
| CN109407654B (en) | Industrial data nonlinear causal analysis method based on sparse deep neural network | |
| CN111047012A (en) | Air quality prediction method based on deep bidirectional long-short term memory network | |
| US20220083839A1 (en) | Accuracy compensation method for discharge caustic alkali concentration measuring device in evaporation process | |
| WO2022021726A1 (en) | Pmu-based power system state estimation performance evaluation method | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN111476278B (en) | Intelligent gas concentration detection system | |
| CN116150897A (en) | Machine tool spindle performance evaluation method and system based on digital twin | |
| CN110222387B (en) | Multi-element drilling time sequence prediction method based on mixed leaky integration CRJ network | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN110597070A (en) | Method for identifying model parameters of thermal power generating unit system | |
| WO2021114320A1 (en) | Wastewater treatment process fault monitoring method using oica-rnn fusion model | |
| CN110309609A (en) | A kind of architecture indoor air quality evaluation method based on rough set and wavelet neural network | |
| CN116432123A (en) | Electric energy meter fault early warning method based on CART decision tree algorithm | |
| CN112819087B (en) | Method for detecting abnormality of BOD sensor of outlet water based on modularized neural network | |
| CN113051806A (en) | Water quality BOD measuring method based on AQPSO-RBF neural network | |
| CN112781644A (en) | Neural network technology-based mining sensor calibration-free method | |
| CN117034808A (en) | Natural gas pipe network pressure estimation method based on graph attention network | |
| CN116662925A (en) | Industrial process soft measurement method based on weighted sparse neural network | |
| CN110990938A (en) | Soft measurement method for rock digging state of tunnel face | |
| CN116109039A (en) | Data-driven anomaly detection and early warning system | |
| CN113836794B (en) | Soft and hard combined fly ash carbon content online monitoring method | |
| CN112116003B (en) | Self-adaptive dynamic fault diagnosis method and system for complex sensor network | |
| CN111881040A (en) | Test data generation method of abstract state model based on recurrent neural network | |
| CN111310907A (en) | Microwave assembly fault diagnosis method, device and equipment |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210511 |