Method for searching harmful gas leakage source based on mobile inspection robot
By using advection-diffusion model and particle filtering combined with least squares method in intelligent patrol robots, the problems of low search efficiency and large positioning error in the prior art are solved, and more efficient and accurate gas leakage is achieved Source location.
Patent Information
- Application Number
- CN202410893487.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2024-11-08
AI Technical Summary
Existing intelligent inspection robots have low efficiency and large positioning errors in searching gas leakage sources. The assumptions are not satisfied mainly due to turbulence and measurement errors, making it difficult to trace the unexpected gas leakage source.
The gas diffusion model based on the advection-diffusion equation is adopted, combined with particle filtering and least squares method, and through regression, the gas leakage source intensity and moving position are predicted, so as to reduce the dependence on the initial range of source intensity and improve search efficiency.
It improves the efficiency and speed of gas leakage source search, reduces positioning errors, and can quickly respond and accurately find gas leakage sources.
Smart Images

Figure CN118918969A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent inspection robots, and in particular to a method for searching for a harmful gas leakage source based on a mobile inspection robot. Background Art
[0002] The leakage of harmful gases in chemical parks will pose a huge threat to the surrounding environment and human health, causing explosions, poisoning and other problems. Once a gas leak occurs, it is of great significance for enterprises to quickly determine the source location and accurately estimate the source information, which can minimize property losses and protect individuals from risks. At present, there are many schemes that propose dangerous source estimation methods based on static air pollution monitoring networks. However, there is a certain randomness in the occurrence of gas leaks, that is, potential gas leaks may occur anywhere, and not all places will have fixed gas monitoring networks pre-installed. Therefore, it is difficult for existing technical solutions to trace unexpected gas leaks. For this reason, mobile intelligent inspection robots have been widely used in chemical parks due to their powerful mobility flexibility and autonomous decision-making capabilities to find unknown leak sources.
[0003] Intelligent inspection robots use various autonomous search strategies to guide searchers to find the source of harmful gas leaks. The convergence algorithm is a gradient-based method for gas leak search that follows the local gradient of chemical concentration. Most of these strategies are based on bionic principles, such as the Zigzag algorithm inspired by the foraging behavior of beetles. In principle, these strategies assume that the concentration distribution in the search area is continuous and smooth, and that searchers can move according to the concentration gradient and wind direction until they reach the source of the leak. However, due to turbulence and measurement errors, this assumption is always difficult to meet in practice, resulting in low efficiency and large error in leak source positioning of existing intelligent inspection robot gas leak source search methods. Summary of the invention
[0004] To solve the above problems, the present application provides a method for searching for a harmful gas leakage source based on a mobile inspection robot, comprising the following steps:
[0005] A gas diffusion model is defined, and a searcher is added to the gas diffusion model; wherein the gas diffusion model is implemented using an advection-diffusion equation;
[0006] Initialize the searcher, determine the running state, starting position and other calculation parameters; wherein the running state includes running and stopping;
[0007] Perform a search operation on the searcher;
[0008] The search operation includes:
[0009] Step S1: Obtain and determine the running state. If the running state is stopped, end the search operation;
[0010] Step S2: If the operating state is operating, predict the intensity of the gas leakage source;
[0011] Step S3: predicting the moving position of the searcher and controlling the moving operation of the searcher;
[0012] Step S4: After the searcher movement operation is completed, the operation status is updated;
[0013] Re-execute step S1.
[0014] Among them, the features that define the gas diffusion model include:
[0015] The harmful gas is released at a stable release rate q through the gas source in the particle diffusion model. The particles of the harmful gas are transported by the wind, and their expansion speed is v. The expansion direction coincides with the positive number of the x-axis. The gas source is located at r = [x s ,y s ] a single source;
[0016] The searcher is a device containing a sensor, and the spherical radius of the sensor is b;
[0017] Adding a searcher to the gas diffusion model means: defining the searcher at the sampling position p k =[x k ,y k ] Collect harmful gas particles, and the unit time step for searching harmful gas particles is k;
[0018] Calculate at sampling position p k =[x k ,y k ]The method for collecting the number of harmful gas particles R is:
[0019] where λ is the location coefficient, and D is the atmospheric diffusion coefficient, τ is the average lifetime of the particle;
[0020] When the sampling interval is t0, the searcher is at the time step u k The average number of particles encountered is expressed as: μ k =R(p k |r,q)t0;
[0021] Sampling position p k Get the actual measured concentration d k The probability is expressed as:
[0022] Through a set of N particle filters Indicates the location of the gas leak source, where is the i-th estimate of the source position at the k-th step, is the association weight;
[0023] Further, initializing the searcher includes:
[0024] Define the running status as running;
[0025] Define the searcher's starting position p1 = [x1, y1], assign k = 0, and define
[0026] Further, the predicted gas leakage source intensity includes:
[0027] The number of steps k increases, that is, k = k + 1, and the historical measured concentration d is loaded k ;
[0028] Update the original particle filter and Updated to
[0029] Update the associated weight of the particle at step k The update method is:
[0030] Get the emission rate corresponding to the state particle;
[0031] According to the actual measurement d k At the sampling position p k probability;
[0032] Get the real-time and historical concentration d of gas leakage at step k k ;
[0033] The emission rate at each location is predicted by least squares to form the source intensity.
[0034] Further, the method for predicting the moving position of the searcher includes:
[0035] The Shannon entropy of the predicted measurement distribution is chosen as the reward function, expressed as:
[0036] In the formula Indicates p k+1 Predicted measurements of location;
[0037] Optimize the reward function to obtain the optimal behavior:
[0038] Get the next step position, expressed as:
[0039] Controlling the searcher movement operation refers to:
[0040] Define the searcher movement direction, represented by U = {left, right, front, back}; get the searcher's action execution at step k, represented by u k , the corresponding optimal behavior is expressed as
[0041] The update operation state means that when the distance between the searcher and the leakage source is less than the specified distance or the unit time step k> the maximum search step number, the update operation state is stopped.
[0042] The present invention is a method for realizing robot search for harmful gas leakage sources through a regression-enhanced convergent leakage source search framework. In the source search strategy, particle filtering is not used, but the emission rate is estimated by the least square method based on historical measurements. Therefore, the search method proposed by the present invention is less dependent on the initial range of source intensity. After being enhanced by the regression method, the dimension of the parameters that need to be estimated in the particle filter is reduced. At the same time, since the least square method is a fast regression technology, the efficiency and speed of the search can be increased by the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 It is a step diagram of a method for searching for a harmful gas leakage source provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0044] The method for searching for harmful gas leaks by a mobile inspection robot proposed in the present invention is a method for searching for harmful gas leak sources by enhancing convergence through regression based on particle filtering and regression methods. Specifically, the search for harmful gas leak sources is realized according to three elements: state, executable action set, and reward function. For the search for gas leak sources: state refers to the prediction of source items (specifically source position and source strength), action set refers to all available actions that the searcher (mobile inspection robot) can choose in the next step, and the reward function can quantify the information gain of the corresponding actions of the searcher. Under this framework, for the searcher, the search for gas leak sources is realized from the following three perspectives:
[0045] 1. Prediction of source items: Based on real-time and historical measurements in a Bayesian framework, the location and intensity of source items are predicted through the posterior probability density function;
[0046] 2. Calculate the information measure of all possible next positions;
[0047] 3. Move to the position where the reward function is maximum based on the source term and information measure.
[0048] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings.
[0049] Figure 1A method step diagram for searching for a harmful gas leakage source by a searcher is provided, wherein the searcher includes an inspection robot and other intelligent devices with sensors.
[0050] like Figure 1 As shown, the following steps are included:
[0051] Step S100: define a gas diffusion model and add a searcher to the gas diffusion model;
[0052] In the present invention, the gas diffusion model is implemented using the advection-diffusion equation. Specifically, the definition of the gas diffusion model includes the following contents:
[0053] The harmful gas is released at a stable release rate q through the gas source in the particle diffusion model. The particles of the harmful gas are transported by the wind, and their expansion speed is v. The expansion direction coincides with the positive number of the x-axis. The gas source is located at r = [x s ,y s ] a single source;
[0054] The searcher is a device including a sensor, and the spherical radius of the sensor is b;
[0055] Adding a searcher to the gas diffusion model means: defining the searcher at the sampling position p k =[x k ,y k ] Collect harmful gas particles, and the unit time step for searching harmful gas particles is k;
[0056] Calculate at sampling position p k =[x k ,y k ]The method for collecting the number of harmful gas particles R is:
[0057]
[0058] where λ is the location coefficient, and D is the atmospheric diffusion coefficient, τ is the average lifetime of the particle;
[0059] When the sampling interval is t0, the searcher is at the time step u k The average number of particles encountered is expressed as:
[0060] μ k =R(p k |r,q) (2).
[0061] In order to simulate the random relationship between the searcher and the emitted particles, the Poisson distribution is used to describe the probability of event counts occurring in a given time interval. k Get the actual measured concentration d k The probability is expressed as:
[0062]
[0063] On the other hand, in the gas leakage source search algorithm proposed in the present invention, the state of the source position corresponding to the kth step is unknown, so the posterior probability distribution function P(r k |d k ) predicts the source location and then updates it according to the Bayesian rule in the following way:
[0064]
[0065] Among them, P(d k |d k-1 )=∫P(d k |r k , d k-1 )P(r k |d k-1 )dr k (5).
[0066] However, the gas leakage source position shows nonlinear characteristics. Therefore, in the present invention, a particle filter under the sequential Monte Carlo framework is used to filter the gas leakage source through a set of N particle filters. Indicates the location of the gas leakage source, using the prior approximate calculation and the posterior probability distribution function P(r k |d k ),in, is the i-th estimate of the source position at the k-th step, is the association weight and satisfies
[0067] Based on this, calculate P(r k |d k ) is as follows: Here, δ(·) is the Dirac function.
[0068] Step S110: Initialize the searcher;
[0069] When initializing the searcher, the operating state, starting position and other calculation parameters are determined;
[0070] Specifically, the running status includes running and stopping, and other calculation parameters include the number of steps k and the weight of the kth step Preset unknown source location r k .
[0071] When initializing the searcher, define the running state as running, define the searcher's starting position p1 = [x1, y1], assign k = 0, and define
[0072] Step S120: performing a search operation on the searcher;
[0073] When performing a search operation, firstly execute step S1: obtain and determine the running state, and if the running state is stopped, end the search operation;
[0074] Otherwise, execute step S2: predict the intensity of the gas leakage source, which specifically includes the following steps:
[0075] The number of steps k increases, that is, k = k + 1; load the historical measured concentration d k ;
[0076] The calculation process for predicting the intensity of a gas leak source includes:
[0077] Update the original particle filter, Updated to
[0078] In this process, update the associated weight of the particle in step k Specifically from Update the weight to the kth step The method is:
[0079] In the leak source search algorithm, it is assumed that: In the formula is the emission rate corresponding to the state particle, is the emissivity d estimated using historical measurements k-1 .
[0080] According to formulas (1) to (3) and actual measurement d k At the sampling position p k The probability algorithm can be obtained,
[0081]
[0082] in,
[0083]
[0084]
[0085] In order to reduce the impact of particle degradation problems, a Markov chain Monte Carlo moving step strategy is used to determine the validity of new samples. If they are valid samples, resampling steps are performed to improve the diversity of samples.
[0086] In the present invention, the emission rate of the gas leakage source is predicted based on historical measurements using least squares. The gas diffusion model formulas (1) and (2) show that when the emission source and the environmental sensor are located at fixed positions and the meteorological conditions are stable, there is a linear relationship between the emission rate q and the average particle number μk, expressed as: μk = a(p k )q(10),
[0087] in:
[0088]
[0089] The coefficient a(pk) represents the unit mass concentration response of the leaked gas at the position pk. The real-time measurement at the jth step is expressed as d j =a(p j )q+e j , where e j is the measurement error.
[0090] Therefore, at the kth step, the real-time and historical concentrations of gas leakage d k Expressed as: d k =a k q+e k (12)
[0091] Where, d k =[d1, d2, …, d k ] T (13)
[0092] a k =[a(p1), a(p2), …, a(p k )] T , e k =[e1, e2, …, e k ] T (14);
[0093] The release rate was estimated by the least squares method to minimize the sum of squared errors, expressed as:
[0094]
[0095] in is the predicted emission rate at the kth step, and the method of predicting the emission rate using least squares is:
[0096]
[0097] However, since the measurement error does not follow the Gaussian distribution, the predicted value of the algorithm in formula (16) is biased. However, the least squares method can still be used in this framework: First, in the proposed algorithm, the estimation of the source strength does not need to be very accurate, because when an unexpected gas leak occurs, the location of the source is more important than the estimated source strength; even a biased estimate of the source strength can provide effective information for the searcher; even if there are some biases in the early stage of the search process, as the searcher gets closer to the source and the monitoring value becomes larger, the estimation bias will gradually decrease; in addition, the low time consumption of the least squares method can meet the requirements of rapid response when a gas leak occurs.
[0098] Step S3: predicting the moving position of the searcher and controlling the moving operation of the searcher;
[0099] First, define the searcher movement direction, denoted as U = {left, right, front, back}; the searcher's action execution at step k is denoted as u k , the corresponding optimal behavior is expressed as
[0100] Secondly, when predicting the moving position, the Shannon entropy of the predicted measurement distribution is selected as the reward function, which is expressed as:
[0101]
[0102] In the formula Indicates p k+1 The predicted measured value of the location, then the probability density distribution Calculated as:
[0103]
[0104] Using particle filtering, formula (21) can be re-expressed using formula (3) as:
[0105]
[0106] in,
[0107]
[0108] Therefore, the reward function is optimized as:
[0109]
[0110] At this point, the optimal behavior can be expressed as:
[0111] In this case, the source rate can be calculated without using least squares and can still be equal to the original value. At the beginning of the search process, the unknown emission rates of various particles are considered to have the same value q0, then:
[0112] in, It can be calculated by combining formula (1) and (3), formula (19) can be rewritten as:
[0113]
[0114] Once the position is determined, move the searcher to the new position:
[0115] Step S4: after the searcher movement operation is completed, the operation status is updated;
[0116] When the distance between the searcher and the leakage source is less than the specified distance (such as 10m) or k> the maximum number of search steps (such as 300 steps), the running state is updated to stop.
[0117] After the running status is updated, step S1 is executed again to enter the next round of search operation, and the running status is continuously judged in subsequent search operations. If the running status is stopped, the search process is terminated.
[0118] Based on the existing convergence algorithm, the present invention proposes a method for an intelligent inspection robot to search for harmful gas leakage sources, namely, a regression-enhanced convergence leakage source search method. In this framework, the emission rate is estimated by using least squares through historical observation data; the probability distribution of the source position is updated using a particle filter; and a subsequent reward function designed based on the maximum entropy sampling principle guides the searcher to finally approach and search for the leakage source. In the source search strategy proposed by the present invention, particle filtering is not used. Instead, the emission rate is estimated by the least squares method based on historical measurements. Therefore, the search method proposed by the present invention is less dependent on the initial range of the source intensity. After being enhanced by the regression method, the dimension of the parameters that need to be estimated in the particle filter is reduced from 3 to 2. At the same time, since the least squares method is a fast regression technology, the search process time can be shortened by the present invention.
[0119] The above disclosures are only several specific embodiments of the present invention; however, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. A method for searching for harmful gas leakage sources based on a mobile inspection robot, characterized in that: The following steps are involved: defining a gas diffusion model and adding a searcher to the gas diffusion model; Initialize the searcher, determine the running state, starting position and other calculation parameters; wherein the running state includes running and stopping; performing a search operation on the searcher; The search operation includes: Step S1: Acquire and determine the running state, if the running state is stopped, end the search operation; Step S2: If the operating state is operating, predicting the intensity of the gas leakage source; Step S3: predicting the moving position of the searcher and controlling the moving operation of the searcher; Step S4: after the searcher movement operation is completed, the operation status is updated; Re-execute step S1; Wherein, the gas diffusion model is implemented using the advection-diffusion equation.
2. The method for searching for a harmful gas leakage source according to claim 1, characterized in that: Defining a gas diffusion model includes: The harmful gas is released at a stable release rate q through the gas source in the particle diffusion model. The particles of the harmful gas are transported by the wind, and their expansion speed is v. The expansion direction coincides with the positive number of the x-axis. The gas source is located at r = [x s ,y s ] a single source; The searcher is a device containing a sensor, and the spherical radius of the sensor is b; Adding the searcher to the gas diffusion model means: defining the searcher at the sampling position p k =[x k ,y k ] Collect harmful gas particles, and the unit time step for searching harmful gas particles is k; Calculate at sampling position p k =[x k ,y k ]The method for collecting the number of harmful gas particles R is: where λ is the location coefficient, and D is the atmospheric diffusion coefficient, τ is the average lifetime of the particle; When the sampling interval is t0, the searcher is at the time step u k The average number of particles encountered is expressed as: μ k =R(p k |r,q)t0; Sampling position p k Get the actual measured concentration d k The probability is expressed as: Through a set of N particle filters Indicates the location of the gas leak source, where is the i-th estimate of the source position at the k-th step, is the association weight.
3. The method for searching for a harmful gas leakage source according to claim 2, characterized in that: The initialization searcher comprises: Defining the operating state as running; Define the searcher's starting position p1 = [x1, y1], assign k = 0, and define 4. The method for searching for a harmful gas leakage source according to claim 2, characterized in that: The predicted gas leakage source intensity includes: The number of steps k increases, that is, k = k + 1, and the historical measured concentration d is loaded k ; Update the original particle filter and Updated to Update the associated weight of the particle at step k The update method is: Get the emission rate corresponding to the state particle; According to the actual measurement d k At the sampling position p k The probability of Get the real-time and historical concentration d of gas leakage at step k k ; The emission rates are predicted by least squares, constituting source intensities.
5. The method for searching for a harmful gas leakage source according to claim 4, characterized in that: The method for predicting the moving position of the searcher comprises: The Shannon entropy of the predicted measurement distribution is chosen as the reward function, expressed as: In the formula Indicates p k+1 Predicted measurements of location; Optimize the reward function to obtain the optimal behavior: Get the next step position, expressed as:
6. The method for searching for a harmful gas leakage source according to claim 5, characterized in that: The control of the searcher movement operation refers to: Define the searcher's moving direction, represented by U = {left, right, front, back}; get the searcher's action execution at step k, represented by u k , the corresponding optimal behavior is expressed as 7. According to the method for searching for the leakage source of harmful gases as described in claim 1, the updating operation state means: when the distance between the searcher and the leakage source is less than the specified distance or the unit time step k> the maximum number of search steps, the updating operation state is stopped.
Citation Information
Cited By
Patrol robot path optimization method and system based on dynamic obstacle avoidance
CN120255529A
Harmful gas distribution detection method and system suitable for battery disassembly
CN120577494A
