A multi-sensor layout optimization method driven by particle swarm optimization algorithm for load performance evaluation
By using particle swarm optimization algorithm and sensor layout index SDI constraint, the layout of multiple types of sensors is optimized, which solves the problems of load recognition accuracy and information redundancy under multiple load fields and achieves more efficient load recognition effect.
Patent Information
- Application Number
- CN202311158047.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-08
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-09-08
AI Technical Summary
Existing sensor layout optimization methods mainly target single-type load fields, making it difficult to effectively identify load information under multiple load fields, resulting in low load identification accuracy and high information redundancy.
The particle swarm optimization algorithm is adopted, and the minimum installation spacing d of the sensors and the sensor layout index SDI are used as constraints to optimize the layout of multiple types of sensors. The sensor positions are optimized by penalty function and fitness function, and the load identification is performed by neural network to improve the load identification accuracy and reduce information redundancy.
It improves the load identification accuracy under multi-load fields, reduces the information redundancy of sensor layout, achieves a more scientific and reasonable sensor distribution, and enhances the performance evaluation of the load identification system.
Smart Images

Figure CN117350096B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sensor layout optimization technology in structural load identification, and particularly to a multi-type sensor layout optimization method for load performance evaluation driven by particle swarm optimization algorithm. This method considers the loaded structure simultaneously affected by multiple load fields such as temperature field and concentrated force load, as well as the problem of simultaneous layout of multiple types of sensors on the structural surface. Under the constraints of two optimization indicators, namely the minimum installation spacing d of sensors and the sensor layout index SDI, the layout of multiple types of sensors is optimized to improve load identification performance and reduce the information redundancy of sensor response data. Background Technology
[0002] Sensor placement is the final step in implementing structural dynamic load identification technology, and a reasonable sensor layout is crucial for obtaining high-precision load information. The optimal design of the number and location of sensors is determined by a combination of factors, including sensor type, structural form, and load identification method. Some engineering structures have strict limitations in terms of mass and size, which significantly restricts the arrangement and number of sensors. Therefore, optimizing the sensor layout of the structure is essential to ensure the best possible load identification performance. In recent years, thanks to the tremendous advancements in computer technology, sensor layout optimization technology has also developed significantly. Machine learning algorithms are widely used to solve sensor layout optimization problems, with representative methods including particle swarm optimization (PSO), genetic algorithms, and convex optimization methods. PSO has the advantage of fast convergence in solving high-dimensional problems, and in this context, sensor layout optimization design based on PSO has been proposed. It is worth noting that most existing sensor layout optimization methods related to load identification are only concerned with force load identification.
[0003] However, with the continuous advancement of technology, the complexity of load conditions on engineering structural systems, especially those in aerospace and marine machinery, is constantly increasing. Engineering structures are often simultaneously affected by multiple load fields, with combined force and thermal load fields being the most common. Therefore, sensor layout optimization methods focused solely on force loads often cannot solve practical problems. For real-world applications, to more accurately identify multiple load fields, it is necessary to acquire various response data of the structure, requiring the installation of multiple types of sensors on the structural surface. Based on this, a sensor optimization method considering the layout of multiple sensor types is needed for performance evaluation of load identification systems. However, existing research mainly focuses on sensor layout optimization for force load identification, with limited research on multi-type sensor layout optimization in multi-field load identification problems. Summary of the Invention
[0004] The technical problem this invention aims to solve is to provide a multi-type sensor layout optimization method driven by particle swarm optimization algorithm for load performance evaluation in the multi-load field load identification problem. This invention fully considers the multi-load field loading characteristics commonly found in practical engineering problems, such as force-thermal coupling, and uses the proposed minimum sensor installation spacing d and sensor layout index SDI as constraints of the optimization model. The resulting sensor layout design achieves better load identification performance and lower information redundancy.
[0005] The technical solution adopted in this invention is: a multi-type sensor layout optimization method driven by particle swarm optimization algorithm for load performance evaluation, the implementation steps of which are as follows:
[0006] Step 1: Based on the optimization objectives of load identification accuracy and information redundancy, the minimum sensor installation spacing limit d and the SDI value (Sensor Distribution Index) are used as constraints. The fitness function is as follows:
[0007]
[0008] Therefore, the optimization objective is to find the minimum value of the fitness function and its corresponding minimum point. Where r... i It is the position vector of the i-th sensor, [r1, r2, ..., r N [ ] represents the position vector of all candidate measurement points. The independent variable of the above function is a vector representing the combination of the positions of each sensor layout, i.e., a "particle":
[0009] X={x1 T x2 T ,…,x k T}
[0010] The penalty value is defined as follows:
[0011]
[0012] Where k is the total number of sensors, f SDI (k) represents the lower limit threshold of SDI when the number of sensors is k, which needs to be determined through preliminary layout optimization experiments. The calculation method of SDI is as follows:
[0013]
[0014] Where A is the total area available for arranging sensors, and D... ij is the minimum distance between the i-th sensor and other sensors, and μ is the average distance of all sensors to the center of the sensor layout.
[0015]
[0016] Among them, (x c ,y c ) represents the center coordinates of the sensor layout, calculated as follows:
[0017]
[0018] The SDI index can measure the degree of sensor dispersion and information redundancy in a sensor layout scheme. The higher the SDI value, the higher the degree of sensor dispersion and the lower the information redundancy in the area.
[0019] The error term in the fitness function represents the load identification error, which is defined as follows:
[0020] (1) If the identification of concentrated dynamic loads is considered, the definition of relative error is adopted:
[0021]
[0022] Where s represents the total number of time points in the dynamic load time history, F identify and F real These represent the load identification value and the actual value, respectively.
[0023] (2) If the identification of concentrated static loads is considered, the definition method is the weighted sum of the identification error of the load application point location and the identification error of the load magnitude:
[0024] error = w|F identify -F real |+(1-w)|R identify -R real |
[0025] Where w is the weight (0 < w < 1); F and R refer to the magnitude of the force load and the position vector of the load application point, respectively; the subscripts identify and real refer to the identified value and the real value, respectively.
[0026] Step 2: Initialize the basic parameters, including:
[0027] (1) Particle population size N, iteration steps T (or other termination conditions). A "particle" is a vector representing a specific sensor layout.
[0028] (2) Sensor types and the range of sensor quantities Where γ = [γ1, γ2, ..., γ n ] T γ z Let z be the number of the z-th type of sensor, where z = 1, 2, ..., n;
[0029] (3) Range of particle motion position [x] min ,x max ] and speed range [v min ,v max ];
[0030] (4) The initial position and initial velocity of the particle;
[0031] (5) Learning factors c1, c2 and inertia weight ω;
[0032] (6) Let P be the optimal layout scheme that the j-th particle has reached in the i-th iteration. j (i), i.e., local optimum; the optimal layout scheme reached by all particles is denoted as G(i), i.e., global optimum. Let p be the local extremum. j (i) = function(P) j (i)) and the global extremum g(i) = function(G(i)), with the initial values of the two extrema set to +∞, that is:
[0033] p1(0)=p2(0)=...=p N (0)=g(0)=+∞;
[0034] (7) Calculate the response data of all candidate measurement points under the load field in advance and store them for later use.
[0035] Step 3: Set the number of sensors γ, substitute in the initialization parameters, and enter the particle swarm optimization algorithm iteration process;
[0036] Step 4: Read in the data of each particle, convert it into a specific sensor layout, and then select the required response data portion from the complete response data set according to the layout. Use load identification methods such as neural networks or model methods to identify the loads and calculate the load identification error and fitness function value. Update the local optimum P of the fitness function in this iteration step. j (i+1):
[0037]
[0038] And the globally optimal G(i+1):
[0039]
[0040] And update the global extremum g(i+1):
[0041]
[0042] Step 5: Update particle position and velocity;
[0043] Vj (i+1)=ω(i)×V j (i)+c1(i)×r1×[P j (i)-X j (i)]+c2(i)×r2×[G(i)-X j (i)]
[0044] X j (i+1)=X j (i)+V j (i+1)
[0045] Among them, V j and X j R1 and R2 represent the velocity and position of the j-th particle, respectively, and are two independent random numbers in [0,1].
[0046] Step Six: Update Inertia Weights and Learning Factors:
[0047]
[0048]
[0049] Where i represents the current iteration number; T max ω represents the maximum number of iterations. start ω end This represents the initial and final values of ω, and is generally set as a parameter ω. start =0.9, ω start =0.4; c1 start c1 end This represents the initial and final values of the acceleration factor c1; c2 start c2 end This represents the initial and final values of the acceleration factor c2.
[0050] Step 7: If the iteration termination condition is met at this point, the optimization process with γ sensors ends; otherwise, return to step 4 to continue the iteration.
[0051] Step 8: Output the optimal layout scheme when the number of sensors is γ. If all values within the range of the number of sensors have been optimized, the program ends, collects and summarizes the optimal layout and its corresponding load recognition performance index data for each number of sensors, draws a chart and determines the optimal number of sensors and sensor layout method, and substitutes it into the load recognition to verify the recognition effect; otherwise, change the number of sensors and return to step 3.
[0052] The advantages of this invention compared to the prior art are:
[0053] This invention provides a multi-type sensor layout optimization method driven by particle swarm optimization (PSO) algorithm for load performance evaluation. It overcomes and improves the limitations of traditional single-type load field identification and single-type sensor layout optimization, thus enhancing the performance evaluation method of load identification systems. The improved PSO algorithm model maintains the advantages of PSO in solving high-dimensional problems while effectively improving local optimization capabilities, global search capabilities, and optimization efficiency. When optimizing the layout of multiple types of sensors for load identification in structural multi-load fields, the method prioritizes identification accuracy. While ensuring sensor installation spacing, it considers the sensor distribution index (SDI), allowing for a more scientific and reasonable evaluation of sensor layout dispersion and information effectiveness, effectively reducing information redundancy in the sensor layout. Attached Figure Description
[0054] Figure 1 This is a flowchart of the multi-type sensor layout optimization method based on particle swarm optimization algorithm for comprehensive performance evaluation of load identification.
[0055] Figure 2 This is a schematic diagram of the neural network load recognition method used in this invention;
[0056] Figure 3 This is a schematic diagram of the iterative principle of particle motion velocity direction used in this invention;
[0057] Figure 4 This is a schematic diagram of the geometric model for optimizing the layout of multiple types of sensors for a rectangular propulsion panel according to the present invention;
[0058] Figure 5 This is a schematic diagram of the finite element analysis unit node distribution and the initial screening test point positions and numbering for optimizing the layout of multiple types of sensors for a rectangular propulsion panel according to the present invention.
[0059] Figure 6 This invention is based on the f determined through experiments before optimizing the layout of multiple sensor types. SDI (k) value as a function of the total number of sensors k;
[0060] Figure 7 This invention summarizes the optimization data for different numbers of sensors after the layout optimization of multiple types of sensors is completed;
[0061] Figure 8 This is the final optimization result of the present invention for optimizing the layout of multiple types of sensors on a rectangular propulsion panel. Detailed Implementation
[0062] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0063] like Figure 1 As shown, this invention proposes a multi-type sensor layout optimization method driven by particle swarm optimization algorithm for load performance evaluation, including the following steps:
[0064] (1) Based on the optimization objectives of load identification accuracy and information redundancy, a fitness function for sensor layout optimization is established using the minimum sensor installation spacing limit d and the SDI value (Sensor Distribution Index) as constraints:
[0065]
[0066] Therefore, the optimization objective is to find the minimum value of the fitness function and its corresponding minimum point. Where r... i It is the position vector of the i-th sensor, [r1, r2, ..., r N [ ] represents the position vector of all candidate measurement points. The independent variable of the above function is a vector representing the combination of the positions of each sensor layout, i.e., a particle.
[0067] X={x1 T x2 T ,…,x k T}
[0068] x i T This refers to the position coordinate vector of the i-th sensor. The penalty value is defined as:
[0069]
[0070] Where k is the total number of sensors of all types, f SDI (k) represents the lower limit threshold of SDI when the number of sensors is k, which needs to be determined through preliminary layout optimization experiments. f can be determined using the following method. SDI The value of (k): First, ignore the SDI value constraint and optimize the sensor layout only under the condition of sensor installation spacing constraint, keeping other processes unchanged. Calculate the corresponding SDI value for the optimal layout obtained under each number of sensors under this condition, and use this SDI value as f. SDI The value of (k). The calculation method for SDI is as follows:
[0071]
[0072] Where A is the total area available for arranging sensors, and D... ij is the minimum distance between the i-th sensor and other sensors, and μ is the average distance of all sensors to the center of the sensor layout.
[0073]
[0074] Among them, (x c ,y c ) represents the center coordinates of the sensor layout, calculated as follows:
[0075]
[0076] The SDI index can measure the degree of sensor dispersion and information redundancy in a sensor layout scheme. The higher the SDI value, the higher the degree of sensor dispersion and the lower the information redundancy in the area.
[0077] The error term in the fitness function represents the load identification error, which is defined as follows:
[0078] (a) If the identification of concentrated dynamic loads is considered, the definition of relative error is adopted:
[0079]
[0080] Where s represents the total number of time points in the dynamic load time history, F identify and F real These represent the load identification value and the actual value, respectively.
[0081] (b) If the identification of concentrated static loads is considered, the definition method is the weighted sum of the load application point location identification error and the load magnitude identification error:
[0082] error = w|F identify -F real |+(1-w)|R identify -R real |
[0083] Where w is the weight (0 < w < 1); F and R refer to the magnitude of the force load and the position vector of the load application point, respectively; the subscripts identify and real refer to the identified value and the real value, respectively.
[0084] (2) Initialize the basic parameters, including:
[0085] (a) Particle population size N, maximum number of iterations T (or other termination conditions). Each particle is a vector representing a specific sensor layout.
[0086] (b) Sensor types and the range of sensor quantities Where γ = [γ1, γ2, ..., γ n ] T γ z Let z be the number of the z-th type of sensor, where z = 1, 2, ..., n;
[0087] (c) Range of particle motion position [x] min ,x max ] and speed range [v min ,v max ];
[0088] (d) The initial position and initial velocity of the particle;
[0089] (e) Learning factors c1, c2 and inertia weight ω;
[0090] (f) Let P be the optimal layout scheme that the j-th particle has reached in the i-th iteration. j (i), i.e., local optimum; the optimal layout scheme reached by all particles is denoted as G(i), i.e., global optimum. Let p be the local extremum. j (i) = function(P) j (i)) and the global extremum g(i) = function(G(i)), with the initial values of the two extrema set to +∞, that is:
[0091] p1(0)=p2(0)=...=p N (0)=g(0)=+∞
[0092] (g) Calculate the response data of all candidate measurement points under the load field in advance and store them for later use.
[0093] (3) Set the number of sensors γ, and substitute in the various initialization basic parameters to enter the particle swarm optimization algorithm iteration process;
[0094] (4) Read in the data of each particle and calculate the fitness function value of the particle. The specific process is as follows: First, the particle is a high-dimensional vector, which contains the position information of each sensor in the sensor layout. After reading in the information of each particle in turn, the sensor coordinate information contained in it is converted into the corresponding finite element node number. Then, according to the node number, the part of the response data to be used is selected from the whole set of response data. Finally, the load is identified by the load identification method such as neural network and the load identification error and fitness function value are calculated. The local optimum P of the fitness function in this iteration step is updated. j (i+1):
[0095]
[0096] And the globally optimal G(i+1):
[0097]
[0098] And update the global extremum g(i+1):
[0099]
[0100] Load identification can be achieved using an RBF neural network, and its flowchart is as follows: Figure 2 As shown, neural networks can perform pattern recognition tasks: a pattern is input into the neural network, and it then outputs an output pattern. In load recognition problems, a certain number of data samples are needed to train the neural network. These data samples consist of two parts: first, the sensor response data matrix, i.e., the input data matrix, which is mainly composed of the response data measured by the sensor measurement points. Each column in the matrix corresponds to one sample. Second, the actual load information matrix, i.e., the output data matrix, which consists of data describing load information, such as the magnitude of the concentrated load, the polynomial fitting coefficient of the distributed load field, and the spatial coordinates of the concentrated load application point. Each column of data in the matrix corresponds to each column in the input response data matrix.
[0101] Training a neural network using the above sample data mainly includes the following steps:
[0102] (a) Obtaining cluster centers C through unsupervised learning r We select I cluster centers from the input samples as initial values; then we input more samples and classify each sample by matching it with its nearest cluster center. This divides the input samples into I different groups. Each time a sample is assigned to a cluster center, that cluster center is updated according to the following rules:
[0103] C r (τ+1)=C r (τ)+μ[X k -C r (τ)]
[0104] Where μ is the learning rate parameter in the interval [0,1]; X k The newly added input sample; τ is the number of iterations.
[0105] (b) Determine the standard deviation parameter σ. This is determined by the maximum pairwise distance d between cluster centers. max The standard deviation parameter is defined as:
[0106]
[0107] Therefore, the activation function at the r-th neuron can be finally determined:
[0108]
[0109] Where r = 0, 1, 2, ..., I, and when i = 0, φ0 ≡ 1 is defined. This special neuron is called a bias neuron; X is the sample input to the neuron.
[0110] (c) Weights are obtained through supervised learning. The samples in the training set consist of two parts: structural response data and actual load values. These are the input and output parts of the sample data, respectively. Let A denote the input and output parts of the v-th sample. v B v When training a neural network, the output is first calculated based on the input response and the weights already defined in the neural network, and this output is denoted as Y. v
[0111] Y v =W·(φ0 φ1(A) v ,C1) φ2(A v ,C2) … φ I (A v C I )) T
[0112] Then, based on the difference D between the output value and the sample value v =Y v -B v To adjust the weight matrix W, the method used here is the "Delta learning rule," where the weight value w in the v-th row and u-th column of matrix W is adjusted. vu The following iterations are performed:
[0113] w vu (t+1)=w vu (t)+αd vu δ v (t)
[0114] Where t is the number of iterations; α is a parameter used to set the learning speed; d vu It is D v The u-th element; δ v (t) represents the activation state; it is 1 if the neuron is activated, and 0 otherwise. The learning process ends and the neural network is completed when the weight matrix W iterates to convergence or the maximum number of iterations is reached.
[0115] (5) Update particle position and velocity. The direction of particle motion updated iteratively depends on three vectors: the inertial velocity direction, the local optimum direction, and the global optimum direction, as follows: Figure 3 As shown. The specific expression is as follows:
[0116] V j (i+1)=ω(i)×V j(i)+c1(i)×r1×[P j (i)-X j (i)]+c2(i)×r2×[G(i)-X j (i)]
[0117] X j (i+1)=X j (i)+V j (i+1)
[0118] Among them, V j and X j Let r1 and r2 represent the velocity and position of the j-th particle, respectively, and r1 and r2 be two independent random numbers in the interval [0,1]. In the formula, V... j (i) is the velocity vector from the previous step, i.e., the inertial velocity; [P j (i)-X j [(i)] represents the difference between the local optimum and the particle position, and its direction is the direction of the local optimum; [G(i)-X j [i] represents the difference between the global optimum and the particle's position, and its direction is the global optimum direction. The three velocity directions together determine the particle's direction of motion.
[0119] Considering that sensor layout optimization requires the sensor position to be located on the element node of the finite element analysis, or within the range of the initial screening candidate measurement points, it is a discrete optimization problem. However, in actual optimization, the calculation result of the above expression generally results in the sensor position falling outside the specified range. Therefore, after each position update, the particle position needs to be corrected, adjusting its iterated position coordinates to the coordinates of the nearest discrete point that meets the sensor position requirements, i.e.:
[0120] finder i stmin{r1-X jl (i+1),r2-X jl (i+1),...,r n -X jl (i+1)}
[0121] thenX jl (i+1)=r i
[0122] Among them, X jl Refers to the j-th particle X j ={x j1 T ,x j2 T ,...,x jl T ,...,x jkT The location information of the l-th sensor contained in}, x jl T =(x jl ,y jl ) T , where represents the coordinates of the l-th sensor.
[0123] If, after the update, the particle's position or velocity exceeds a pre-defined range, then the particle's position or velocity will be reset to a random value within the position or velocity range.
[0124] (6) Update the inertia weights and learning factors:
[0125]
[0126]
[0127] Where i represents the current iteration number; T max ω represents the maximum number of iterations. start ω end This represents the initial and final values of ω, and is generally set as a parameter ω. start =0.9, ω start =0.4; c1 start c1 end This represents the initial and final values of the acceleration factor c1; c2 start c2 end This represents the initial and final values of the acceleration factor c2.
[0128] In general, the learning factor and inertia weight in particle swarm optimization algorithms are often constants. This definition is simple and easy to implement; however, it has the disadvantage that while the algorithm has good convergence speed and global search capability in the early stages of the search, it easily gets trapped in local optima in the later stages due to insufficient particle updates, and the solution accuracy decreases. Therefore, here we define the learning factor and inertia weight as time-varying variables, ensuring better solution accuracy for particles in the later stages of the search and helping them escape local optima.
[0129] For the learning factor, in the early stage, a larger c1 value and a smaller c2 value are selected to enable particles to enhance their "self-awareness" ability and weaken their "social awareness" ability, which is conducive to enhancing the global search ability. In the later stage of the search, the c1 value is decreased while the c2 value is increased. This allows more particles to learn towards the "overall optimum" and fewer particles to learn towards the "local optimum", thereby strengthening the global search ability of the particles.
[0130] For the inertia weight ω, the above update method makes ω decrease as the number of iterations increases. In the early stage of the iteration, ω decreases slowly to ensure the local optimization ability of the algorithm. In the later stage of the algorithm iteration, ω decreases faster to improve the global search ability of the algorithm and finally find the global optimal solution of the problem.
[0131] Updating the parameters of the optimization algorithm helps improve local optimization capabilities, global search capabilities, and overall optimization efficiency.
[0132] (7) If the iteration termination condition is met at this time, the optimization process with the number of sensors γ ends; otherwise, return to step (4) to continue the iteration.
[0133] (8) Output the optimal layout scheme when the number of sensors is γ. If all values within the range of the number of sensors have completed the optimization steps, then end the program, collect and summarize the optimal layout and its corresponding load recognition performance index data under each number of sensors, draw a chart and determine the optimal number of sensors and sensor layout method, and substitute it into the load recognition to verify the recognition effect; otherwise, change the number of sensors and return to step (3).
[0134] Example:
[0135] To gain a fuller understanding of the features of this invention and its applicability to practical engineering, this invention addresses, for example... Figure 4 The rectangular propulsion panel shown is designed with optimized strain gauge and temperature sensor layout. This structure is a 2m × 2m rectangle with a thickness of 0.01m. In the finite element analysis, it is divided into 20 × 20 elements. The material's elastic modulus E = 200MPa and Poisson's ratio μ = 0.3. The lower edge of the rectangular region is fixed, and a concentrated load F = 583N is applied. The load application point coordinates are (0.6, -0.1), in meters. The temperature field of the structure is expressed as T = e x+y +293.15, in K. The effect of gravity is not considered.
[0136] Consider the layout problem of two types of sensors coexisting on the surface of a structure: strain gauges, ranging from 3 to 9; and temperature sensors, ranging from 6 to 10. Set the particle population size to 20 and the maximum number of iterations to 150.
[0137] Figure 5 This diagram illustrates the finite element meshing method and element node numbering for the structure. The nodes in the diagram represent the range of potential measurement points for the sensor. The lower limit threshold of SDI, calculated after preliminary sensor layout optimization experiments, is shown below. Figure 6 The following table is provided for reference in subsequent algorithm optimization.
[0138]
[0139] Figure 7 The table below shows the optimal layout scheme and its load recognition performance under different combinations of sensor numbers, with an error weight of w = 0.5:
[0140]
[0141] The final optimal sensor layout is as follows: Figure 8 As shown in the figure, the load recognition performance of this layout scheme is verified in the following table.
[0142]
[0143] In summary, this invention proposes a multi-sensor layout optimization method driven by particle swarm optimization (PSO) algorithm for load performance evaluation. First, a mathematical model is established to define the optimization objective and criteria of the structure in the load identification problem. Then, the values of various parameters in the PSO algorithm and the scope and constraints of sensor layout optimization are considered. Next, based on the PSO algorithm, a two-layer optimization strategy is implemented for multiple sensor types: the outer layer optimizes the number of sensors, while the inner layer optimizes the layout for a specified number of sensors. In addition to considering the necessary constraint of the minimum sensor installation spacing *d*, a new SDI index is introduced to measure the information redundancy of the sensor layout. Finally, the optimization calculation data is collected and summarized to determine the optimal number of sensors and their corresponding layout, and this layout is verified for load identification.
[0144] The above are merely specific steps of the present invention and do not constitute any limitation on the scope of protection of the present invention; it can be extended to the field of multi-load field load identification and multi-type sensor layout optimization design. All technical solutions formed by equivalent transformation or equivalent substitution fall within the scope of protection of the present invention.
[0145] The parts of this invention not described in detail are well-known to those skilled in the art.
Claims
1. A multi-sensor layout optimization method for load performance evaluation driven by particle swarm optimization algorithm, characterized in that... The implementation steps are as follows: Step 1: Based on the optimization objectives of load identification accuracy and information redundancy, using the minimum sensor installation spacing limit d and the sensor distribution index SDI value as constraints, the fitness function is: Therefore, the optimization objective is to find the minimum value of the fitness function and its corresponding minimum point, where, It is the position vector of the i-th sensor. Given the position vectors of all candidate measurement points, the independent variable of the above function is a vector representing the combination of the positions of each sensor layout, i.e., a particle. The penalty item is the penalty value; the error item is the load identification error, which is defined as follows: (1) If the identification of concentrated dynamic loads is considered, the definition of relative error is adopted: in, This indicates the total number of time points in the dynamic load time history. and These represent the identified load value and the actual load value, respectively. (2) If the identification of concentrated static loads is considered, the definition method is the weighted sum of the identification error of the load application point location and the identification error of the load magnitude: in, These are weights; F and R refer to the magnitude of the force load and the position vector of the load application point, respectively; the subscripts identify and real refer to the identified value and the true value, respectively. Step 2: Initialize the basic parameters, including: (1) Particle population size N, iteration steps T or other termination conditions; a particle is a vector representing a specific sensor layout. (2) Sensor types and the range of sensor quantities ,in , Let z be the number of the z-th type of sensor. ; (3) Range of particle motion position With speed range ; (4) The initial position and initial velocity of the particle; (5) Learning factor of particle swarm optimization algorithm and inertia weight ; (6) Let the optimal layout scheme reached by the j-th particle in the i-th iteration be denoted as . That is, local optimum; the optimal layout scheme reached by all particles is denoted as . That is, the global optimum; denote the local extremum. and global extrema Set the initial values of the two extreme values to ,Right now: ; (7) Calculate the response data of all candidate measurement points under the load field in advance and store them for later use; Step 3: Set the number of sensors Then, substitute the initialization parameters and enter the particle swarm optimization algorithm iteration process; Step 4: Read in the data of each particle, convert it into a specific sensor layout, and then select the required portion of the response data from the entire response data set according to the layout. Use a neural network or model-based load identification method to identify the loads and calculate the load identification error and fitness function value. Update the local optimum of the fitness function in this iteration step. : and global optimal : And update the global extremum. : Step 5: Update particle position and velocity; in, and These represent the velocity and position of the j-th particle, respectively. These are two independent random numbers in the range [0,1]. Step Six: Update Inertia Weights and learning factors ; Where i represents the current iteration number; Indicates the maximum number of iterations; , express The initial and final values are generally set by parameters. , ; , Indicates acceleration factor The initial and final values; , Indicates acceleration factor The initial and final values; Step 7: If the iteration termination condition is met at this point, then the number of sensors terminated is... If the process is not optimized, return to step four and continue iterating. Step 8: The number of output sensors is If the optimal layout scheme is determined at any given time, and all values within the range of the number of sensors have been optimized, then the program ends, collects and summarizes the optimal layout and corresponding load recognition performance index data for each number of sensors, draws a chart and determines the optimal number of sensors and sensor layout method, and substitutes it into the load recognition to verify the recognition effect; otherwise, the number of sensors is changed and the program returns to step three. The optimization constraints in step one are the minimum sensor installation spacing limit d and the sensor distribution index SDI. Where k is the number of sensors; This is the lower limit threshold of SDI when the number of sensors is k, which needs to be determined through preliminary layout optimization experiments; the calculation method of SDI is as follows: in, It is the total area available for arranging sensors. It is the minimum distance between the s-th sensor and other sensors. It is the average distance from all sensors to the center of the sensor layout: in, These are the center coordinates of the sensor layout, calculated as follows: ; In step two, multiple ranges of sensor quantities were set, and the optimization results under different combinations of multiple sensor quantities were considered in the multi-load field problem of load identification. In step eight, the optimal solutions under different sensor quantities were compared, realizing a two-layer optimization strategy for sensor layout optimization.
2. The multi-sensor layout optimization method for load performance evaluation driven by particle swarm optimization algorithm according to claim 1, characterized in that: In step six, the particle swarm optimization algorithm was improved. This improved algorithm updates the learning factor during the iteration process. and inertia weight : This update method ensures the algorithm's local optimization capability, enhances its global search capability, and improves optimization efficiency.
Citation Information
Patent Citations
Method for optimizing support vector machine based on Particle Swarm Optimization
CN106682682A
Entity material and truss-like microstructure material integrated non-probability reliability topology optimization method
CN108763778A
