State estimation and bit rate allocation collaborative optimization method for resource-constrained unmanned system

Abstract

The invention discloses a state estimation and bit rate allocation collaborative optimization method for a resource-constrained unmanned system, and the method comprises the steps: firstly building an unmanned system dynamics and sensor measurement model, introducing a quantization error model, determining the index relation between the quantization error model and a bit rate, and fusing bounded noise and quantization errors; then designing hybrid scheduling in static and dynamic segments, transmitting the two segments by adopting polling and event triggering protocols, and forming a complete measurement sequence by virtue of a scheduling matrix and a zero-order holder; designing a rolling time domain estimator, and constructing a cost function fusing multiple information to solve optimal estimation; and finally, constructing a bit rate allocation optimization problem by taking the cost function as a target, and solving by using a particle swarm algorithm to realize dynamic allocation under the limitation of the total bit rate. According to the scheme, high-precision, high-robustness and real-time state estimation of the unmanned system is realized under the condition that airborne computing resources and the total bit rate are strictly limited.

Description

Technical Field

[0001] This invention relates to unmanned system control technology, and in particular to a method for co-optimization of state estimation and bit rate allocation for resource-constrained unmanned systems. Background Technology

[0002] When unmanned systems perform tasks in complex and unknown environments, accurate estimation of their various states is necessary to ensure their safety and mission completion. However, due to the unique challenges posed by complex and unknown environments, state estimation for unmanned systems faces a series of technical difficulties, mainly including the following aspects:

[0003] Existing unmanned systems mostly employ fixed bit rate allocation or simple round-robin (RRP) protocols. This static resource allocation method cannot adaptively adjust to dynamically changing mission scenarios, environments, and the real-time perceived value of each sensor node (e.g., a sensor's data suddenly becoming exceptionally important). The direct consequence is that high-value data cannot be transmitted in a timely manner due to insufficient resources, while low-value data continuously consumes valuable communication bandwidth. While some event-triggered protocols (ETPs) offer some flexibility, they are insufficient in ensuring system stability. Unmanned systems require a scheduling mechanism that guarantees both real-time performance and flexibility.

[0004] In digital communication, quantization error is unavoidable. Existing methods treat quantization error as fixed noise, failing to establish a quantitative mathematical relationship between the allocation of communication resources (such as bit rate) and the accuracy of the final state estimation. This design approach, which separates quantization error from communication resources, causes the optimal allocation of communication resources to lose its direction in serving the final estimation goal.

[0005] To address the uncertainties introduced by quantization errors and communication scheduling, some advanced estimation algorithms (such as optimization algorithms considering complex constraints) can theoretically provide high-precision estimates. However, their computational complexity far exceeds the capacity of the limited onboard computing resources of unmanned systems, failing to meet the real-time requirements of state estimation. Conversely, some simple estimation algorithms with low computational cost and real-time operation suffer severe performance degradation and insufficient accuracy when dealing with quantization errors and intermittent data loss. This puts existing technologies in a dilemma: either they cannot perform the calculations effectively, or their accuracy is poor.

[0006] Therefore, there is an urgent need for an integrated solution that can deeply integrate communication scheduling, state estimation and bit rate allocation. Summary of the Invention

[0007] To address the aforementioned shortcomings, the present invention aims to propose a collaborative optimization method for state estimation and bit rate allocation in resource-constrained unmanned systems, thereby achieving high-precision, robust, and real-time state estimation of unmanned systems under strict constraints on onboard computing resources and total bit rate.

[0008] To achieve this objective, the present invention adopts the following technical solution:

[0009] A method for co-optimizing state estimation and bit rate allocation in resource-constrained unmanned systems includes the following steps:

[0010] S100. Establish the dynamic model and sensor measurement model of the unmanned system, including the state equation and measurement equation; describe the quantized measurement value through the component encoding and decoding process, introduce the quantization error model, establish the quantitative relationship between the quantization error and the allocated bit rate, clarify the exponential relationship between the quantization error limit and the bit rate, and finally integrate the bounded noise and quantization error augmentation.

[0011] S200: Design a hybrid scheduling mechanism to divide sensor nodes into static and dynamic segments; the static segment uses a polling protocol to ensure basic data transmission; the dynamic segment uses a normalized priority static event triggering protocol to select key nodes for transmission based on the real-time changes in node data; the scheduling matrix enables flexible control and recording of data transmission, and combines a zero-order hold to fill in missing data, forming a complete measurement sequence usable at the estimation end;

[0012] S300: Design a rolling time-domain estimator for quantization measurement and communication scheduling. Construct a simplified cost function that explicitly integrates quantization error, scheduling information and historical data. Obtain the optimal state estimate by solving the compensated measurement value and analytical solution method.

[0013] S400. Using the cost function of MHE as the optimization objective, an optimization problem with bit rate allocation as the decision variable is constructed and solved using the particle swarm optimization algorithm. By designing a fitness function and a constraint penalty term, dynamic bit rate allocation under the total bit rate constraint is achieved.

[0014] Preferably, in step S100:

[0015] A dynamic model of the unmanned system for state estimation is established, referring to a conventional UAV model; the state equation is described as follows:

[0016] ;

[0017] Wherein, the state vector Location in the northeast coordinate system Attitude angle (roll) , looking up ,yaw ), linear velocity and angular velocity ; This represents process noise, characterizing unmodeled dynamics and external disturbances;

[0018] The sensor measurement equation is described as follows:

[0019] ;

[0020] Matrix A and C are determined by system parameters and are known; the measurement vector... Composed of the outputs of various sensors; This scheme represents the measurement noise. and For unknown but bounded noise, i.e., UBB noise satisfies:

[0021] ;

[0022] ;

[0023] For encoding and decoding, we adopt a component-based encoding and decoding process, namely CBCDP; the measurement after quantization encoding and decoding can be expressed as:

[0024] ;

[0025] in, This represents the total number of nodes, and the node quantization error. and the overall quantization error vector It can be represented as follows:

[0026] ;

[0027] ;

[0028] The error limits are further expressed as follows:

[0029] ;

[0030] ;

[0031] in, This indicates that, given the scaling parameter, it is the first... The absolute value of the maximum range of each sensor; the specific value depends on the design of each sensor. Indicates the first Bit rate of each node;

[0032] The above quantization error limits show an exponential relationship between them and the allocated bit rate:

[0033] ;

[0034] Among them, communication resources are provided. And the direct influencing factors of performance estimation Based on the direct quantitative relationship between them, an optimization problem is constructed to systematically allocate bit rate resources;

[0035] Quantization introduces quantization error. To simplify the subsequent derivation and to follow the principle of concentrating all bounded perturbations into a single constraint uncertainty vector, we will... , and Augmented to a vector And impose a joint norm bound, This represents the dimension of node i. Indicates the total dimension of the measurement;

[0036] The augmented and redefined system model, along with the quantized and decoded measurement dynamics, can be obtained as follows:

[0037] ;

[0038] in, ; ; .

[0039] Preferably, step S200 includes the following steps:

[0040] S210, the communication protocol basis of the hybrid scheduling protocol;

[0041] S211. Modify the original FlexRay protocol, retain the communication cycle structure of static and dynamic segments, and ignore the symbol window and network idle time;

[0042] S212, Protocol Segmentation and Node Grouping:

[0043] The FlexRay communication cycle is divided into static and dynamic segments, corresponding to two sets of sensor nodes:

[0044] Static segment corresponding set The scheduling is done using the Round Robin Protocol (RRP).

[0045] Dynamic segment corresponding set The redesigned static event triggering protocol, SETP, is used for scheduling.

[0046] S220 and FlexRay protocol scheduling mechanisms;

[0047] S221. Initialization: Set the scheduling matrix Initialize it as an all-zero matrix, indicating that all nodes are initially in an unscheduled state;

[0048] S222, Static Segment Scheduling, also known as RRP: Traversing the static segment node group Determine the transmission node at time k based on RRP. ,

[0049] ;

[0050] Then, a unit matrix block is allocated to it in the scheduling matrix. At this time, the node measurement information that the static node can successfully transmit is expressed by the following formula:

[0051] ;

[0052] S223, Dynamic Segment Scheduling (SETP): Traversing Dynamic Segment Node Groups Initialize the set of dynamic segment nodes that have obtained transmission rights. Calculate the competition error for each node. :

[0053] ;

[0054] in, This represents the measurement value that was successfully decoded and transmitted from node j before time k.

[0055] And node normalization priority:

[0056] ;

[0057] in, This is a weight matrix; weight values ​​are assigned as needed. The characteristic range is designed according to actual needs, and can be the expected measurement value, the sensor range, etc.

[0058] Nodes exceeding the normalization threshold are granted transmission rights.

[0059] ;

[0060] Simultaneously update the set of dynamic segment nodes that have obtained transmission rights. ,

[0061] ;

[0062] And allocate an identity matrix block to it in the scheduling matrix.

[0063] ;

[0064] The node measurement information that a dynamic node can successfully transmit is expressed by the formula (15).

[0065] ;

[0066] S224. Combined Scheduling Matrix: Combines the scheduling results of static and dynamic segments into a final block diagonal scheduling matrix. ;

[0067] ;

[0068] The measurement information of the node that has been successfully transmitted at this time can be expressed as:

[0069] ;

[0070] S230, the specific operation of the scheduling matrix and the zero-order hold;

[0071] At the front end of the estimator, a zero-order hold (ZOH) is used to fill in the missing node measurements with the values ​​from the previous successful transmission of that node, ultimately yielding a received measurement value from the estimator. as follows:

[0072] ;

[0073] in, yes After the data segment is quantized and encoded at any given time, substituting formula (8c) into formula (16) will yield the following expression for the measurement received at the estimator front end after noise interference augmentation:

[0074] ;

[0075] In the formula, the data received by the estimator is... It is a complex function that includes system state. Measurement noise Quantization error Scheduling matrix and historical data .

[0076] Furthermore, in step S300:

[0077] The measurement used in the estimation is a complex function containing scheduling information, quantized measurement values, and historical data. Therefore, the cost function in the classic MHE objective function also needs to be modified accordingly to accurately estimate the state. The modified objective function is as follows:

[0078] ;

[0079] ;

[0080] The cost function needs to satisfy the following constraints:

[0081] ;

[0082] ;

[0083] And noise constraint formula (3) and quantization error constraint formula (6);

[0084] in, ;

[0085] This represents the known information in the measurement cost term. This indicates the compensated measurement value. For the scheduling matrix, yes The covariance matrix of the known prior estimate error at time t, where the diagonal elements are determined based on the physical properties of the state variables and prior knowledge; The metric matrix representing the interference level of each sensor node, and its inverse matrix The weight matrix in the cost function is used to penalize measurement bias. The structure should be a block diagonal matrix, corresponding to the grouping of sensor nodes: the smaller the interference boundary, the smaller the measurement uncertainty of that node. The larger the value of the corresponding diagonal element, the better. The smaller the diagonal elements, the lower the penalty weight assigned to the measurement channel of that node in the cost function. Specifically, the... Nodes The design is as follows:

[0086] ;

[0087] in, It is the first The upper limit of the measurement noise boundary at the node, and the other term in the denominator is the square of the quantization error. ;

[0088] This approach ensures that measurement noise and quantization error are weighed at the same power, consistent with the physical meaning of the MHE weight design. This weight matrix integrates the effects of measurement noise and quantization error, enabling the estimator to adaptively adjust the confidence level of the data at each node.

[0089] Finally, the overall interference penalty weight metric matrix can be expressed as: ;

[0090] In equation (18), the objective function of MHE lacks a process cost term because the prior of this scheme is... The smooth update mechanism used is The state evolution is implicitly constrained by using the optimal estimate from the previous moment as a priori.

[0091] In the case of UBB, the goal of MHE is usually not to minimize the norm of process noise, but to ensure the final boundedness of the estimation error through boundedness analysis; the influence of process noise is implicit in the state transition model and is indirectly handled through prior terms and measurement terms.

[0092] Unlike recursive filtering methods that require state augmentation to clarify delays caused by analog communication, rolling time-domain estimation naturally adapts to these effects by using a fixed-length measurement window; at each time k, MHE utilizes the received measurement sequence and scheduling information The structural delay in the static segment and the zero-order preservation behavior in the dynamic segment are inherently captured in this measurement sequence; Effectively extract "new" information from each measurement, while historical values It is considered a known constant within the estimation window; this method avoids the need for state augmentation while fully considering the scheduling effect of the FlexRay protocol.

[0093] This patent uses the analytical solution to the MHE problem, which can be used to obtain the solution in... The optimal estimate for time is as follows:

[0094] ;

[0095] The analytical solution is the result of solving the optimization problem of formula (18). This analytical solution avoids the complex numerical optimization process and is suitable for real-time operation under the limited onboard computing power of unmanned systems. , ;

[0096] The calculation process of the state estimation module can be summarized as follows: at each time step... This estimator uses data within a fixed-length N+1 time window and solves the aforementioned optimization problem to obtain the result from... arrive The optimal state estimation sequence at time t.

[0097] Furthermore, in step S400:

[0098] S410, Construction optimization problem;

[0099] To achieve optimal bit rate allocation, the MHE-PSO collaborative optimization algorithm is proposed, and its optimization problem is:

[0100] ;

[0101] ;

[0102] ,

[0103] ;

[0104] in, It is the objective function of the MHE part (18). This is a penalty term for the total bit rate constraint. It is a sufficiently large penalty coefficient, chosen to be several orders of magnitude larger than the typical MHE cost value, such as The objective function ensures that any violation of the total bit rate constraint will be severely punished. Ensure that there are enough bits available for data encoding after considering the data ID;

[0105] This optimization problem is solved through the following steps;

[0106] The fitness function for each particle is defined as:

[0107] ;

[0108] in, , Using particles The MHE cost function represents the bit rate allocation evaluation.

[0109] This function is a specific implementation of the optimization objective of formula (24). Formula (24) directly minimizes the MHE objective function. The MHE objective function is mathematically used as a proxy index for state estimation error. Minimizing the objective function will also indicate a reduction in state estimation error. After PSO, a bit rate allocation scheme with smaller estimation error will be obtained. This allows PSO to directly optimize the MHE objective function and more directly reflect the estimation performance of the overall system.

[0110] S420 and MHE-PSO collaborative optimization algorithm;

[0111] S421. Initialization: Randomly initialize particle positions. and speed Initialize your personal best position and individual optimal fitness Global best position and global best fitness ;

[0112] S422, Fitness Assessment: For each particle Using the objective function Calculate fitness and impose penalties for violations of constraints;

[0113] ;

[0114] In the formula, correspond Penalty items correspond ;

[0115] S423, Update Individual and Global Optimum: Update the individual optimal position of each particle based on its fitness value. and global optimal position ;

[0116] For the particles obtained in S422 Corresponding fitness function Update;

[0117] 1) Update the individual optimality: If the fitness function of the current particle i is... Less than the individual's optimal fitness So, let's set the best position for each person. Assigned to the current particle Location Then, the individual's optimal fitness Assigned to the current particle fitness The formula is expressed as follows:

[0118] if ,So ;

[0119] 2) Update global optimum: If the current particle If the fitness of particle i is less than the global best fitness, then the global best position is assigned to the position of the current particle i, and the global best fitness is assigned to the fitness of the current particle i. The formula is expressed as follows:

[0120] if ,So ;

[0121] S424, Velocity and Position Update: Adjust particle velocity using inertial, cognitive, and social components, then update particle velocity and position;

[0122] 1) Speed ​​update: using inertia Cognitive and social weight and the generated random numbers To adjust particle speed The formula is expressed as follows:

[0123] ;

[0124] 2) Position update: Based on the particle velocity obtained above The formula for updating particle position is expressed as follows:

[0125] ;

[0126] S425, Forced Constraints: Ensure the bit rate is an integer value through appropriate rounding and clamping operations. And bit rate greater than minimum value ;

[0127] S426. Termination: First, update the inertia weights in each iteration. ,from linearly decreasing to The expression is as follows:

[0128] ;

[0129] Finally, repeat steps S422-S425 until the maximum number of iterations is reached. After iteration After this, a globally optimal position is obtained. ,this The bit rate allocation scheme for a given location is considered the optimal bit rate allocation scheme. .

[0130] One of the above technical solutions includes the following beneficial effects: The communication-estimation collaborative optimization architecture designed in this patent can achieve deep collaboration between communication and estimation requirements. Compared with existing methods, this patent, through the joint design of improved FlexRay protocol hybrid scheduling and quantization coding, enables communication scheduling to directly serve state estimation, and can adaptively select key nodes for data transmission through thresholds, effectively alleviating network congestion and improving communication efficiency and overall estimation performance. Attached Figure Description

[0131] Figure 1 This is a schematic diagram of the overall logic of the patent;

[0132] Figure 2 It is an improved FlexRay protocol communication cycle structure;

[0133] Figure 3 It is an estimation process for a network system based on the improved FlexRay protocol communication transmission;

[0134] Figure 4 This is a schematic diagram of a rolling time-domain estimation (MHE) timing window that includes scheduling information;

[0135] Figure 5It is the MHE-PSO collaborative optimization architecture. Detailed Implementation

[0136] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0137] A method for co-optimizing state estimation and bit rate allocation in resource-constrained unmanned systems includes the following steps:

[0138] S100. Establish the dynamic model and sensor measurement model of the unmanned system, including the state equation and measurement equation; describe the quantized measurement value through the component encoding and decoding process, introduce the quantization error model, establish the quantitative relationship between the quantization error and the allocated bit rate, clarify the exponential relationship between the quantization error limit and the bit rate, and finally integrate the bounded noise and quantization error augmentation.

[0139] S200: Design a hybrid scheduling mechanism to divide sensor nodes into static and dynamic segments; the static segment uses a polling protocol to ensure basic data transmission; the dynamic segment uses a normalized priority static event triggering protocol to select key nodes for transmission based on the real-time changes in node data; the scheduling matrix enables flexible control and recording of data transmission, and combines a zero-order hold to fill in missing data, forming a complete measurement sequence usable at the estimation end;

[0140] S300: Design a rolling time-domain estimator for quantization measurement and communication scheduling. Construct a simplified cost function that explicitly integrates quantization error, scheduling information and historical data. Obtain the optimal state estimate by solving the compensated measurement value and analytical solution method.

[0141] S400. Using the cost function of MHE as the optimization objective, an optimization problem with bit rate allocation as the decision variable is constructed and solved using the particle swarm optimization algorithm. By designing a fitness function and a constraint penalty term, dynamic bit rate allocation under the total bit rate constraint is achieved.

[0142] This patented communication-estimation collaborative optimization architecture enables deep collaboration between communication and estimation requirements. Compared to existing methods, this patent, through a joint design of improved FlexRay protocol hybrid scheduling and quantization coding, allows communication scheduling to directly serve state estimation. It can adaptively select key nodes for data transmission via thresholds, effectively alleviating network congestion and improving communication efficiency and overall estimation performance. The lightweight rolling time-domain estimator designed for quantization measurement and communication scheduling in this patent can achieve high-precision state estimation even with incomplete communication, missing measurements, and limited onboard computing resources. Compared to existing methods, this patent's design, by explicitly processing quantization errors and historical data and employing analytical solution methods, can effectively suppress accumulated errors under limited onboard computing resources, ensuring the convergence and accuracy of estimation while meeting the real-time requirements of unmanned systems for state estimation. The MHE-PSO collaborative optimization algorithm designed in this patent can dynamically allocate the bit rate of each sensor node based on the state estimation error. Compared with existing methods, this patent achieves the overall optimization of bit rate allocation scheme and estimation performance by using MHE to evaluate estimation performance in real time and combining it with PSO to dynamically optimize bit rate allocation, thereby improving the state estimation accuracy of unmanned systems in bit rate-constrained environments.

[0143] like Figure 1 As shown, in step S100:

[0144] A dynamic model of the unmanned system for state estimation is established, referring to a conventional UAV model; the state equation is described as follows:

[0145] ;

[0146] Wherein, the state vector Location in the northeast coordinate system Attitude angle (roll) , looking up ,yaw ), linear velocity and angular velocity ; This represents process noise, characterizing unmodeled dynamics and external disturbances;

[0147] The sensor measurement equation is described as follows:

[0148] ;

[0149] Matrix A and C are determined by system parameters and are known; the measurement vector... Composed of the outputs of various sensors; This scheme represents the measurement noise. and For unknown but bounded (UBB) noise, the following conditions are met:

[0150] ;

[0151] ;

[0152] For encoding and decoding, we adopt a component-based encoding and decoding process, namely CBCDP; the measurement after quantization encoding and decoding can be expressed as:

[0153]

[0154] in, This represents the total number of nodes, and the node quantization error. and the overall quantization error vector It can be represented as follows:

[0155] ;

[0156] ;

[0157] The error limits are further expressed as follows:

[0158] ;

[0159] ;

[0160] in, This represents the absolute value of the maximum range of the i-th sensor, given the scaling parameter. The specific value depends on the design of each sensor. This represents the bit rate of the i-th node;

[0161] The above quantization error limits show an exponential relationship between them and the allocated bit rate:

[0162] ;

[0163] Among them, communication resources are provided. And the direct influencing factors of performance estimation Based on the direct quantitative relationship between them, an optimization problem is constructed to systematically allocate bit rate resources;

[0164] Quantization introduces quantization error. To simplify the subsequent derivation and to follow the principle of concentrating all bounded perturbations into a single constraint uncertainty vector, we will... , and Augmented to a vector And impose a joint norm bound, This represents the dimension of node i. Indicates the total dimension of the measurement;

[0165] The augmented and redefined system model, along with the quantized and decoded measurement dynamics, can be obtained as follows:

[0166] ;

[0167] in, ; ; .

[0168] By establishing an exponential relationship between the quantization error bound and the allocated bit rate, the quantitative impact of communication resource allocation on state estimation performance is clarified, providing a direct theoretical basis for subsequent optimization problems. This allows bit rate allocation to be based on quantitative predictions of estimation performance, rather than relying on experience or trial and error, thus achieving a direct correlation between communication resources and estimation accuracy. Simultaneously, by fusing bounded noise and quantization error augmentation into a single vector, the system model structure is simplified, the complexity of subsequent estimator design is reduced, and it facilitates the unified handling of multiple disturbance sources.

[0169] like Figure 2 As shown, step S200 includes the following steps:

[0170] S210, the communication protocol basis of the hybrid scheduling protocol;

[0171] S211. Modify the original FlexRay protocol, retain the communication cycle structure of static and dynamic segments, and ignore the symbol window and network idle time;

[0172] S212, Protocol Segmentation and Node Grouping:

[0173] The FlexRay communication cycle is divided into static and dynamic segments, corresponding to two sets of sensor nodes:

[0174] Static segment corresponding set The scheduling is done using the Round Robin Protocol (RRP).

[0175] Dynamic segment corresponding set The redesigned static event triggering protocol, SETP, is used for scheduling.

[0176] S220 and FlexRay protocol scheduling mechanisms;

[0177] S221. Initialization: Set the scheduling matrix Initialize it as an all-zero matrix, indicating that all nodes are initially in an unscheduled state;

[0178] S222, Static Segment Scheduling, also known as RRP: Traversing the static segment node group Determine the transmission node at time k based on RRP. ,

[0179] ;

[0180] Then, a unit matrix block is allocated to it in the scheduling matrix. At this time, the node measurement information that the static node can successfully transmit is expressed by the following formula:

[0181] ;

[0182] S223, Dynamic Segment Scheduling (SETP): Traversing Dynamic Segment Node Groups Initialize the set of dynamic segment nodes that have obtained transmission rights. Calculate the competition error for each node. :

[0183] ;

[0184] in, Indicates time Previously from the node Measurement values ​​successfully decoded and transmitted.

[0185] And node normalization priority:

[0186] ;

[0187] in, This is a weight matrix; weight values ​​are assigned as needed. The characteristic range is designed according to actual needs, and can be the expected measurement value, the sensor range, etc.

[0188] Nodes exceeding the normalization threshold are granted transmission rights.

[0189] ;

[0190] Simultaneously update the set of dynamic segment nodes that have obtained transmission rights. ,

[0191] ;

[0192] And allocate an identity matrix block to it in the scheduling matrix.

[0193] ;

[0194] The node measurement information that a dynamic node can successfully transmit is expressed by formula (15):

[0195] ;

[0196] S224. Combined Scheduling Matrix: Combines the scheduling results of static and dynamic segments into a final block diagonal scheduling matrix. ;

[0197] ;

[0198] The measurement information of the node that has been successfully transmitted at this time can be expressed as:

[0199] ;

[0200] S230, the specific operation of the scheduling matrix and the zero-order hold;

[0201] At the front end of the estimator, a zero-order hold (ZOH) is used to fill in the missing node measurements with the values ​​from the previous successful transmission of that node, ultimately yielding a received measurement value from the estimator. as follows:

[0202] ;

[0203] in, yes The data of the entire data segment after time-quantization encoding. Let represent the total dimension of the measurement. Substituting formula (8c) into formula (16) yields the following expression for the measurement received at the estimator front end after noise-induced augmentation:

[0204] ;

[0205] In the formula, the data received by the estimator is... It is a complex function that includes system state. Measurement noise Quantization error Scheduling matrix and historical data .

[0206] To address the real-time requirements of resource-constrained unmanned systems in complex environments, the FlexRay protocol was improved. It divides the communication cycle into static and dynamic segments, using polling and normalized priority static event-triggered protocols for scheduling, respectively. This hybrid scheduling mechanism adaptively selects key nodes for transmission based on the real-time changes in sensor node data, effectively avoiding network congestion and improving communication efficiency. Simultaneously, it ensures the necessary data transmission for state estimation and relies on a zero-order hold to supplement the complete measurement information required by the estimator, thus providing a stable and timely data supply for core functional modules such as the backend state estimator.

[0207] In addition, in step S300:

[0208] The measurement used in the estimation is a complex function containing scheduling information, quantized measurement values, and historical data. Therefore, the cost function in the classic MHE objective function also needs to be modified accordingly to accurately estimate the state. The modified objective function is as follows:

[0209] ;

[0210] ;

[0211] The cost function needs to satisfy the following constraints:

[0212] ;

[0213] ;

[0214] And noise constraint formula (3) and quantization error constraint formula (6);

[0215] in,

[0216] ;

[0217] This represents the known information in the measurement cost term. This indicates the compensated measurement value. For the scheduling matrix, yes The covariance matrix of the known prior estimate error at time t, where the diagonal elements are determined based on the physical properties of the state variables and prior knowledge; The metric matrix representing the interference level of each sensor node, and its inverse matrix The weight matrix in the cost function is used to penalize measurement bias. The structure should be a block diagonal matrix, corresponding to the grouping of sensor nodes: the smaller the interference boundary, the smaller the measurement uncertainty of that node. The larger the value of the corresponding diagonal element, the better. The smaller the diagonal elements, the lower the penalty weight assigned to the measurement channel of that node in the cost function. Specifically, the... Nodes The design is as follows:

[0218] ;

[0219] in, It is the first The upper limit of the measurement noise boundary at the node, and the other term in the denominator is the square of the quantization error. ;

[0220] This approach ensures that measurement noise and quantization error are weighed at the same power, consistent with the physical meaning of the MHE weight design. This weight matrix integrates the effects of measurement noise and quantization error, enabling the estimator to adaptively adjust the confidence level of the data at each node.

[0221] Finally, the overall interference penalty weight metric matrix can be expressed as: ;

[0222] In equation (18), the objective function of MHE lacks a process cost term because the prior of this scheme is... The smooth update mechanism used is The state evolution is implicitly constrained by using the optimal estimate from the previous moment as a priori.

[0223] In the case of UBB, the goal of MHE is usually not to minimize the norm of process noise, but to ensure the final boundedness of the estimation error through boundedness analysis; the influence of process noise is implicit in the state transition model and is indirectly handled through prior terms and measurement terms.

[0224] Unlike recursive filtering methods that require state augmentation to clarify delays caused by analog communication, rolling time-domain estimation naturally adapts to these effects by using a fixed-length measurement window; at each time k, MHE utilizes the received measurement sequence and scheduling information The structural delay in the static segment and the zero-order preservation behavior in the dynamic segment are inherently captured in this measurement sequence; Effectively extract "new" information from each measurement, while historical values It is considered a known constant within the estimation window; this method avoids the need for state augmentation while fully considering the scheduling effect of the FlexRay protocol.

[0225] This patent uses the analytical solution to the MHE problem, which can be used to obtain the solution in... The optimal estimate for time is as follows:

[0226] ;

[0227] The analytical solution is the result of solving the optimization problem of formula (18). This analytical solution avoids the complex numerical optimization process and is suitable for real-time operation under the limited onboard computing power of unmanned systems. ,

[0228] ;

[0229] The rolling time-domain estimation timing window and the information needed within the window, such as... Figure 4 As shown.

[0230] The calculation process of the state estimation module can be summarized as follows: at each time step... This estimator uses data within a fixed-length N+1 time window and solves the aforementioned optimization problem to obtain the result from... arrive The optimal state estimation sequence at time t.

[0231] A rolling time-domain estimator using compensated measurements was designed. By explicitly incorporating the quantization error model and communication scheduling sequence into the optimization problem, the estimator can effectively overcome the interference caused by data quantization and communication delay. It obtains the optimal state estimate through analytical solution, avoiding complex numerical optimization processes, reducing computational burden, and ensuring estimation accuracy and convergence while meeting real-time requirements under limited onboard computing resources. This provides reliable internal state information for the system to achieve accurate decision-making and behavior planning.

[0232] The rolling time-domain estimator window length N here is a preset positive integer and a key adjustable parameter. Its value is determined by balancing the system state dimension, the desired estimation accuracy, and the available computing and storage resources of the system. There is no very precise standard, and it is not fixed. Instead, it is adjusted according to the actual system requirements.

[0233] In the optimization process of the MHE estimator, formulas (3), (6), (19), and (20) are all constraints on the cost function. The noise boundedness constraint defined by formula (3) and the quantization error boundedness constraint defined by formula (6) together constitute the theoretical basis for the estimator design, which is to ensure the final boundedness of the estimation error.

[0234] This patent transforms a complex nonlinear optimization problem into a rapidly computable linear algebraic problem by simplifying the cost function and deriving the analytical solution. This is akin to finding a highway (analytical solution) leading to the destination (high-precision estimation), replacing the original, uncertain, and arduous mountain path (numerical iterative optimization), thus simultaneously satisfying the dual requirements of real-time performance and high accuracy.

[0235] In addition, in step S400:

[0236] S410, Construction optimization problem;

[0237] To achieve optimal bit rate allocation, the MHE-PSO collaborative optimization algorithm is proposed, and its optimization problem is:

[0238] ;

[0239] ;

[0240] ,

[0241] ;

[0242] in, It is the objective function of the MHE part (18). This is a penalty term for the total bit rate constraint. It is a sufficiently large penalty coefficient, chosen to be several orders of magnitude larger than the typical MHE cost value, such as The objective function ensures that any violation of the total bit rate constraint will be severely punished. Ensure that there are enough bits available for data encoding after considering the data ID;

[0243] This optimization problem is solved through the following steps;

[0244] The fitness function for each particle is defined as:

[0245] ;

[0246] in, , Using particles The MHE cost function represents the bit rate allocation evaluation.

[0247] This function is a specific implementation of the optimization objective of formula (24). Formula (24) directly minimizes the MHE objective function. The MHE objective function is mathematically used as a proxy index for state estimation error. Minimizing the objective function will also indicate a reduction in state estimation error. After PSO, a bit rate allocation scheme with smaller estimation error will be obtained. This allows PSO to directly optimize the MHE objective function and more directly reflect the estimation performance of the overall system.

[0248] S420 and MHE-PSO collaborative optimization algorithm;

[0249] S421. Initialization: Randomly initialize particle positions. and speed Initialize your personal best position and individual optimal fitness Global best position and global best fitness ;

[0250] S422, Fitness Assessment: For each particle Using the objective function Calculate fitness and impose penalties for violations of constraints;

[0251] ;

[0252] In the formula, correspond Penalty items correspond ;

[0253] S423, Update Individual and Global Optimum: Update the individual optimal position of each particle based on its fitness value. and global optimal position ;

[0254] For the particles obtained in S422 Corresponding fitness function Update;

[0255] 1) Update the individual optimality: If the fitness function of the current particle i is... Less than the individual's optimal fitness So, let's set the best position for each person. Assigned to the current particle Location Then, the individual's optimal fitness Assigned to the current particle fitness The formula is expressed as follows:

[0256] if ,So ;

[0257] 2) Update the global optimum: If the fitness of the current particle i is less than the global optimum fitness, then the global optimum position is assigned to the position of the current particle i, and the global optimum fitness is assigned to the fitness of the current particle i. The formula is expressed as follows:

[0258] if ,So ;

[0259] S424, Velocity and Position Update: Adjust particle velocity using inertial, cognitive, and social components, then update particle velocity and position;

[0260] 1) Speed ​​update: using inertia Cognitive and social weight and the generated random numbers To adjust particle speed The formula is expressed as follows:

[0261] ;

[0262] 2) Position update: Based on the particle velocity obtained above The formula for updating particle position is expressed as follows:

[0263] ;

[0264] S425, Forced Constraints: Ensure the bit rate is an integer value through appropriate rounding and clamping operations. And bit rate greater than minimum value ;

[0265] S426. Termination: First, update the inertia weights in each iteration. ,from linearly decreasing to The expression is as follows:

[0266] ;

[0267] Finally, repeat steps S422-S425 until the maximum number of iterations is reached. After iteration After this, a globally optimal position is obtained. ,this The bit rate allocation scheme for a given location is considered the optimal bit rate allocation scheme. .

[0268] Collaborative optimization frameworks, such as Figure 5 As shown, a deep coupling between MHE state estimation and PSO bit rate allocation is achieved: at the objective level, PSO directly uses the objective function of MHE as the optimization objective, unifying the requirements of total communication bit rate allocation and state estimation performance; at the information flow level, a closed-loop mechanism of "bit rate allocation → state estimation → performance feedback → bit rate allocation" is established; at the time scale level, the fast estimation cycle of MHE and the slow optimization cycle of PSO form a hierarchical coordination; at the constraint processing level, hard communication constraints are flexibly integrated into the optimization objective through a penalty function. This multi-level collaborative mechanism breaks the limitations of traditional separate design, enabling the system to autonomously coordinate internal bit rates in a dynamic network environment, achieving an overall optimal balance between communication efficiency and estimation accuracy.

[0269] The proposed MHE-PSO collaborative optimization algorithm uses the cost function of the rolling time-domain estimator as the optimization objective, constructs an optimization problem with bit rate allocation as the decision variable, and solves it using the particle swarm optimization algorithm, achieving dynamic allocation under total bit rate constraints. Through the design of the fitness function and constraint penalty term, the algorithm enables bit rate allocation to directly serve the optimization of state estimation performance, improving the overall estimation accuracy of unmanned systems under bit rate-constrained environments.

[0270] The technical principles of the present invention have been described above with reference to specific embodiments. These descriptions are merely for explaining the principles of the invention and should not be construed as limiting the scope of protection of the invention in any way. Based on this explanation, those skilled in the art can readily conceive of other specific embodiments of the invention without inventive effort, and these embodiments will all fall within the scope of protection of the present invention.

Claims

1. A method for coordinated optimization of state estimation and bit rate allocation in resource-constrained unmanned systems, characterized in that, Includes the following steps: S100. Establish the dynamic model and sensor measurement model of the unmanned system, including the state equation and measurement equation; describe the quantized measurement value through the component encoding and decoding process, introduce the quantization error model, establish the quantitative relationship between the quantization error and the allocated bit rate, clarify the exponential relationship between the quantization error limit and the bit rate, and finally integrate the bounded noise and quantization error augmentation. S200: Design a hybrid scheduling mechanism to divide sensor nodes into static and dynamic segments; Static segments use a polling protocol to ensure basic data transmission; The dynamic segment adopts a static event triggering protocol with normalized priority, and selects key nodes for transmission based on the real-time changes in node data; the scheduling matrix enables flexible control and recording of data transmission, and combines a zero-order hold to fill in missing data, forming a complete measurement sequence usable at the estimation end; S300: Design a rolling time-domain estimator for quantization measurement and communication scheduling. Construct a simplified cost function that explicitly integrates quantization error, scheduling information and historical data. Obtain the optimal state estimate by solving the compensated measurement value and analytical solution method. S400. Using the cost function of MHE as the optimization objective, an optimization problem with bit rate allocation as the decision variable is constructed and solved using the particle swarm optimization algorithm. By designing a fitness function and a constraint penalty term, dynamic bit rate allocation under the total bit rate constraint is achieved.

2. The method for coordinated optimization of state estimation and bit rate allocation for resource-constrained unmanned systems according to claim 1, characterized in that, In step S100: A dynamic model of the unmanned system for state estimation is established, referring to a conventional UAV model; the state equation is described as follows: ； Wherein, the state vector Location in the northeast coordinate system Attitude angle (roll) , looking up ,yaw ), linear velocity and angular velocity ; This represents process noise, characterizing unmodeled dynamics and external disturbances; The sensor measurement equation is described as follows: ； Matrix A and C are determined by system parameters and are known; the measurement vector... Composed of the outputs of various sensors; This scheme represents the measurement noise. and For unknown but bounded noise, i.e., UBB noise satisfies: ； ； For encoding and decoding, we adopt a component-based encoding and decoding process, namely CBCDP; the measurement after quantization encoding and decoding can be expressed as: ； in, This represents the total number of nodes, and the node quantization error. and the overall quantization error vector It can be represented as follows: ； ； The error limits are further expressed as follows: ； ； in, Indicates that given the scaling parameter, it is the first... The absolute value of the maximum range of each sensor; the specific value depends on the design of each sensor. Indicates the first Bit rate of each node; The above quantization error limits show an exponential relationship between them and the allocated bit rate: ； Among them, communication resources are provided. And the direct influencing factors of performance estimation Based on the direct quantitative relationship between them, an optimization problem is constructed to systematically allocate bit rate resources; Quantization introduces quantization error. To simplify the subsequent derivation and to follow the principle of concentrating all bounded perturbations into a single constraint uncertainty vector, we will... , and Augmented to a vector And impose a joint norm bound, Represents a node dimensionality Indicates the total dimension of the measurement; The augmented and redefined system model, along with the quantized and decoded measurement dynamics, can be obtained as follows: ； in, ; ; .

3. The method for coordinated optimization of state estimation and bit rate allocation for resource-constrained unmanned systems according to claim 2, characterized in that, Step S200 includes the following steps: S210, the communication protocol basis of the hybrid scheduling protocol; S211. Modify the original FlexRay protocol, retain the communication cycle structure of static and dynamic segments, and ignore the symbol window and network idle time; S212, Protocol Segmentation and Node Grouping: The FlexRay communication cycle is divided into static and dynamic segments, corresponding to two sets of sensor nodes: Static segment corresponding set The scheduling is done using the Round Robin Protocol (RRP). Dynamic segment corresponding set The redesigned static event triggering protocol, SETP, is used for scheduling. S220 and FlexRay protocol scheduling mechanisms; S221. Initialization: Set the scheduling matrix Initialize it as an all-zero matrix, indicating that all nodes are initially in an unscheduled state; S222, Static Segment Scheduling, also known as RRP: Traversing the static segment node group Determine the transmission node at time k based on RRP. , ； Then, a unit matrix block is allocated to it in the scheduling matrix. At this time, the node measurement information that the static node can successfully transmit is expressed by the following formula: ； S223, Dynamic Segment Scheduling (SETP): Traversing Dynamic Segment Node Groups Initialize the set of dynamic segment nodes that have obtained transmission rights. Calculate the competition error for each node. : ； in, This represents the measurement value that was successfully decoded and transmitted from node j before time k. And node normalization priority: ； in, This is a weight matrix; weight values ​​are assigned as needed. The feature range is designed according to actual needs, such as the desired measurement value, the sensor range, etc. Nodes exceeding the normalization threshold are granted transmission rights. ； Simultaneously update the set of dynamic segment nodes that have obtained transmission rights. , ； And allocate an identity matrix block to it in the scheduling matrix. ； The node measurement information that a dynamic node can successfully transmit is expressed by the formula (15). ； S224. Combined Scheduling Matrix: Combines the scheduling results of static and dynamic segments into a final block diagonal scheduling matrix. ; ； The measurement information of the node that has been successfully transmitted at this time can be expressed as: ； S230, the specific operation of the scheduling matrix and the zero-order hold; At the front end of the estimator, a zero-order hold (ZOH) is used to fill in the missing node measurements with the values ​​from the previous successful transmission of that node, ultimately yielding a received measurement value from the estimator. as follows: ； in, yes After time-quantization encoding, the data of the entire data segment, when substituted into formula (16) using formula (8c), yields the following expression for the measurement received at the estimator front end after noise interference augmentation: ； In the formula, the data received by the estimator is... It is a complex function that includes system state. Measurement noise Quantization error Scheduling matrix and historical data .

4. The method for coordinated optimization of state estimation and bit rate allocation for resource-constrained unmanned systems according to claim 3, characterized in that, In step S300: The measurement used in the estimation is a complex function containing scheduling information, quantized measurement values, and historical data. Therefore, the cost function in the classic MHE objective function also needs to be modified accordingly to accurately estimate the state. The modified objective function is as follows: ； ； The cost function needs to satisfy the following constraints: ； ； And noise constraint formula (3) and quantization error constraint formula (6); in, ； This represents the known information in the measurement cost term. This indicates the compensated measurement value. For the scheduling matrix, yes The covariance matrix of the known prior estimate error at time t, where the diagonal elements are determined based on the physical properties of the state variables and prior knowledge; The metric matrix representing the interference level of each sensor node, and its inverse matrix The weight matrix in the cost function is used to penalize measurement bias. The structure should be a block diagonal matrix, corresponding to the grouping of sensor nodes: the smaller the interference boundary, the smaller the measurement uncertainty of that node. The larger the value of the corresponding diagonal element, the better. The smaller the diagonal elements, the lower the penalty weight assigned to the measurement channel of that node in the cost function; specifically, the... Nodes The design is as follows: ； in, It is the first The upper limit of the measurement noise boundary at the node, and the other term in the denominator is the square of the quantization error. ; This approach ensures that measurement noise and quantization error are weighed at the same power, which is consistent with the physical meaning of the MHE weight design. The weight matrix integrates the effects of measurement noise and quantization error, enabling the estimator to adaptively adjust the confidence level of the data at each node. Finally, the overall interference penalty weight metric matrix can be expressed as: ; In equation (18), the objective function of MHE lacks a process cost term because the prior of this scheme is... The smooth update mechanism used is The state evolution is implicitly constrained by using the optimal estimate from the previous moment as a priori. In the case of UBB, the goal of MHE is usually not to minimize the norm of process noise, but to ensure the final boundedness of the estimation error through boundedness analysis; the influence of process noise is implicit in the state transition model and is indirectly handled through prior terms and measurement terms. Unlike recursive filtering methods that require state augmentation to clarify delays caused by analog communication, rolling time-domain estimation naturally adapts to these effects by using a fixed-length measurement window; at each time k, MHE utilizes the received measurement sequence and scheduling information The structural delay in the static segment and the zero-order preservation behavior in the dynamic segment are inherently captured in this measurement sequence; Effectively extract "new" information from each measurement, while historical values It is considered a known constant within the estimation window; this method avoids the need for state augmentation while fully considering the scheduling effect of the FlexRay protocol. This patent uses the analytical solution to the MHE problem, which can be used to obtain the solution in... The optimal estimate for time is as follows: ； The analytical solution is the result of solving the optimization problem of formula (18). This analytical solution avoids the complex numerical optimization process and is suitable for real-time operation under the limited onboard computing power of unmanned systems. ， ； The calculation process of the state estimation module can be summarized as follows: at each time step... This estimator uses data within a fixed-length N+1 time window and solves the aforementioned optimization problem to obtain the result from... arrive The optimal state estimation sequence at time t.

5. The method for coordinated optimization of state estimation and bit rate allocation for resource-constrained unmanned systems according to claim 3, characterized in that, In step S400: S410, Construction optimization problem; To achieve optimal bit rate allocation, the MHE-PSO collaborative optimization algorithm is proposed, and its optimization problem is: ； ； ， ； in, It is the objective function of the MHE part (18). This is a penalty term for the total bit rate constraint. It is a sufficiently large penalty coefficient, chosen to be several orders of magnitude larger than the typical MHE cost value, such as The objective function ensures that any violation of the total bit rate constraint will be severely punished. Ensure that there are enough bits available for data encoding after considering the data ID; This optimization problem is solved through the following steps; The fitness function for each particle is defined as: ； in, , Using particles The MHE cost function represents the bit rate allocation evaluation. This function is a specific implementation of the optimization objective of formula (24). Formula (24) directly minimizes the MHE objective function. The MHE objective function is mathematically used as a proxy index for state estimation error. Minimizing the objective function will also indicate a reduction in state estimation error. After PSO, a bit rate allocation scheme with smaller estimation error will be obtained. This allows PSO to directly optimize the MHE objective function and more directly reflect the estimation performance of the overall system. S420 and MHE-PSO collaborative optimization algorithm; S421. Initialization: Randomly initialize particle positions. and speed Initialize your personal best position and individual optimal fitness Global best position and global best fitness ; S422, Fitness Evaluation: For each particle i, use the objective function... Calculate fitness and impose penalties for violations of constraints; ； In the formula, correspond Penalty items correspond ; S423, Update Individual and Global Optimum: Update the individual optimal position of each particle based on its fitness value. and global optimal position ; For the particles obtained in S422 Corresponding fitness function Update; 1) Update the individual optimality: If the fitness function of the current particle i is... Less than the individual's optimal fitness So, let's set the best position for each person. The value is assigned to the position of the current particle i. Then, the individual's optimal fitness The value is assigned to the fitness of the current particle i. The formula is expressed as follows: if ,So ; 2) Update the global optimum: If the fitness of the current particle i is less than the global optimum fitness, then the global optimum position is assigned to the position of the current particle i, and the global optimum fitness is assigned to the fitness of the current particle i. The formula is expressed as follows: if ,So ; S424, Velocity and Position Update: Adjust particle velocity using inertial, cognitive, and social components, then update particle velocity and position; 1) Speed ​​update: using inertia Cognitive and social weight and the generated random numbers To adjust particle speed The formula is expressed as follows: ； 2) Position update: Based on the particle velocity obtained above The formula for updating particle position is expressed as follows: ； S425, Forced Constraints: Ensure the bit rate is an integer value through appropriate rounding and clamping operations. And bit rate greater than minimum value ; S426. Termination: First, update the inertia weights in each iteration. ,from linearly decreasing to The expression is as follows: ； Finally, repeat steps S422-S425 until the maximum number of iterations is reached. After iteration After this, a globally optimal position is obtained. ,this The bit rate allocation scheme for a given location is considered the optimal bit rate allocation scheme. .

Images