CN116976774A - Logistics information management method and system based on artificial intelligence - Google Patents

Abstract

The invention discloses a logistics information management method and system based on artificial intelligence, wherein the method comprises the following steps: the method comprises the steps of data acquisition, efficient noise reduction based on optimized singular value decomposition, feature extraction based on PCA and efficient and accurate logistics trend prediction. The invention belongs to the technical field of logistics information management, in particular to a logistics information management method and system based on artificial intelligence, wherein the scheme adopts singular value decomposition based on standard deviation optimization to perform data noise reduction treatment, and the window length is dynamically adjusted according to actual conditions so as to adapt to noise reduction requirements of different data types; and optimizing an initialization process by adopting a ROL strategy and a DE algorithm, and optimizing a position updating method of the optimization parameters by considering a method of combining the optimization in the group with the global optimization while using an SFLA algorithm to perform the position updating of the worst optimization parameters.

Description

Logistics information management method and system based on artificial intelligence

Technical Field

The invention belongs to the technical field of logistics information management, and particularly relates to a logistics information management method and system based on artificial intelligence.

Background

And the logistics information management system performs reasonable resource allocation and transportation plan making according to the logistics trend prediction result in a short period in the future, so that logistics efficiency is improved and cost is reduced. However, the existing logistics information management system has the technical problems that the acquired data contain noise, so that the accuracy of the data is reduced, the actual situation cannot be truly reflected by the data, different data have different noise degrees, and the noise reduction requirements of the different data cannot be met by adopting a fixed window length; the optimization parameter initialization position is unreasonable to select, so that the performance of an algorithm is reduced, the calculation cost of model training and prediction is high, and the model is easy to fall into the local optimum.

Disclosure of Invention

Aiming at the problems that the acquired data contains noise, so that the accuracy of the data is reduced, the data cannot truly reflect the actual situation, different data have different noise degrees, the fixed window length cannot adapt to the noise reduction requirements of the different data, the data noise reduction processing is carried out by adopting the singular value decomposition based on standard deviation optimization, the size of the window length is dynamically adjusted according to the actual situation, so as to adapt to the noise reduction requirements of the different data types, the main characteristics of the data are extracted, the important information in the data is reserved, the redundancy and the noise are removed, the data is more accurate and reliable, the noise reduction effect is further improved, and the accuracy of the data is improved; aiming at the technical problems that the performance of an algorithm is reduced, the calculation cost of model training and prediction is high, and the model is easy to fall into local optimum due to unreasonable selection of an optimization parameter initialization position, the ROL strategy and DE algorithm optimization position initialization process is adopted in the scheme, the search range is enlarged, the diversity is increased, and the method of combining the optimal in a group with the global optimum is considered when the SFLA algorithm is used for carrying out worst optimization parameter position update, so that the local development and the global search capability of the algorithm are coordinated, the probability of premature convergence of the algorithm is reduced, the fall into local optimum is avoided, and the system performance is improved.

The technical scheme adopted by the invention is as follows: the invention provides a logistics information management method based on artificial intelligence, which comprises the following steps:

step S1: data acquisition, namely acquiring object flow prediction data, wherein the object flow prediction data comprises historical logistics data, historical sales data, historical supply chain data, historical economic data and corresponding labels, and the corresponding labels comprise ascending, descending and keeping stable;

step S2: performing high-efficiency noise reduction on the data based on optimized singular value decomposition, performing data noise reduction processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusting the window length according to actual conditions so as to adapt to noise reduction requirements of different data types, removing noise and reserving important information in the data;

step S3: performing feature extraction based on PCA, performing data standardization processing by using a Z-score standardization method, and performing feature extraction by using a PCA method to form a feature data set;

step S4: and (3) carrying out efficient and accurate flow trend prediction, optimizing an initialization process by adopting an ROL strategy and a DE algorithm, carrying out worst optimization parameter position update by using an SFLA algorithm, simultaneously, carrying out optimization on a position update method of optimization parameters by considering a method combining intra-group optimization and global optimization, constructing an efficient and accurate flow trend prediction model, and predicting the flow trend in a short period in the future.

Further, in step S2, based on the collected traffic prediction data, a time sequence is constructed for each data type and data denoising processing is performed, and the efficient denoising of the data based on the optimized singular value decomposition specifically includes the following steps:

step S21: data initialization, presetting a window length S=2 and a step length lambda=1;

step S22: constructing a time sequence, arranging historical logistics data according to a time sequence to obtain a time sequence, wherein the formula is as follows:

P（n1）=[p ₁ ，p ₂ ，…，p _N1 ]；

wherein P (N1) is a time series of the historical logistics data, and N1 is the number of the historical logistics data;

step S23: constructing a track matrix, wherein the number of rows of the track matrix is the window length S, the number of columns is D, and the following formula is used:

D=N1-S+1；

；

wherein P is _SD Is a track matrix;

step S24: singular value decomposition, the steps are as follows:

step S241: the covariance matrix is constructed using the following formula:

Q=P _SD ×P _SD ^T ；

where Q is the covariance matrix, P _SD ^T Is a track matrix P _SD X is the multiplication operator;

step S242: calculating eigenvalue and eigenvector, and calculating eigenvalue beta of covariance matrix Q ₁ ，β ₂ ，…，β _S （β ₁ ≥β ₂ ≥…≥β _S 0) and corresponding orthogonal feature vector C ₁ ，C ₂ ，…，C _S The formula used is as follows:

d=max{i:β _i >0}，i=1，2，…，s；

；

；

P _SD =P _SD，1 +…+P _SD，d ；

in the method, in the process of the invention, Is a singular value, d is the number of non-zero eigenvalues, M _i Is the right feature matrix, P _SD，i Is a first-class matrix, M _i ^T Is the right feature matrix M _i Is a transposed matrix of (a);

step S25: the contribution value is calculated using the following formula:

；

wherein ω is a contribution value;

step S26: noise reduction, arranging the contribution values in order from big to small, and selecting the feature vectors C corresponding to the first b contribution values _i And a characteristic value beta _i Reconstructing a trajectory matrix P _SD Obtaining a track matrix P after noise reduction _SD ' then P _SD ' diagonal average as a time series R of length N1 _i (n 1) obtaining a time sequence P1 (n 1) of the noise-reduced historical logistics data, wherein the following formula is used:

S ^* =min（S，D）；

D ^* =max（S，D）；

；

；

R _i （n1）=[r ₁ ，r ₂ ，…，r _N1 ]；

；

wherein P1 (n 1) is a time series of noise-reduced historical logistics data, S ^* Is the smaller of the number of rows and columns of the track matrix, D ^* Is the larger of the number of rows and columns of the track matrix, f _i，j Is a track matrix P _SD Element f of (a) _i，j ^* Is a track matrix P _SD ' element after reconstruction, r _k Is a track matrix P _SD The kth diagonal average in';

step S27: calculating a singular value standard deviation, calculating singular values of a time sequence of the noise-reduced historical logistics data according to the methods in the step S23 and the step S24, and calculating the singular value standard deviation, wherein the following formula is used:

；

Wherein sigma is the singular value standard deviation, and mean (beta) is the mean of the eigenvalues beta;

step S28: the window length is updated using the following formula:

S’=S+λ；

wherein S' is the updated window length;

step S29: window length determination, when S'. Ltoreq.Returning to the step S23, and carrying out noise reduction again based on the updated window length to obtain a series of time sequences and singular value standard deviations of the noise-reduced historical logistics data; when S'>/>When the method is used, the singular value standard deviation is sequenced according to the sequence from large to small, the time sequence of the noise-reduced historical logistics data obtained by the window length processing corresponding to the maximum singular value standard deviation is selected, and the time sequence of the noise-reduced historical logistics data is used as the time sequence of the final output noise-reduced historical logistics data;

step S210: determining output, and processing the historical sales data, the historical supply chain data and the historical economic data by adopting the same method from the step S21 to the step S29 respectively to obtain a time sequence of the historical sales data, a time sequence of the historical supply chain data and a time sequence of the historical economic data after noise reduction, wherein the time sequence is used as a time sequence of each data type finally output after noise reduction.

Further, in step S3, the feature extraction based on PCA is based on the time sequence of each data type and the corresponding label after the noise reduction processing, and the time sequence and the corresponding label are combined into a noise reduction data set, and the data normalization processing is performed by using the Z-score normalization method, and then the feature extraction is performed by using the PCA method, so as to form a feature data set.

Further, in step S4, the efficient and accurate flow trend prediction specifically includes the following steps:

step S41: constructing a training data set and a test data set, randomly selecting 70% of sample data from the characteristic data set as the training data set, and the rest 30% of sample data as the test data set;

step S42: initializing, namely presetting the total number N3 of optimized parameters, the second maximum iteration number T2, the space dimension A2 of the optimized parameters, a third evaluation threshold value xi 4 and a mutation scale factor H _b Probability of crossover H _c The step factor constant o, the punishment factor F, the fault-tolerant parameter E and the search space range of the Gaussian kernel function key parameter phi in the flow trend prediction model, wherein the lower boundary of the search space is LB2, the upper boundary of the search space is UB2, and the parameters (F, E, phi) represent the positions of the optimization parameters;

step S43: the initial matrix is constructed as follows:

step S431: the first optimization parameter position initialization method uses the following formula:

G1 _i，a2 ⁰ =rand（0，1）×（UB2 _a2 -LB2 _a2 ）+LB2 _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

wherein G1 _i，a2 ⁰ Is the initialization position of the first ith optimization parameter in the a 2-dimensional space, and rand (0, 1) is a random number between 0 and 1;

step S432: the second optimization parameter position initialization method uses the following formula:

G2 _i，a2 ⁰ =LB2 _a2 +UB2 _a2 -rand（0，1）×G _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

Wherein G2 _i，a2 ⁰ Is the initialization position of the second ith optimization parameter in the a 2-dimensional space, LB2 _a2 And UB2 _a2 Is the lower and upper boundaries of the a 2-dimensional search space, G _a2 Is a random position in the a 2-dimensional search space, i.e., G _a2 ∈[LB2 _a2 ，UB2 _a2 ]；

Step S433: initializing position combination, combining step S431 and stepCombining the initialization positions of the optimization parameters generated in step S432, and forming a set G based on the initialization positions of the (2 XN 3) optimization parameters ⁰ The formula used is as follows:

G ⁰ =（G ₁ ⁰ ，G ₂ ⁰ ，…，G _2×N3 ⁰ ）；

step S434: constructing a flow trend prediction model, calling an SVR function based on current parameters (F, E, phi) by using a python imported sklearn library, training a material flow trend prediction model based on a training data set, and predicting sample data of a test data set by using the trained material flow trend prediction model;

step S435: calculating an optimal fitness value and a second global optimal position, sequencing all optimization parameters according to the second fitness value in order from small to large, selecting the lowest fitness value as the optimal fitness value, and taking the position of the corresponding optimization parameter as a second global optimal position G _best ⁰ The formula used is as follows:

；

wherein L is the fitness value of the optimization parameter, y _j Is a real tag that is not a real tag, Is a predictive tag, N4 is the number of sample data in the test dataset;

step S436: position mutation, three different optimization parameters are randomly selected from 2×n3 and k +.m +.n +.i, the formula used is as follows:

V _i =G _i ⁰ +H _b ×（G _best ⁰ -G _k ⁰ +G _m ⁰ -G _n ⁰ ），i=1，2，…，2×N3；

wherein V is _i Is the position after mutation of the ith optimization parameter;

step S437: the position is crossed, and the formula is as follows:

；

wherein O is _i，a2 Is the position of the ith optimization parameter after the a2 nd dimensional space is crossed;

step S438: and (3) position replacement, namely calculating the fitness value of the optimization parameter after position crossing by adopting the same method in the step S435, and performing position replacement by adopting the following formula:

；

wherein G is _i Is the position of the ith optimization parameter after replacement, L (O _i ) Is O _i Is adapted to the degree of adaptation of L (G) _i ⁰ ) Is G _i ⁰ Is a fitness value of (a);

step S439: selecting optimization parameters, namely calculating fitness values, optimal fitness values and second global optimal positions of the optimization parameters after position replacement by adopting the same method in the step S435, sequencing the fitness values in a sequence from small to large, selecting optimization parameters corresponding to the first N3 low fitness values, and constructing an initial matrix based on a greedy strategy;

step S44: the location update parameters are calculated using the following formula:

；

；

；

；

；

wherein η1, η2, η3 and η4 are 4 different position update step size methods, R _B Is a vector of random numbers subject to normal distribution, R _L Is a vector of random numbers based on a levy distribution,is a multiplication operator item by item, CF is an adaptive parameter controlling the update step length, and t2 is the current iteration number;

step S45: the first location update uses the following formula:

；

in the method, in the process of the invention,is the position of the ith optimization parameter after the first position update;

step S46: and (3) updating the position for the second time, calculating the fitness value, the optimal fitness value and the second global optimal position of the optimized parameters after the position updating by adopting the same method in the step S435, sequencing the optimized parameters according to the fitness value in a sequence from small to large, dividing the optimized parameters into z groups based on a circular allocation mode, and updating the worst optimized parameter position, namely the position with the highest fitness value, in each group, wherein the formula is as follows:

；

；

；

；

in the method, in the process of the invention,、/>and->Is a method for updating the worst optimized parameter position in 3 different groups, G _best，i Is the optimal optimized parameter position in group i, G _worst，i Is the worst optimized parameter position in group i,/->Is->Is adapted to the value of->Is->Is adapted to the value of->Is->Is a fitness value of (a);

step S47: and (3) a third position update, wherein the maximum value and the minimum value in all dimensions of the current ith optimization parameter are adjusted to be the upper boundary and the lower boundary of the search space, and the third position update is performed in the adjusted dynamic search space, and the following formula is used:

；

；

In the method, in the process of the invention,is the position after the second position update of the ith optimization parameter,/for the second time>Is the position after the third position update of the ith optimization parameter, u _i Is the midpoint of the dynamic search space for the ith optimization parameter;

step S48: the location determination is performed using the following formula:

；

in the method, in the process of the invention,is the position determined after the iteration,/->Is->Is used for the adaptation value of the (a),is->Is a fitness value of (a);

step S49: updating the optimal fitness value and the second global optimal position, and calculating the fitness value, the optimal fitness value and the second global optimal position of the optimization parameter after the position determination by adopting the same method in the step S435;

step S410: the model determines that when the optimal fitness value is lower than a third evaluation threshold value xi 4, the flow trend prediction model is based on the current parameter and the step S411 is carried out; otherwise, if the second maximum iteration number T2 is reached, go to step S43; otherwise go to step S44;

step S411: and (3) predicting the flow trend, namely acquiring real-time flow data, sales data, supply chain data and economic data, inputting the real-time flow data, the sales data, the supply chain data and the economic data into a flow trend prediction model, and predicting the flow trend in a short period in the future based on a label output by the flow trend prediction model.

The invention provides an artificial intelligence-based logistics information management system which comprises a data acquisition module, a data noise reduction module, a feature extraction module and a logistics trend prediction module;

the data acquisition module acquires object flow prediction data, wherein the object flow prediction data comprises historical object flow data, historical sales data, historical supply chain data, historical economic data and corresponding labels, the corresponding labels comprise ascending, descending and keeping stable, and the object flow prediction data is sent to the data noise reduction module;

the data denoising module receives the flow prediction data sent by the data acquisition module, performs data denoising processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusts the window length according to actual conditions so as to adapt to denoising requirements of different data types, removes noise, retains important information in the data, forms a time sequence after the data is subjected to the denoising, and sends the time sequence after the data types are subjected to the denoising to the feature extraction module;

the feature extraction module receives the time sequence after the different data types sent by the data noise reduction module are subjected to drying, performs data standardization processing by using a Z-score standardization method, performs feature extraction by using a PCA method to form a feature data set, and sends the feature data set to the logistics trend prediction module;

The flow trend prediction module receives the characteristic data set sent by the characteristic extraction module, optimizes the initialization process by adopting an ROL strategy and a DE algorithm, and optimizes the position updating method of the optimization parameters by considering a method combining the optimization parameters in the group and the global optimization while carrying out the position updating of the worst optimization parameters by using an SFLA algorithm, so as to construct a high-efficiency accurate flow trend prediction model and predict the flow trend in a short period in the future.

By adopting the scheme, the beneficial effects obtained by the invention are as follows:

(1) Aiming at the technical problems that the acquired data contains noise, the accuracy of the data is reduced, the data cannot truly reflect the actual situation, and different data have different noise degrees, and the fixed window length cannot adapt to the noise reduction requirements of different data, the data noise reduction processing is carried out by adopting the singular value decomposition based on standard deviation optimization, the size of the window length is dynamically adjusted according to the actual situation, so that the noise reduction requirements of different data types are met, the main characteristics of the data are extracted, the important information in the data is reserved, redundancy and noise are removed, the data is more accurate and reliable, the noise reduction effect is further increased, and the accuracy of the data is improved.

(2) Aiming at the technical problems that the performance of an algorithm is reduced, the calculation cost of model training and prediction is high, and the model is easy to fall into local optimum due to unreasonable selection of an optimization parameter initialization position, the ROL strategy and DE algorithm optimization position initialization process is adopted in the scheme, the search range is enlarged, the diversity is increased, and the method of combining the optimal in a group with the global optimum is considered when the SFLA algorithm is used for carrying out worst optimization parameter position update, so that the local development and the global search capability of the algorithm are coordinated, the probability of premature convergence of the algorithm is reduced, the fall into local optimum is avoided, and the system performance is improved.

Drawings

FIG. 1 is a schematic flow chart of a logistics information management method based on artificial intelligence;

FIG. 2 is a schematic diagram of an artificial intelligence-based logistics information management system;

FIG. 3 is a flow chart of step S2;

fig. 4 is a flow chart of step S4;

FIG. 5 is a schematic diagram of an optimization parameter search location;

FIG. 6 is a graph of an optimization parameter search.

The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In the description of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," "outer," and the like indicate orientation or positional relationships based on those shown in the drawings, merely to facilitate description of the invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus should not be construed as limiting the invention.

Referring to fig. 1, the present invention provides a logistic information management method based on artificial intelligence, which includes the following steps:

step S1: data acquisition, namely acquiring object flow prediction data, wherein the object flow prediction data comprises historical logistics data, historical sales data, historical supply chain data, historical economic data and corresponding labels, and the corresponding labels comprise ascending, descending and keeping stable;

Step S2: performing high-efficiency noise reduction on the data based on optimized singular value decomposition, performing data noise reduction processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusting the window length according to actual conditions so as to adapt to noise reduction requirements of different data types, removing noise and reserving important information in the data;

step S3: performing feature extraction based on PCA, performing data standardization processing by using a Z-score standardization method, and performing feature extraction by using a PCA method to form a feature data set;

step S4: and (3) carrying out efficient and accurate flow trend prediction, optimizing an initialization process by adopting an ROL strategy and a DE algorithm, carrying out worst optimization parameter position update by using an SFLA algorithm, simultaneously, carrying out optimization on a position update method of optimization parameters by considering a method combining intra-group optimization and global optimization, constructing an efficient and accurate flow trend prediction model, and predicting the flow trend in a short period in the future.

In step S2, based on the collected traffic prediction data, a time sequence is constructed and data denoising processing is performed for each data type, and the efficient denoising processing for the data based on the optimized singular value decomposition specifically includes the following steps:

Step S21: data initialization, presetting a window length S=2 and a step length lambda=1;

step S22: constructing a time sequence, arranging historical logistics data according to a time sequence to obtain a time sequence, wherein the formula is as follows:

P（n1）=[p ₁ ，p ₂ ，…，p _N1 ]；

wherein P (N1) is a time series of the historical logistics data, and N1 is the number of the historical logistics data;

step S23: constructing a track matrix, wherein the number of rows of the track matrix is the window length S, the number of columns is D, and the following formula is used:

D=N1-S+1；

；

wherein P is _SD Is a track matrix;

step S24: singular value decomposition, the steps are as follows:

step S241: the covariance matrix is constructed using the following formula:

Q=P _SD ×P _SD ^T ；

where Q is the covariance matrix, P _SD ^T Is a track matrix P _SD X is the multiplication operator;

step S242: calculating eigenvalue and eigenvector, and calculating eigenvalue beta of covariance matrix Q ₁ ，β ₂ ，…，β _S （β ₁ ≥β ₂ ≥…≥β _S 0) and corresponding orthogonal feature vector C ₁ ，C ₂ ，…，C _S The formula used is as follows:

d=max{i:β _i >0}，i=1，2，…，s；

；

；

P _SD =P _SD，1 +…+P _SD，d ；

in the method, in the process of the invention,is a singular value, d is the number of non-zero eigenvalues, M _i Is the right feature matrix, P _SD，i Is a first-class matrix, M _i ^T Is the right feature matrix M _i Is a transposed matrix of (a);

step S25: the contribution value is calculated using the following formula:

；

wherein ω is a contribution value;

step S26: noise reduction, arranging the contribution values in order from big to small, and selecting the feature vectors C corresponding to the first b contribution values _i And a characteristic value beta _i Reconstructing a trajectory matrix P _SD Obtaining a track matrix P after noise reduction _SD ' then P _SD ' diagonal average as a time series R of length N1 _i (n 1) obtaining a time sequence P1 (n 1) of the noise-reduced historical logistics data, wherein the following formula is used:

S ^* =min（S，D）；

D ^* =max（S，D）；

；

；

R _i （n1）=[r ₁ ，r ₂ ，…，r _N1 ]；

；

wherein P1 (n 1) is a time series of noise-reduced historical logistics data, S ^* Is the smaller of the number of rows and columns of the track matrix, D ^* Is the larger of the number of rows and columns of the track matrix, f _i，j Is a track matrix P _SD Element f of (a) _i，j ^* Is a track matrix P _SD ' element after reconstruction, r _k Is a track matrix P _SD The kth diagonal average in';

step S27: calculating a singular value standard deviation, calculating singular values of a time sequence of the noise-reduced historical logistics data according to the methods in the step S23 and the step S24, and calculating the singular value standard deviation, wherein the following formula is used:

；

wherein sigma is the singular value standard deviation, and mean (beta) is the mean of the eigenvalues beta;

step S28: the window length is updated using the following formula:

S’=S+λ；

wherein S' is the updated window length;

step S29: window length determination, when S'. Ltoreq.Returning to the step S23, and carrying out noise reduction again based on the updated window length to obtain a series of time sequences and singular value standard deviations of the noise-reduced historical logistics data; when S' >/>When the singular value standard deviation is changed from large to smallThe sequence of the noise-reduced historical logistics data is sequenced, a time sequence of the noise-reduced historical logistics data, which is obtained through window length processing corresponding to the maximum singular value standard deviation, is selected and used as a time sequence of the noise-reduced historical logistics data which is finally output;

step S210: determining output, and processing the historical sales data, the historical supply chain data and the historical economic data by adopting the same method from the step S21 to the step S29 respectively to obtain a time sequence of the historical sales data, a time sequence of the historical supply chain data and a time sequence of the historical economic data after noise reduction, wherein the time sequence is used as a time sequence of each data type after noise reduction of final output.

By executing the above operation, aiming at the technical problems that the acquired data contains noise, the accuracy of the data is reduced, the data cannot truly reflect the actual situation, and different data have different noise degrees, and the fixed window length cannot adapt to the noise reduction requirements of different data.

Referring to fig. 1, in step S3, the feature extraction based on PCA is based on a time sequence of each data type and a corresponding label after the noise reduction process, and the time sequence and the corresponding label are combined into a noise reduction data set, and the data normalization process is performed by using a Z-score normalization method, and then the feature extraction is performed by using a PCA method, so as to form a feature data set.

In a fourth embodiment, referring to fig. 1 and 4, the present embodiment is based on the above embodiment, and in step S4, the efficient and accurate flow trend prediction specifically includes the following steps:

step S41: constructing a training data set and a test data set, randomly selecting 70% of sample data from the characteristic data set as the training data set, and the rest 30% of sample data as the test data set;

step S42: initializing, namely presetting the total number N3 of optimized parameters, the second maximum iteration number T2, the space dimension A2 of the optimized parameters, a third evaluation threshold value xi 4 and a mutation scale factor H _b Probability of crossover H _c The step factor constant o, the punishment factor F, the fault-tolerant parameter E and the search space range of the Gaussian kernel function key parameter phi in the flow trend prediction model, wherein the lower boundary of the search space is LB2, the upper boundary of the search space is UB2, and the parameters (F, E, phi) represent the positions of the optimization parameters;

Step S43: the initial matrix is constructed as follows:

step S431: the first optimization parameter position initialization method uses the following formula:

G1 _i，a2 ⁰ =rand（0，1）×（UB2 _a2 -LB2 _a2 ）+LB2 _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

wherein G1 _i，a2 ⁰ Is the initialization position of the first ith optimization parameter in the a 2-dimensional space, and rand (0, 1) is a random number between 0 and 1;

step S432: the second optimization parameter position initialization method uses the following formula:

G2 _i，a2 ⁰ =LB2 _a2 +UB2 _a2 -rand（0，1）×G _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

wherein G2 _i，a2 ⁰ Is the initialization position of the second ith optimization parameter in the a 2-dimensional space, LB2 _a2 And UB2 _a2 Is the lower and upper boundaries of the a 2-dimensional search space, G _a2 Is a random position in the a 2-dimensional search space, i.e., G _a2 ∈[LB2 _a2 ，UB2 _a2 ]；

Step S433: combining the initialization positions, combining the initialization positions of the optimization parameters generated in step S431 and step S432, and forming a set G based on the initialization positions of (2 XN 3) optimization parameters ⁰ The formula used is as follows:

G ⁰ =（G ₁ ⁰ ，G ₂ ⁰ ，…，G _2×N3 ⁰ ）；

step S434: constructing a flow trend prediction model, calling an SVR function based on current parameters (F, E, phi) by using a python imported sklearn library, training a material flow trend prediction model based on a training data set, and predicting sample data of a test data set by using the trained material flow trend prediction model;

step S435: calculating an optimal fitness value and a second global optimal position, sequencing all optimization parameters according to the second fitness value in order from small to large, selecting the lowest fitness value as the optimal fitness value, and taking the position of the corresponding optimization parameter as a second global optimal position G _best ⁰ The formula used is as follows:

；

wherein L is the fitness value of the optimization parameter, y _j Is a real tag that is not a real tag,is a predictive tag, N4 is the number of sample data in the test dataset;

step S436: position mutation, three different optimization parameters are randomly selected from 2×n3 and k +.m +.n +.i, the formula used is as follows:

V _i =G _i ⁰ +H _b ×（G _best ⁰ -G _k ⁰ +G _m ⁰ -G _n ⁰ ），i=1，2，…，2×N3；

wherein V is _i Is the position after mutation of the ith optimization parameter;

step S437: the position is crossed, and the formula is as follows:

；

wherein O is _i，a2 Is the position of the ith optimization parameter after the a2 nd dimensional space is crossed;

step S438: and (3) position replacement, namely calculating the fitness value of the optimization parameter after position crossing by adopting the same method in the step S435, and performing position replacement by adopting the following formula:

；

wherein G is _i Is the position of the ith optimization parameter after replacement, L (O _i ) Is O _i Is adapted to the degree of adaptation of L (G) _i ⁰ ) Is G _i ⁰ Is a fitness value of (a);

step S439: selecting optimization parameters, namely calculating fitness values, optimal fitness values and second global optimal positions of the optimization parameters after position replacement by adopting the same method in the step S435, sequencing the fitness values in a sequence from small to large, selecting optimization parameters corresponding to the first N3 low fitness values, and constructing an initial matrix based on a greedy strategy;

Step S44: the location update parameters are calculated using the following formula:

；

；

；/>

；

；

wherein η1, η2, η3 and η4 are 4 different position update step size methods, R _B Is a vector of random numbers subject to normal distribution, R _L Is a vector of random numbers based on a levy distribution,is a multiplication operator item by item, CF is an adaptive parameter controlling the update step length, and t2 is the current iteration number;

step S45: the first location update uses the following formula:

；

in the method, in the process of the invention,is the position of the ith optimization parameter after the first position update;

step S46: and (3) updating the position for the second time, calculating the fitness value, the optimal fitness value and the second global optimal position of the optimized parameters after the position updating by adopting the same method in the step S435, sequencing the optimized parameters according to the fitness value in a sequence from small to large, dividing the optimized parameters into z groups based on a circular allocation mode, and updating the worst optimized parameter position, namely the position with the highest fitness value, in each group, wherein the formula is as follows:

；

；

；

；

in the method, in the process of the invention,、/>and->Is a method for updating the worst optimized parameter position in 3 different groups, G _best，i Is the optimal optimized parameter position in group i, G _worst，i Is the worst optimized parameter position in group i,/- >Is->Is adapted to the value of->Is->Is adapted to the value of->Is->Is a fitness value of (a);

step S47: and (3) a third position update, wherein the maximum value and the minimum value in all dimensions of the current ith optimization parameter are adjusted to be the upper boundary and the lower boundary of the search space, and the third position update is performed in the adjusted dynamic search space, and the following formula is used:

；/>

；

in the method, in the process of the invention,is the position after the second position update of the ith optimization parameter,/for the second time>Is the third order bit of the ith optimization parameterSetting the updated position, u _i Is the midpoint of the dynamic search space for the ith optimization parameter;

step S48: the location determination is performed using the following formula:

；

in the method, in the process of the invention,is the position determined after the iteration,/->Is->Is used for the adaptation value of the (a),is->Is a fitness value of (a);

step S49: updating the optimal fitness value and the second global optimal position, and calculating the fitness value, the optimal fitness value and the second global optimal position of the optimization parameter after the position determination by adopting the same method in the step S435;

step S410: the model determines that when the optimal fitness value is lower than a third evaluation threshold value xi 4, the flow trend prediction model is based on the current parameter and the step S411 is carried out; otherwise, if the second maximum iteration number T2 is reached, go to step S43; otherwise go to step S44;

Step S411: and (3) predicting the flow trend, namely acquiring real-time flow data, sales data, supply chain data and economic data, inputting the real-time flow data, the sales data, the supply chain data and the economic data into a flow trend prediction model, and predicting the flow trend in a short period in the future based on a label output by the flow trend prediction model.

By executing the operations, the optimization parameter initialization position is selected unreasonably, so that the performance of an algorithm is reduced, the calculation cost of model training and prediction is increased, and the model is easy to fall into local optimum.

Fifth embodiment, referring to fig. 5 and 6, the embodiment is based on the above embodiment, and in fig. 5, a process of continuously updating the location of the optimization parameter until the second global optimal location is found is shown; in fig. 6, the ordinate is the fitness value, and the abscissa is the iteration number, and shows the change process that the fitness value gradually decreases with the change of the iteration number, so that the optimization parameters are close to a better search area, the prediction effect of the algorithm is greatly improved, and the problem that a better global solution cannot be found due to the fact that the local optimization is trapped is avoided.

In a sixth embodiment, referring to fig. 2, the embodiment is based on the above embodiment, and the logistics information management system based on artificial intelligence provided by the invention includes a data acquisition module, a data noise reduction module, a feature extraction module and a logistics trend prediction module;

the data acquisition module acquires object flow prediction data, wherein the object flow prediction data comprises historical object flow data, historical sales data, historical supply chain data, historical economic data and corresponding labels, the corresponding labels comprise ascending, descending and keeping stable, and the object flow prediction data is sent to the data noise reduction module;

the data denoising module receives the flow prediction data sent by the data acquisition module, performs data denoising processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusts the window length according to actual conditions so as to adapt to denoising requirements of different data types, removes noise, retains important information in the data, forms a time sequence after the data is subjected to the denoising, and sends the time sequence after the data types are subjected to the denoising to the feature extraction module;

the feature extraction module receives the time sequence after the different data types sent by the data noise reduction module are subjected to drying, performs data standardization processing by using a Z-score standardization method, performs feature extraction by using a PCA method to form a feature data set, and sends the feature data set to the logistics trend prediction module;

The flow trend prediction module receives the characteristic data set sent by the characteristic extraction module, optimizes the initialization process by adopting an ROL strategy and a DE algorithm, and optimizes the position updating method of the optimization parameters by considering a method combining the optimization parameters in the group and the global optimization while carrying out the position updating of the worst optimization parameters by using an SFLA algorithm, so as to construct a high-efficiency accurate flow trend prediction model and predict the flow trend in a short period in the future.

It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.

Claims (6)

1. The logistics information management method based on artificial intelligence is characterized by comprising the following steps of: the method comprises the following steps:

step S1: data acquisition, namely acquiring object flow prediction data, wherein the object flow prediction data comprises historical logistics data, historical sales data, historical supply chain data, historical economic data and corresponding labels, and the corresponding labels comprise ascending, descending and keeping stable;

step S2: performing high-efficiency noise reduction on the data based on optimized singular value decomposition, performing data noise reduction processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusting the window length according to actual conditions so as to adapt to noise reduction requirements of different data types, removing noise and reserving important information in the data;

Step S3: performing feature extraction based on PCA, performing data standardization processing by using a Z-score standardization method, and performing feature extraction by using a PCA method to form a feature data set;

step S4: and (3) carrying out efficient and accurate flow trend prediction, optimizing an initialization process by adopting an ROL strategy and a DE algorithm, carrying out worst optimization parameter position update by using an SFLA algorithm, simultaneously, carrying out optimization on a position update method of optimization parameters by considering a method combining intra-group optimization and global optimization, constructing an efficient and accurate flow trend prediction model, and predicting the flow trend in a short period in the future.

2. The logistics information management method based on artificial intelligence of claim 1, wherein: in step S4, the efficient and accurate flow trend prediction specifically includes the following steps:

step S41: constructing a training data set and a test data set, randomly selecting 70% of sample data from the characteristic data set as the training data set, and the rest 30% of sample data as the test data set;

step S42: initializing, presetting the total number N3 of optimized parameters and the second mostLarge iteration number T2, optimized parameter space dimension A2, third evaluation threshold ζ4, mutation scale factor H _b Probability of crossover H _c The step factor constant o, the punishment factor F, the fault-tolerant parameter E and the search space range of the Gaussian kernel function key parameter phi in the flow trend prediction model, wherein the lower boundary of the search space is LB2, the upper boundary of the search space is UB2, and the parameters (F, E, phi) represent the positions of the optimization parameters;

step S43: the initial matrix is constructed as follows:

step S431: the first optimization parameter position initialization method uses the following formula:

G1 _i，a2 ⁰ =rand（0，1）×（UB2 _a2 -LB2 _a2 ）+LB2 _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

wherein G1 _i，a2 ⁰ Is the initialization position of the first ith optimization parameter in the a 2-dimensional space, and rand (0, 1) is a random number between 0 and 1;

step S432: the second optimization parameter position initialization method uses the following formula:

G2 _i，a2 ⁰ =LB2 _a2 +UB2 _a2 -rand（0，1）×G _a2 ，i=1，2，…，N3，a2=1，2，…，A2；

wherein G2 _i，a2 ⁰ Is the initialization position of the second ith optimization parameter in the a 2-dimensional space, LB2 _a2 And UB2 _a2 Is the lower and upper boundaries of the a 2-dimensional search space, G _a2 Is a random position in the a 2-dimensional search space, i.e., G _a2 ∈[LB2 _a2 ，UB2 _a2 ]；

Step S433: combining the initialization positions, combining the initialization positions of the optimization parameters generated in step S431 and step S432, and forming a set G based on the initialization positions of (2 XN 3) optimization parameters ⁰ The formula used is as follows:

G ⁰ =（G ₁ ⁰ ，G ₂ ⁰ ，…，G _2×N3 ⁰ ）；

Step S434: constructing a flow trend prediction model, calling an SVR function based on current parameters (F, E, phi) by using a python imported sklearn library, training a material flow trend prediction model based on a training data set, and predicting sample data of a test data set by using the trained material flow trend prediction model;

step S435: calculating an optimal fitness value and a second global optimal position, sequencing all optimization parameters according to the second fitness value in order from small to large, selecting the lowest fitness value as the optimal fitness value, and taking the position of the corresponding optimization parameter as a second global optimal position G _best ⁰ The formula used is as follows:

；

wherein L is the fitness value of the optimization parameter, y _j Is a real tag that is not a real tag,is a predictive tag, N4 is the number of sample data in the test dataset;

step S436: position mutation, three different optimization parameters are randomly selected from 2×n3 and k +.m +.n +.i, the formula used is as follows:

V _i =G _i ⁰ +H _b ×（G _best ⁰ -G _k ⁰ +G _m ⁰ -G _n ⁰ ），i=1，2，…，2×N3；

wherein V is _i Is the position after mutation of the ith optimization parameter;

step S437: the position is crossed, and the formula is as follows:

；

wherein O is _i，a2 Is the position of the ith optimization parameter after the a2 nd dimensional space is crossed;

step S438: and (3) position replacement, namely calculating the fitness value of the optimization parameter after position crossing by adopting the same method in the step S435, and performing position replacement by adopting the following formula:

；

Wherein G is _i Is the position of the ith optimization parameter after replacement, L (O _i ) Is O _i Is adapted to the degree of adaptation of L (G) _i ⁰ ) Is G _i ⁰ Is a fitness value of (a);

step S439: selecting optimization parameters, namely calculating fitness values, optimal fitness values and second global optimal positions of the optimization parameters after position replacement by adopting the same method in the step S435, sequencing the fitness values in a sequence from small to large, selecting optimization parameters corresponding to the first N3 low fitness values, and constructing an initial matrix based on a greedy strategy;

step S44: the location update parameters are calculated using the following formula:

；

；

；

；

；

wherein η1, η2, η3 and η4 are 4 different position update step size methods, R _B Is a vector of random numbers subject to normal distribution, R _L Is based on the random distribution of LevyNumber of the vector quantity is used to determine the vector quantity,is a multiplication operator item by item, CF is an adaptive parameter controlling the update step length, and t2 is the current iteration number;

step S45: the first location update uses the following formula:

；

in the method, in the process of the invention,is the position of the ith optimization parameter after the first position update;

step S46: and (3) updating the position for the second time, calculating the fitness value, the optimal fitness value and the second global optimal position of the optimized parameters after the position updating by adopting the same method in the step S435, sequencing the optimized parameters according to the fitness value in a sequence from small to large, dividing the optimized parameters into z groups based on a circular allocation mode, and updating the worst optimized parameter position, namely the position with the highest fitness value, in each group, wherein the formula is as follows:

；

；

；

；

In the method, in the process of the invention,、/>and->Is a method for updating the worst optimized parameter position in 3 different groups, G _best，i Is the optimal optimized parameter position in group i, G _worst，i Is the worst optimized parameter position in group i,/->Is->Is adapted to the value of->Is->Is adapted to the value of->Is->Is a fitness value of (a);

step S47: and (3) a third position update, wherein the maximum value and the minimum value in all dimensions of the current ith optimization parameter are adjusted to be the upper boundary and the lower boundary of the search space, and the third position update is performed in the adjusted dynamic search space, and the following formula is used:

；

；

in the method, in the process of the invention,is the position after the second position update of the ith optimization parameter,/for the second time>Is the position after the third position update of the ith optimization parameter, u _i Is the midpoint of the dynamic search space for the ith optimization parameter;

step S48: the location determination is performed using the following formula:

；

in the method, in the process of the invention,is the position determined after the iteration,/->Is->Is adapted to the value of->Is thatIs a fitness value of (a);

step S49: updating the optimal fitness value and the second global optimal position, and calculating the fitness value, the optimal fitness value and the second global optimal position of the optimization parameter after the position determination by adopting the same method in the step S435;

step S410: the model determines that when the optimal fitness value is lower than a third evaluation threshold value xi 4, the flow trend prediction model is based on the current parameter and the step S411 is carried out; otherwise, if the second maximum iteration number T2 is reached, go to step S43; otherwise go to step S44;

Step S411: and (3) predicting the flow trend, namely acquiring real-time flow data, sales data, supply chain data and economic data, inputting the real-time flow data, the sales data, the supply chain data and the economic data into a flow trend prediction model, and predicting the flow trend in a short period in the future based on a label output by the flow trend prediction model.

3. The logistics information management method based on artificial intelligence of claim 1, wherein: in step S2, based on the collected traffic prediction data, for each data type, a time sequence is constructed and data denoising processing is performed, and the efficient denoising of the data based on the optimized singular value decomposition specifically includes the following steps:

step S21: data initialization, presetting a window length S=2 and a step length lambda=1;

step S22: constructing a time sequence, arranging historical logistics data according to a time sequence to obtain a time sequence, wherein the formula is as follows:

P（n1）=[p ₁ ，p ₂ ，…，p _N1 ]；

wherein P (N1) is a time series of the historical logistics data, and N1 is the number of the historical logistics data;

step S23: constructing a track matrix, wherein the number of rows of the track matrix is the window length S, the number of columns is D, and the following formula is used:

D=N1-S+1；

；

wherein P is _SD Is a track matrix;

step S24: singular value decomposition, the steps are as follows:

Step S241: the covariance matrix is constructed using the following formula:

Q=P _SD ×P _SD ^T ；

where Q is the covariance matrix, P _SD ^T Is a track matrix P _SD X is the multiplication operator;

step S242: calculating eigenvalue and eigenvector, and calculating eigenvalue beta of covariance matrix Q ₁ ，β ₂ ，…，β _S （β ₁ ≥β ₂ ≥…≥β _S 0) and corresponding orthogonal feature vector C ₁ ，C ₂ ，…，C _S The formula used is as follows:

d=max{i:β _i >0}，i=1，2，…，s；

；

；

P _SD =P _SD，1 +…+P _SD，d ；

in the method, in the process of the invention,is a singular value, d is the number of non-zero eigenvalues, M _i Is the right feature matrix, P _SD，i Is a first-class matrix, M _i ^T Is the right feature matrix M _i Is a transposed matrix of (a);

step S25: the contribution value is calculated using the following formula:

；

wherein ω is a contribution value;

step S26: noise reduction, arranging the contribution values in order from big to small, and selecting the feature vectors C corresponding to the first b contribution values _i And a characteristic value beta _i Reconstructing a trajectory matrix P _SD Obtaining a track matrix P after noise reduction _SD ' then P _SD ' diagonal average as a time series R of length N1 _i (n 1) obtaining a time sequence P1 (n 1) of the noise-reduced historical logistics data, wherein the following formula is used:

S ^* =min（S，D）；

D ^* =max（S，D）；

；

；

R _i （n1）=[r ₁ ，r ₂ ，…，r _N1 ]；

；

wherein P1 (n 1) is a time series of noise-reduced historical logistics data, S ^* Is the smaller of the number of rows and columns of the track matrix, D ^* Is the larger of the number of rows and columns of the track matrix, f _i，j Is a track matrix P _SD Element f of (a) _i，j ^* Is a track matrix P _SD ' element after reconstruction, r _k Is a track matrix P _SD The kth diagonal average in';

step S27: calculating a singular value standard deviation, calculating singular values of a time sequence of the noise-reduced historical logistics data according to the methods in the step S23 and the step S24, and calculating the singular value standard deviation, wherein the following formula is used:

；

wherein sigma is the singular value standard deviation, and mean (beta) is the mean of the eigenvalues beta;

step S28: the window length is updated using the following formula:

S’=S+λ；

wherein S' is the updated window length;

step S29: window length determination, when S'. Ltoreq.Returning to the step S23, and carrying out noise reduction again based on the updated window length to obtain a series of time sequences and singular value standard deviations of the noise-reduced historical logistics data; when S'>/>When the method is used, the singular value standard deviation is sequenced according to the sequence from large to small, the time sequence of the noise-reduced historical logistics data obtained by the window length processing corresponding to the maximum singular value standard deviation is selected, and the time sequence of the noise-reduced historical logistics data is used as the time sequence of the final output noise-reduced historical logistics data;

step S210: determining output, and processing the historical sales data, the historical supply chain data and the historical economic data by adopting the same method from the step S21 to the step S29 respectively to obtain a time sequence of the historical sales data, a time sequence of the historical supply chain data and a time sequence of the historical economic data after noise reduction, wherein the time sequence is used as a time sequence of each data type after noise reduction of final output.

4. The logistics information management method based on artificial intelligence of claim 1, wherein: in step S3, the feature extraction based on PCA is based on the time sequence of each data type and the corresponding label after the noise reduction processing, and the time sequence and the corresponding label are combined into a noise reduction data set, and the data normalization processing is performed by using a Z-score normalization method, and then the feature extraction is performed by using a PCA method, so as to form a feature data set.

5. An artificial intelligence based logistics information management system for implementing the artificial intelligence based logistics information management method as set forth in any one of claims 1 to 4, wherein: the system comprises a data acquisition module, a data noise reduction module, a feature extraction module and a logistics trend prediction module.

6. The artificial intelligence based logistics information management system of claim 5, wherein: the data acquisition module acquires object flow prediction data, wherein the object flow prediction data comprises historical object flow data, historical sales data, historical supply chain data, historical economic data and corresponding labels, the corresponding labels comprise ascending, descending and keeping stable, and the object flow prediction data is sent to the data noise reduction module;

The data denoising module receives the flow prediction data sent by the data acquisition module, performs data denoising processing by adopting singular value decomposition based on standard deviation optimization, dynamically adjusts the window length according to actual conditions so as to adapt to denoising requirements of different data types, removes noise, retains important information in the data, forms a time sequence after the data is subjected to the denoising, and sends the time sequence after the data types are subjected to the denoising to the feature extraction module;

the feature extraction module receives the time sequence after the different data types sent by the data noise reduction module are subjected to drying, performs data standardization processing by using a Z-score standardization method, performs feature extraction by using a PCA method to form a feature data set, and sends the feature data set to the logistics trend prediction module;

the flow trend prediction module receives the characteristic data set sent by the characteristic extraction module, optimizes the initialization process by adopting an ROL strategy and a DE algorithm, and optimizes the position updating method of the optimization parameters by considering a method combining the optimization parameters in the group and the global optimization while carrying out the position updating of the worst optimization parameters by using an SFLA algorithm, so as to construct a high-efficiency accurate flow trend prediction model and predict the flow trend in a short period in the future.