Fracturing propping agent storage scheduling method

Abstract

The invention discloses a fracturing proppant storage scheduling method, which constructs a usage amount prediction model, predicts the usage amount on the basis of the historical usage record of a pressure proppant, and provides a prediction result for storage workers to assist the storage workers in replenishment, so that the requirements on the storage workers can be effectively reduced, and the storage efficiency is improved. In addition, using influence factors of the fracturing propping agent can be used as identification factors, so that more accurate demand prediction is achieved, meanwhile, real-time storage scheduling information can be received, the storage robot is controlled to conduct intelligent scheduling, the storage scheduling efficiency can be effectively improved, and compared with manual scheduling in the prior art, a large amount of time can be saved.

Description

Technical Field

[0001] The invention relates to the technical field of fracturing proppant scheduling, and in particular to a fracturing proppant storage scheduling method. Background Art

[0002] Fracturing proppant is a ceramic particle product with high fracturing strength. It is mainly used for downhole support in oil fields to increase the production of oil and gas. It is an environmentally friendly product. Fracturing technology, as an emerging and effective way to transform oil and gas fields, has been well known and widely used. Fracturing technology uses artificially applied ultra-high pressure to create cracks of varying degrees in the rock formation, and then injects fluid into the cracks to form a channel that is easy for oil and gas to flow. In order to keep the cracks created after fracturing open, granular fracturing proppant is usually injected into the fluid.

[0003] When fracturing proppants are stored and dispatched, they are often picked up manually using forklifts and replenished by staff based on experience, which results in high operational requirements and low picking efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a fracturing proppant storage scheduling method to solve the problems existing in the prior art.

[0005] The present invention is achieved through the following technical solutions:

[0006] A fracturing proppant storage scheduling method, comprising:

[0007] A usage prediction model corresponding to the fracturing proppant is constructed by using a neural network model, and a network parameter of the usage prediction model is initialized by using a chaotic mapping mechanism to obtain individual network parameters, wherein the network parameters include weight parameters and threshold parameters of the neural network model;

[0008] Obtain the fitness value corresponding to each network parameter individual, and take the network parameter individual with the largest fitness value as the leader, and divide the remaining network parameter individuals into explorers and followers according to the preset ratio;

[0009] A neighborhood search algorithm is used to update the leader to obtain an updated leader; a collaborative search algorithm is used to update the explorer to obtain an updated explorer; based on the leader and the explorer, an adaptive search algorithm is used to update the follower to obtain an updated follower;

[0010] Select the target network parameter individual with the largest fitness value from the updated leader, the updated explorer, and the updated follower, and take the target network parameter individual as the global optimal value;

[0011] Repeat the updating of the leader, explorer and follower until the maximum number of training times is reached, and use the global optimal value as the final network parameter of the usage prediction model to obtain the trained usage prediction model; during the training process, the leader, explorer and follower are re-divided after each training cycle; each training cycle includes multiple training processes;

[0012] Collect the historical usage records of the pressure proppant corresponding to each target user, and use the trained usage prediction model to identify the historical usage records to obtain the predicted usage corresponding to each target user, and send the predicted usage corresponding to each target user to the corresponding work terminal of the warehouse staff to assist the warehouse staff in replenishing the stock;

[0013] After sending the predicted usage corresponding to each target user to the work terminal corresponding to the warehouse staff, obtaining the real-time warehouse scheduling information transmitted by the work terminal, the real-time warehouse scheduling information includes the type and quantity of the fracturing proppant;

[0014] According to the real-time warehouse scheduling information and the current warehouse information, the warehouse robot's picking route is planned, and the warehouse robot is controlled to pick up goods according to the established picking route to complete the storage scheduling of fracturing proppants and improve the warehouse efficiency; the current warehouse information includes the type, quantity and shelf number of the stored fracturing proppants.

[0015] In a possible implementation, the fitness value corresponding to each network parameter individual is obtained, and the network parameter individual with the largest fitness value is taken as the leader, and the remaining network parameter individuals are divided into explorers and followers according to a preset ratio, including:

[0016] The inverse of the error function of the usage prediction model is used as the fitness function, and based on the fitness function, the fitness value corresponding to each individual network parameter is obtained;

[0017] The network parameter individual with the largest fitness value is taken as the leader, M network parameter individuals among the remaining network parameter individuals are taken as explorers, and N-1-M network parameter individuals are taken as followers; where N represents the total number of network parameter individuals.

[0018] In a possible implementation, a neighborhood search algorithm is used to update the leader to obtain an updated leader, including:

[0019] Obtain the historical optimal value of the leader in the historical training process and the global optimal value in the current update process, and obtain the first target vector of the historical optimal value corresponding to the leader minus the global optimal value in the current update process;

[0020] Obtaining the norm of the first target vector, and using a random number α generated according to a normal distribution N(0,1) to attenuate the norm of the first target vector to obtain a first coefficient;

[0021] After multiplying the global optimal value in the current updating process by the first coefficient, a first updated value is obtained;

[0022] Add the first updated value to the global optimal value in the current update process to obtain the updated leader.

[0023] In a possible implementation, the collaborative search algorithm is used to update the explorer to obtain an updated explorer, including:

[0024] Get the historical optimal values ​​of all explorers during historical training;

[0025] Randomly select an explorer from all explorers to get the target explorer;

[0026] Randomly select an explorer from all explorers except the target explorer, obtain the collaborator corresponding to the target explorer, and determine the historical optimal value corresponding to the collaborator;

[0027] Get the mean between the historical optimal value of the target explorer and the historical optimal value of the collaborator to obtain the mean vector;

[0028] Based on the current number of training times, the step factor is obtained, and the step factor is attenuated using a random number α generated according to the normal distribution N(0,1) to obtain the second coefficient;

[0029] After multiplying the mean vector by the second coefficient, a second update value is obtained, and the mean vector and the second update value are added to obtain an updated target explorer;

[0030] Traverse all explorers and get the updated explorers.

[0031] In a possible implementation, based on the leader and the explorer, an adaptive search algorithm is used to update the followers to obtain updated followers, including:

[0032] Obtain the leader's reverse learning individuals;

[0033] Determine the first decision interval (0,0.33], the second decision interval (0.33,0.66], and the third decision interval (0.66,1), and randomly generate a decision factor between (0,1) for each follower;

[0034] When the decision factor is in the first decision interval, the follower is guided and updated based on the leader to obtain an updated follower;

[0035] When the decision factor is in the second decision interval, a random explorer is determined for the follower, and based on the determined explorer, the follower is guided and updated to obtain an updated follower;

[0036] When the decision factor is in the third decision interval, the follower is guided and updated based on the reverse learning individual to obtain an updated follower.

[0037] In a possible implementation, when the decision factor is in the first decision interval, the leader is used as the basis to guide and update the follower, and the updated follower is:

[0038] When the decision factor is in the first decision interval, the global optimal value minus the follower's historical optimal value is used to obtain the second target vector;

[0039] Obtaining the norm of the second target vector, dividing the norm of the second target vector by the interval length of the network parameter to obtain a third coefficient; wherein the interval length is used to represent the upper limit minus the lower limit of the network parameter;

[0040] The fourth coefficient is determined based on the fitness of the follower during the previous two training processes;

[0041] According to the third coefficient and the fourth coefficient, obtaining a first update step factor of the follower;

[0042] The first update step factor of the follower is attenuated by using a random number α generated by a normal distribution N(0,1) to obtain a fifth coefficient;

[0043] Multiply the average of the global optimal value and the historical optimal value of the follower by the fifth coefficient to obtain the third update value corresponding to the follower;

[0044] The average of the global optimal value and the historical optimal value of the follower is added to the third updated value to obtain the updated follower.

[0045] In a possible implementation, when the decision factor is in the second decision interval, an explorer is randomly determined for the follower, and based on the determined explorer, the follower is guided and updated, and the updated follower is obtained as follows:

[0046] When the decision factor is in the second decision interval, the historical optimal value corresponding to the determined explorer is subtracted from the historical optimal value of the follower to obtain the third target vector;

[0047] Obtaining a norm of the third target vector, and dividing the norm of the third target vector by the interval length of the network parameter to obtain a sixth coefficient; wherein the interval length is used to represent the upper limit minus the lower limit of the network parameter;

[0048] The seventh coefficient is determined based on the fitness of the follower during the previous two training processes;

[0049] Obtaining a second update step factor of the follower according to the sixth coefficient and the seventh coefficient;

[0050] The second update step factor of the follower is attenuated using a random number α generated according to a normal distribution N(0,1) to obtain an eighth coefficient;

[0051] Multiply the average of the determined historical optimal value corresponding to the explorer and the historical optimal value of the follower by the eighth coefficient to obtain a fourth update value corresponding to the follower;

[0052] The average of the determined historical optimal value corresponding to the explorer and the historical optimal value of the follower is added to the fourth update value to obtain an updated follower.

[0053] In a possible implementation, when the decision factor is in the third decision interval, the follower is guided and updated based on the reverse learning individual, and the updated follower is:

[0054] When the decision factor is in the third decision interval, the fourth target vector is obtained by subtracting the historical optimal value of the follower from the reverse learning individual;

[0055] Obtaining a norm of a fourth target vector, and dividing the norm of the fourth target vector by the interval length of the network parameter to obtain a ninth coefficient; wherein the interval length is used to represent an upper limit minus a lower limit of the network parameter;

[0056] The tenth coefficient is determined based on the fitness of the follower during the previous two training processes;

[0057] According to the ninth coefficient and the tenth coefficient, obtaining a third update step factor of the follower;

[0058] The third update step factor of the follower is attenuated using a random number α generated by normal distribution N(0,1) to obtain the eleventh coefficient;

[0059] Multiply the average of the historical optimal values ​​of the reverse learning individual and the follower by the eleventh coefficient to obtain the fifth update value corresponding to the follower;

[0060] The average of the historical optimal values ​​of the reverse learning individual and the follower is added to the fifth update value to obtain the updated follower.

[0061] In a possible implementation, according to the real-time storage scheduling information and the current storage information, the picking route of the storage robot is planned, and the storage robot is controlled to pick up the goods according to the predetermined picking route to complete the storage scheduling of the fracturing proppant, including:

[0062] Determine the pickup nodes that the warehouse robot needs to traverse based on real-time warehouse scheduling information and current warehouse information;

[0063] Based on the picking nodes that the storage robot needs to traverse, a picking path code is obtained; the first element in the picking path code is the location of the storage robot, and the last element in the picking path code is the shipping location;

[0064] Repeatedly obtain multiple different pickup path codes to obtain multiple pickup path codes;

[0065] Obtain the driving time corresponding to each pickup path code, where the driving time is used to represent the time required for the warehouse robot to move forward at a constant speed according to the pickup path code;

[0066] Determine the pickup path code with the shortest travel time among multiple pickup path codes to obtain the target pickup path code;

[0067] Perform mutation and crossover operations on the pickup path code to obtain an updated pickup path code;

[0068] Re-determine the target pickup path code based on the original pickup path code and the updated pickup path code;

[0069] Determine whether the driving time corresponding to the target pickup path code has not decreased in N rounds. If so, output the target pickup path code. Otherwise, eliminate half of the codes with the largest driving time in the original pickup path code and the updated pickup path code, and return to the steps of performing mutation and crossover operations.

[0070] According to the current storage information, the storage robot is controlled to pick up goods at each picking node in the order of the target picking path code to complete the storage scheduling of fracturing proppant.

[0071] In a possible implementation, picking up goods at each picking up node includes:

[0072] Scan the RFID information or QR code information of the fracturing proppant storage scheduling to obtain the current storage information of the target storage fracturing proppant;

[0073] Based on the current storage information of the target storage fracturing proppant, the storage robot is controlled to pick up the goods according to the requirements of the real-time storage scheduling information.

[0074] A fracturing proppant storage scheduling method provided by the present invention constructs a usage prediction model, which can predict the usage based on the historical usage records of the pressure proppant, and provide the prediction results to the warehouse staff to assist the warehouse staff in replenishing the stock, which can effectively reduce the requirements for the warehouse staff, and can also use the factors affecting the use of fracturing proppant as identification factors to achieve more accurate demand prediction. At the same time, it can receive real-time warehouse scheduling information and control the warehouse robot for intelligent scheduling, which can effectively improve the warehouse scheduling efficiency and save a lot of time compared with the manual scheduling of the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0076] Figure 1 A schematic flow chart of a method for storage and scheduling of fracturing proppants provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0078] Example

[0079] like Figure 1 As shown, an embodiment of the present invention provides a method for storage scheduling of fracturing proppant, comprising:

[0080] S1. A usage prediction model corresponding to the fracturing proppant is constructed by using a neural network model, and a network parameter of the usage prediction model is initialized by using a chaotic mapping mechanism to obtain individual network parameters, wherein the network parameters include weight parameters and threshold parameters of the neural network model.

[0081] Optionally, the neural network model can be set to a convolutional neural network or a BP (Back Propagation) neural network. In this embodiment, the BP neural network is preferably used as the usage prediction model to identify the input matrix data or vector data to obtain the predicted usage.

[0082] In addition to using the chaotic mapping mechanism to initialize the network parameters of the usage prediction model, the network parameters can also be randomly generated between the upper limit and the lower limit of the network parameters.

[0083] S2. Obtain the fitness value corresponding to each network parameter individual, and take the network parameter individual with the largest fitness value as the leader, and divide the remaining network parameter individuals into explorers and followers according to a preset ratio.

[0084] The total number of input matrix data or vector data L can be preset, and the network parameter individuals are applied to the usage prediction model. The historical data of the previous L days is used as input to obtain the actual output of the usage prediction model, and the historical data of the L+1th day is used as the expected output. The error function value can be obtained based on the actual output and the expected output, and the fitness value can be obtained by taking the inverse of the error function value. It is worth noting that in order to avoid the denominator of the fitness value being zero, a very small constant can be added, such as the fitness value can be 1 / (error function value+0.00001).

[0085] Optionally, in addition to selecting the historical usage records of the pressure proppant as training data, the usage influencing factors of the fracturing proppant in the previous L days can also be used as input data, and the historical data of the L+1th day is used as the expected output to predict the usage model, thereby further improving the usage prediction effect.

[0086] S3. Use a neighborhood search algorithm to update the leader to obtain an updated leader. Use a collaborative search algorithm to update the explorer to obtain an updated explorer. Based on the leader and the explorer, use an adaptive search algorithm to update the follower to obtain an updated follower.

[0087] This embodiment uses a neighborhood search algorithm, a collaborative search algorithm, and an adaptive search algorithm, which can effectively improve the training effect of the usage prediction model, thereby making the final predicted usage more in line with reality and effectively reducing the requirements for the usage prediction model.

[0088] S4. Select the target network parameter individual with the largest fitness value from the updated leader, the updated explorer, and the updated follower, and take the target network parameter individual as the global optimal value.

[0089] S5. Repeat the updating of the leader, explorer, and follower until the maximum number of training times is reached, and use the global optimal value as the final network parameter of the usage prediction model to obtain a trained usage prediction model. During the training process, the leader, explorer, and follower are re-divided after each training cycle. Each training cycle includes multiple training processes.

[0090] For example, the maximum number of training times is 1,000, and the leaders, explorers, and followers are re-divided every 10 times, but the historical optimal value of each individual still needs to be recorded to facilitate subsequent updates.

[0091] S6. Collect the historical usage records of the pressure proppant corresponding to each target user, and use the trained usage prediction model to identify the historical usage records to obtain the predicted usage corresponding to each target user, and send the predicted usage corresponding to each target user to the work terminal corresponding to the warehouse staff to assist the warehouse staff in replenishing the stock.

[0092] By predicting the usage of fracturing proppant, the experience requirements of warehouse staff can be effectively reduced, thereby effectively assisting warehouse staff in replenishing stocks.

[0093] S7. After sending the predicted usage corresponding to each target user to the work terminal corresponding to the warehouse staff, obtain the real-time warehouse scheduling information transmitted by the work terminal, wherein the real-time warehouse scheduling information includes the type and quantity of the fracturing proppant.

[0094] Because the real-time storage scheduling information includes the type and quantity of the fracturing proppant, the goods can be automatically distributed according to the storage information of various fracturing proppant in the current storage information. It is worth noting that when the fracturing proppant is put into storage, it forms a single packaged product, and each product is provided with a QR code or RFID (Radio Frequency Identification) tag, which records the type and quantity (or weight) of the single product.

[0095] The current warehouse information may include the storage location, type, and quantity (or weight) of each product, so that it is possible to determine which storage points to pick up the goods from based on the total demand.

[0096] S8. According to the real-time storage scheduling information and the current storage information, the storage robot's pickup route is planned, and the storage robot is controlled to pick up the goods according to the predetermined pickup route, so as to complete the storage scheduling of the fracturing proppant and improve the storage efficiency. The current storage information includes the type, quantity and shelf number of the stored fracturing proppant.

[0097] The warehouse robot can be a single intelligent robot or a combination of multiple robots. It can travel along a prescribed route, scan QR codes or RFID tags, and then automatically pick up and deliver goods based on machine vision technology.

[0098] A fracturing proppant storage scheduling method provided by the present invention constructs a usage prediction model, which can predict the usage based on the historical usage records of the pressure proppant, and provide the prediction results to the warehouse staff to assist the warehouse staff in replenishing the stock, which can effectively reduce the requirements for the warehouse staff, and can also use the factors affecting the use of fracturing proppant as identification factors to achieve more accurate demand prediction. At the same time, it can receive real-time warehouse scheduling information and control the warehouse robot for intelligent scheduling, which can effectively improve the warehouse scheduling efficiency and save a lot of time compared with the manual scheduling of the prior art.

[0099] In a possible implementation, the fitness value corresponding to each network parameter individual is obtained, and the network parameter individual with the largest fitness value is taken as the leader, and the remaining network parameter individuals are divided into explorers and followers according to a preset ratio, including:

[0100] The inverse of the error function of the usage prediction model is used as the fitness function, and based on the fitness function, the fitness value corresponding to each individual network parameter is obtained.

[0101] The network parameter individual with the largest fitness value is taken as the leader, M network parameter individuals among the remaining network parameter individuals are taken as explorers, and N-1-M network parameter individuals are taken as followers, where N represents the total number of network parameter individuals.

[0102] In a possible implementation, a neighborhood search algorithm is used to update the leader to obtain an updated leader, including:

[0103] Get the historical optimal value of leader i in the historical training process And the global optimal value in the current update process And obtain the historical optimal value corresponding to the leader Subtract the global optimal value in the current update process The first target vector.

[0104] In the first update process, the leader is the global optimal value, but as the update proceeds, individuals better than the leader may emerge. In order to ensure the overall training effect, the leader is always updated near the global optimal value.

[0105] Get the norm of the first target vector The norm of the first target vector is attenuated by a random number α generated by the normal distribution N(0,1), and the first coefficient is obtained.

[0106] The global optimal value in the current update process Multiply by the first coefficient Afterwards, the first updated value is obtained

[0107]

[0108] The first updated value and the global optimal value in the current update process Add them together to get the updated leader

[0109] In a possible implementation, the collaborative search algorithm is used to update the explorer to obtain an updated explorer, including:

[0110] Get the historical optimal values ​​of all explorers during historical training Among them, j represents the explorer, j = 1, 2, ..., J, J represents the total number of explorers, p represents the historical optimal value, and t represents the current number of training times.

[0111] Randomly select an explorer from all explorers and get the target explorer

[0112] In addition to the target explorer Randomly select an explorer from all explorers except The corresponding collaborator And determine the historical optimal value corresponding to the collaborator

[0113] Get Target Explorer The historical best value The historical optimal value corresponding to the collaborator The mean between , get the mean vector

[0114] Based on the current number of training times, obtain the step factor a1×(x max -x min )×exp(-30t / T) 10 , and use the random number α generated by the normal distribution N(0,1) to decay the step factor a1×(x max -x min )×exp(-30t / T) 10 , and obtain the second coefficient α×a1×(x max -x min )×exp(-30t / T) 10 . Among them, a1 represents a random number between (0,1), x max Represents the maximum value of the network parameter, x min represents the minimum value of the network parameter, and T represents the maximum number of training times.

[0115] After multiplying the mean vector by the second coefficient, we get the second updated value And the mean vector Add to the second update value to get the updated target explorer It is worth noting that this embodiment adopts the method of directly combining the mean vector with (x max -x min ), in this case, the upper and lower limits of each dimension are the same, so they can be directly multiplied. If the upper and lower limits of a dimension are different, each dimension needs to be extracted and updated according to the formula shown in the updated target explorer. In the BP neural network, the weight and threshold are between (0,1), so this embodiment does not consider the different upper and lower limits of the dimension.

[0116] Traverse all explorers and get the updated explorers.

[0117] In a possible implementation, based on the leader and the explorer, an adaptive search algorithm is used to update the followers to obtain updated followers, including:

[0118] Get the leader's reverse learning individual x dmax represents the upper limit of the d-th dimension parameter, x dmin represents the lower limit of the d-th dimension parameter, represents the d-th dimension parameter in leader i, d = 1, 2, ..., D, D represents the total number of network dimensions, Represents the d-th dimension parameter in the reverse learning individual.

[0119] The first decision interval (0, 0.33], the second decision interval (0.33, 0.66] and the third decision interval (0.66, 1) are determined, and a decision factor Q between (0, 1) is randomly generated for each follower. It is worth noting that the decision intervals can also be divided into different lengths.

[0120] When the decision factor Q is in the first decision interval, the leader is used as a basis to guide and update the follower to obtain an updated follower.

[0121] When the decision factor Q is in the second decision interval, an explorer is randomly determined for the follower, and based on the determined explorer, the follower is guided and updated to obtain an updated follower.

[0122] When the decision factor Q is in the third decision interval, the follower is guided and updated based on the reverse learning individual to obtain an updated follower.

[0123] In a possible implementation, when the decision factor Q is in the first decision interval, the leader is used as a basis to guide and update the followers, and the updated followers are:

[0124] When the decision factor Q is in the first decision interval, the global optimal value is adopted. Subtract the follower's historical best value Get the second target vector in, Represents the historical optimal value of the nth follower during the tth training process.

[0125] Get the norm of the second target vector Take the norm of the second target vector Divide by the interval length x of the network parameter max -x min , and get the third coefficient The interval length is used to represent the upper limit minus the lower limit of the network parameter.

[0126] The fourth coefficient is determined based on the fitness of the follower during the previous two training processes.

[0127] Optionally, the method for determining the fourth coefficient may include: in each round of updating, obtaining the fitness value of the follower after three updating modes.

[0128] During the t-th round of training, for each updating method, the fitness value F1 during the t-1th round of training and the fitness value F2 during the t-2th round of training are obtained.

[0129] According to the fitness value F1 and fitness value F2 corresponding to each updating method, the fitness difference values ​​corresponding to the three updating methods can be obtained, namely ζ1, ζ2 and ζ3. Among them, ζ1 represents the fitness difference corresponding to the guided update based on the leader, ζ2 represents the fitness difference corresponding to the guided update based on the determined explorer, and ζ3 represents the fitness difference corresponding to the guided update based on the reverse learning individual.

[0130] Therefore, according to ζ1, ζ2 and ζ3, the fourth coefficient φ can be obtained as: It is worth mentioning that, according to the above formula, it can be found that each follower corresponds to a fourth coefficient.

[0131] According to the third coefficient And the fourth coefficient φ, obtain the first update step factor of the follower Among them, φ max represents the largest fourth coefficient among all followers, and N represents the total number of followers.

[0132] The random number α generated according to the normal distribution N(0,1) is used to attenuate the first update step factor β1 of the follower to obtain the fifth coefficient α×β1.

[0133] The global optimal value The historical best value with followers The average Multiply by the fifth coefficient to get the third update value corresponding to the follower

[0134] Add the average of the global optimal value and the follower's historical optimal value to the third updated value to get the updated follower

[0135]

[0136] In a possible implementation, when the decision factor Q is in the second decision interval, an explorer is randomly determined for the follower, and based on the determined explorer, the follower is guided and updated, and the updated follower is obtained as follows:

[0137] When the decision factor Q is in the second decision interval, the historical optimal value corresponding to the determined explorer is adopted Subtract the follower's historical best value Get the third target vector

[0138] Get the norm of the third target vector And take the norm of the third target vector Divide by the interval length x of the network parameter max -x min , and get the sixth coefficient The interval length is used to represent the upper limit minus the lower limit of the network parameter.

[0139] The seventh coefficient is determined based on the fitness of the follower in the previous two training processes. The seventh coefficient is the same as the fourth coefficient and will not be repeated here.

[0140] According to the sixth coefficient and the seventh coefficient, the second update step factor of the follower is obtained.

[0141]

[0142] The random number α generated according to the normal distribution N(0,1) is used to attenuate the second update step factor of the follower to obtain the eighth coefficient α×β2.

[0143] The average of the historical optimal value corresponding to the determined explorer and the historical optimal value of the follower Multiply by the eighth coefficient α×β2 to get the fourth update value corresponding to the follower

[0144] The average of the historical optimal value corresponding to the determined explorer and the historical optimal value of the follower is added to the fourth update value to obtain the updated follower

[0145] In a possible implementation, when the decision factor Q is in the third decision interval, the follower is guided and updated based on the reverse learning individual, and the updated follower is:

[0146] When the decision factor Q is in the third decision interval, the reverse learning individual Subtract the follower's historical best value Get the fourth target vector

[0147] Get the norm of the fourth target vector And take the norm of the fourth target vector Divide by the interval length x of the network parameter max -x min , and we get the ninth coefficient The interval length is used to represent the upper limit minus the lower limit of the network parameter.

[0148] The tenth coefficient is determined based on the fitness of the follower in the previous two training processes. The tenth coefficient is the same as the fourth coefficient and will not be repeated here.

[0149] According to the ninth coefficient and the tenth coefficient, the third update step factor of the follower is obtained.

[0150]

[0151] The third update step factor of the follower is attenuated using a random number α generated according to the normal distribution N(0,1) to obtain the eleventh coefficient α×β3.

[0152] The average of the historical optimal values ​​of the reverse learning individual and the follower Multiply by the eleventh coefficient α×β3 to get the fifth update value corresponding to the follower

[0153] Add the average of the historical optimal values ​​of the reverse learning individual and the follower to the fifth update value to obtain the updated follower

[0154] Through the algorithm described in this embodiment, it is possible to balance global search and local search well and accelerate the convergence speed of the network through collaborative search and adaptive search on the basis of ensuring the training effect.

[0155] In a possible implementation, according to the real-time storage scheduling information and the current storage information, the picking route of the storage robot is planned, and the storage robot is controlled to pick up the goods according to the predetermined picking route to complete the storage scheduling of the fracturing proppant, including:

[0156] According to the real-time warehouse scheduling information and the current warehouse information, determine the pickup nodes that the warehouse robot needs to traverse.

[0157] For example, the current warehouse scheduling information requires B units of weight of Class A products. The current warehouse information includes the location, category, and weight of each product. Assuming that the packaging weight of each product is B / 5, 5 Class A products and their locations can be found based on the previous warehouse information to meet the warehouse scheduling requirements. The found location is the pickup node that needs to be reached.

[0158] It is worth noting that only integer quantity inputs or integer multiples of the weight of each product are accepted.

[0159] Based on the pickup nodes that the storage robot needs to traverse, a pickup path code is obtained. The first element in the pickup path code is the location of the storage robot, and the last element in the pickup path code is the shipping location.

[0160] Repeatedly obtain multiple different pickup path codes to obtain multiple pickup path codes. For example, the starting point can be represented by 0, the end point can be represented by 1, and other pickup nodes are assigned values ​​in sequence (assuming there are 5 nodes, then 2, 3, 4, 5, 6), then the pickup path code can be 0234561 or 0245361, so the travel path can be obtained.

[0161] The driving time corresponding to each pickup path code is obtained, and the driving time is used to represent the time required for the storage robot to move forward at a constant speed according to the pickup path code.

[0162] The pickup path code with the shortest travel time is determined among multiple pickup path codes to obtain the target pickup path code.

[0163] Perform mutation and crossover operations on the pickup path code to obtain an updated pickup path code.

[0164] Mutation and crossover operations are conventional operations of genetic algorithms, and this embodiment will not be described in detail. The data processing process after the mutation and crossover operations is mainly described. After the mutation and crossover operations are performed, the same values ​​may exist. At this time, the priority of the mutation position and the crossover position is increased, the mutation position and the crossover position are retained, and the positions of other nodes are modified. For example, 0234561 is obtained after mutation to 0335561. It can be found that the second position is mutated from 2 to 3, and the fourth position is mutated from 4 to 5. Then the data is processed from the front to the back. The second position is the mutation position and is retained as 3. The value of the third position is repeated with the value of the second position and has not been mutated. It should be changed to 2, and the same applies to the subsequent. It can also be processed from the back to the front. According to the reverse order, it can be found that the values ​​of the third position before and after the mutation are both 5, and the fourth position changes from 4 to 5. Then the value of the fourth position is retained and the third position is changed to 4.

[0165] To summarize, other positions that are the same as the mutation position are transformed into other values ​​to ensure that the values ​​of each pickup node appear.

[0166] Based on the original pickup path code and the updated pickup path code, the target pickup path code is re-determined.

[0167] Determine whether the driving time corresponding to the target pickup path code has not decreased in N rounds. If so, output the target pickup path code. Otherwise, eliminate half of the codes with the largest driving time in the original pickup path code and the updated pickup path code, and return to the steps of performing mutation and crossover operations.

[0168] According to the current storage information, the storage robot is controlled to pick up goods at each picking node in the order of the target picking path code to complete the storage scheduling of fracturing proppant.

[0169] In a possible implementation, picking up goods at each picking up node includes:

[0170] Scan the RFID information or QR code information of the fracturing proppant storage scheduling to obtain the current storage information of the target storage fracturing proppant.

[0171] Based on the current storage information of the target storage fracturing proppant, the storage robot is controlled to pick up the goods according to the requirements of the real-time storage scheduling information.

[0172] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for storage and scheduling of fracturing proppants, characterized in that: include: A usage prediction model corresponding to the fracturing proppant is constructed by using a neural network model, and a network parameter of the usage prediction model is initialized by using a chaotic mapping mechanism to obtain individual network parameters, wherein the network parameters include weight parameters and threshold parameters of the neural network model; Obtain the fitness value corresponding to each network parameter individual, and take the network parameter individual with the largest fitness value as the leader, and divide the remaining network parameter individuals into explorers and followers according to the preset ratio; A neighborhood search algorithm is used to update the leader to obtain an updated leader; The explorer is updated by using a collaborative search algorithm to obtain an updated explorer; Based on the leader and the explorer, an adaptive search algorithm is used to update the followers to obtain updated followers; Select the target network parameter individual with the largest fitness value from the updated leader, the updated explorer, and the updated follower, and take the target network parameter individual as the global optimal value; Repeat the updating of the leader, explorer, and follower until the maximum number of training times is reached, and use the global optimal value as the final network parameter of the usage prediction model to obtain the trained usage prediction model; During the training process, the leaders, explorers, and followers are reclassified after each training cycle. Each training cycle includes multiple training processes. Collect the historical usage records of the pressure proppant corresponding to each target user, and use the trained usage prediction model to identify the historical usage records to obtain the predicted usage corresponding to each target user, and send the predicted usage corresponding to each target user to the corresponding work terminal of the warehouse staff to assist the warehouse staff in replenishing the stock; After sending the predicted usage corresponding to each target user to the work terminal corresponding to the warehouse staff, obtaining the real-time warehouse scheduling information transmitted by the work terminal, the real-time warehouse scheduling information includes the type and quantity of the fracturing proppant; According to the real-time warehouse scheduling information and the current warehouse information, the warehouse robot's picking route is planned, and the warehouse robot is controlled to pick up goods according to the established picking route to complete the storage scheduling of fracturing proppants and improve the warehouse efficiency; the current warehouse information includes the type, quantity and shelf number of the stored fracturing proppants.

2. The method for storage and scheduling of fracturing proppant according to claim 1, characterized in that: Get the fitness value corresponding to each network parameter individual, and take the network parameter individual with the largest fitness value as the leader, and divide the remaining network parameter individuals into explorers and followers according to the preset ratio, including: The inverse of the error function of the usage prediction model is used as the fitness function, and based on the fitness function, the fitness value corresponding to each individual network parameter is obtained; The network parameter individual with the largest fitness value is taken as the leader, M network parameter individuals among the remaining network parameter individuals are taken as explorers, and N-1-M network parameter individuals are taken as followers; where N represents the total number of network parameter individuals.

3. The method for storage and scheduling of fracturing proppant according to claim 1, characterized in that: The leader is updated using a neighborhood search algorithm to obtain an updated leader, including: Obtain the historical optimal value of the leader in the historical training process and the global optimal value in the current update process, and obtain the first target vector of the historical optimal value corresponding to the leader minus the global optimal value in the current update process; Obtaining the norm of the first target vector, and using a random number α generated according to a normal distribution N(0,1) to attenuate the norm of the first target vector to obtain a first coefficient; After multiplying the global optimal value in the current updating process by the first coefficient, a first updated value is obtained; Add the first updated value to the global optimal value in the current update process to obtain the updated leader.

4. The method for storage and scheduling of fracturing proppant according to claim 3, characterized in that: The collaborative search algorithm is used to update the explorer to obtain an updated explorer, including: Get the historical optimal values ​​of all explorers during historical training; Randomly select an explorer from all explorers to get the target explorer; Randomly select an explorer from all explorers except the target explorer, obtain the collaborator corresponding to the target explorer, and determine the historical optimal value corresponding to the collaborator; Get the mean between the historical optimal value of the target explorer and the historical optimal value of the collaborator to obtain the mean vector; Based on the current number of training times, the step factor is obtained, and the step factor is attenuated using a random number α generated according to the normal distribution N(0,1) to obtain the second coefficient; After multiplying the mean vector by the second coefficient, a second update value is obtained, and the mean vector and the second update value are added to obtain an updated target explorer; Traverse all explorers and get the updated explorers.

5. The method for storage and scheduling of fracturing proppant according to claim 3, characterized in that: Based on the leader and the explorer, an adaptive search algorithm is used to update the followers to obtain updated followers, including: Obtain the leader's reverse learning individuals; Determine the first decision interval (0,0.33], the second decision interval (0.33,0.66], and the third decision interval (0.66,1), and randomly generate a decision factor between (0,1) for each follower; When the decision factor is in the first decision interval, the follower is guided and updated based on the leader to obtain an updated follower; When the decision factor is in the second decision interval, a random explorer is determined for the follower, and based on the determined explorer, the follower is guided and updated to obtain an updated follower; When the decision factor is in the third decision interval, the follower is guided and updated based on the reverse learning individual to obtain an updated follower.

6. The method for storage and scheduling of fracturing proppant according to claim 5, characterized in that: When the decision factor is in the first decision interval, the leader is used as the basis to guide and update the followers, and the updated followers are: When the decision factor is in the first decision interval, the global optimal value minus the follower's historical optimal value is used to obtain the second target vector; Obtaining the norm of the second target vector, dividing the norm of the second target vector by the interval length of the network parameter to obtain a third coefficient; wherein the interval length is used to represent the upper limit minus the lower limit of the network parameter; The fourth coefficient is determined based on the fitness of the follower during the previous two training processes; According to the third coefficient and the fourth coefficient, obtaining a first update step factor of the follower; The first update step factor of the follower is attenuated by using a random number α generated by a normal distribution N(0,1) to obtain a fifth coefficient; Multiply the average of the global optimal value and the historical optimal value of the follower by the fifth coefficient to obtain the third update value corresponding to the follower; The average of the global optimal value and the historical optimal value of the follower is added to the third updated value to obtain the updated follower.

7. The method for storage and scheduling of fracturing proppant according to claim 5, characterized in that: When the decision factor is in the second decision interval, a random explorer is determined for the follower, and the follower is guided and updated based on the determined explorer, and the updated follower is: When the decision factor is in the second decision interval, the historical optimal value corresponding to the determined explorer is subtracted from the historical optimal value of the follower to obtain the third target vector; Obtaining a norm of the third target vector, and dividing the norm of the third target vector by the interval length of the network parameter to obtain a sixth coefficient; wherein the interval length is used to represent the upper limit minus the lower limit of the network parameter; The seventh coefficient is determined based on the fitness of the follower during the previous two training processes; According to the sixth coefficient and the seventh coefficient, obtaining a second update step factor of the follower; The second update step factor of the follower is attenuated using a random number α generated according to a normal distribution N(0,1) to obtain an eighth coefficient; Multiply the average of the determined historical optimal value corresponding to the explorer and the historical optimal value of the follower by the eighth coefficient to obtain a fourth update value corresponding to the follower; The average of the determined historical optimal value corresponding to the explorer and the historical optimal value of the follower is added to the fourth update value to obtain an updated follower.

8. The method for storage and scheduling of fracturing proppant according to claim 5, characterized in that: When the decision factor is in the third decision interval, the follower is guided and updated based on the reverse learning individual, and the updated follower is: When the decision factor is in the third decision interval, the fourth target vector is obtained by subtracting the historical optimal value of the follower from the reverse learning individual; Obtaining a norm of a fourth target vector, and dividing the norm of the fourth target vector by the interval length of the network parameter to obtain a ninth coefficient; wherein the interval length is used to represent an upper limit minus a lower limit of the network parameter; The tenth coefficient is determined based on the fitness of the follower during the previous two training processes; According to the ninth coefficient and the tenth coefficient, obtaining a third update step factor of the follower; The third update step factor of the follower is attenuated using a random number α generated by normal distribution N(0,1) to obtain the eleventh coefficient; Multiply the average of the historical optimal values ​​of the reverse learning individual and the follower by the eleventh coefficient to obtain the fifth update value corresponding to the follower; The average of the historical optimal values ​​of the reverse learning individual and the follower is added to the fifth update value to obtain the updated follower.

9. The method for storage and scheduling of fracturing proppant according to claim 1, characterized in that: According to the real-time storage scheduling information and current storage information, the warehouse robot's pickup route is planned, and the warehouse robot is controlled to pick up goods according to the established pickup route to complete the fracturing proppant storage scheduling, including: Determine the pickup nodes that the warehouse robot needs to traverse based on real-time warehouse scheduling information and current warehouse information; Based on the picking nodes that the storage robot needs to traverse, a picking path code is obtained; the first element in the picking path code is the location of the storage robot, and the last element in the picking path code is the shipping location; Repeatedly obtain multiple different pickup path codes to obtain multiple pickup path codes; Obtain the driving time corresponding to each pickup path code, where the driving time is used to represent the time required for the warehouse robot to move forward at a constant speed according to the pickup path code; Determine the pickup path code with the shortest travel time among multiple pickup path codes to obtain the target pickup path code; Perform mutation and crossover operations on the pickup path code to obtain an updated pickup path code; Re-determine the target pickup path code based on the original pickup path code and the updated pickup path code; Determine whether the driving time corresponding to the target pickup path code has not decreased in N rounds. If so, output the target pickup path code. Otherwise, eliminate half of the codes with the largest driving time in the original pickup path code and the updated pickup path code, and return to the steps of performing mutation and crossover operations. According to the current storage information, the storage robot is controlled to pick up goods at each picking node in the order of the target picking path code to complete the storage scheduling of fracturing proppant.

10. The method for storage and scheduling of fracturing proppant according to claim 9, characterized in that: Pick up goods at various pickup points, including: Scan the RFID information or QR code information of the fracturing proppant storage scheduling to obtain the current storage information of the target storage fracturing proppant; Based on the current storage information of the target storage fracturing proppant, the storage robot is controlled to pick up the goods according to the requirements of the real-time storage scheduling information.

Images