CN103712599B - A kind of relative height measurement mechanism and method - Google Patents

Abstract

The invention discloses a kind of relative height measurement mechanism, the array gas pressure sensor module, power module, LCD MODULE, the serial communication modular that comprise host computer, microprocessor module and be connected respectively with microprocessor module.Adopt the method that described measurement mechanism is measured, in host computer, complete genetic algorithm optimization BP neural network and gained weights, threshold value and relative height computing formula are transferred to microprocessor module; Array gas pressure sensor module gathers air pressure and temperature transfers to microprocessor module, microprocessor module calculates relative height after processing the air pressure received and temperature, result is transferred to LCD MODULE carries out showing, host computer stores.The present invention, by software and hardware combining, is conducive to restraint speckle, recovery, enhancing and extraction useful signal, has higher accuracy and stability.

Description

A kind of relative height measurement mechanism and method

Technical field

The present invention relates to a kind of relative height measurement mechanism and method, particularly a kind of relative height measurement mechanism based on genetic algorithm optimization BP neural network and method.

Background technology

In daily life and production run, the relative height vertical height of any two points (in the space) is for being indispensable basic parameter us, such as, relative height is lived at us, building, science and technology, even military affairs all have a wide range of applications, as field exploration, building ground elevation carrection, unmanned robot and the navigation of target missile height etc.In general, the measurement for relative height has three kinds of conventional methods usually, the first: traditional machinery is directly measured; The second: the measurement utilizing GPS; The third: is based on the indirect inspection of pneumatically-sensed electronic device.The limited precision of traditional mechanical measurement device, volume is large, carries inconvenience.The measurement of GPS can reach good accuracy requirement, but cost is higher.Comparatively speaking, the electronic device based on pneumatically-sensed relative height has to be applied widely.

Present stage, for utilizing air pressure to calculate relative height, people generally adopt standard pressure altitude formula.But there is larger defective in standard pressure altitude conversion formula, such as, atmospheric pressure around measurement point does not meet the atmospheric state of ideal standard, and (pressure is 1013.25hPa, temperature is 15 DEG C), adopt standard pressure altitude formula comparatively large by the impact of the extraneous factors such as temperature, wind speed, humidity, thus cause elevation carrection to produce larger error.In addition, be all generally adopt single-sensor to measure, because certain some air pressure may to alter a great deal and sensor dispatches from the factory the slightly difference such as parameter, these factors all add measuring error.

Summary of the invention

Technical matters to be solved by this invention is to provide a kind of relative height measurement mechanism and method, is conducive to restraint speckle, recovery, enhancing and extraction useful signal, has higher accuracy and stability.

The present invention is for solving the problems of the technologies described above by the following technical solutions:

A kind of relative height measurement mechanism, the array gas pressure sensor module, power module, LCD MODULE, the serial communication modular that comprise host computer, microprocessor module and be connected with microprocessor module respectively; Described host computer is connected with described microprocessor module by described serial communication modular; Described power module is that modules is powered.

Adopt a measuring method for relative height measurement mechanism described above, comprise the following steps:

Step 1, described microprocessor module sends startup command to described array gas pressure sensor module, described array gas pressure sensor module gathers air pressure and the temperature data of n different testing location, and transferring to described microprocessor module, the air pressure received and temperature data are sent in described host computer by described microprocessor module;

Step 2, described host computer sets up BP neural network by the air pressure that receives and temperature data, being optimized BP neural network according to the principle of genetic algorithm and completing its training study process, the relative height computing formula obtained be sent in described microprocessor module, concrete steps are as follows:

201, using the atmospheric pressure value P that collects and temperature value t as input quantity P _kthe parameter of=(P, t); Choose a reference mark, record the relative height △ H of each testing location, it can be used as output quantity C _kthe parameter of=(△ H); The input quantity P of same testing location _kwith output quantity C _kthere is mapping corresponding relation, obtain the test data set that n has above-mentioned mapping relations thus, it can be used as test sample book stored in database; N is natural number, and K is any one group in n group test sample book, K=1,2 ..., n;

202, be normalized test sample book, all data are all converted into the number between [0,1] by employing minimax method, and normalization formula is as follows:

χ _k=（χ _k-χ _min）/(χ _max-χ _min)

In formula, χ _minfor the minimum number in data sequence; χ _maxfor the maximum number in data sequence;

According to normalization formula by input quantity P _kwith output quantity C _kall convert normalized input quantity P ' to _kwith output quantity C ' _k, finally obtain normalized test sample book and database;

203, according to the mapping corresponding relation of input quantity and output quantity in the test sample book after normalization, set up the BP neural network of three etale topology structures, specific as follows:

The input layer of setting BP neural network is 2, and output layer neuron is 1, and hidden layer neuron is 6; Wherein the activation function of input layer and hidden layer all chooses tansig type function, and output layer activation function chooses pureline type function; Connection weights between input layer and hidden layer are W _ij, the threshold value of hidden layer is θ _j, the connection weights between hidden layer and output layer are W _jq, the threshold value of output layer is θ _q; I is input layer number, i=1,2; J is node in hidden layer, j=1,2 ..., 6; Q is output layer nodes, q=1;

Tansig type function is defined as

$f\left(\mathrm{\χ}\right)=\frac{2}{1+\exp (-{2S}_j)}-1$

In formula, S _jthe input of j hidden layer node, a _ifor the matrix that atmospheric pressure value and temperature value are formed;

Pureline type function is linear function, and wherein, independent variable is the output of hidden layer node, and dependent variable is the output of output layer node;

204, using the neural network of BP described in step 3 as blackbox model, set up with the input variable of the input quantity in test sample book described in step 1 for BP neural network, relative height value using correspondence as the BP neural network prediction model of output variable;

205, random initializtion is carried out to the weights W in BP neural network prediction model and threshold value θ, makes BP neural network model have the most basic predicted condition; W, θ are respectively W _ijand W _jq, θ _jand θ _qgeneral name;

206, from normalized database, transfer input quantity P ' _k, as the input variable of BP neural network prediction model with most fundamental forecasting condition, thus obtain and input variable P ' _kthe prediction output quantity Y ' of the relative height that the BP neural network prediction model that mapping pair is answered exports _kdata group;

207, the threshold value determined by test sample book normalized described in step 202 and the scope of weights, according to initial threshold value and the weights of the principle Optimized BP Neural Network forecast model of genetic algorithm, specific as follows:

A. using the threshold value of random initializtion many groups BP neural network prediction model and the weights initial population as genetic algorithm, Population Size rule of thumb presets, and encodes to this initial population, and setting largest optimization algebraically was 50 generations;

B. the ideal adaptation degree function of the initialization population sample of the genetic algorithm of Optimized BP Neural Network forecast model initial threshold and weights is configured to:

$F=\frac{k}{\sqrt{\underset{j=1}{\overset{N}{\mathrm{\Σ}}}(T_j-Y_j)}}$

In formula, N is the node that network exports, N=1; T _jfor the desired output of a BP neural network jth node; Y _jfor the prediction of a jth node exports; K is coefficient, k=1;

The numerical value of the fitness F of each individuality in initial population is calculated according to ideal adaptation degree function formula;

C. according to roulette method, namely based on the selection strategy of fitness ratio, from the multiple individuality of initialization population, select multiple optimization individualities that corresponding ideal adaptation degree F numerical value is relatively large, carry out intersecting, mutation operation, the new individuality of generation is as first filial generation weights W ₁with threshold value θ ₁the individuality of population, specific as follows:

Interlace operation adopts real number bracketing method, a kth chromosome a _kwith l chromosome a _lin j' position interlace operation method be:

a _kj'=a _kj'(1-b)+a _i'j'b；

a _lj'=a _lj'(1-b)+a _kj'b；

In formula, a _kj'for a kth chromosomal j' position; a _ij'be i-th chromosomal j' position; a _lj'be l chromosomal j' position; B is the random number between [0,1];

Mutation operation is the jth of individuality ' ' ' ' the individual gene a that chooses i-th _i' ' _j' ' make a variation, be specially:

When r is more than or equal to 0.5,

a _i''j''=a _i''j''+(a _i''j''-a _max)*f(g)；

When r is less than 0.5,

a _i''j''=a _i''j''+(a _min-a _i''j'')*f(g)；

Wherein, r is the random number between [0,1]; a _maxand a _minbe respectively gene a _i' ' _j' ' the upper bound and lower bound; F (g)=r ₂(1-g/G _max); r ₂it is a random value; G is current iteration number of times; G _maxit is maximum evolution number of times;

D. first filial generation weights W is used ₁with threshold value θ ₁the new individual individuality replacing initial population of population, repeat the genetic algorithm of step a to step c to the optimizing process of first filial generation population at individual, until ideal adaptation degree function described in step b is basicly stable constant or when reaching the evolutionary generation of setting, terminate optimizing process, and obtain initialization complete last in generation population individuality, namely obtain the initial weight W that initialization is complete ₂with threshold value θ ₂;

208, to optimize complete initial weight and threshold value BP neural network prediction model carry out training study, specific as follows;

I. the test sample book transferred from normalized database, input quantity P ' wherein _kwith output quantity C ' _khave and map corresponding relation really; With the input quantity P ' of identical test sample book _kbring in BP neural network prediction model, and by weights W complete for initialization ₂with threshold value θ ₂as new initial threshold and the weights of BP neural network; The prediction output quantity Y ' now obtained by BP neural network _j, this predicts output quantity Y ' _jwith relative height desired throughput T _jbetween there is error;

II. set up the error back propagation model of BP neural network, construct one by relative desired throughput T _jwith the prediction output quantity Y ' of BP neural network prediction model _jthe objective function formed: the process of this minimization of object function is made to be exactly the process of BP neural network error back propagation;

III. carry out computing with the objective function of gradient descent method to BP neural network error back propagation, make the initial weight W of BP neural network prediction model ₂with threshold value θ ₂optimize further; According to gradient descent method, with the increase of iterations, the error of objective function will progressively reduce, until error meets the accuracy requirement preset, terminate the optimizing process of initial threshold and weights;

IV. through step III calculating repeatedly and reduce error, when error finally meets the accuracy requirement preset, obtain weights W and the threshold value θ of one group of optimum, using weights W now and threshold value θ as the final weights and threshold of BP neural network prediction model, the training study of BP neural network prediction model completes;

209, by the BP neural network prediction model completing training study, obtain prediction output quantity and relative height, and its computing formula and model final weights W, threshold value θ are transferred in described microprocessor module; Relative height formula is as follows:

△H=purelin{tansig(p*W _1j+t*W _2j+θ _j)*W _j1+θ _q}

In formula, W _1jand W _2jbe respectively air pressure P and temperature t corresponding by the weights of hidden layer to input layer; W _3jfor output layer is to the weights of hidden layer; θ _qfor output layer threshold value;

Step 3, array-type sensor module described in described microprocessor module order reads compensating parameter, and is transmitted back in described microprocessor module by the air pressure of collection and temperature value and compensating parameter; Described microprocessor module is compensated air pressure and temperature value by compensating parameter, averages, and be normalized mean value after the air pressure after compensation and temperature value are removed maximin;

Step 4, applies the relative height computing formula that described microprocessor module receives, and calculates corresponding normalized compensation air pressure and temperature value obtains relative height value; The relative height value calculated is carried out renormalization process, obtains actual relative height;

Step 5, the air pressure of renormalization, temperature and relative height after compensating are sent to that host computer carries out storing, LCD MODULE shows by described microprocessor module respectively.

As a preferred embodiment of the present invention, described microprocessor module selects the microcontroller of MSP430F149 model.

As a preferred embodiment of the present invention, described array gas pressure sensor module selects BMP085 model baroceptor to form the square formation of 3*3.

As a preferred embodiment of the present invention, described LCD MODULE selects LCD12864 model.

As a preferred embodiment of the present invention, described serial communication modular selects RS232 serial communication chip.

As a preferred embodiment of the present invention, described power module selects 7133 voltage stabilizing chips.

The present invention adopts above technical scheme compared with prior art, has following technique effect:

(1) hardware utilize BMP085 baroceptor to form the square formation of 3*3, adopt array multi-point average metering system, effectively eliminate measuring error that the defect due to sensor individuals itself brings and decrease the stochastic error or reproducibility error that produce due to sensor chip creep, significantly reducing the impact of year drift;

(2) software adopts the BP neural network algorithm based on genetic algorithm optimization carry out temperature drift and nonlinear compensation to measuring-signal, inhibit the impact of noise, enhance useful signal, optimize measurement data;

(3) control treatment is carried out by the MSP430F149 microcontroller of low-power consumption, form a set of more accurate relative height measurement mechanism, be conducive to restraint speckle, recovery, enhancing and extraction useful signal, measure based on single baroceptor than traditional and utilize the measuring system of standard pressure altitude formulae discovery relative height to have higher accuracy and stability, also can be extended to fields of measurement widely;

(4) in view of the low cost of BMP085 baroceptor, the high precision of this device and stability make it have higher commercial value.

Accompanying drawing explanation

Fig. 1 is structure drawing of device of the present invention.

Fig. 2 is the peripheral circuit diagram of BMP085 baroceptor.

Fig. 3 is the floor map of array baroceptor.

Fig. 4 is power module circuitry figure.

Fig. 5 is method flow diagram of the present invention.

Fig. 6 is the topology diagram of BP neural network.

Fig. 7 is the method flow diagram of genetic algorithm optimization BP neural network.

Fig. 8 is the relative height comparison diagram that relative height actual value, output relative height value of the present invention and standard pressure altitude computing formula calculate.

Fig. 9 is the error of output relative height value of the present invention and the error comparison diagram of standard pressure altitude computing formula calculating.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail:

As shown in Figure 1, a kind of relative height measurement mechanism, the array gas pressure sensor module, power module, LCD MODULE, the serial communication modular that comprise host computer, microprocessor module and be connected respectively with microprocessor module; Described host computer is connected with described microprocessor module by described serial communication modular.

As shown in Figure 2, the integrated digital baroceptor BMP085 of temperature compensation capability that what described gas pressure sensor module selected BOSCH company to produce have, the supply voltage of this digital pressure sensing chip is 1.8V ~ 3.6V, it has super low-power consumption, high precision (under low-power consumption mode, resolution is 0.06hPa) and high reliability, its economy and applicability all meet the requirement of the design, the output of BMP085 is directly proportional to additional pressure, measurement range is 300 ~ 1100hPa, calibration numeral completely exports, after completing sampling, directly data are sent to MSP430F149 by I2C bus.

As shown in Figure 3, described array gas pressure sensor module selects BMP085 baroceptor to form the square formation of 3*3.

The signal disturbing caused to eliminate power supply unstable, the present invention selects 7133 chips and peripheral circuit to form source of stable pressure, and Absorbable organic halogens exports 5V and 3.3V and powers to each module, as shown in Figure 4.

As shown in Figure 5, a kind of measuring method adopting relative height measurement mechanism described above, comprises the following steps:

Step 1, after device for opening power supply, whole system carries out initialization, comprises clock initialization, port initialization, liquid crystal display initialization, serial ports initialization, BMP085 sensor initializing and closes house dog; MSP430F149 is by port analog I ²c port communications, I ²sCL with SDA of C is connected with the IO port of microcontroller respectively, the low and high level Simulation with I of IO port ²the sequential of C communication, makes I ²c communication can effectively work;

Step 2, MSP430F149 microcontroller sends enabling signal to BMP085 sensor; Sensor carries out the measurement of air pressure and temperature, and transfers to MSP430F149 microcontroller after converting uncompensated air pressure and temperature value to numerical signal by built-in A/D converter; The air pressure received and temperature data are sent in host computer by MSP430F149 microcontroller;

Step 3, described host computer sets up BP neural network by the air pressure that receives and temperature data, being optimized BP neural network according to the principle of genetic algorithm and completing its training study process, the relative height computing formula obtained be sent in described microprocessor module, concrete steps are as follows:

301, using the atmospheric pressure value P that collects and temperature value t as input quantity P _kthe parameter of=(P, t); Choose a reference mark, record the relative height △ H of each testing location, it can be used as output quantity C _kthe parameter of=(△ H); The input quantity P of same testing location _kwith output quantity C _kthere is mapping corresponding relation, obtain the test data set that n has above-mentioned mapping relations thus, it can be used as test sample book stored in database; N is natural number, and K is any one group in n group test sample book, K=1,2 ..., n;

302, be normalized test sample book, all data are all converted into the number between [0,1] by employing minimax method, and normalization formula is as follows:

χ _k=（χ _k-χ _min）/(χ _max-χ _min)

In formula, χ _minfor the minimum number in data sequence; χ _maxfor the maximum number in data sequence;

According to normalization formula by input quantity P _kwith output quantity C _kall convert normalized input quantity P ' to _kwith output quantity C ' _k, finally obtain normalized test sample book and database;

303, according to the mapping corresponding relation of input quantity and output quantity in the test sample book after normalization, set up the BP neural network of three etale topology structures, as shown in Figure 6, specific as follows:

The input layer of setting BP neural network is 2, and output layer neuron is 1, and hidden layer neuron is 6; Wherein the activation function of input layer and hidden layer all chooses tansig type function, and output layer activation function chooses pureline type function; Connection weights between input layer and hidden layer are W _ij, the threshold value of hidden layer is θ _j, the connection weights between hidden layer and output layer are W _jq, the threshold value of output layer is θ _q; I is input layer number, i=1,2; J is node in hidden layer, j=1,2 ..., 6; Q is output layer nodes, q=1;

Tansig type function is defined as

$f\left(\mathrm{\χ}\right)=\frac{2}{1+\exp (-{2S}_j)}-1$

In formula, S _jthe input of j hidden layer node, a _ifor the matrix that atmospheric pressure value and temperature value are formed;

Pureline type function is linear function, and wherein, independent variable is the output of hidden layer node, and dependent variable is the output of output layer node;

304, using the neural network of BP described in step 3 as blackbox model, set up with the input variable of the input quantity in test sample book described in step 1 for BP neural network, relative height value using correspondence as the BP neural network prediction model of output variable;

305, random initializtion is carried out to the weights W in BP neural network prediction model and threshold value θ, makes BP neural network model have the most basic predicted condition; W, θ are respectively W _ijand W _jq, θ _jand θ _qgeneral name;

306, from normalized database, transfer input quantity P ' _k, as the input variable of BP neural network prediction model with most fundamental forecasting condition, thus obtain and input variable P ' _kthe prediction output quantity Y ' of the relative height that the BP neural network prediction model that mapping pair is answered exports _kdata group;

307, the threshold value determined by test sample book normalized described in step 2 and the scope of weights, according to initial threshold value and the weights of the principle Optimized BP Neural Network forecast model of genetic algorithm, as shown in Figure 7, and complete the training study of BP neural network prediction model; Specific as follows:

A. using the threshold value of random initializtion many groups BP neural network prediction model and the weights initial population as genetic algorithm, Population Size rule of thumb presets, and encodes to this initial population, and setting largest optimization algebraically was 50 generations;

B. the ideal adaptation degree function of the initialization population sample of the genetic algorithm of Optimized BP Neural Network forecast model initial threshold and weights is configured to:

$F=\frac{k}{\sqrt{\underset{j=1}{\overset{N}{\mathrm{\Σ}}}(y_j-o_j)}}$

In formula, N is the node that network exports, N=1; y _jfor the desired output of a BP neural network jth node; o _jfor the prediction of a jth node exports; K is coefficient, k=1;

The numerical value of the fitness F of each individuality in initial population is calculated according to ideal adaptation degree function formula;

C. according to roulette method, namely based on the selection strategy of fitness ratio, from the multiple individuality of initialization population, select multiple optimization individualities that corresponding ideal adaptation degree F numerical value is relatively large, carry out intersecting, mutation operation, the new individuality of generation is as first filial generation weights W ₁with threshold value θ ₁the individuality of population, specific as follows:

Interlace operation adopts real number bracketing method, a kth chromosome a _kwith l chromosome a _lin j' position interlace operation method be:

a _kj'=a _kj'(1-b)+a _i'j'b；

a _lj'=a _lj'(1-b)+a _kj'b；

In formula, a _kj'for a kth chromosomal j' position; a _ij'be i-th chromosomal j' position; a _lj'be l chromosomal j' position; B is the random number between [0,1];

Mutation operation is the jth of individuality ' ' ' ' the individual gene a that chooses i-th _i' ' _j' ' make a variation, be specially:

When r is more than or equal to 0.5,

a _i''j''=a _i''j''+(a _i''j''-a _max)*f(g)；

When r is less than 0.5,

a _i''j''=a _i''j''+(a _min-a _i''j'')*f(g)；

Wherein, r is the random number between [0,1]; a _maxand a _minbe respectively gene a _i' ' _j' ' the upper bound and lower bound; F (g)=r ₂(1-g/G _max); r ₂it is a random value; G is current iteration number of times; G _maxit is maximum evolution number of times;

D. first filial generation weights W is used ₁with threshold value θ ₁the new individual individuality replacing initial population of population, repeat the genetic algorithm of step a to step c to the optimizing process of first filial generation population at individual, until ideal adaptation degree function described in step b is basicly stable constant or when reaching the evolutionary generation of setting, terminate optimizing process, and obtain initialization complete last in generation population individuality, namely obtain the initial weight W that initialization is complete ₂with threshold value θ ₂;

308, to optimize complete initial weight and threshold value BP neural network prediction model carry out training study, specific as follows;

I. the test sample book transferred from normalized database, input quantity P ' wherein _kwith output quantity C ' _khave and map corresponding relation really; With the input quantity P ' of identical test sample book _kbring in BP neural network prediction model, and by weights W complete for initialization ₂with threshold value θ ₂as new initial threshold and the weights of BP neural network.The prediction output quantity Y ' now obtained by BP neural network _j, this predicts output quantity Y ' _jwith relative height desired throughput T _jbetween there is error;

II. set up the error back propagation model of BP neural network, construct one by relative desired throughput T _jwith the prediction output quantity Y ' of BP neural network prediction model _jthe objective function formed: the process of this minimization of object function is made to be exactly the process of BP neural network error back propagation;

III. carry out computing with the objective function of gradient descent method to BP neural network error back propagation, make the initial weight W of BP neural network prediction model ₂with threshold value θ ₂optimize further; According to gradient descent method, with the increase of iterations, the error of objective function will progressively reduce, until error meets the accuracy requirement preset, terminate the optimizing process of initial threshold and weights;

IV. through step g calculating repeatedly and reduce error, when error finally meets the accuracy requirement preset, obtain weights W and the threshold value θ of one group of optimum, using weights W now and threshold value θ as the final weights and threshold of BP neural network prediction model, the training study of BP neural network prediction model completes;

309, by the BP neural network prediction model completing training study, obtain prediction output quantity and relative height, and its computing formula and model final weights W, threshold value θ are transferred in described microprocessor module; Relative height formula is as follows:

△H=purelin{tansig(p*W _1j+t*W _2j+θ _j)*W _j1+θ _q}

In formula, W _1jand W _2jbe respectively air pressure P and temperature t corresponding by the weights of hidden layer to input layer; W _3jfor output layer is to the weights of hidden layer; θ _qfor output layer threshold value;

Step 4, MSP430F149 microcontroller order BMP085 sensor reads E ²11 compensating parameters of PROM, air pressure and temperature value and compensating parameter are sent in MSP430F149 microcontroller to BMP085 by MSP430F149 microcontroller; Described MSP430F149 microcontroller is compensated air pressure and temperature value by compensating parameter, averages, and be normalized mean value after the air pressure after compensation and temperature value are removed maximin;

Step 5, the relative height computing formula that application MSP430F149 microcontroller receives, calculates corresponding normalized compensation air pressure and temperature value obtains relative height value; The relative height value calculated is carried out renormalization process, obtains actual relative height;

Step 6, the air pressure of renormalization, temperature and relative height after compensating are sent to that host computer carries out storing, LCD MODULE shows by MSP430F149 microcontroller respectively.

Below to be further described the present invention in conjunction with example:

Using Nanjing Information engineering Univ's Electronics and Information Engineering institute subject building (below referred to as the large No. two subject buildings of letter, south) as measuring object.The large No. two subject buildings one of letter, south have five floor, and every one deck is on average approximately 4.04m, and overall height is about 20.2323m.In order to the convenience measured, select stair to be the place tested, using the ground of ground floor as test basic point (namely testing zero point), have chosen 26 place's relative height testing locations.Often locate testing location take 8 times measure sample (comprising temperature and atmospheric pressure value), obtain atmospheric pressure value herein and temperature after being gone by the maxima and minima of atmospheric pressure value often locating to measure sample (wherein atmospheric pressure value is accurate to 0.001hPa except averaging, temperature is accurate to 0.01 DEG C), and the relative height △ H(recording each testing location is accurate to 0.0001m).

The temperature recorded by testing location respectively and atmospheric pressure value are as the input quantity P measuring sample _k, using relative height value as the output quantity C measuring sample _k, show that corresponding weights and threshold is according to step 3:

$W_{1j}=\left[\begin{array}{c}2.2079\\ 1.4675\\ -2.5283\\ -1.0323\\ -0.2514\\ 1.878\end{array}\right],W_{2j}=\left[\begin{array}{c}0.2608\\ -0.7638\\ 0.1838\\ 1.9342\\ 0.6537\\ 0.7811\end{array}\right],W_{j1}=\left[\begin{array}{c}-2.1073\\ 2.9851\\ 0.6531\\ 2.8472\\ -2.3171\\ 1.0144\end{array}\right],{\mathrm{\θ}}_j=\left[\begin{array}{c}2.5758\\ -1.2952\\ -0.3681\\ 2.2179\\ -0.4702\\ 1.5427\end{array}\right],{\mathrm{\θ}}_q=0.3334$

By the weights and threshold of gained and computing formula △ H=purelin{tansig (p*W _1j+ t*W _2j+ θ _j) * W _j1+ θ _qstored in MSP430F149 microcontroller.

Other 18 places apparatus of the present invention being placed into the large No. two subject buildings of letter, south are measured, and relative height value and air pressure and temperature are write down.

According to barometric leveling principle, under standard atmospheric conditions, standard pressure altitude computing formula is as follows:

$H=\frac{T_0}{\mathrm{\β}}[{\left(\frac{P}{P_0}\right)}^{-\mathrm{\βR}/g}-1]+\mathrm{Ho}$

In formula, R=287.05287m ²/ (s ²k), air special gas constant is represented; G=9.80665m ²/ s ²represent free-fall acceleration; β represents the vertical rate of temperature, highly when 0 ~ 11km, and β=-0.0065K/m; T ₀, P ₀, H ₀distinguish the temperature of reference planes, atmospheric pressure value and height, under standard atmosphere condition (namely pressure is 1013.25hPa, and temperature is 15 DEG C), mean sea level gets following value: P ₀=1013.25hPa, T ₀=288.15K, H ₀be the height of mean sea level, be defined as 0m.

Corresponding parameter value is substituted in standard pressure altitude computing formula, can obtain:

$H=-4433.8[{\left(\frac{P}{1013.25}\right)}^{0.190263102}-1]$

Can the atmospheric pressure value of measurement be brought into criterion calculation formula like this thus obtain relative height value.

Finally the relative height that measured relative height actual value, output relative height value of the present invention and standard pressure altitude computing formula calculate is depicted as broken line graph, as shown in Figure 8.The relative height value that output relative height value of the present invention and standard pressure altitude computing formula calculate and actual value obtain error and are depicted as broken line graph, as shown in Figure 9.

As seen from the experiment, the error of the relative height value that the present invention exports obviously reduces, and has higher precision relative to the calculated value of standard pressure altitude computing formula.

The above; be only the embodiment in the present invention; but protection scope of the present invention is not limited thereto; any people being familiar with this technology is in the technical scope disclosed by the present invention; the conversion or replacement expected can be understood; all should be encompassed in and of the present inventionly comprise within scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (6)

1. adopt a measuring method for relative height measurement mechanism, array gas pressure sensor module, power module, LCD MODULE, serial communication modular that this device comprises host computer, microprocessor module and is connected with microprocessor module respectively; Described host computer is connected with described microprocessor module by described serial communication modular; Described power module is that modules is powered, and it is characterized in that, measuring method comprises the following steps:

Step 1, described microprocessor module sends startup command to described array gas pressure sensor module, described array gas pressure sensor module gathers air pressure and the temperature data of n different testing location, and transferring to described microprocessor module, the air pressure received and temperature data are sent in described host computer by described microprocessor module;

Step 2, described host computer sets up BP neural network by the air pressure that receives and temperature data, being optimized BP neural network according to the principle of genetic algorithm and completing its training study process, the relative height computing formula obtained be sent in described microprocessor module, concrete steps are as follows:

201, using the atmospheric pressure value P that collects and temperature value t as input quantity P _kthe parameter of=(P, t); Choose a reference mark, record the relative height △ H of each testing location, it can be used as output quantity C _kthe parameter of=(△ H); The input quantity P of same testing location _kwith output quantity C _kthere is mapping corresponding relation, obtain the test data set that n has above-mentioned mapping relations thus, it can be used as test sample book stored in database; N is natural number, and K is any one group in n group test sample book, K=1,2 ..., n;

202, be normalized test sample book, all data are all converted into the number between [0,1] by employing minimax method, and normalization formula is as follows:

χ _k＝(χ _k-χ _min)/(χ _max-χ _min)

In formula, χ _minfor the minimum number in data sequence; χ _maxfor the maximum number in data sequence;

According to normalization formula by input quantity P _kwith output quantity C _kall convert normalized input quantity P ' to _kwith output quantity C ' _k, finally obtain normalized test sample book and database;

203, according to the mapping corresponding relation of input quantity and output quantity in the test sample book after normalization, set up the BP neural network of three etale topology structures, specific as follows:

The input layer of setting BP neural network is 2, and output layer neuron is 1, and hidden layer neuron is 6; Wherein the activation function of input layer and hidden layer all chooses tansig type function, and output layer activation function chooses pureline type function; Connection weights between input layer and hidden layer are W _ij, the threshold value of hidden layer is θ _j, the connection weights between hidden layer and output layer are W _jq, the threshold value of output layer is θ _q; I is input layer number, i=1,2; J is node in hidden layer, j=1,2 ..., 6; Q is output layer nodes, q=1;

Tansig type function is defined as

$f\left(\mathrm{\χ}\right)=\frac{2}{1+\exp (-{2S}_j)}-1$

In formula, S _jthe input of j hidden layer node, a _ifor the matrix that atmospheric pressure value and temperature value are formed;

Pureline type function is linear function, and wherein, independent variable is the output of hidden layer node, and dependent variable is the output of output layer node;

204, using the neural network of BP described in step 203 as blackbox model, set up with the input variable of the input quantity in test sample book described in step 1 for BP neural network, relative height value using correspondence as the BP neural network prediction model of output variable;

205, random initializtion is carried out to the weights W in BP neural network prediction model and threshold value θ, makes BP neural network model have the most basic predicted condition; W, θ are respectively W _ijand W _jq, θ _jand θ _qgeneral name;

206, from normalized database, transfer input quantity P ' _k, as the input variable of BP neural network prediction model with most fundamental forecasting condition, thus obtain and input variable P ' _kthe prediction output quantity Y ' of the relative height that the BP neural network prediction model that mapping pair is answered exports _kdata group;

207, the threshold value determined by test sample book normalized described in step 202 and the scope of weights, according to initial threshold value and the weights of the principle Optimized BP Neural Network forecast model of genetic algorithm, and complete the training study of BP neural network prediction model; Specific as follows:

A. using the threshold value of random initializtion many groups BP neural network prediction model and the weights initial population as genetic algorithm, Population Size rule of thumb presets, and encodes to this initial population, and setting largest optimization algebraically was 50 generations;

B. the ideal adaptation degree function of the initialization population sample of the genetic algorithm of Optimized BP Neural Network forecast model initial threshold and weights is configured to:

$F=\frac{k}{\sqrt{\underset{j=1}{\overset{N}{\mathrm{\Σ}}}(T_j-Y_j^{,})}}$

In formula, N is the node that network exports, N=1; T _jfor the desired output of a BP neural network jth node; Y' _jfor the prediction of a jth node exports; K is coefficient, k=1;

The numerical value of the fitness F of each individuality in initial population is calculated according to ideal adaptation degree function formula;

C. according to roulette method, namely based on the selection strategy of fitness ratio, from the multiple individuality of initialization population, select multiple optimization individualities that corresponding ideal adaptation degree F numerical value is relatively large, carry out intersecting, mutation operation, the new individuality of generation is as first filial generation weights W ₁with threshold value θ ₁the individuality of population, specific as follows:

Interlace operation adopts real number bracketing method, a kth chromosome a _kwith l chromosome a _lin j' position interlace operation method be:

a _kj'＝a _kj'(1-b)+a _i'j'b；

a _lj'＝a _lj'(1-b)+a _kj'b；

In formula, a _kj'for a kth chromosomal j' position; a _i'j'be i-th chromosomal j' position; a _lj'be l chromosomal j' position; B is the random number between [0,1];

Mutation operation chooses i-th " jth of individuality " individual gene a _i" _j" make a variation, be specially:

When r is more than or equal to 0.5,

a _i” _j”＝a _i” _j”+(a _i” _j”-a _max)*f(g)；

When r is less than 0.5,

a _i” _j”＝a _i” _j”+(a _min-a _i” _j”)*f(g)；

Wherein, r is the random number between [0,1]; a _maxand a _minbe respectively gene a _i" _j" the upper bound and lower bound; F (g)=r ₂(1-g/G _max); r ₂it is a random value; G is current iteration number of times; G _maxit is maximum evolution number of times;

D. first filial generation weights W is used ₁with threshold value θ ₁the new individual individuality replacing initial population of population, repeat the genetic algorithm of step a to step c to the optimizing process of first filial generation population at individual, until ideal adaptation degree function described in step b is basicly stable constant or when reaching the evolutionary generation of setting, terminate optimizing process, and obtain initialization complete last in generation population individuality, namely obtain the initial weight W that initialization is complete ₂with threshold value θ ₂;

208, to optimize complete initial weight and threshold value BP neural network prediction model carry out training study, specific as follows;

I. the test sample book transferred from normalized database, input quantity P ' wherein _kwith output quantity C ' _khave and map corresponding relation really; With the input quantity P ' of identical test sample book _kbring in BP neural network prediction model, and by weights W complete for initialization ₂with threshold value θ ₂as new initial threshold and the weights of BP neural network; The prediction output quantity Y ' now obtained by BP neural network _j, this predicts output quantity Y ' _jwith relative height desired throughput T _jbetween there is error;

II. set up the error back propagation model of BP neural network, construct one by relative desired throughput T _jwith the prediction output quantity Y ' of BP neural network prediction model _jthe objective function formed: the process of this minimization of object function is made to be exactly the process of BP neural network error back propagation;

III. carry out computing with the objective function of gradient descent method to BP neural network error back propagation, make the initial weight W of BP neural network prediction model ₂with threshold value θ ₂optimize further; According to gradient descent method, with the increase of iterations, the error of objective function will progressively reduce, until error meets the accuracy requirement preset, terminate the optimizing process of initial threshold and weights;

IV. through step III calculating repeatedly and reduce error, when error finally meets the accuracy requirement preset, obtain weights W and the threshold value θ of one group of optimum, using weights W now and threshold value θ as the final weights and threshold of BP neural network prediction model, the training study of BP neural network prediction model completes;

209, by the BP neural network prediction model completing training study, obtain prediction output quantity and relative height, and its computing formula and model final weights W, threshold value θ are transferred in described microprocessor module; Relative height formula is as follows:

△H＝purelin{tansig(p*W _1j+t*W _2j+θ _j)*W _j1+θ _q}

In formula, W _1jand W _2jbe respectively air pressure P and temperature t corresponding by the weights of hidden layer to input layer; W _j1for output layer is to the weights of hidden layer; θ _qfor output layer threshold value;

Step 3, array-type sensor module described in described microprocessor module order reads compensating parameter, and is transmitted back in described microprocessor module by the air pressure of collection and temperature value and compensating parameter; Described microprocessor module is compensated air pressure and temperature value by compensating parameter, averages, and be normalized mean value after the air pressure after compensation and temperature value are removed maximin;

Step 4, applies the relative height computing formula that described microprocessor module receives, and calculates corresponding normalized relative height value; The relative height value calculated is carried out renormalization process, obtains actual relative height;

Step 5, the air pressure of renormalization, temperature and relative height after compensating are sent to that host computer carries out storing, LCD MODULE shows by described microprocessor module respectively.

2. measuring method according to claim 1, is characterized in that, described microprocessor module selects the microcontroller of MSP430F149 model.

3. measuring method according to claim 1, is characterized in that, described array gas pressure sensor module selects BMP085 model baroceptor to form the square formation of 3*3.

4. measuring method according to claim 1, is characterized in that, described LCD MODULE selects LCD12864 model.

5. measuring method according to claim 1, is characterized in that, described serial communication modular selects RS232 serial communication chip.

6. measuring method according to claim 1, is characterized in that, described power module selects 7133 voltage stabilizing chips.