JPH01500714A - Distributed Kalman filter - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Distributed Kalman filter Technical field The present invention relates to a data estimation method and apparatus, and particularly to a data estimation method and apparatus using distributed data processing technology. Regarding distributed Kalman filters.

Background technique Kalman filter technique mainly approximates state parameters in dynamic systems has been developed for. These Kalman filters are control devices that cannot be measured in real time. It has been used in many applications such as systems. Kalman filter becomes important One such field of technology is avionics.

Kalman filters are increasingly required for tactical aircraft avionics systems. Ru. This requirement is impacting the performance of navigational subsystems. Currently, airplanes are flying Multiple gyroscopes to detect various parameters required for row control Inertial systems such as the Suspended Attitude and Head Reference System (SAHR5) with accelerometers and accelerometers It uses a navigation system. Other systems send navigation signals no matter where they are located. A series of 18 artificial satellites orbiting the Earth on two sides in a day in six different orbital planes. It is a Global Positioning System (GPS) using stars and three spare satellites.

Each of these systems has unique advantages and disadvantages.

A system that combines an inertial navigation system and GPS can create optimal navigation. It has been determined. The title of the document by Mr. Cox (Cow Jr.) is “G with an inertial navigation system. "Integral of PS," Journal of the Aeronautical Society of Japan, Vol. 25, No. 2, 1978, pp. 236-245. , the authors disclose the use of an integral WGPS-inertial navigation filter structure. cock Mr. Su has a higher-order Kalman algorithm that provides problems while his filter is running at the desired speed. It is based on algorithm. Stalza at NAECO8 1984 (5ta rts) et al.'s Cooperative Mixture J in rGPs/5AHR5 May 33, 1984 On pages 9-348, we introduce an integral calculus combining GPS and 5AHRS systems. A Mann filter is disclosed. However, implementing the integral Kalman filter t; No description of the model is disclosed. Productivity of the sensors listed in these systems Standard Kalman filter techniques are used. However, the error behavior In the evolution of the mathematical description, the dimensions of the Kalman filter states increase significantly and the system It shifts to a higher-order model of the stem. Furthermore, the state of each estimation (approximate) error is The large number of uncertain variables involved requires enormous computational power and memory.

Modern systems on the subject of real-time Kalman filters in integral navigation problems The community literature includes two types of studies that deal with large-scale state estimation algorithms. . Considerable effort has been made in research to reduce the order of Kalman filters. . Typically, this effort involves partitioning the system state and filter determinants, resulting in several sets of is introduced into the sub-Kalman filter by rewriting the filter formula into a lower-order equality of There is. Asserting a reduced 1-state that has computational significance in applications suggests that the filter There is a risk that the performance of the data may be degraded.

Another study is a distributed Kalman interface in which all subsystems and their measurements are interconnected. It is a filter. The basic concept is to decompose a large system into subsystems and then The purpose of the system is to operate the subsystems subdivided so that -i . Although this dispersion filter is stable, state estimation is not suitable. Furthermore, mutual contact There is no mechanism to force connections, and there are no viable algorithms for large systems. .

Demonstration of development The present invention receives instrument error status data from at least one sensor device processor. , a sensor state processor for calculating sensor equipment error data, and the sensor device processor. coupled to a sensor state processor that further receives system error state data from the system; Calculate the system error data and provide this system error data to the sensor device processor. A system state processor that outputs the desired system data and reduces error data. and means for providing negative feedback to at least one sensor device processor. A system with at least one measuring device for forming data and device data. Distributed Kalmanf processing signals from at least one sensor device per stem oriented towards filters.

The present invention provides a sensor state processor, from at least one sensor device processor. Receive equipment error status data, calculate sensor equipment error data, and perform system status processing. receiving system error status data from the sensor device processor; calculating system error data and transmitting the system error data to the sensor device processor; output the desired system data and output the error data to at least one section. Distributed data is negatively fed back to the device processor and is formed by a Kalman filter to identify at least one measuring device for forming system data and device data of a distributed device processing signals from at least one sensor device per system having It also provides data processing methods.

The present invention uses a distributed Kalman filter (DKF) using distributed data processing technology. oriented. The DKF of the present invention is an integral multisensor such as 5AHR5-GPS. Particularly useful for system systems. This DKF offers improved accuracy, speed and reliability It has various advantages of allowing greater computing power and solving the burden on computing time. provide. The DKF of the present invention has great advantages in sensor systems for various devices. This is a general-purpose filter that can be used. Outside of navigation, the distributed Kalman filter is Radar, image processing in all areas where noise problems exist in determining real-time measurements. It can be used for data processing in optical, optical, TV or any system. DKF is used Possible devices include airplanes, spacecraft, amphibious ships, televisions and cameras. these are merely examples and do not limit the use of DKF to those cited above.

Typically, sensor systems navigate vehicles, identify targets, or coordinate cameras. Contains lf1 or higher sensors that collect the data necessary for the operation of the device, such as I'm here. The necessary data is usually supplied in bulk. For example, for navigation may consist of position, velocity, and orientation. These are called system states . Furthermore, the operation of the sensor itself consists of various states. In navigation, e.g. , the sensor is a gyroscope with states including alignment, coupling and drift That's fine. These are called device states. Errors are due to accurate measurement and data collection is always present in the sensor system because it is affected by noise.

The DKF estimates the error for all states and then provides negative feedback to the data acquisition processor. , continuous measurement corrections are made to compensate for errors.

In particular, the DKF of the present invention processes signals from at least one of the systems to system and equipment data.

A distributed Kalman filter receives instrument error information from at least one sensor device processor. includes a sensor condition processor that receives condition data and calculates sensor instrument error data. There is. This sensor state processing involves receiving system state data from the sensor state processor. A system state processor is coupled to the system to receive and calculate system error data.

A system and condition processor provides system error data to a sensor device processor. Ma In addition, DKF outputs desired system data I2 and processes error data to the sensor device. It also includes means for providing negative feedback to the device.

A distributed system is one in which the computations done in each processor are each performed using the joint resources of the component machines. is formed as any configuration of two or more processors with dedicated memory. between processors The amount of communication depends on the nature of the multi-sensor system. Operation of the system in each processor is determined by the communication requirements, and is required for the necessary connections between software and effective communication. provides the signal to be used. Software processed in a distributed computing system is It consists of function modules with intensive type programs.

In addition to the above-mentioned normal implementation of DKF, three other implementations are applicable to the GPS-5AHRS environment. It is disclosed in 5AHR5 and GPS systems each incorporate the invention. with corresponding equipment and system errors indicated in the multiplex conditions detailed in the specification. . 1 system is a 5A system in which the DKF includes a GPS status processor and a system status processor. It is a GPS navigation device that supports HR3. The GPS processor may e.g. Calculate the distance and distance change rate (differential distance) for the four satellites from the error shift. form the data to be used. This data can be obtained from GPS as described in the regular DKF Supplied through the state handler and system state handler.

The second system is required by DKF to estimate only the error in 5AHR3. and these errors are negatively fed back to recalibrate only the 5AHR3. 5AHR5 navigation device. GPS position and speed measurements both use 5AHRS Supplied through. The third system uses distributed processing techniques to DKF has a 5AHR3 state handler and a G It is a mixed 5AHR3/GPS navigator that includes both a PS state analyzer and a PS status analyzer. GPS is The 5AHR3 provides distance measurements and satellite data, along with orientation data. device) provides the converted acceleration and velocity. GPS navigation equipment for signal regathering using this information. 5AHR3 uses GPS-5AHRS and the speed is Updated for instrument alignment and calibration.

Brief description of the drawing Figure 1 is a block diagram of a conventional integral Kalman filter, and Figure 2 is a block diagram of a conventional integral Kalman filter. The block diagram of the distributed Kalman filter (DKF) shown in Figure 3 is 5AHR5/GP. A block diagram showing the DKF in S environment, Figure 4 is a mixed 5AHR5/GPS system. Figure 5 shows the block diagram of the DKF in the 5AHR5 supported GPS system. The block diagram of the DKF in the 5AHRS system with GPS support is shown in Figure 6. The block diagram of DKF, Figure 7a, is the system model of the conventional standard Kalman filter. Dell's block diagram, Figure 7b, shows the DKF for the 5AHR5'/GPS mixed system. FIG. 2 is a block diagram of a system model.

BEST MODE FOR CARRYING OUT THE INVENTION Referring to the drawings, FIG. FIG. 2 is a block diagram showing a filter array. Sensors 1 and 2 were later replaced by Kalman filters. The error status signals provided to the terminals 1 and 2, respectively, are calculated. Usually the sensor l and 2 calculates both system error and sensor error. Kalman filter l and 2 After processing the system errors, these errors form a Kalman function that further processes the system errors. is supplied to filter 3. This type of situation is a decentralized multirate Kalman filter appears to be a candidate. This prior art system has the same system error. Since the difference is processed by Kalman filters 1 and 2, it is redundant. Furthermore, Karman Integration of Kalman filters 1 and 2 by filter 3 reduces the reliability of calculations. There is.

In the present invention, a Kalman filter (DKF) is used to detect single dispersion due to device (sensor) error. Used to handle differential and systematic errors, increasing the amount of error conditions that can be handled ing. As shown in FIG. 2, the DKF 10 has at least two types of processors, It includes a processor 12 for equipment errors and a processor 14 for system errors. Second The DKF 10 shown in the figure is capable of calculating multiple status signals received from multiple sensors. The sensor device processor 16 is coupled to a system having a single sensor device processor 16. This sensor device Processor 16 processes the sensor data into DKF as long as it is processed by state processors 12 and 14. Sent to IO. Sensor data is typically input to equipment status processor 12. The system error is processed through the processor 12 to process the system state. 14. The processor 14 measures a system error that is negatively fed back to the processor 12. Calculate. These system and equipment errors are accounted for by making necessary adjustments to the incoming condition signals. Negative feedback is provided to the sensor device processor 16.

The advantage of this arrangement is that the instrument error handler can filter out system state errors. Although there is no burden and only equipment errors are filtered, the system state processor The system error received from the server is filtered. Therefore, the redundancy of the system error processor is Eliminating long operations requires less computation time and memory.

Furthermore, the size of the hardware required to adapt the system can be reduced and Can be applied to time manipulation.

In another embodiment of the invention, a distributed Kalman filter is used for two sensor systems. used to integrate systems. In Figure 3, the hanging posture self-surrender reference system integrated data from GPS (SAHR5) and Global Positioning System (GPS) DKF18 is shown.

The 5AHRS system includes the airplane's speed and acceleration as inputs. inertia book Body velocity and acceleration data are sensed by an array of tilted inertial components. Sen Redundancy algorithms are designed to select signals, isolate faults, and monitor performance. will be accomplished. Sensor compensation components such as bias, scale 7 actor and body curvature are calibrated. and the sensor information is decoded along the orthogonal body axes. These orthogonal velocity data is the local level by correcting the effects of the earth's velocity and the airplane's angular velocity relative to the earth's surface. Obtaining the angular velocity of the airplane with respect to the coordinate frame. These velocities are determined by the direction cosine and the related It is used to guide (differentiate) the vehicle attitude and head position.

The inertial body axial acceleration is converted to the local level frame and the gravitational and Coriolis acceleration are integrated to compensate for the effects of degree and obtain local level velocities. The level speed is Divide by the Earth's radius to determine the angular velocity to compensate for the measured angular velocity of inertia. .

The first-order compensation of the 5AHR5 processor 20 transforms the airplane's coordinate system into a local level coordinate system. Determination of the directional cosine determinant associated with the stem. The resulting data is particularly free from isolated errors. is not accurate enough.

The more stringent accuracy requirements for 5AHR5 mean that the actual filter uses sensory output and This is combined with external reference data to roughly estimate the source of error.

The basis of the GPS system is the information transmitted from each artificial satellite. This information is for people This is the celestial position of the satellite and the time of signal transmission. The transition time is determined by the clock bias ( bias) is proportional to distance, excluding offset, and is the user's measurement of distance from the transmitting satellite. Has a fixed value. These measurements are called pseudoranges because of the clock bias.

The four simultaneous pseudorange measurements are the three components of the four unknown so-called principal positions. , is sufficient to allow the clock bias to be decoded. Error at initial position Knowing the influence of the received satellite signal and the initial time of the approximate Doppler shift of the received satellite signal is broadcast on a well-defined carrier frequency and usually tracked with a PLL (phase-locked loop) frequency. Gradual increase in tracking error and gradual decrease in detection sensitivity are locked will be a complete loss. Eliminates loss of lock, improves Doppler estimates, and improves In order to reduce the time between I can't stand it.

As shown in FIG. It includes an S-sensor status processor 26 and a common system status processor 28. This 5 The AHR5 state processor 24 calculates the instrument error of the 5AHRS system, but the The system error data is passed to system processor 28. Similarly, the GPS status analyzer is , calculates the equipment error of the GPS system, but calculates the system error by the system processor 28. pass through. This system error processor 28 stores system error data of 5A each. Provide negative feedback to HRS and GPS systems. DKF is 5AHR3 and GPS error is returned to each sensor processor 20 and 22. DKF is rolling, pitching Requested data including , self-heading, northward velocity, eastward velocity, vertical velocity, latitude, longitude and altitude Provide output.

Figure 4 shows how DKFlg is used to combine data from four processors. 2 shows another embodiment of the present invention. 5AHR5 and GPS processors 20 and 22 Besides, a reference sensor processor 30 and a satellite data processor 32 are formed. child The reference sensor processor 30 includes a magnetic self-heading base that determines pressure, altitude and true airspeed. Contains quasi-sensors. Critical self-heading error in the presence of sensory error of 5AHR3 In order to ensure that the external magnetic self-heading reference (7 lux valve) is selected. child These seven lux valves are used to create long-term, accurate self-propulsion. 7 la The Tux valve data and the gyro-driven self-heading data are combined via a filter. In this way, short-term and long-term self-heading accuracy can be achieved. 5AHRS algorithm Vertical velocity calculations require external reference to ensure stable velocity data . Accelerometer bias and incomplete gravity corrections result in a fairly short period of unbounded vertical Causes speed. Pressure altimeters can be used to isolate vertical velocity errors. is necessary. These local level velocities are transferred to each other on the earth's surface to calculate the angular velocity. used for. These angular velocities are calculated using the airplane fuselage to compensate for the measured angular velocities. It is transformed (mapped) into a projection view along the axis j1. Attitude error and velocity error are reference signals. without true airspeed and in the presence of some component error, including Sycora oscillations. I'm reading.

The processor 24 calculates 3 differentiated (!l!alF) from the 5AHRS sensor error model. It includes three types of states. Gyro error models can be categorized into the following five types. .

Scale factor error, 3 states Positioning misalignment error, 6 states Bias error, 3 states Mass unbalance drift error, 3 states Arbitrary noise error, 3 states Accelerometer models can be described in the following classifications:

Scale factor error, 3 states Positioning misalignment error, 6 states Bias error, 3 states Arbitrary noise error, 3 states A comprehensive network of satellites consists of at least four different satellites located on or around the Earth. It is arranged so that it lies on the local horizontal plane of almost every point in its vicinity. These four types of health The choice of stars greatly influences the accuracy of navigational positions. Satellite data processor 32 is preferred. Select a new satellite. The satellite selection algorithm consists of the following four steps.

1st step: Selection of the first satellite on the largest elevation angle, 2nd step: 1 from the first satellite Selection of the second satellite within a 10 degree range, third satellite located in the optimal direction based on the third layer Determination of stars, selection of the last satellite with the characteristics of minimum geometric dilution of the fourth stage stratification accuracy.

The satellite motion algorithm determines the position of each satellite by the satellite motion equation. child These equations are defined by the right ascension, elliptical inclination and latitude of the ascending node (pol It can be expressed in the Euler-Hill style, which is a coordinate system. These are Euler Hill The position vector of the satellite in the rotational mode is inertially fixed t; Let there be an orthogonal determinant to convert to Cartesian coordinates. The purpose of this algorithm is to Perturbed acceleration related to angular velocity vector and eccentricity vector on Elahill rotating frame The objective is to develop a lag-lung equation for the satellite motion, and to calculate the distance and and distance change rate.

The processor 26 includes 10 states derived from the GPS sensor error model. Ru. These include three distance measurement states, three distance change rate states, and one state and one clock rate of change state. The processor 28 determines the attitude and position of the airplane. and nine states derived from velocity.

The reconfiguration data handling system 34 provides fault monitoring, fault isolation, configuration selection, and data processing. Contains algorithms for forming data normalizations. Furthermore, analytical test calculations are It is designed to minimize the hardware requirements of the body. Normalization calculation process is the final output parameter that uses the best approximation data to derive the output parameters. This is a parameter calculation.

The GPS receiver provides pseudorange and delta range measurements and satellite data. 5AH R5 provides acceleration and velocity which are converted to navigation frame and attitude data. GPS Navigation equipment uses this signal to detect areas outside the range of the antenna or in poor terrain conditions each time the signal is output. Silent interval # due to high speed maneuver: Follow up and recollect the signal. 5AHR3 is , utilizes GPS position and velocity updates to perform instrument alignment and calibration.

Accurate position fix values from satellite data only prevent long-term inertial error growth Rather, various inertial errors can be estimated in real time and therefore compensated for. Fi The router error model is a GPS-assisted SAHRS error model with 10 elements. is obtained by increasing the state vector of . These 10 elements are G PS correlation error distance, distance change rate, clock bias, and clock change Show the rate.

ing. There are 46 error models for the entire state, and the update interval is 1 second.

Figure 5 shows the DKFlg arranged as a GPS-assisted SA HRS navigator. are doing. One way to eliminate long-term error growth from 5AHR5 is to use GPS. by periodically resetting the user's position coordinates using these precise fixed points. Ru. This configuration requires a DKF that approximates / of the error in 5AHR5, These errors are fed back negatively and only 5AHR5 is recalibrated. GPS location and speed Both degree measurements are provided to 5AHR3. Then the system updates for 1 second Provides an updated state that propagates sequentially through time 50 times up to the period of the period. 36 states The filter is installed on the GPS-assisted 5AHR3 navigation set. These error conditions includes 6 types of acceleration errors, 9 types of gyro errors, 12 types of accelerometer and gyro position It includes alignment errors and nine types of system errors.

The system in Figure 6 is implemented as a 5AHR5 supported GPS navigation device. It shows FlB. The GPS receiver is 4g from the Dogler shift of the carrier frequency. provides the data necessary to calculate the distance and rate of change in distance to satellite I;

Two important types of errors occur in measuring distance and rate of change of distance. The first error is the satellite This is due to the user's clock not being fully synchronized with the system's clock system. $ The error of 2 is due to the oscillator frequency error relative to the satellite's transmission frequency.

5AHR5 provides acceleration and velocity to assist the receiver in the PLL. D KF is formed in a two-step process. In the first stage, GPS pseudo-range measurements and speed input are performed. Estimate position from force.

The second stage uses the distance change rate measurements, the output from the first stage, and the acceleration input. . The filter style requires 16 error states, which correspond to 4 distance measurements, 4 measured distance change rate, three types of gyro bias values, and GPS receiver clock bias values. are the deviation value and bias change rate. The remainder of the distance measurement is calculated five times per second.

The measurement vector is based on 5AHR5 calculation which can be calculated 50 times per second.

The specification of the algorithm, outside of the specification of the analytical approximation algorithm, Addresses are also specified in the specifications for execution procedures such as detecting and responding to white noise. There is. This specification process incorporates these algorithms into the system's software procedures ( map) and modify the environment between them when the target device executes. It contains things that interact to satisfy the real-time constraints of the problem.

Symbols and descriptors in the following discussion are defined as follows. Mi's secondary For the first update time of the stem, X, , - state vector, 219. −Measurement base vector, vlvk - white measurement noise vector, W, - human white noise vector Kutle.

from the j-th subsystem state vector to the i-th subsystem state vector state transition determinant (matrix), H, , km, first subsystem state vector linear connection determinant from the vector to the jth subsystem measurement vector.

In a distributed Kalman filter, the starting point is an open set system of standard Kalman equations. derived from the system model. Therefore, it is divided into desired subsystems. The system is written in the linear vector equation where wk is the system noise and the covariance (Cov) is a zero-mean white noise process with .

Co　v　(who　w,)　−Qkδ++(E[wh]−0(2) However, Ql is a non-negative finite determinant, δ, 9. is the Dirac delta function.

The descriptor is an open set filter update time argument with k,j>O. system etc. Because the equation is often referred to as a system model, it provides the basic information to try to make decisions. It is listed.

The state vector (xh: k>O) is observed by the following noisy means: .

zk-HTkXk+vk″ (3) However, the measurement noise, V, is a zero-means white noise process with the following equation:

Cov (who v+) = RkJk+j″E[vh]-0’ (4) R3 is It is a non-negative finite determinant.

The measurement formula is called an observation model. For simplicity, W and V are as follows: Assume that they are uncorrelated.

Cov (w,, v4) = 0 For all j and k (5) The initial value of X is the following formula is an arbitrary variable.

E[Xo]-x. and Var　(Xo)　-Po　(6) Of course, the following can also be used. It is assumed.

Cow (X,, wk) = 0 for all k (5) The comprehensive state vector Xk is , X++h is the sensor 1 state vector, X2. k is the sensor 2 state vector and It can be partitioned into three substate vectors, where Xl, , is the sensor 3 state vector.

This scheme is shown in the Mi diagram and forms a distributed computing system model. minutes One of the differences between distributed jobs and traditional jobs is that jobs run latently on separate processes. It is the process of creating coherence in a set of inputs.

Therefore, These equations are expanded and rewritten into the following three types of separation systems.

Sensor-1 state space formula % formula % (10) Sensor-2 state space formula +l to X2+” Fzz+hX 2., +F2++hX,,,+ F zx + hX s + h + W lk (12) Z 2. h! Hz: +h’X ! l　h+　H+z+h”X　l+k　+　Hsx+hTX　!+k　+　V　 ik (13) Sensor-3 state space equation %formula%() Since the sensors X, , k and X, , k are almost independent, the following equation is obtained.

F rz = F 21 = O, Hzl = Hrz = O (16) Standard Kalmanf A filter is a linear discrete time finite dimensional system. These equations can be written as follows: Summarized.

The filter is initialized with:

xol−+−x,,p,I　−,m　Pa　(+7) These approximate values are as follows It is.

Cross*-1l-(Fk-KkH-”) Odor hlhr+K*Zk(I S)P　h+1Ik-F　k[P　1k-x-P　klh-+Hk(Hh”P　 hlh~IHk10Rk)-'Hk'P hl-xl Fh'+Q h (+9 ) The measurement value update formula is as follows.

K *-F *P 1m-+Hk (Hk Ding P hlh-+Hh+Rh)- ’ (2G)Q ih-+-Q klk-1+P hlk-rHh(Hh”P 1k-rHk+R,)-'(Zh-Hh"Qhlh-i)(21)P Jh-P ih-+-P ih-+Hh (Hh-P lh-+Hh+ Rk) -’Hk”P hlh-+ (H) However, based on the observation of one set of ranks, Zh=(Z+, Z!+ Zs,...zhl (23) Rhl k-, xEt xhl Zk-+1 (H) Cross kl k-E [Xkl Zk] (25) Standard A further expansion of the Kalman filter is that the three nonlinear side filters are no longer linear. obtained, and its performance is different than the original one. For some, the substate vector Tor partitions may diverge and are valid if well formed for other selections. It is essentially unused. Stability of distributed Kalman filter for basic selection of certain coordinates In order to ensure, one important characteristic is that the individual processors achieve a global impact. , code and data5, which interact to complete the approximate task. The goal is to ensure that.

These subsystem models are as follows.

X i+ *l-f ink (x tube) + Wink (ta) Z,, , m h i, k (X ink) + v ink (H) However, fh and The function is nonlinear, and i = 1.2 and 3, where - is the partial rate of change (derivative). be.

These approximate values are introduced to derive a clear sub-optimal filter for each model. be done.

f +, h (x h) = f +, h (Rlh) 10F Link (cross h-x hlh) ten --- (N) h +, h (X h) = h +, h (ex hlh)" Hz, h(X h-cross di-+) + --- (30) Therefore, the model is as follows It is.

+1” to X i+ (Ft tube) nk (31) z,,,x (H+++h)x,,+v,,+ Y ink (32) However, U l, h-f +, h (cross 1 + k l k) - F l l + k cross 1+klk+F+z, h cross 2+klk+F 13+ cross s, hlh (33) U z, h = f t, h (cross z + hL) - F xz * h cross! +klk+ F x r + h intersection r * h : h + Fz 3 * * intersection s, hL (34) U3 ・耘-f　3・k(交　1・kh)-F　ss・k　3・klk1　Fs+,h Cross 1 + klk” F 3! + Cross z, hL (35) Y +, i-h r, hc Cross 1+kl&-1)-HII+ and Cross 1+klk-1+H□1. Cross 2 + kl k -1" H31 + k Q3 + k l k -1 (") Y2・k "h2・h(Rz・klh-+) H22・hQx・klk-10H+z+hQ l+klk-1+H3! +lt intersection s, hlh-1(37)Y 5-h=h s, h (cross 3+klk-1)'! 3+ to S+klk-1÷H,... to history 1 +klk-□→Cross H23+7. 1 to -8 (3B) extended Kalman filter equation is as follows.

History i + klk-intersection, hlb-+ + L ink [Z,, k (Htl, h Cross,. klk-114 ink-1+hX i-1+klk-1”’ In +-2 ob intersection 53. Word 1, -+ (39) Cross + + k l k -1-F i + * k cross 1 + klk + F l + i-” +-1nklkikX 10F...-2・k history・-z, 1k(40)L　,,h-P　,,hlh-+H ,,h(Hi i, h'P +, it-+H+ i +k + Rink)-' (41) P Liklk= P l+:ni-I P ink h-IHLi nk [H+ ++h''Phi, -2H,,,,+R,,,)-'Hin K to Ph1h-+ (42) Pick, iL-F ++, kP i, ihF t+, h”1Q ink (43) Figure 7a shows the standard Kalman fit shown in Figure 1. The diagram shows Ruta's continuous system model. Approximate state 1′lL vector must be modeled by Eq.

X-FX+w The measurement relationship related to the noisy slippage I and Z with respect to the state vector] and LeX is as follows. It must be a formula and must be Jl.

Z = HX + v The processing method in the Bouyannel 40 is as follows: "Noise vector W" [W I, x′2+W3]”, and this vector is expressed in the consolidator 42 as In multiplier 44, by the linear connection matrix F in channel 46 Congruent with the multiplied state vector X “” [Xi, Xt, Xs]” and the current state vector x　　　[x　r　　+　Intersecting2.intersecting,]　Calculate the derivative of 0.

The output passes through an integrator 48 to calculate the current state vector X. this current situation The state vector may be passed through channel 46 for re-input to consolidator 42. , channel 4 as input to the multiplier multiplied by the linear connection matrix H It may remain at 0.) The output of the multiplexer (switcher) 44 is are combined in a combiner 42 to approximate the state vector.

The output of the multiplexer 50 is converted into a white noise permutation V"'[ V I + V 2 * V s ]”, the current state vector Z” [21, Z2. Z3]” is generated. Based on the system model in Figure 7a, , the equations of the standard Kalman filter are shown in Equations 17 to 25.

Figure 7b shows that the input white noise vector W, is in channel 60 and W2 is Indicates a method of distributed engineering where W3 is in channel 62 and W3 is in channel 64. ing. The processor 24 in FIG. 7b is located within the combiner 66 and within the channel 70. The previous state vector is multiplied in multiplier 68 by the linear connection matrix Fll of multiplier 7 by the torque XI and the linear connection matrix F' in channel 74. The previous state vector X, multiplied in 2 and the linear connection matrix in channel 78 The previous state vector X, which is multiplied within 972 by the It shows the input white noise component w, which has been synchronized. The output from the combiner 66 is The rate of change of the current state vector is calculated. This vector intersection is the integrator 80. to generate the current state vector X1. This current state vector, X, is It passes through channel 70, is multiplied by Fr+, and is sent to consolidator 66 in channel 60. is re-inputted and also multiplied by the linear connection matrix H8° in the multiplier 82. and further supplied to processors 26 and 28. X from the processor 28 is a channel is multiplied by HI3 in multiplier 88 of block 90. Channel 60.90 and 86? The sum of the outputs of The current measurement component 2. Calculate.

The processor 26 of FIG. Matrix F2! The previous state vector X2 multiplied in the multiplier 96 by , the linear connectivity matrix F! in channel 102. In multiplier 100 by 3 the multiplied state vector X, and the linear connection matrix in channel 106 congruent with the previous state vector X1 multiplied in the multiplier 104 by the vector F21 The input white noise component W is shown in FIG. The output from the combiner 94 is currently The change rate amount 2 of the state vector is calculated. This vector intersection is the integrator 108. to generate the current state vector X2. This current state vector X2 is Pass channel 98 and F. is multiplied by is input and also multiplied by the linear connection matrix H1 in the multiplier 110. , further supplied to processors 24 and 28. XI from processor 24 is channel It is multiplied by H21 in the multiplier 112 of 114. vector X from processor 28, is multiplied by HI3 in multiplier 116 of channel 118. channel 64 ．． The sum of the outputs from 11g and 114 is stored in the measurement white in the consolidator 120. and the current measurement component z2.

The processor 28 of FIG. The previous state vector multiplied in multiplier 124 by the connection matrix F33 X3 and the linear connection matrix F in channel 130. Nyotte multiplier 12B the previous state vector X2 multiplied within and the linear connection matrix within channel 134 the previous state vector X multiplied in multiplier 132 by box F31, and The combined input white noise component W is shown. The output from the combiner 122 is , the rate of change amount of the current state vector. This vector intersection is the integrator 1 36 to generate the current state vector, X,. This current state vector X, is passed through channel 126 and multiplied by F33 to obtain the congruence in channel 62. 122 and is also provided to multiplier processors 24 and 26. Processor xl from 24 is multiplied by H31 in multiplier 140 of channel 142.

Vector X2 from processor 26 is input to H3! in multiplier 144 of channel 146. is multiplied by The sum of the outputs from channels 62, 142 and 146 is 148 with the measured white noise sequence v3 to obtain the current measured component z3 Calculate. Based on the system model shown in Figure 7b, distributed Kalman filter The equations are executed according to equations 39 to 43. Dotted lines and nodes indicates an optional selection.

The system model in Figure 7b is implemented across multiple physical devices that communicate with each other. This figure shows the operation of the DKF. The DKF algorithm is improved as a final result. operating on system errors to remove errors from the system that form the performance It will be done.

The advantage of the DKF according to the present invention is that compared to conventional methods and devices, the operation is approximately 78% faster. , and a 57% reduction in the required calculator memory. Mixed 5AHR 3/GPS system, this results in good short-term performance of the 5AHR5 and This is the result of optimal combination with the long-term performance of PS.

ingredient FIG. 7A FtG, 7B international search report

Claims (1)

[Claims] 1. receiving instrument error status data from at least one sensor device processor (16); and a sensor state processor (12) for calculating sensor equipment error data; a sensor state processor further receiving system error state data from the sensor device processor; The system error data is calculated by combining the system error data (12) with a system status processor (14) that provides data to a sensor device processor (16); Output desired system data and process error data to at least one sensor device specific system data and equipment data with means for negative feedback to the device (16); at least one per system with at least one measuring device to form a A distributed Kalman filter that processes signals from two sensor devices. 2. The system may be a Suspended Attitude Head Reference System (SAHRS) or a Comprehensive Orientation System. The dispersion according to claim 1, which is a navigation system such as GPS (GPS) type Kalman filter. 3. Both the SAHRS and GPS are coupled to the distributed Kalman filter. A distributed Kalman filter according to claim 2. 4. The distributed Kalman filter network includes a SAHRS sensor state processor (24) and a GPS sensor status processor (26), the SAHRS and GP S is coupled to the system state processor (28). Distributed Kalman filter according to item 2 or 3. 5. The sensor state processor (24, 26) and the system state processor (28) are the noisy input signal to the first sensor current state vector and the system current state vector, respectively. a first means (66, 94, 122), and the sensor current state vector is obtained by integrating these differential sensor vectors. means (80, 108, 122) for calculating the second sensor current state vector; Claims 1 to 4, comprising means for converging with the first conjoining means. Distributed Kalman filter according to any of the above. 6. The sensor state processor (24, 26) and the system state processor (28) are a first and a second system current state vector, respectively, before merging with said first merging means; a first means (68, 96, 128 ) means and a second means (76, 100, 132), conjoined to the first conjoint means. before third means (72) for multiplying said second current state vector by a third matrix; , 104, 124). 7. The sensor state processor (24, 26) and the system state processor (28) are a second congruence congruent of at least two types of current state vectors, each with a noise vector; means (92, 120, 148), comprising means (92, 120, 148); The component according to claim 5 or 6, wherein the vectors are combined within the second combining means. Distributed Kalman filter. 8. The second merging means determines the sensor and system current state base before merging. The first matrix (82, 11, 146), the second matrix ( 88, 116, 138) and the third matrix (84, 112, 144) The distributed Kalman filter according to claim 5 or 6, further comprising means for. 9. In the sensor status processor, the device fault is detected from at least one sensor device processor. receive difference status data and calculate sensor equipment error data; In the system state processor, the system error state data is transmitted from the sensor device processor. calculate system error data, supplying the system error data to the sensor device processor; Output desired system data and process error data to at least one sensor device Distributed data that is negatively fed back to the device is formed by a Kalman filter and A system with at least one measuring device for forming data and device data. A distributed data processing method that processes signals from at least one sensor device per system. . 10. The system may be a Suspended Attitude Head Reference System (SAHRS) or a Comprehensive The person according to claim 9 which is a navigation system such as a positioning system (GPS) Law. 11. Both the SAHRS and GPS are coupled to the distributed Kalman filter. 11. The method according to claim 10. 12. The step of receiving and calculating equipment and system error conditions includes the step of: , the noisy input signal is connected to the first sensor current state vector and the system current state vector. A differential sensor vector is calculated congruently with the state vector, and this differential sensor vector integrating the sensor current state vector to generate the sensor current state vector; Claims 9 and 10 further comprising the step of merging the two sensor current state vectors. The method according to item 1 or item 11. 13. the first and second system current state vectors prior to the combination with the first combination means; The first and second matrices are multiplied by the first and second matrices, and the first conjoint means the second current state vector multiplied by a third matrix. The method according to item 12. 14. In the second concatenation means, the current state vector with the noise vector is reduced. The current measurement signal is calculated by combining at least two types, and the signal before the combination is calculated by the second combination means. each of the sensor and system current state vectors to a first, second and third matrix. 14. A method as claimed in claim 12 or claim 13, comprising the step of multiplying by lix.