Large-scale pipeline grouping and laying sequence optimization method and system based on machine learning, and storage medium

Abstract

The invention discloses a large-scale pipeline laying sequence optimization method and system based on machine learning and a storage medium, and belongs to the field of pipeline system design. Comprising the following steps of 1, dividing extracted information data according to a space region, and constructing a quadtree; 2, performing large-scale pipeline grouping according to a Mean Shift clustering algorithm; 3, setting a pipeline laying sequence optimization rule; and 4, introducing reinforcement learning to carry out pipeline sequence optimization, and calculating a Q value. According to the method, the pipeline laying sequence can better meet the actual engineering requirements, the later algorithm complexity is reduced, the algorithm robustness is improved, and the capability of adapting to the dynamic and complex pipeline laying environment is improved.

Description

Technical Field

[0001] The present invention belongs to the field of pipeline system design, and specifically relates to a large-scale pipeline grouping and laying sequence optimization method, system and storage medium based on machine learning. Background Art

[0002] Pipeline systems are widely used in complex equipment and industrial systems such as aerospace, ships, and automobiles, and have a significant impact on the design cycle, performance, and reliability of equipment. Pipeline laying, as an important part of pipeline system design, has received extensive attention from the theoretical and engineering communities. In recent years, large-scale pipeline laying problems have mostly focused on solving the optimization of laying paths, and there has been little research on pipeline grouping and laying sequence optimization. Therefore, the invention proposes to achieve large-scale pipeline grouping and laying sequence optimization based on machine learning.

[0003] The key to pipeline grouping and laying sequence optimization lies in the design of the algorithm. In order to improve the efficiency and quality of complex equipment pipeline layout and enhance the digitalization and intelligence level of pipeline layout, some theoretical explorations and research have been done on the pipeline laying sequence optimization problem at home and abroad, including pipeline laying sequence optimization methods based on the Kriging model, overall coding strategies for intelligent optimization, collaborative evolution methods, and optimized laying sequences. However, these methods still have shortcomings in the face of large-scale data. For example, efficiency issues are not taken into account when facing large-scale laying data, and it is difficult to establish accurate mathematical models when facing complex models. To address these problems, it is extremely important to study a method that considers large-scale pipeline grouping and laying sequence optimization. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a large-scale pipeline grouping and laying sequence planning method, system and storage medium based on machine learning. This method constructs a tree data structure for a large-scale pipeline candidate point set based on the quadtree region division characteristics, and introduces the Mean Shift clustering algorithm to realize large-scale pipeline grouping. When querying neighborhood pipeline candidate points, a query search is performed through the quadtree to improve the pipeline laying preprocessing rate. Considering the complexity of subsequent algorithms, this method also introduces reinforcement learning to realize pipeline sequence optimization, which greatly improves the robustness of the algorithm, can adapt to large-scale complex and dynamic environments, reduce the complexity of the subsequent pipeline laying main algorithm, improve the overall laying efficiency and accuracy, and reduce the errors caused by prior knowledge. It has important practical value in the field of pipeline laying.

[0005] In order to achieve the purpose of the present invention, the present invention is implemented by the following technical solutions:

[0006] In one aspect, the present invention provides a large-scale pipeline grouping and laying sequence optimization method based on machine learning, comprising the following steps:

[0007] Step 1: Divide the extracted pipeline information data into spatial regions and construct a quadtree;

[0008] Step 2: Group large-scale pipelines according to the Mean Shift clustering algorithm;

[0009] Step 3: Set the pipeline laying sequence optimization rules;

[0010] Step 4: Optimize the pipeline laying sequence based on reinforcement learning and calculate the Q value.

[0011] Preferably, the step 1 specifically includes the following steps:

[0012] 1.1) Initialize and create the root node, representing the entire two-dimensional space area;

[0013] 1.2) Recursively start from the root node, determine which area the coordinate point belongs to and divide it into this node, and mark the coordinate as visited;

[0014] 1.3) If the node capacity exceeds the threshold, the node is split and the spatial area is divided into four quadrants again;

[0015] 1.4) Repeat steps 1.2) and 1.3) until the leaf node depth is reached or the region information does not exceed the maximum capacity of the node, stop the region division, and delete the coordinate information of the deepest leaf node;

[0016] 1.5) When all coordinates are marked as visited, the space division is completed;

[0017] 1.6) Insertion operation: re-query which deepest leaf node area range each coordinate class belongs to, and re-store the coordinate information therein to facilitate subsequent search for point sets by area and improve the efficiency of querying sub-node information.

[0018] Preferably, the step 2 specifically includes the following steps:

[0019] 2.1) Randomly select the center point C from the original data;

[0020] 2.2) Find all candidate points within the bandwidth of the center point C according to the quadtree, recorded as set M;

[0021] 2.3) Calculate the offset vector from the center point to each element vector in the set M;

[0022] 2.4) The center point moves along the direction of the offset vector;

[0023] 2.5) Repeat steps 2.2), 2.3), and 2.4) until the size of the offset vector meets the set threshold;

[0024] 2.6) Repeat steps 2.1), 2.2), 2.3), 2.4), and 2.5) until all candidate points have been visited;

[0025] 2.7) According to the access frequency of each point, the class with the highest frequency is taken as the current point set class;

[0026] Specifically, the formula for solving the offset vector in step 2.3) is:

[0027]

[0028] Where x is the center point, x i is a point within the bandwidth, n is the number of points within the bandwidth, h is the bandwidth of the kernel function, and g(x) is the negative derivative of the kernel function.

[0029] Specifically, the center point moving formula described in step 2.4) is:

[0030]

[0031] in Indicates the center point The t-th position update coordinates.

[0032] Preferably, the step 3 specifically includes the following steps:

[0033] 3.1) Constraint f1: interference between adjacent pipeline laying sequences should be avoided as much as possible;

[0034] 3.2) Optimize the target f2 and give a higher reward to the pipeline sequence with a smaller angle between the two pipelines;

[0035] 3.3) Optimize the target f3 and give priority to giving higher reward values ​​to adjacent pipeline sequences that do not interfere with each other;

[0036] Specifically, the reward calculation formula described in step 3.2) is:

[0037] Reward=f1+α1·f2+α1·f3

[0038] Among them, α1 is the inertia weight one measuring the angle between the two pipelines, and α2 is the inertia weight two measuring the distance between the two pipelines;

[0039] Specifically, constraint f1 is a reward value that measures whether a pipeline collides. If a collision occurs, p is 0, otherwise it is 1:

[0040]

[0041] Specifically, constraint f2 is used to measure the angle between the two pipelines, and the calculation formula is:

[0042]

[0043] in is the direction vector of pipeline 1, is the direction vector of pipeline 2.

[0044] Specifically, constraint f3 measures the distance between the two pipelines:

[0045]

[0046] Among them, P1 represents the distance between the starting points of the pipeline; P2 represents the distance between the end points of the pipeline.

[0047] Preferably, the step 4 specifically includes the following steps:

[0048] Using the Q-learning algorithm, the specific steps are:

[0049] 4.1) Initialize the Q table, learning rate, discount factor Epsililon-greedyz value, and number of iterations;

[0050] 4.2) Randomly generate an initial state state, in the present invention, the state is the current pipeline sequence number;

[0051] 4.3) Select an action according to the strategy. In the present invention, the action is the sequence number of the remaining pipelines;

[0052] 4.4) Execute the action and calculate the reward value;

[0053] 4.5) Go to the next state according to the action;

[0054] 4.6) Update the Q table according to the calculated reward value;

[0055] 4.7) Repeat operations 4.2), 4.3), and 4.4) until the maximum number of iterations is reached;

[0056] 4.8) Obtain the optimal pipeline laying sequence according to the Q table;

[0057] Specifically, the Q table update formula described in step 4.6) is:

[0058] Q(s,a)←Q(s,a)+α[reward+γMax a' Q(s',a')-Q(s,a)]

[0059] Where Q(s,a) represents the current state s and the action a taken, Q(s',a') represents the new state s' reached after state a; Reward represents the feedback given by the environment from state s to state s'; α represents the learning rate; γ represents the discount rate; Max a' Q(s',a') means choosing the action with the largest Q value from all the actions a' that s' can take.

[0060] On the other hand, the present invention provides a computing system, including a processor and a memory for storing a program executable by the processor, wherein when the processor executes the program stored in the memory, the above-mentioned large-scale pipeline laying sequence optimization method based on machine learning is implemented.

[0061] On the other hand, the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the above-mentioned large-scale pipeline laying sequence optimization method based on machine learning.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. The present invention proposes a Mean Shift clustering algorithm improved by quadtree, which is applied to large-scale pipeline grouping. The Mean Shift clustering algorithm has a large computational overhead for large data sets and needs to traverse each data point to calculate its kernel density estimate. Therefore, an improved Mean-Shift algorithm is designed to improve the clustering speed of mean clustering by using the regional partitioning characteristics of quadtree.

[0064] 2. The present invention proposes a laying sequence planning method, that is, to achieve pipeline laying sequence optimization through Q-learning. Q-learning, as a type of machine learning, makes decisions through interactive learning with the environment. It is suitable for dynamic and complex new environments and can achieve stable laying sequence solutions for complex pipeline laying environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A distribution diagram of a quadtree region division and candidate point query by region according to an embodiment of the present invention;

[0066] Figure 2 A comparison chart of data processing speeds of the improved and original Mean Shift clustering algorithms according to an embodiment of the present invention;

[0067] Figure 3 This is a clustering effect diagram of an embodiment of the present invention;

[0068] Figure 4 A cluster projection diagram of an embodiment of the present invention;

[0069] Figure 5 Comparison of the laying effects caused by different pipeline laying orders in one embodiment of the present invention Figure 1 ;

[0070] Figure 6 Comparison of the laying effects caused by different pipeline laying orders in one embodiment of the present invention Figure 2 ;

[0071] Figure 7 This is a diagram showing the calculation results of the optimal pipeline sequence calculated by MATLAB according to an embodiment of the present invention;

[0072] Figure 8 This is a UG simulation result diagram of the optimal pipeline sequence of an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to specifically explain the advantages and purposes of the present invention, the following embodiments are described in detail. Figures 1 to 8 , a simplified aircraft engine piping model is used as an example for further detailed explanation and description. It should be noted that this example is only for explaining one example of the invention, not all examples.

[0074] In an embodiment of the present invention, a large-scale pipeline laying sequence optimization method based on machine learning is shown in the flowchart of the method. Figure 7 As shown, the following steps are included:

[0075] Step 1: Divide the extracted pipeline information data into spatial regions and construct a quadtree;

[0076] 1.1) Initialize and create the root node, representing the entire two-dimensional space area;

[0077] 1.2) Recursively start from the root node, determine which area the coordinate point belongs to and divide it into this node, and mark the coordinate as visited;

[0078] 1.3) If the node capacity exceeds the threshold, the node is split and the spatial area is divided into four quadrants again;

[0079] 1.4) Repeat steps 2.2) and 2.3) until the leaf node depth is reached or the region information does not exceed the maximum capacity of the node, stop the region division, and delete the coordinate information of the deepest leaf node;

[0080] 1.5) When all coordinates are marked as visited, the space division is completed;

[0081] 1.6) Insertion operation: re-query which deepest leaf node area range each coordinate class belongs to, and re-store the coordinate information therein to facilitate subsequent search for point sets by area and improve the efficiency of querying sub-node information.

[0082] First, Siemens NX is used to perform CAD modeling of the space environment for laying aircraft engines. The starting point coordinates and pipeline end point coordinates are extracted through Siemens NX / Grip and stored in a TXT document. The stored TXT file is then imported into MATLAB, and a quadtree is constructed for the data information in MATLAB. First, the root node is initialized to represent the entire two-dimensional space area. Starting from the root node, recursion is performed to determine which area the coordinate point belongs to and divide it into this node. The coordinate is marked as visited. If the node capacity exceeds the threshold, the node is split and the space area is divided into four quadrants again. Repeat the steps until the leaf node depth is reached or the area information does not exceed the maximum capacity of the node. Stop the area division and delete the coordinate information of the deepest leaf node. When all coordinates are marked as visited, the space division is completed. Step 2: Large-scale pipeline grouping is performed according to the Mean Shift clustering algorithm, specifically:

[0083] 2.1) Randomly select the center point C from the original data;

[0084] 2.2) Find all candidate points within the bandwidth of the center point C according to the quadtree, recorded as set M;

[0085] 2.3) Calculate the offset vector from the center point to each element vector in the set M;

[0086] 2.4) The center point moves along the direction of the offset vector;

[0087] 2.5) Repeat steps 2.2), 2.3), and 2.4) until the size of the offset vector meets the set threshold;

[0088] 2.6) Repeat steps 2.1), 2.2), 2.3), 2.4), and 2.5) until all candidate points have been visited;

[0089] 2.7) According to the access frequency of each point, the class with the highest frequency is taken as the current point set class;

[0090] Furthermore, the formula for solving the offset vector described in step 2.3 is:

[0091]

[0092] Where x is the center point, x i is a point within the bandwidth, n is the number of points within the bandwidth, h is the bandwidth of the kernel function, and g(x) is the negative derivative of the kernel function.

[0093] Furthermore, the center point movement formula described in step 2.4 is:

[0094]

[0095] in Indicates the center point The t-th position update coordinates.

[0096] Initialize the cluster center point, search all pipeline coordinate points within the bandwidth by area through the quadtree, divide the point set M in the range into the current class, and then calculate the offset vector based on the point set M to update the cluster center point calculation method as described in 1.4). When all pipeline coordinate points are classified, the clustering algorithm ends, and pipelines of the same category are placed on the same layer to avoid collision with pipelines of the next category.

[0097] Step 3: Set the pipeline laying sequence optimization rules, specifically:

[0098] The optimization rules for constructing pipeline laying sequences should include: 3.1) Constraint f1, interference between adjacent pipeline laying sequences should be avoided as much as possible; 3.2) Optimization target f2, giving a higher reward to pipeline sequences with a smaller angle between two pipelines; 3.3) Optimization target f3, giving a higher reward to adjacent pipeline sequences that do not interfere with each other;

[0099] Furthermore, the reward calculation formula described in step 3.2 is:

[0100] Reward=f1+α1·f2+α1·f (3)

[0101] Among them, α1 is the inertia weight one measuring the angle between the two pipelines, and α2 is the inertia weight two measuring the distance between the two pipelines;

[0102] Furthermore, constraint f1 is a reward value to measure whether a pipeline collides. If a collision occurs, p is 0, otherwise

[0103]

[0104] Furthermore, constraint f2 is used to measure the angle between the two pipelines, and the calculation formula is:

[0105]

[0106]

[0107] in is the direction vector of pipeline 1, is the direction vector of pipeline 2.

[0108] Furthermore, constraint f3 is used to measure the distance between the two pipelines:

[0109]

[0110] Among them, P1 represents the distance between the starting points of the pipeline; P2 represents the distance between the end points of the pipeline.

[0111] Step 4: Optimize the pipeline laying sequence through reinforcement learning Q-learning, specifically:

[0112] The specific steps of Q-learning are:

[0113] 4.1) Initialize the Q table, learning rate, discount factor Epsililon-greedyz value, and number of iterations;

[0114] 4.2) Randomly generate the initial state state, in this article the state is the current pipeline number;

[0115] 4.3) Select action according to the strategy. In this paper, the action is the sequence number of the remaining pipelines;

[0116] 4.4) Execute the action and calculate the reward value;

[0117] 4.5) Go to the next state according to the action;

[0118] 4.6) Update the Q table according to the calculated reward value;

[0119] 4.7) Repeat operations 4.2), 4.3), and 4.4) until the maximum number of iterations is reached;

[0120] 4.8) Obtain the optimal pipeline laying sequence according to the Q table;

[0121] Furthermore, the Q table update formula described in step 4.6 is:

[0122] Q(s,a)←Q(s,a)+α[reward+γMax a' Q(s',a')-Q(s,(8)

[0123] Where Q(s,a) represents the current state s and the action a taken, Q(s',a') represents the new state s' reached after state a; Reward represents the feedback given by the environment from state s to state s'; α represents the learning rate; γ represents the discount rate; Max a' Q(s',a') means choosing the action with the largest Q value from all the actions a' that s' can take.

[0124] First, various parameters are set. The present invention sets the maximum number of iterations to 500, the learning factor learning-rate to 0.1, the discount factor to 0.9, and the exploration rate to 0.8; randomly select a pipeline sequence from the pipeline code; the remaining pipeline codes are designed as actions, and a decision action is randomly selected to calculate the reward value generated by this action. The reward value is calculated as described in step 3.3), and the Q table is updated by the reward value. The update formula is as described in step 4.6); repeat the operation. When the maximum number of iterations is reached, the action with the highest reward value from a certain state to the next state is found through the Q table, and finally the optimal pipeline sequence is obtained.

[0125] In the embodiment of the present invention, Figures 1 to 8 As shown, in order to more intuitively demonstrate and verify the effectiveness and accuracy of the large-scale pipeline laying sequence optimization method based on machine learning proposed in the present invention, a simulation verification was carried out.

[0126] The optimal sequence for pipeline laying is calculated by MATLAB simulation software, specifically:

[0127] The optimal pipeline sequence obtained in step 4 above is used for pipeline laying through subsequent algorithms. The final pipeline laying result is stored in a TXT format file and the pipeline is materialized through Siemens NX, as shown in the following figure. Figure 8 As shown in the figure, the effectiveness of the large-scale pipeline laying sequence optimization method based on machine learning is verified.

[0128] In the embodiment of the present invention, a simplified aircraft engine CAD model is taken as an example to verify the feasibility of the present invention by using Q-learning;

[0129] After selecting the starting point and end point, Figure 7 As shown in the figure, the pipeline laying sequence calculated in MATLAB is 1-2-3-6-5-4-8-7, as shown in the figure. Figure 8 As shown, the laying effect of the pipeline can be clearly seen, verifying the effectiveness of the present invention.

[0130] In an embodiment of the present invention, a computer-readable storage medium stores a program, and when the program is executed by a processor, the large-scale pipeline laying sequence optimization method based on machine learning as described above is implemented.

[0131] In an embodiment of the present invention, a computing system includes a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the large-scale pipeline laying sequence optimization method based on machine learning as described above is implemented.

[0132] In summary, the present invention uses a simplified aircraft engine CAD model as an example to verify the effectiveness and feasibility of the control method proposed in the present invention. In this example, the present invention implements pipeline grouping through the Mean Shift clustering algorithm improved by quadtree, improves the processing speed of large-scale pipeline quantity, and implements pipeline sequence planning through Q-learning, which greatly reduces the complexity of the algorithm and creates a good prerequisite for the subsequent pipeline laying algorithm. The present invention uses Mean Shift in machine learning and Q-learning algorithms to implement large-scale pipeline grouping and laying sequence planning. If other algorithms such as DBSCAN, K-mean, deep reinforcement learning, etc. are used to apply the technical framework of the present invention for solution, it is still applicable and still within the claims of the present invention.

[0133] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A large-scale pipeline grouping and laying sequence optimization method based on machine learning, characterized in that: The following steps are involved: Step 1: Divide the extracted pipeline information data into spatial regions and construct a quadtree; Step 2: Group large-scale pipelines according to the Mean Shift clustering algorithm; Step 3: Set the pipeline laying sequence optimization rules; Step 4: Optimize the pipeline laying sequence based on reinforcement learning and calculate the Q value.

2. The large-scale pipeline grouping and laying sequence optimization method based on machine learning according to claim 1 is characterized in that: The step 1 specifically comprises the following steps: 1.1) Initialize and create the root node, representing the entire two-dimensional space area; 1.2) Recursively start from the root node, determine which area the coordinate point belongs to and divide it into this node, and mark the coordinate as visited; 1.3) If the node capacity exceeds the threshold, the node is split and the spatial area is divided into four quadrants again; 1.4) Repeat steps 1.2) and 1.3) until the leaf node depth is reached or the region information does not exceed the maximum capacity of the node, stop the region division, and delete the coordinate information of the deepest leaf node; 1.5) When all coordinates are marked as visited, the space division is completed; 1.6) Insertion operation: re-query which deepest leaf node area range each coordinate class belongs to, and re-store the coordinate information therein to facilitate subsequent search for point sets by area and improve the efficiency of querying sub-node information.

3. The large-scale pipeline grouping and laying sequence optimization method based on machine learning according to claim 1 is characterized in that: The step 2 specifically includes the following steps: 2.1) Randomly select the center point C from the original data; 2.2) Find all candidate points within the bandwidth of the center point C according to the quadtree, recorded as set M; 2.3) Calculate the offset vector from the center point to each element vector in the set M; 2.4) The center point moves along the direction of the offset vector; 2.5) Repeat steps 2.2), 2.3), and 2.4) until the size of the offset vector meets the set threshold; 2.6) Repeat steps 2.1), 2.2), 2.3), 2.4), and 2.5) until all candidate points have been visited; 2.7) According to the access frequency of each point, the class with the highest frequency is taken as the current point set class; Specifically, the formula for solving the offset vector in step 2.3) is: Where x is the center point, x i is a point within the bandwidth, n is the number of points within the bandwidth, h is the bandwidth of the kernel function, and g(x) is the negative derivative of the kernel function. Specifically, the center point moving formula described in step 2.4) is: in Indicates the center point The t-th position update coordinates.

4. The large-scale pipeline grouping and laying sequence optimization method based on machine learning according to claim 1 is characterized in that: The step 3 specifically comprises the following steps: 3.1) Constraint f1: interference between adjacent pipeline laying sequences should be avoided as much as possible; 3.2) Optimize the target f2 and give a higher reward to the pipeline sequence with a smaller angle between the two pipelines; 3.3) Optimize the target f3 and give priority to giving higher reward values ​​to adjacent pipeline sequences that do not interfere with each other; Specifically, the reward calculation formula described in step 3.2) is: Reward=f1+α1·f2+α1·f3 Among them, α1 is the inertia weight one measuring the angle between the two pipelines, and α2 is the inertia weight two measuring the distance between the two pipelines; Specifically, constraint f1 is a reward value that measures whether a pipeline collides. If a collision occurs, p is 0, otherwise it is 1: Specifically, constraint f2 is used to measure the angle between the two pipelines, and the calculation formula is: in is the direction vector of pipeline 1, is the direction vector of pipeline 2. Specifically, constraint f3 measures the distance between the two pipelines: Among them, P1 represents the distance between the starting points of the pipeline; P2 represents the distance between the end points of the pipeline.

5. The large-scale pipeline grouping and laying sequence optimization method based on machine learning according to claim 1 is characterized in that: The step 4 specifically comprises the following steps: Using the Q-learning algorithm, the specific steps are: 4.1) Initialize the Q table, learning rate, discount factor Epsililon-greedyz value, and number of iterations; 4.2) Randomly generate an initial state state, in the present invention, the state is the current pipeline sequence number; 4.3) Select an action according to the strategy. In the present invention, the action is the sequence number of the remaining pipelines; 4.4) Execute the action and calculate the reward value; 4.5) Go to the next state according to the action; 4.6) Update the Q table according to the calculated reward value; 4.7) Repeat operations 4.2), 4.3), and 4.4) until the maximum number of iterations is reached; 4.8) Obtain the optimal pipeline laying sequence according to the Q table; Specifically, the Q table update formula described in step 4.6) is: Q(s,a)←Q(s,a)+α[reward+γMax a' Q(s',a')-Q(s,a)] Where Q(s,a) represents the current state s and the action a taken, Q(s',a') represents the new state s' reached after state a; Reward represents the feedback given by the environment from state s to state s'; α represents the learning rate; γ represents the discount rate; Max a' Q(s',a') means choosing the action with the largest Q value from all the actions a' that s' can take.

6. A computing system, characterized in that: The invention comprises a processor and a memory for storing a program executable by the processor, and when the processor executes the program stored in the memory, the method as claimed in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, and when the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images