Aerial material transaction data analysis method

Abstract

The invention discloses an aerial material transaction data analysis method, which belongs to the technical field of data processing, and is characterized in that a worker is allowed to customize historical aerial material transaction data and corresponding transaction abnormal labels, and then the custom historical aerial material transaction data and the corresponding transaction abnormal labels are learned through a multi-machine learning method; according to the method, the optimal machine learning model is determined, and finally automatic analysis of the aerial material transaction data is performed through the optimal machine learning model, so that a data analysis task established by a worker can be effectively realized, the worker is assisted to improve the working efficiency, and the data analysis is more systematic.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular relates to an aviation material transaction data analysis method. Background Art

[0002] Aviation material trading, also known as aviation equipment trading, refers to the buying and selling of aircraft, helicopters, drones and other aircraft and their parts, spare parts, avionics equipment, aviation interiors and other aviation-related products. The aviation material trading market involves a large amount of transaction data, which includes key factors such as the price, supply and demand relationship, and quality information of aviation materials. Traditional aviation material trading data analysis methods often rely on manual experience, lack of systematicity and accuracy, and also lead to problems such as long processing time and low work efficiency. Therefore, an efficient and accurate data analysis method is needed to optimize the aviation material trading process. Summary of the invention

[0003] The present invention provides an aviation material transaction data analysis method, which is used to solve the problems of lack of systematicity and low work efficiency caused by relying on manual experience to process aviation material transaction data in the prior art.

[0004] A method for analyzing aviation material transaction data, comprising: Acquire the historical aviation material transaction data uploaded in advance by the staff and the transaction abnormality labels corresponding to the historical aviation material transaction data, and pre-process the historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data to obtain the pre-processed historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data; Calling at least one machine learning model from a preset model library, and training the at least one machine learning model through the pre-processed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, so as to obtain an optimal machine learning model; After the optimal machine learning model is deployed, the current aviation material transaction data uploaded by the staff is obtained, and the current aviation material transaction data is identified through the optimal machine learning model to determine the predicted probability of transaction anomalies, and when the predicted probability of transaction anomalies is greater than the preset threshold, the transaction anomaly result is determined.

[0005] Furthermore, the historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data are preprocessed to obtain the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, including: formatting and normalizing the historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data.

[0006] Furthermore, the pre-set model library includes CNN model, BP neural network model, LSTM model, SVM model, ResNet model, RBF neural network model and RNNs model; Calling at least one machine learning model from a pre-set model library includes: obtaining a machine learning model calling instruction input by a staff member, and determining at least one machine learning model corresponding to the machine learning model calling instruction.

[0007] Furthermore, at least one machine learning model is trained by using the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain an optimal machine learning model, including: For any called machine learning model, the hyperparameters to be optimized of the machine learning model are initialized and encoded into vectors, and a plurality of different parameter vectors are determined; According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vector is divided into a first group, a second group, and a third group according to the fitness value of the parameter vector; For the first parameter vectors in the first population, information is exchanged between the first parameter vectors according to the fitness, to obtain the first parameter vector after an update; For the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain the second parameter vector after an update; For the third parameter vector in the third population, an adaptive rapid development strategy is used to conduct a guided search on the third parameter vector to obtain the third parameter vector after an update; For a first parameter vector after an update, a second parameter vector after an update, and a third parameter vector after an update, a global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector, and the third parameter vector to obtain the first parameter vector, the second parameter vector, and the third parameter vector after the global search; Determine whether the training end condition is met. If so, determine the optimal parameter vector based on the first parameter vector, the second parameter vector, and the third parameter vector after the global search, and use the parameters contained in the optimal parameter vector as the final parameters of the machine learning model to obtain the machine learning model after training. Otherwise, return to the population division step; Traverse all called machine learning models to obtain at least one trained machine learning model, and determine the optimal machine learning model based on the at least one trained machine learning model.

[0008] Further, according to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vector is divided into a first group, a second group and a third group according to the fitness value of the parameter vector, including: According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vectors are sorted in descending order of the fitness value to obtain the sorted parameter vectors; Based on the sorted parameter vectors, all parameter vectors are divided into the first group, the second group and the third group in the ratio of 1:8:1.

[0009] Furthermore, for the first parameter vectors in the first population, information exchange is performed between the first parameter vectors according to the fitness to obtain the first parameter vector after an update, including: For any first parameter vector in the first population, randomly match the first parameter vector with another first parameter vector to obtain the information exchange vector corresponding to each first parameter vector; According to the first parameter vector and the information exchange vector corresponding to the first parameter vector, the information exchange factor is obtained as follows:

[0010] in, represents the information exchange factor, Indicates i The fitness value corresponding to the first parameter vector, Indicates i The fitness value corresponding to the information exchange vector corresponding to the first parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value sign; According to the information exchange factor, the first parameter vector is updated, and the first parameter vector after one update is obtained as follows:

[0011] in, Indicates t During the training i The first parameter vector, i =1,2,…,L1, L1 represents the number corresponding to the first parameter vector, Represents the first parameter vector after an update , It means that the mean is 0 and the variance is Gaussian distributed random numbers.

[0012] Furthermore, for the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain an updated second parameter vector, including: For the first parameter vector in the first population, determining the optimal parameter vector according to the fitness value of the first parameter vector; For any second parameter vector in the second population, randomly match a first parameter vector to obtain a learning parameter vector corresponding to each second parameter vector; According to the fitness value corresponding to the optimal parameter vector and the fitness value corresponding to the learning parameter vector, the first learning factor and the second learning factor are obtained as follows:

[0013]

[0014] in, represents the first learning factor, represents the second learning factor, represents the fitness value corresponding to the mth second parameter vector, Represents the fitness value corresponding to the learning parameter vector corresponding to the mth second parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value symbol, Represents the fitness value corresponding to the optimal parameter vector, and exp represents the natural constant e An exponential function with base ; According to the optimal parameter vector, the learning parameter vector, the first learning factor and the second learning factor, a local search is performed on the second parameter vector, and the second parameter vector after one update is obtained as follows:

[0015] in, Indicates m The second parameter vector, Represents the second parameter vector after an update , m =1,2,…,L2, L2 represents the number corresponding to the second parameter vector, represents the first random number between (0,1), represents the second random number between (0,1), represents the second parameter vector The corresponding learning parameter vector, represents the optimal parameter vector.

[0016] Furthermore, for the third parameter vector in the third population, an adaptive rapid development strategy is used to conduct a guided search on the third parameter vector to obtain an updated third parameter vector, including: For any third parameter vector in the third population, randomly match a second parameter vector and a first parameter vector to obtain a first guided parameter vector and a second guided parameter vector; According to the current number of training times, the adaptive inertia weight is obtained as:

[0017] in, represents the adaptive inertia weight, T represents the preset maximum number of training times, represents the maximum inertia weight, Indicates the minimum inertia weight; According to the adaptive inertia weight, the first guided parameter vector and the second guided parameter vector, a guided search is performed on the third parameter vector, and the third parameter vector after one update is obtained as follows:

[0018] in, Indicates t During the training process n A third parameter vector, Represents the third parameter vector after an update , represents the third random number uniformly distributed in [0,1], represents the fourth random number uniformly distributed in [0,1], represents the first guided parameter vector, represents the second guided parameter vector.

[0019] Furthermore, the global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector, and the third parameter vector, and the first parameter vector, the second parameter vector, and the third parameter vector after the global search are obtained, including: Get the Levy flight step length:

[0020] Among them, s represents the Levy flight step length, U represents the first Levy flight factor, and V represents the second Levy flight factor. represents the Levy flight constant, and U represents the mean of 0 and the variance of V represents a random number generated by a normal distribution with a mean of 0 and a variance of 1. represents the normal distribution factor, and , represents the gamma function, sin represents the sine function, represents pi; According to the Levi flight step length, the flight step length is obtained as:

[0021] in, Indicates the flight step length; Randomly generate a flight step length corresponding to each dimension parameter of the third parameter vector, and form a vector of the flight step lengths corresponding to all dimension parameters to obtain a flight step length vector; According to the flight step vector, a global search is performed on the first parameter vector, the second parameter vector and the third parameter vector, and the first parameter vector, the second parameter vector and the third parameter vector after the global search are obtained as follows:

[0022] in, Indicates t During the training process k target parameter vectors, the target parameter vectors are the first parameter vector after one update, the second parameter vector after one update, and the third parameter vector after one update, represents the target parameter vector after global search, k =1,2,…,K, K represents the total number of target parameter vectors, represents the step size control factor, Represents the flight step vector.

[0023] Further, judging whether the training end condition is met includes: judging whether the current training times is greater than or equal to the maximum training times, if so, determining that the training end condition is met, otherwise determining that the training end condition is not met.

[0024] The present invention provides an aviation material transaction data analysis method, which allows staff to customize historical aviation material transaction data and corresponding transaction exception labels, and then learn the customized historical aviation material transaction data and corresponding transaction exception labels through multiple machine learning methods to determine the optimal machine learning model. Finally, the aviation material transaction data is automatically analyzed through the optimal machine learning model, which can effectively achieve the staff's established data analysis tasks, assist the staff in improving work efficiency, and make data analysis more systematic. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0026] Figure 1 The present invention provides a flowchart of an aviation material transaction data analysis method.

[0027] Figure 2 A flowchart for obtaining an optimal machine learning model provided by an embodiment of the present invention.

[0028] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0030] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0031] like Figure 1 As shown, an aviation material transaction data analysis method provided by an embodiment of the present invention includes: S11, obtaining historical aviation material transaction data uploaded in advance by a staff member and transaction abnormality labels corresponding to the historical aviation material transaction data, and preprocessing the historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data to obtain the preprocessed historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data; In the embodiment of the present invention, by having the staff upload data and corresponding labels, the data identification tasks set by the staff can be realized, so that the data processing is standardized and the data processing efficiency is improved. For example, the staff can input the type, specification, price, transaction volume, supplier information, purchaser information, etc. of aviation materials, and the transaction abnormality label can include a normal state label and different abnormal state labels, so that the data relationship can be learned.

[0032] Preprocessing of historical aviation material transaction data and abnormal transaction labels corresponding to the historical aviation material transaction data may include: removing duplicate values, missing values, and normalization processing, etc. However, the data input by the staff is generally relatively regular data, so only normalization processing can be performed.

[0033] S12, calling at least one machine learning model from a preset model library, and training the at least one machine learning model through the pre-processed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, so as to obtain an optimal machine learning model; Multiple machine learning models can be set in the pre-set model library. For different data, the recognition effects of different machine learning models may be different. Therefore, multiple machine learning models can be called for training, and finally the optimal machine learning model with the best effect can be determined, thereby assisting staff to better complete data processing tasks.

[0034] Training at least one machine learning model through preprocessed historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data can include: using the preprocessed historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data as data support, and using gradient descent method, particle swarm algorithm, genetic algorithm and other intelligent optimization algorithms to train multiple machine learning models one by one.

[0035] S13. After the optimal machine learning model is deployed, the current aviation material transaction data uploaded by the staff is obtained, and the current aviation material transaction data is identified through the optimal machine learning model to determine the predicted probability of transaction anomalies, and when the predicted probability of transaction anomalies is greater than a preset threshold, the transaction anomaly result is determined.

[0036] After the optimal machine learning model identifies the current aviation material transaction data, it outputs the probabilities of multiple transaction anomaly labels set by the staff. When the predicted probability of a certain label is greater than the preset threshold, it can be determined that it belongs to the corresponding transaction anomaly label, that is, the transaction anomaly result is obtained.

[0037] An aviation material transaction data analysis method provided by the present invention allows staff to customize historical aviation material transaction data and corresponding transaction exception labels, and then learn the customized historical aviation material transaction data and corresponding transaction exception labels through multiple machine learning methods to determine the optimal machine learning model. Finally, the aviation material transaction data is automatically analyzed through the optimal machine learning model, which can effectively achieve the data analysis tasks set by the staff, assist the staff to improve work efficiency, and make data analysis more systematic.

[0038] In an embodiment of the present invention, historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data are preprocessed to obtain the preprocessed historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data, including: formatting and normalizing the historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data to obtain the preprocessed historical aviation material transaction data and transaction anomaly labels corresponding to the historical aviation material transaction data.

[0039] Format normalization can include: 1) converting text data into digital data; 2) filling the data with zeros before the data for the same type of data with inconsistent lengths to ensure the consistency of data length without affecting the data size. 3) using unique codes to represent data that only exists in a few or limited numbers (such as aviation material types).

[0040] In an embodiment of the present invention, the pre-set model library includes a CNN (Convolutional Neural Network) model, a BP (Back Propagation) neural network model, an LSTM (Long Short-Term Memory) model, an SVM model (Support Vector Machine), a ResNet (Residual Network) model, an RBF (Radial Basis Function) neural network model, and an RNNs (Recurrent Neural Networks) model; It is worth noting that the above-mentioned machine learning models are only some examples of this embodiment. Only a part of the machine learning models can be used to build a model library, or other machine learning models can be used to build a model library.

[0041] Calling at least one machine learning model from a pre-set model library includes: obtaining a machine learning model calling instruction input by a staff member, and determining at least one machine learning model corresponding to the machine learning model calling instruction.

[0042] like Figure 2 As shown, at least one machine learning model is trained by preprocessing the historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain the optimal machine learning model, including: S21. For any called machine learning model, initialize and encode the hyperparameters to be optimized of the machine learning model into a vector, and determine a plurality of different parameter vectors; For example, when the selected machine learning model is a CNN model, the hyperparameters to be optimized of the machine learning model are weights and thresholds, and both weights and thresholds have upper and lower limits, generally (0,1). Therefore, they can be randomly initialized within the upper and lower limits of the parameters to be optimized and encoded into a vector to obtain a parameter vector, which is then initialized multiple times to obtain multiple different parameter vectors.

[0043] S22, obtaining a fitness value corresponding to each parameter vector according to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, and dividing the parameter vector into a first group, a second group, and a third group according to the fitness value of the parameter vector; According to the historical aviation material transaction data after preprocessing and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector can be obtained, which can include: after applying the parameters contained in the parameter vector to the machine learning model, the historical aviation material transaction data after preprocessing can be used as the output, and the transaction anomaly labels corresponding to the historical aviation material transaction data can be used as the expected output, and the error function value can be obtained, and then the reciprocal of the error function value can be obtained to obtain the fitness value. However, the error function value may be 0 in theory, which will result in the inability to take the reciprocal and the fitness value being infinite, so a constant (such as 0.001) can be set, and the set constant can be added to the error function value, and then the reciprocal is taken to obtain the fitness value.

[0044] S23, for the first parameter vectors in the first population, exchanging information between the first parameter vectors according to the fitness, to obtain an updated first parameter vector; S24, for the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain an updated second parameter vector; S25, for the third parameter vector in the third population, adopt an adaptive rapid development strategy to conduct a guided search on the third parameter vector to obtain the third parameter vector after an update; S26, for the first parameter vector after one update, the second parameter vector after one update, and the third parameter vector after one update, a global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector, and the third parameter vector to obtain the first parameter vector, the second parameter vector, and the third parameter vector after the global search; S27, judging whether the training end condition is met, if so, determining the optimal parameter vector according to the first parameter vector, the second parameter vector and the third parameter vector after the global search, and taking the parameters contained in the optimal parameter vector as the final parameters of the machine learning model to obtain the machine learning model after training, otherwise returning to the step of population division; S28. Traverse all called machine learning models to obtain at least one trained machine learning model, and determine the optimal machine learning model based on at least one trained machine learning model.

[0045] Determining the optimal machine learning model based on at least one trained machine learning model may include: testing each trained machine learning model using a test set based on at least one trained machine learning model, and taking the machine learning model with the highest recognition accuracy as the optimal machine learning model.

[0046] In the process of optimizing the parameters of the machine learning model, the prior art often falls into the local optimum, resulting in the inability to accurately identify data and the inability to accurately complete the data identification task set by the staff. Therefore, the embodiment of the present invention provides a new training method that can effectively perform global search and improve the search accuracy of the algorithm, so that the data identification task set by the staff can be completed more accurately.

[0047] In an embodiment of the present invention, according to the historical aviation material transaction data after preprocessing and the transaction abnormality label corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vector is divided into a first group, a second group and a third group according to the fitness value of the parameter vector, including: According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vectors are sorted in descending order of the fitness value to obtain the sorted parameter vectors; Based on the sorted parameter vectors, all parameter vectors are divided into the first group, the second group and the third group in the ratio of 1:8:1.

[0048] In order to ensure that no decimals appear when the first population, the second population, and the third population are divided, the total number of parameter vectors can be set to an integer multiple of 10.

[0049] In the embodiment of the present invention, for the first parameter vectors in the first population, information exchange is performed between the first parameter vectors according to the fitness to obtain the first parameter vector after an update, including: For any first parameter vector in the first population, randomly match the first parameter vector with another first parameter vector to obtain the information exchange vector corresponding to each first parameter vector; According to the first parameter vector and the information exchange vector corresponding to the first parameter vector, the information exchange factor is obtained as follows:

[0050] in, represents the information exchange factor, Indicates i The fitness value corresponding to the first parameter vector, Indicates i The fitness value corresponding to the information exchange vector corresponding to the first parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value sign; According to the information exchange factor, the first parameter vector is updated, and the first parameter vector after one update is obtained as follows:

[0051] in, Indicates t During the training process i The first parameter vector, i =1,2,…,L1, L1 represents the number corresponding to the first parameter vector, Represents the first parameter vector after an update , It means that the mean is 0 and the variance is Gaussian distributed random numbers.

[0052] Since the fitness values ​​of the first parameter vectors are relatively large and cannot learn from other vectors, information is exchanged through different first parameter vectors to achieve the search of the currently known optimal area, improve the algorithm's ability to find the optimal solution, and ensure the algorithm's search efficiency.

[0053] In the embodiment of the present invention, for the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain the second parameter vector after an update, including: For the first parameter vector in the first population, determining the optimal parameter vector according to the fitness value of the first parameter vector; For any second parameter vector in the second population, randomly match a first parameter vector to obtain a learning parameter vector corresponding to each second parameter vector; According to the fitness value corresponding to the optimal parameter vector and the fitness value corresponding to the learning parameter vector, the first learning factor and the second learning factor are obtained as follows:

[0054]

[0055] in, represents the first learning factor, represents the second learning factor, represents the fitness value corresponding to the mth second parameter vector, Represents the fitness value corresponding to the learning parameter vector corresponding to the mth second parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value symbol, Represents the fitness value corresponding to the optimal parameter vector, and exp represents the natural constant e An exponential function with base ; According to the optimal parameter vector, the learning parameter vector, the first learning factor and the second learning factor, a local search is performed on the second parameter vector, and the second parameter vector after one update is obtained as follows:

[0056] in, Indicates m The second parameter vector, Represents the second parameter vector after an update , m =1,2,…,L2, L2 represents the number corresponding to the second parameter vector, represents the first random number between (0,1), represents the second random number between (0,1), represents the second parameter vector The corresponding learning parameter vector, represents the optimal parameter vector.

[0057] The position-based adaptive learning strategy provided by the embodiment of the present invention can not only enable the second parameter vector to exchange information, but also learn the information of the optimal position, thereby improving the solution space exploration capability and ensuring the search speed of the algorithm. At the same time, adaptive search is achieved based on the position of the second parameter vector.

[0058] In the embodiment of the present invention, for the third parameter vector in the third population, an adaptive rapid development strategy is used to conduct a guided search on the third parameter vector to obtain the third parameter vector after an update, including: For any third parameter vector in the third population, randomly match a second parameter vector and a first parameter vector to obtain a first guided parameter vector and a second guided parameter vector; According to the current number of training times, the adaptive inertia weight is obtained as:

[0059] in, represents the adaptive inertia weight, T represents the preset maximum number of training times, represents the maximum inertia weight, Indicates the minimum inertia weight; According to the adaptive inertia weight, the first guided parameter vector and the second guided parameter vector, a guided search is performed on the third parameter vector, and the third parameter vector after one update is obtained as follows:

[0060] in, Indicates t During the training process n A third parameter vector, Represents the third parameter vector after an update , represents the third random number uniformly distributed in [0,1], represents the fourth random number uniformly distributed in [0,1], represents the first guided parameter vector, represents the second guided parameter vector.

[0061] The adaptive rapid development strategy provided by the embodiment of the present invention can quickly guide the search of parameter vectors with poor positions in the early stage of the algorithm and improve the search accuracy of the algorithm in the later stage of the algorithm.

[0062] In an embodiment of the present invention, a global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector, and the third parameter vector, and the first parameter vector, the second parameter vector, and the third parameter vector after the global search are obtained, including: Get the Levy flight step length:

[0063] Among them, s represents the Levy flight step length, U represents the first Levy flight factor, and V represents the second Levy flight factor. represents the Levy flight constant, and U represents the mean of 0 and the variance of V represents a random number generated by a normal distribution with a mean of 0 and a variance of 1. represents the normal distribution factor, and , represents the gamma function, sin represents the sine function, represents pi; According to the Levi flight step length, the flight step length is obtained as:

[0064] in, Indicates the flight step length; Randomly generate a flight step length corresponding to each dimension parameter of the third parameter vector, and form a vector of the flight step lengths corresponding to all dimension parameters to obtain a flight step length vector; According to the flight step vector, a global search is performed on the first parameter vector, the second parameter vector and the third parameter vector, and the first parameter vector, the second parameter vector and the third parameter vector after the global search are obtained as follows:

[0065] in, Indicates t During the training process k target parameter vectors, the target parameter vectors are the first parameter vector after one update, the second parameter vector after one update, and the third parameter vector after one update, represents the target parameter vector after global search, k =1,2,…,K, K represents the total number of target parameter vectors, represents the step size control factor, Represents the flight step vector.

[0066] The improved global search strategy of Levy flight provided in the embodiment of the present invention has stronger randomness and can effectively assist the algorithm to jump out of the local optimal solution. However, in order to ensure the search speed of the algorithm, after the global search, the fitness value of the target parameter vector after the global search can also be verified. If the fitness value increases, the global search is retained, otherwise the global search is rejected.

[0067] Optionally, after each search, the parameter vector may be processed for out-of-bounds errors so that the parameters are always within a valid range.

[0068] In an embodiment of the present invention, determining whether the training end condition is met includes: determining whether the current training times is greater than or equal to the maximum training times, if so, determining that the training end condition is met, otherwise determining that the training end condition is not met.

[0069] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for analyzing aviation material transaction data, characterized in that: include: Acquire the historical aviation material transaction data uploaded in advance by the staff and the transaction abnormality labels corresponding to the historical aviation material transaction data, and pre-process the historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data to obtain the pre-processed historical aviation material transaction data and the transaction abnormality labels corresponding to the historical aviation material transaction data; Calling at least one machine learning model from a preset model library, and training the at least one machine learning model through the pre-processed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, so as to obtain an optimal machine learning model; After the optimal machine learning model is deployed, the current aviation material transaction data uploaded by the staff is obtained, and the current aviation material transaction data is identified through the optimal machine learning model to determine the predicted probability of transaction anomalies, and when the predicted probability of transaction anomalies is greater than the preset threshold, the transaction anomaly result is determined.

2. The aviation material transaction data analysis method according to claim 1, characterized in that: Preprocessing the historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, including: formatting and normalizing the historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data.

3. The aviation material transaction data analysis method according to claim 1, characterized in that: The pre-set model library includes CNN model, BP neural network model, LSTM model, SVM model, ResNet model, RBF neural network model and RNNs model; Calling at least one machine learning model from a pre-set model library includes: obtaining a machine learning model calling instruction input by a staff member, and determining at least one machine learning model corresponding to the machine learning model calling instruction.

4. The aviation material transaction data analysis method according to claim 1, characterized in that: At least one machine learning model is trained by using the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data to obtain an optimal machine learning model, including: For any called machine learning model, the hyperparameters to be optimized of the machine learning model are initialized and encoded into vectors, and a plurality of different parameter vectors are determined; According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vector is divided into a first group, a second group, and a third group according to the fitness value of the parameter vector; For the first parameter vectors in the first population, information is exchanged between the first parameter vectors according to the fitness, to obtain the first parameter vector after an update; For the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain the second parameter vector after an update; For the third parameter vector in the third population, an adaptive rapid development strategy is used to conduct a guided search on the third parameter vector to obtain the third parameter vector after an update; For a first parameter vector after an update, a second parameter vector after an update, and a third parameter vector after an update, a global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector, and the third parameter vector to obtain the first parameter vector, the second parameter vector, and the third parameter vector after the global search; Determine whether the training end condition is met. If so, determine the optimal parameter vector based on the first parameter vector, the second parameter vector, and the third parameter vector after the global search, and use the parameters contained in the optimal parameter vector as the final parameters of the machine learning model to obtain the machine learning model after training. Otherwise, return to the population division step; Traverse all called machine learning models to obtain at least one trained machine learning model, and determine the optimal machine learning model based on the at least one trained machine learning model.

5. The aviation material transaction data analysis method according to claim 4, characterized in that: According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vector is divided into a first group, a second group and a third group according to the fitness value of the parameter vector, including: According to the preprocessed historical aviation material transaction data and the transaction anomaly labels corresponding to the historical aviation material transaction data, the fitness value corresponding to each parameter vector is obtained, and the parameter vectors are sorted in descending order of the fitness value to obtain the sorted parameter vectors; Based on the sorted parameter vectors, all parameter vectors are divided into the first group, the second group and the third group in the ratio of 1:8:

1.

6. The aviation material transaction data analysis method according to claim 5, characterized in that: For the first parameter vectors in the first population, information exchange is performed between the first parameter vectors according to the fitness, and the first parameter vector after an update is obtained, including: For any first parameter vector in the first population, randomly match the first parameter vector with another first parameter vector to obtain the information exchange vector corresponding to each first parameter vector; According to the first parameter vector and the information exchange vector corresponding to the first parameter vector, the information exchange factor is obtained as follows: in, represents the information exchange factor, Indicates i The fitness value corresponding to the first parameter vector, Indicates i The fitness value corresponding to the information exchange vector corresponding to the first parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value sign; According to the information exchange factor, the first parameter vector is updated, and the first parameter vector after one update is obtained as follows: in, Indicates t During the training i The first parameter vector, i =1,2,…,L1, L1 represents the number corresponding to the first parameter vector, Represents the first parameter vector after an update , It means that the mean is 0 and the variance is Gaussian distributed random numbers.

7. The aviation material transaction data analysis method according to claim 6, characterized in that: For the second parameter vector in the second population, a position-based adaptive learning strategy is used to perform a local search on the second parameter vector to obtain the second parameter vector after an update, including: For the first parameter vector in the first population, determining the optimal parameter vector according to the fitness value of the first parameter vector; For any second parameter vector in the second population, randomly match a first parameter vector to obtain a learning parameter vector corresponding to each second parameter vector; According to the fitness value corresponding to the optimal parameter vector and the fitness value corresponding to the learning parameter vector, the first learning factor and the second learning factor are obtained as follows: in, represents the first learning factor, represents the second learning factor, represents the fitness value corresponding to the mth second parameter vector, Represents the fitness value corresponding to the learning parameter vector corresponding to the mth second parameter vector, represents the noise term, and Less than 0.0001, || indicates the absolute value symbol, Represents the fitness value corresponding to the optimal parameter vector, and exp represents the natural constant e An exponential function with base ; According to the optimal parameter vector, the learning parameter vector, the first learning factor and the second learning factor, a local search is performed on the second parameter vector, and the second parameter vector after one update is obtained as follows: in, Indicates m The second parameter vector, Represents the second parameter vector after an update , m =1,2,…,L2, L2 represents the number corresponding to the second parameter vector, represents the first random number between (0,1), represents the second random number between (0,1), represents the second parameter vector The corresponding learning parameter vector, represents the optimal parameter vector.

8. The aviation material transaction data analysis method according to claim 7, characterized in that: For the third parameter vector in the third population, an adaptive rapid development strategy is used to guide the search of the third parameter vector to obtain the third parameter vector after an update, including: For any third parameter vector in the third population, randomly match a second parameter vector and a first parameter vector to obtain a first guided parameter vector and a second guided parameter vector; According to the current number of training times, the adaptive inertia weight is obtained as: in, represents the adaptive inertia weight, T represents the preset maximum number of training times, represents the maximum inertia weight, Indicates the minimum inertia weight; According to the adaptive inertia weight, the first guided parameter vector and the second guided parameter vector, a guided search is performed on the third parameter vector, and the third parameter vector after one update is obtained as follows: in, Indicates t During the training process n A third parameter vector, Represents the third parameter vector after an update , represents the third random number uniformly distributed in [0,1], represents the fourth random number uniformly distributed in [0,1], represents the first guided parameter vector, represents the second guided parameter vector.

9. The aviation material transaction data analysis method according to claim 8, characterized in that: The global search strategy improved by Levy flight is used to perform a global search on the first parameter vector, the second parameter vector and the third parameter vector, and the first parameter vector, the second parameter vector and the third parameter vector after the global search are obtained, including: Get the Levy flight step length: Among them, s represents the Levy flight step length, U represents the first Levy flight factor, and V represents the second Levy flight factor. represents the Levy flight constant, and U represents the mean of 0 and the variance of V represents a random number generated by a normal distribution with a mean of 0 and a variance of 1. represents the normal distribution factor, and , represents the gamma function, sin represents the sine function, represents pi; According to the Levi flight step length, the flight step length is obtained as: in, Indicates the flight step length; Randomly generate a flight step length corresponding to each dimension parameter of the third parameter vector, and form a vector of the flight step lengths corresponding to all dimension parameters to obtain a flight step length vector; According to the flight step vector, a global search is performed on the first parameter vector, the second parameter vector and the third parameter vector, and the first parameter vector, the second parameter vector and the third parameter vector after the global search are obtained as follows: in, Indicates t During the training k target parameter vectors, the target parameter vectors are the first parameter vector after one update, the second parameter vector after one update, and the third parameter vector after one update, represents the target parameter vector after global search, k =1,2,…,K, K represents the total number of target parameter vectors, represents the step size control factor, Represents the flight step vector.

10. The aviation material transaction data analysis method according to claim 9, characterized in that: Determining whether the training end condition is met includes: determining whether the current training times is greater than or equal to the maximum training times, if so, determining that the training end condition is met, otherwise determining that the training end condition is not met.

Images