CN101266145A - Atmospheric pressure altimeter for measuring using multilayer perception machine nerval net and the measurement method - Google Patents

Abstract

The invention discloses an atmospheric pressure altimeter for measurement by use of a multilayer perception neural network, which includes static press and a temperature measurement unit, the static press and the temperature measurement unit is orderly connected to an A/D transverter unit, a center process unit, a communication interface unit, which are respectively connected to a power module, the static press and the temperature measurement unit include a static sensor and a temperature sensor, the center process includes microprocessor and data process unit of the multilayer perception neural network. The benefit effect of the invention is that time shift and a temperature time are avoided, learning function is provided and automatically learned through the neural network, difference of real-time precision is under 0.5%, volume is small, weight is light, and reliability is high.

Description

Take into account measuring method with the barometer altitude that the realization of Multilayer Perception machine neural network is measured

Technical field

The invention belongs to field of measuring technique, relate to a kind of barometric altimeter that utilizes the Multilayer Perception machine neural network to realize and measuring, the invention still further relates to the measuring method of this barometric altimeter.

Background technology

The measurement of height has crucial effect for the safe flight of aircraft with automatic control.By the principle branch, barometric altimeter mainly contains vapour-pressure type and radio formula.Barometric altimeter is according to atmospheric composition and characteristics, and we know the static pressure P of air _sMaximum on the ground, along with being index law, increase highly reduces.By measuring air pressure P _s, indirect measuring height.In addition because aerial temperature variation is big, and the pressure transducer temperature influence is bigger, need compensate by the output of the temperature sensor output to pressure transducer for this reason.Radio altimeter is the principle work that utilizes radio-wave reflection.During measurement, transmitter is launched same radiowave with receiver simultaneously earthward through emitting antenna, and receiver will successively receive by direct next electric wave of transmitter and the echo after ground return, and two bundle electric waves have the mistiming.If electric wave is not interfered in transport process, the mistiming is proportional to tested height.Measure the mistiming, highly also just known.

Barometric altimeter is also invented and designed in many scholars and research institution both at home and abroad, wherein ETA Sa Manufacture Horlogere S.'s design has a kind of barometric altimeter that comprises temperature compensating device, this altitude gauge rely on self time quantum, correct circuit, date and predetermined weather value and select reference temperature value among this altitude gauge train value and determine highly from being included in to be stored in, because temperature is not a The real time measure, pressure transducer can not get real-time correction like this, calculates like this and has very mistake.

Owing to the concern complexity of height with atmospheric pressure, influence factor is many, be difficult to directly describe the relation of atmospheric pressure to geometric height with an explicit function, by traditional method for designing, suppose that gas is ideal gas, utilizes surface fitting method, separate calculated altitude, because the influence of the drift of sensor self and stability, the height-precision that calculates can change along with the variation of time, and precision can be very not high.

Summary of the invention

The purpose of this invention is to provide a kind of barometric altimeter that utilizes the Multilayer Perception machine neural network realize to measure, solved that prior art exists the time float, temperature floats phenomenon, and the not high problem of real-time accuracy.

Another object of the present invention provides the measuring method of above-mentioned barometric altimeter.

The technical solution adopted in the present invention is, a kind of barometric altimeter that utilizes the Multilayer Perception machine neural network to realize and measuring, comprise static pressure valve and data processing equipment, be provided with static pressure and temperature measurement unit in the static pressure valve, static pressure and temperature measurement unit comprise static pressure transducer and temperature sensor, and static pressure transducer is made up of presser sensor pipe and sensing head, the presser sensor pipe is inserted in the valve, sensing head is near an end of static pressure valve bottom, and temperature sensor places place, static pressure valve bottom

Data processing equipment comprises A/D converter unit, CPU (central processing unit) and the communication interface unit that connects successively, and A/D converter unit, CPU (central processing unit) and communication interface unit are connected with power module respectively; Static pressure is connected with the A/D converter unit with temperature measurement unit;

CPU (central processing unit) comprises the CPU microprocessor, is pre-loaded into multi-layer perception neural network computing software in this CPU microprocessor.

Another technical scheme of the present invention is, utilizes above-mentioned barometric altimeter to carry out the method for highly measuring, may further comprise the steps,

Step 1 is introduced ambient atmos by a static pressure valve, by being arranged at static pressure and the temperature analog signal that static pressure transducer in the static pressure valve and temperature sensor measurement obtain current state;

Step 2, static pressure that the last step was obtained and temperature analog signal are sent into A/D converter and are carried out the A/D conversion, convert above-mentioned static pressure and temperature analog signal to digital signal;

Step 3, static pressure that the last step was obtained and temperature digital signal are sent in the CPU (central processing unit) and are handled, obtain the pressure and temperature signal of digital quantity, in the CPU microprocessor, finish the temperature compensation of pressure, the Multilayer Perception machine neural network that pressure and temperature value after will compensating is again sent into CPU microprocessor stored carries out intelligence and resolves, and obtains the height value of current state;

Step 4 is exported by communication interface unit the height value that the last step obtains with signal.

The invention has the beneficial effects as follows, can overcome time drift and temperature is floated, have self-learning function, and the self study by neural network, the error of real-time accuracy can reach below 0.5%, has that volume is little, in light weight, high reliability features.

Description of drawings

Fig. 1 is the module connection diagram of the embodiment of the invention;

Fig. 2 is the structural representation of static pressure of the present invention and temperature measurement unit embodiment, and wherein a is the structural representation of static pressure and temperature measurement unit, and b is the structural representation of the static pressure transducer of apparatus of the present invention.

Fig. 3 is the realization circuit diagram of the microprocessor AD μ C845 of the embodiment of the invention;

Fig. 4 is the schematic flow sheet of the multi-layer perception neural network implementation method of the embodiment of the invention.

Among the figure, 10. static pressure valve, 20. static pressure and temperature measurement unit, 30.A/D converter, 40. CPU (central processing unit), 50. communication interface, 60. power modules, the 1. import of static pressure valve, 2. static pressure transducer, 3. temperature sensor, 4. static pressure valve bottom, 5. presser sensor pipe, 6. sensing head.

Embodiment

The present invention is described in detail below in conjunction with the drawings and specific embodiments.

Owing to the concern complexity of height with atmospheric pressure, influence factor is many, be difficult to directly to describe the relation of atmospheric pressure, and based on the barometric altimeter of Multilayer Perception machine neural network precision height not only to geometric height with an explicit function, and can overcome altitude gauge the time float, temperature floats phenomenon.

Figure 1 shows that the module connection diagram of apparatus of the present invention.Static pressure is connected with A/D converter 30, CPU (central processing unit) 40, communication interface 50 successively with temperature measurement unit 20, and aforementioned each unit also is connected with power module 60 respectively, accepts the power supply of power module 60.

Static pressure and temperature measurement unit 20 comprise static pressure transducer 2 and temperature sensor 3.

CPU (central processing unit) 40 comprises microprocessor CPU, is pre-loaded into multi-layer perception neural network computing software in this CPU microprocessor.

Fig. 2 a is depicted as the static pressure of apparatus of the present invention and the structural representation of temperature measurement unit, and Fig. 2 b is depicted as the structural representation of the static pressure transducer 2 of apparatus of the present invention.Static pressure valve 10 is provided with static pressure valve import 1, static pressure transducer 2 and temperature sensor 3 are installed in the static pressure valve import 1, static pressure transducer 2 adopts the pressure transducer of TO-8 encapsulation, the presser sensor pipe 5 of static pressure transducer 2 is inserted in the valve, the sensing head 6 of static pressure transducer 2 is near the position of static pressure valve bottom 4, and both junctions will seal, and temperature sensor 3 places valve bottom 4.Ambient atmos enters static pressure valve import 1 by connecting pipe, the static pressure of gas and temperature are measured to external world respectively by the static pressure transducer 2 in the static pressure valve import 1 and temperature sensor 3, when static pressure and temperature measurement unit 20 under the situation of power module 60 normal power supplies, static pressure transducer 2 and temperature sensor 3 are exported the voltage analog signal that is directly proportional with the autosensitization amount separately.

A/D converter 30 is 24 A/D converters of two-way that the microprocessor for CPU (central processing unit) 40 is equipped with, A/D figure place and slewing rate can be provided with by software, pressure signal and temperature signal are carried out the A/D conversion, and the output digital signal directly enters CPU (central processing unit) 40.

CPU (central processing unit) 40 comprises CPU microprocessor and ANN-MLP Multilayer Perception machine neural network, the CPU microprocessor at a high speed 8 for being with 2 tunnel 24 A/D converters, the Chip Microcomputer A μ C845 of the program storage of 62K, realize circuit as shown in Figure 3, simulated pressure signal p1, + p1 and temperature T+, T-is through C6, the A/D converter ADC0 that carries by single-chip microcomputer after the C7 filtering, ADC1 changes, obtain the pressure and temperature signal of digital quantity, in single-chip microcomputer, finish the temperature compensation of pressure, pressure and temperature signal after will compensating is again sent into the multi-layer perception neural network structure that has trained of single-chip microcomputer stored, resolves through the intelligence of Multilayer Perception machine neural network and obtains accurate height value.The supply voltage of Single Chip Microcomputer (SCM) system is+5V _DC,+5V _DCThrough the filter network generation+5V that forms by C1 C2 C3 C4 C5 (capacitance adopts 0.1pF) _DCAVDD use for single-chip microcomputer, pressure and temperature sensor, the voltage reference+2.5V of A/D converter is provided by REF2925, the oscillator of single-chip microcomputer work is made up of OSC30 (32.768K) and two capacitor C 8C9 (capacitance employing 30pF) in addition.

Communication interface 50 adopts MAX488, under the situation of power module 60 normal power supplies, the Transistor-Transistor Logic level that the microprocessor serial ports of CPU (central processing unit) 40 is exported is converted to the baud rate output of RS485 fiduciary level with 57600bit/s, and the refreshing frequency of output data is 100Hz.

Power module 60 adopts the wide power design, in DC voltage 5～32V interval, after the input voltage process DC/DC transducer LT1933, output+5V DC voltage, this voltage is again through two linear voltage stabilization transducer MAX1615, export digital voltage+5V and aanalogvoltage+5V respectively, wherein one the tunnel for microprocessor 40 work uses, and another road voltage supply power and temperature measurement unit 20 and communication interface 50 work are used.

As shown in Figure 4, be the schematic flow sheet of the multi-layer perception neural network implementation method of the embodiment of the invention, three layers of perceptron neural network of network using, ground floor neuron number is 2, transition function adopts the S type function; Second layer neuron number is 10, and transition function adopts linear function; The 3rd layer of neuron number is 1, and transition function adopts the S type function.P=[p _sT] ^TInput for network; ω ¹Weights for the ground floor neural network; b ¹Threshold value for the ground floor neural network; f ¹Transition function for the ground floor neural network; a ¹Output for the ground floor neural network; ω ²Weights for second layer neural network; b ²Threshold value for second layer neural network; f ²Transition function for second layer neural network; a ²Output for second layer neural network; ω ³Be the weights of three-layer neural network; b ³It is the threshold value of three-layer neural network; f ³It is the transition function of three-layer neural network; a ³Be the output of three-layer neural network, h=a ³

The training sample of Multilayer Perception machine neural network of the present invention extracts foundation, and the one, according to the atmosphere static pressure given among the air standard HB6127-86 and the data of geometric height; The 2nd, measured data of experiment.Adopt this two kinds of methods to obtain the sample storehouses, the output that promptly records sensor under the normal pressure and temperature condition is as the input sample, and under this condition in the air standard determined height value as output sample.The MOBP method is adopted in the training of Multilayer Perception machine neural network of the present invention, obtains the weights and the threshold value of each layer of the mathematical model of Multilayer Perception machine neural network after the training.

Embodiment:

The little process chip AD μ of the high speed C845 that the embodiment of the invention selects for use U.S. ADI company to produce, static pressure transducer is selected S17-030A (measurement range: 0～30Psi) for use, temperature sensor is selected LM50 (measurement range :-55～125 ℃ for use, 0.5 ℃ of precision), the simulating signal of sensor output converts the CPU that digital signal is sent into AD μ C845 chip to by A/D converter, in CPU, finish the temperature correction of pressure transducer, obtain the multilayer neural network mathematical model in conjunction with above-mentioned training again, finally obtain accurate barometer altitude value, by the form output digital value of level conversion (MAX488) with the RS485 interface.

(1) utilizes the data of atmosphere static pressure, temperature and geometric height given among the air standard HB6127-86, with the input of atmosphere static pressure and temperature as neural network, height programs under the Matlab environment neural network is trained as the output of neural network.

(2) with output error 10 ^-5As final goal, after training reaches target, obtain the weights and the threshold value of each layer of neural network, the result is as follows:

${\mathrm{\ω}}^1={\left[\begin{array}{cc}-3.890605812898889e+000& -6.964923417245192e+000\\ -1.905649286947199e+000& 9.033754486929974e+000\end{array}\right]}_{2\×2}$

${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="A20081001795500142.png"\; state="\{\{state\}\}">b</figure-callout>}}^1={\left[\begin{array}{c}7.379029894239634e+000\\ -7.690280177312385e+000\end{array}\right]}_{2\&times;1}$

${\mathrm{\&omega;}}^2={\left[\begin{array}{cc}-1.300805910545332e+000& -5.127511489031426e-001\\ 5.173620379975948e-001& -1.145472690129816e-001\\ 8.484287720101860e-001& -8.686125857567048e-002\\ -1.268964012903185e+000& -2.281354282066237e+000\\ 2.543370620580513e-001& -7.457387219913907e-002\\ -8.367824932490380e-002& 9.044134480449404e-001\\ 4.955695982876381e-001& 7.053788597910656e-001\\ 1.488672484125829e+000& 1.177380343117734e+000\\ -5.626171965418318e-001& -1.454629653106387e+000\\ -7.463466616328988e-001& -1.269891701482028e+000\end{array}\right]}_{10\&times;2}$

${\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="A20081001795500132.png,A20081001795500142.png"\; state="\{\{state\}\}">b</figure-callout>}}^2={\left[\begin{array}{c}-8.697183758856503e-002\\ -2.720694408633647e-002\\ -1.357389395813382e+000\\ 8.016650200136481e-001\\ 5.790389194812009e-001\\ -1.180919373989656e+000\\ -1.515577863705230e+000\\ -3.736384303778968e-001\\ 1.085846167021961e+000\\ -7.970150388845445e-001\end{array}\right]}_{10\&times;1}$

${\mathrm{\&omega;}}^3={\left[\begin{array}{ccc}-1.640897399612558e+000& -3.713497377608849e-001& 9.749896375978796e-001\\ -2.980320920637806e+000& -3.901663744432288e-001& 1.198825501893435e+000\\ 1.594775893938393e+000& 1.808063840293959e+000& -2.248986637484770e+000\\ -8.509854899235951e-001& & \end{array}\right]}_{1\&times;10}$

b ³＝[5.869054496112409e-001] _1×1

(3) so far the mathematical model of network obtains:

If: P=[p _sT] ^TBe input, p _sBe the actual measurement pressure of pressure transducer, t is the actual measurement temperature of temperature sensor.

P is input to the Multilayer Perception machine neural network, substitution ground floor weights ω ¹With threshold value b ¹, the ground floor of network is output as:

$a^1=\frac{1}{1+e^{-(p*{\mathrm{\&omega;}}^1+b^1})}$

Again with a ¹As the input of second layer neural network, substitution second layer weights ω ²With threshold value b ², network is output as:

a ²＝a ¹*ω ²+b ²

Again with a ²As the input of three-layer neural network, the 3rd layer of weights ω of substitution ³With threshold value b ³, network is output as:

$h=a^3=\frac{1}{1+e^{-(a^2*{\mathrm{\&omega;}}^3+b^3)}}$

Be respectively as measured value: P when pressure transducer and temperature sensor _C=92076.4p _a, t=25 ℃

Utilize above-mentioned formula to calculate:

h＝8.041171609062708e+002m

Theoretical value is: 800m, error is: 4.117m

Result of implementation shows, utilize the Nonlinear Mapping relation of neural network, can realize approaching to arbitrary function, the method that the present invention adopts hardware and software to combine has realized the design of high precision barometric altimeter, the degree of accuracy of barometric altimeter is apparently higher than the barometric altimeter that adopts the curved surface fitting method design, and make barometric altimeter because self performance descends or the weights of neural network and threshold values are adjusted in the drift of time by off-line learning and do not influence the degree of accuracy of output, the variation that conforms, accomplished real intelligence, the elevation information that can be used for airflight object and mountaineer is easily measured.

Claims (3)

1, a kind of barometric altimeter that utilizes the Multilayer Perception machine neural network to realize and measuring is characterized in that, comprises static pressure valve (10) and data processing equipment,

Be provided with static pressure and temperature measurement unit (20) in the described static pressure valve (10), static pressure and temperature measurement unit (20) comprise static pressure transducer (2) and temperature sensor (3), static pressure transducer (2) is made up of presser sensor pipe (5) and sensing head (6), presser sensor pipe (5) is inserted in the valve, sensing head (6) is near an end of static pressure valve bottom (4), temperature sensor (3) places static pressure valve bottom (4) to locate

Described data processing equipment comprises A/D converter unit (30), CPU (central processing unit) (40) and the communication interface unit (50) that connects successively, described A/D converter unit (30), CPU (central processing unit) (40) and communication interface unit (50) are connected with power module (60) respectively

Described static pressure is connected with A/D converter unit (30) with temperature measurement unit (20),

Described CPU (central processing unit) (40) comprises the CPU microprocessor, is pre-loaded into multi-layer perception neural network computing software in this CPU microprocessor.

2, a kind of described barometric altimeter of claim 1 that utilizes carries out the method for highly measuring, and it is characterized in that this method may further comprise the steps,

Step 1 is introduced ambient atmos by a static pressure valve (10), by being arranged at static pressure and the temperature analog signal that interior static pressure transducer (2) of static pressure valve (10) and temperature sensor (3) measure current state;

Step 2, static pressure that the last step was obtained and temperature analog signal are sent into A/D converter (30) and are carried out the A/D conversion, convert above-mentioned static pressure and temperature analog signal to digital signal;

Step 3, static pressure that the last step was obtained and temperature digital signal are sent in the CPU (central processing unit) (40) and are handled, obtain the pressure and temperature signal of digital quantity, in the CPU microprocessor, finish the temperature compensation of pressure, the Multilayer Perception machine neural network that pressure and temperature value after will compensating is again sent into CPU microprocessor stored carries out intelligence and resolves, and obtains the height value of current state;

Step 4 is exported by communication interface unit (50) height value that the last step obtains with signal.

3, in accordance with the method for claim 2, it is characterized in that in the described step 3, the Multilayer Perception machine neural network carries out the method that intelligence is resolved, specifically comprise,

Step 1, set up the Multilayer Perception machine neural network,

Set up an input layer, an output layer and several hidden layers, each hidden layer is made up of a plurality of neurons, and the Multilayer Perception machine neural network mathematical model of foundation is:

a ^m+1＝f ^m+1(ω ^m+1a ^m+b ^m+1) m＝0，1，2，

Wherein: a ^mBe the output of network,

f ^mBe the transition function of network, wherein $f^1=f^3=\frac{1}{1+e^{-a^{m+1}}},$ f ²＝a ^m+1，

ω ^mBe the weights of network, b ^mBe the threshold value of network, m is the number of plies of network,

The output P=[p that is input as the pressure and temperature sensor of network _sT] ^T, network is output as height value h=a ³,

The ground floor neuron number of described Multilayer Perception machine neural network is 2, and transition function adopts the S type function; Second layer neuron number is 10, and transition function adopts linear function; The 3rd layer of neuron number is 1, and transition function adopts the S type function,

Step 2, the Multilayer Perception machine neural network that the last step was set up are trained, and obtain the weights and the threshold value of this each layer of Multilayer Perception machine neural network,

With the output that records sensor under the normal pressure and temperature condition as the input sample, with determined height value in the air standard under this condition as output sample, the Multilayer Perception machine neural network that the last step sets up is trained, obtain the weights ω of each layer of mathematical model of Multilayer Perception machine neural network ^mWith threshold value b ^m,

Step 3, input static pressure and temperature signal are resolved, obtain height value,

With P=[p _sT] ^TBe input, p _sBe the actual measurement pressure of static pressure transducer (2), t is the actual measurement temperature of temperature sensor (3),

P is input to the Multilayer Perception machine neural network, substitution ground floor weights ω ¹With threshold value b ¹, the ground floor output a of network ¹For:

$a^1=\frac{1}{1+e^{-(p{*\mathrm{\&omega;}}^1+b^1})},$

Again with a ¹As the input of second layer neural network, substitution second layer weights ω ²With threshold value b ², the second layer output a of network ²For:

a ²＝a ¹*ω ²+b ²，

Again with a ²As the input of three-layer neural network, the 3rd layer of weights ω of substitution ³With threshold value b ³, the 3rd layer of output a of network ³For:

$h=a^3=\frac{1}{1+e^{-(a^2*{\mathrm{\&omega;}}^3+b^3)}}$

Promptly obtain the height value h of current state.

Images