AI-based intelligent identification method for the warehousing industry

Abstract

The present invention relates to the field of warehouse management technology, and in particular to an AI-based intelligent identification method for the warehouse industry, including: reducing the dimension of each parameter vector to obtain a reduced dimension vector, hierarchically clustering all samples to obtain a hierarchical clustering tree, screening out the best layer in the hierarchical clustering tree according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree, adjusting the weight of the weak classifier in the Adaboost algorithm by the difference between the reduced dimension vector of the sample and the reduced dimension vector of other samples with the same label, and combining several neural networks in combination with the adjusted Adaboost algorithm to identify the state of goods transported by a warehouse transport vehicle. The present invention ensures the efficiency and accuracy of processing real-time data, and can effectively improve the accuracy of identifying the transportation state of warehouse goods.

Description

Technical Field

[0001] The present invention relates to the technical field of warehouse management, and in particular to an AI-based intelligent identification method for the warehouse industry. Background Art

[0002] In the cargo transportation and distribution link of the warehousing industry, it is usually necessary to identify the transportation status of the cargo to ensure the transportation safety of the cargo. Usually, the transportation status of the cargo is identified based on a variety of data during the transportation process.

[0003] In the existing methods for identifying the transportation status of goods, neural networks are usually used as classifiers to classify various data in the cargo transportation process to identify the transportation status of the goods. However, the classification effect of a single neural network is highly dependent on the training effect of the neural network. Therefore, the Adaboost algorithm is used to linearly combine multiple neural networks as classifiers to improve the classification and recognition effect of the neural network. However, the weights of weak classifiers in the Adaboost algorithm are prone to large changes, resulting in low accuracy of the classification results and inability to accurately identify the transportation status of stored goods. Summary of the invention

[0004] The present invention provides an AI-based intelligent identification method for the warehousing industry to solve existing problems.

[0005] The AI-based intelligent identification method for the warehousing industry of the present invention adopts the following technical solutions:

[0006] An embodiment of the present invention provides an AI-based intelligent identification method for the warehousing industry, the method comprising the following steps:

[0007] Obtain the parameter vector formed by several types of data at each moment in the process of transporting goods by the storage and transportation vehicle, and label the parameter vector at each moment according to the transportation status of the goods during the transportation process;

[0008] Each parameter vector is reduced in dimension to obtain a reduced dimension vector of the parameter vector. A reduced dimension vector is taken as a sample. All samples are hierarchically clustered to obtain a hierarchical clustering tree. The best layer in the hierarchical clustering tree is selected according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree.

[0009] According to the difference between the dimension reduction vectors of all clusters in the best layer and the samples in each cluster and the dimension reduction vectors of other samples with the same label, the centrality of each sample is obtained, and the weight of the weak classifier in the Adaboost algorithm is adjusted by the centrality to obtain the adjusted Adaboost algorithm;

[0010] Combined with the adjusted Adaboost algorithm, several neural networks are combined to obtain an integrated neural network, which is used to identify the status of goods transported by storage and transportation vehicles.

[0011] Furthermore, the hierarchical clustering tree is obtained by performing hierarchical clustering on all samples, and the best layer in the hierarchical clustering tree is screened out according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree, including the specific method of:

[0012] The dimensionality reduction vectors at all times are clustered using a hierarchical clustering algorithm, and the clustering results obtained by the hierarchical clustering are recorded as a hierarchical clustering tree;

[0013] The reduced-dimensional vector corresponding to the parameter vector with label 0 is taken as a positive sample, and the reduced-dimensional vector corresponding to the parameter vector with label 1 is taken as a negative sample; according to the number of samples with different labels in each layer of the hierarchical clustering tree, the clusters in each layer are divided into positive clusters and negative clusters;

[0014] Record any layer in the hierarchical clustering tree as the target layer, and obtain the weighted variance v1 of the positive samples in the target layer and the weighted variance v2 of the negative samples in the target layer;

[0015] Get the variance v of all dimensionality reduction vectors in the target layer, record v-v1-v2 as the positive and negative separability of the target layer, and record the layer with the largest positive and negative separability in the hierarchical clustering tree as the best layer.

[0016] Furthermore, the method of dividing the clusters in each layer into positive clusters and negative clusters according to the number of samples with different labels in each layer of the hierarchical clustering tree includes the following specific methods:

[0017] Any cluster in any layer of the hierarchical clustering tree is recorded as the target cluster, and the number of positive samples and the number of negative samples in the target cluster are obtained. When the number of positive samples in the target cluster is greater than the number of negative samples, the target cluster is recorded as a positive cluster. When the number of negative samples in the target cluster is greater than the number of positive samples, the target cluster is recorded as a negative cluster.

[0018] Furthermore, the method of obtaining the weighted variance v1 of the positive samples in the target layer and the weighted variance v2 of the negative samples in the target layer includes:

[0019] The ratio of the number of positive samples in the positive cluster to the number of all samples in the positive cluster is recorded as the sample purity of the positive cluster. The variance of all dimensionality reduction vectors in each positive cluster is weighted by the sample purity of each positive cluster to obtain the weighted variance v1 of the positive samples in the target layer. Similarly, the weighted variance v2 of the negative samples in the target layer is obtained.

[0020] Furthermore, the centrality of each sample is obtained according to the difference between the dimensionality reduction vectors of all clusters in the optimal layer and the samples in each cluster and the dimensionality reduction vectors of other samples with the same label, including the specific method of:

[0021] The partition centrality of the sample is obtained based on the difference between the dimensionality reduction vectors of the samples in all clusters in the best layer and the dimensionality reduction vectors of other samples with the same label;

[0022] According to the difference between the dimension reduction vector of each sample in each cluster in the best layer and the dimension reduction vector of other samples with the same label, the cluster centrality of the sample is obtained;

[0023] The centrality of the samples is obtained based on the partition centrality and the intra-cluster centrality.

[0024] Furthermore, the method of obtaining the partition centrality of the samples according to the difference between the dimensionality reduction vectors of the samples in all clusters in the optimal layer and the dimensionality reduction vectors of other samples with the same label includes the following specific methods:

[0025] The set consisting of all positive samples in the best layer is recorded as the positive set, and the set consisting of all negative samples in the best layer is recorded as the negative set. The combination of the dimensionality reduction vector corresponding to any positive sample R in the positive set and the dimensionality reduction vector corresponding to each other positive sample is obtained, and recorded as the first positive sample group of the positive sample R. The ratio of the minimum value to the maximum value in the first positive sample group is recorded as the first relative ratio of the positive sample R. The cumulative sum of all first relative ratios of the positive sample R is recorded as the overall relative ratio of the positive sample R. The ratio of the overall relative ratio of any positive sample R in the positive set to the maximum overall relative ratio of all positive samples is obtained, and recorded as the partition centrality of the positive sample R. By analogy, the partition centrality of each negative sample is obtained.

[0026] Furthermore, the method of obtaining the intra-cluster centrality of the sample according to the difference between the dimension reduction vector of the sample in each cluster in the optimal layer and the dimension reduction vector of other samples with the same label includes the following specific methods:

[0027] The combination of the dimensionality reduction vector corresponding to any positive sample R in the best layer and the dimensionality reduction vector corresponding to each other sample in the cluster to which it belongs is recorded as the second positive sample group of the positive sample R, the ratio of the minimum value to the maximum value in the second positive sample group of the positive sample R is recorded as the second relative ratio of the positive sample R, the cumulative value of all second relative ratios of the positive sample R in the cluster to which it belongs is recorded as the intra-cluster similarity of the positive sample R, the ratio of the intra-cluster similarity of the positive sample R to the maximum intra-cluster similarity of all positive samples in the cluster to which the positive sample R belongs is recorded as the intra-cluster centrality of the positive sample R, and so on, to obtain the intra-cluster centrality of each negative sample.

[0028] Furthermore, the adjusted Adaboost algorithm is combined to combine several neural networks to obtain an integrated neural network, and the integrated neural network is used to identify the state of the goods transported by the storage and transportation vehicle, including the specific method of:

[0029] Construct several BP neural networks, use the set formed by several labeled dimension reduction vectors as the data set to train all BP neural networks, obtain the trained BP neural networks, and combine all the trained BP neural networks into an integrated neural network by combining the adjusted Adaboost algorithm;

[0030] An integrated neural network is used to identify the status of a storage transport vehicle when transporting goods. If the recognition result is 0, the status is normal; if the recognition result is 1, the status is abnormal.

[0031] Furthermore, the method of labeling the parameter vector at each moment according to the transportation status of the goods during transportation includes the following specific methods:

[0032] A label is given to the parameter vector at each moment. When the cargo transported by the transport vehicle is in normal condition at any moment, 0 is used as the label of the parameter vector at the corresponding moment. When the cargo transported by the transport vehicle is damaged at any moment, 1 is used as the label of the parameter vector at the corresponding moment.

[0033] Furthermore, the method of reducing the dimension of each parameter vector to obtain a reduced-dimensional vector of the parameter vector includes the following specific methods:

[0034] The PCA algorithm is used to perform dimension reduction processing on the parameter vector at each moment, and the result after the dimension reduction of the parameter vector is recorded as the dimension reduction vector of the parameter vector.

[0035] The beneficial effects of the technical solution of the present invention are: by reducing the dimension of the parameter vector through the PCA algorithm, the main features in the parameter vector can be extracted, and redundant information can be removed, thereby simplifying the complexity of the parameter vector and improving its effectiveness and interpretability; clustering the parameter vector after dimension reduction using the hierarchical clustering algorithm, screening out the best layer, and effectively ensuring the accuracy of subsequent classification of positive and negative samples, and by calculating the centrality of the sample, the weight of the weak classifier in the Adaboost algorithm can be adjusted, so that when classifying samples with larger centrality, the weight reduction degree is moderate, and the stability and accuracy of the classification process are improved. Combined with the adjusted Adaboost algorithm and multiple neural networks, an integrated neural network is obtained, and the advantages of integrated learning can be used to improve the recognition accuracy of the state of goods transported by storage and transportation vehicles, and the efficiency and accuracy of processing real-time data are guaranteed by the steps of dimension reduction, clustering, and weight adjustment, and by improving the Adaboost algorithm, the recognition accuracy of the state of goods can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 This is a flowchart of the steps of the AI-based intelligent identification method for the warehousing industry of the present invention;

[0038] Figure 2 A flow chart of intelligent identification for the warehousing industry provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of the AI-based intelligent identification method for the warehousing industry proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0040] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0041] The specific scheme of the AI-based intelligent identification method for the warehousing industry provided by the present invention is described in detail below with reference to the accompanying drawings.

[0042] See also Figure 1 , which shows a flowchart of a method for intelligent identification of the warehousing industry based on AI according to an embodiment of the present invention, the method comprising the following steps:

[0043] Step S001: Obtain a parameter vector formed by several types of data at each moment in the process of transporting goods by a storage transport vehicle, and mark a label for the parameter vector at each moment according to the transportation status of the goods during the transportation process.

[0044] It should be noted that in industry warehousing, in order to analyze and identify the status of goods during the warehousing process, thereby ensuring warehousing safety; vibration data is used to reflect the operating conditions of transport vehicles in the warehouse, and abnormal vibration data indicates that there is a fault in the transportation process and needs to be repaired or replaced; secondly, since some goods are very sensitive to temperature, abnormal temperatures can be identified in time by monitoring temperature data to ensure that the temperature conditions in the warehouse meet the storage requirements of the goods; in addition, some goods are very sensitive to humidity, so a high humidity environment may cause such goods to become damp or even moldy. By monitoring humidity data, measures can be taken to adjust the humidity in the warehouse to protect the goods from damage. By comprehensively analyzing vibration, temperature and humidity data, a more comprehensive understanding of the environmental conditions inside the warehouse can be achieved, and potential abnormalities can be identified and handled in time, thereby improving the efficiency of warehousing management and the safety of goods. In addition, this embodiment uses advanced data analysis technology and artificial intelligence algorithms to process and analyze these data, which can further improve the level of intelligence in warehousing management.

[0045] Specifically, first, a parameter vector formed by several types of data at each moment in the process of transporting goods by the storage and transportation vehicle is obtained.

[0046] As an embodiment, the method for obtaining the parameter vector is: obtaining the vibration data of the storage and transport vehicle during the cargo transportation process and the temperature data and humidity data of the environment in which the storage and transport vehicle is located, wherein each data point in the vibration data, temperature data and humidity data corresponds to a time point, and the numerical value of each data point represents the vibration amplitude, temperature value and humidity value respectively, and the vibration amplitude, temperature value and humidity value corresponding to the vibration data, temperature data and humidity data at the same time are combined into a three-dimensional vector, which is recorded as the parameter vector at the corresponding time.

[0047] Then, a label is added to the parameter vector at each moment. When the cargo transported by the transport vehicle is in normal condition at any moment, 0 is used as the label of the parameter vector at the corresponding moment. When the cargo transported by the transport vehicle is damaged at any moment, 1 is used as the label of the parameter vector at the corresponding moment.

[0048] It should be noted that the vibration data, temperature data and humidity data are acquired using a vibration sensor, a temperature sensor and a humidity sensor respectively.

[0049] So far, the parameter vector formed by the vibration data, temperature data and humidity data and the label of each parameter vector are obtained through the above method.

[0050] Step S002: Reduce the dimension of each parameter vector to obtain a reduced dimension vector of the parameter vector, take a reduced dimension vector as a sample, perform hierarchical clustering on all samples to obtain a hierarchical clustering tree, and screen out the best layer in the hierarchical clustering tree based on the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree.

[0051] Specifically, in step 2.1, the PCA algorithm is used to perform dimension reduction processing on the parameter vector at each moment, and the result after the dimension reduction of the parameter vector is recorded as the dimension reduction vector of the parameter vector.

[0052] It should be noted that the reduced dimensionality vector obtained after dimensionality reduction by the PCA algorithm is a scalar, wherein the PCA (Principal Component Analysis, PCA) algorithm is an existing linear dimensionality reduction algorithm, so it is not described in detail in this embodiment; and this embodiment reduces the sample vector into a scalar by using the PCA algorithm, with the aim of removing redundant information in the parameter vector, so as to simplify the complexity of the parameter vector and improve the validity and interpretability of the parameter vector.

[0053] Step 2.2, reduce the dimension of each parameter vector to obtain a reduced dimension vector of the parameter vector, take a reduced dimension vector as a sample, perform hierarchical clustering on all samples to obtain a hierarchical clustering tree, and screen out the best layer in the hierarchical clustering tree according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree, specifically including:

[0054] The reduced-dimensional vectors at all times are clustered using a hierarchical clustering algorithm, and the clustering results obtained by the hierarchical clustering are recorded as a hierarchical clustering tree; the best layer among all layers of the hierarchical clustering tree is obtained, and the specific method for obtaining the best layer is:

[0055] First, the dimension reduction vector corresponding to the parameter vector with label 0 is taken as the positive sample, and the dimension reduction vector corresponding to the parameter vector with label 1 is taken as the negative sample. The positive samples and negative samples are collectively referred to as samples. Any cluster in any layer of the hierarchical clustering tree is recorded as the target cluster, and the number of positive samples and the number of negative samples in the target cluster are obtained. When the number of positive samples in the target cluster is greater than the number of negative samples, the target cluster is recorded as a positive cluster. When the number of negative samples in the target cluster is greater than the number of positive samples, the target cluster is recorded as a negative cluster.

[0056] It should be noted that when the number of positive samples and negative samples in a cluster is the same, the cluster is not considered and does not participate in the subsequent analysis and calculation process.

[0057] It should be noted that, in this embodiment, the parameter vector corresponding to each dimensionality reduction vector corresponds to a label, so the positive samples and negative samples are collectively referred to as samples, which also means that a dimensionality reduction vector in this embodiment is a sample.

[0058] Then, the proportion of the number of positive samples in the positive cluster to the number of all samples in the positive cluster is recorded as the sample purity of the positive cluster, and any layer in the hierarchical clustering tree is recorded as the target layer, and the weighted variance of the positive samples in the target layer is obtained. As an embodiment, the specific calculation method is:

[0059]

[0060] Among them, v1 represents the weighted variance of the positive cluster; s i represents the sample purity of the i-th positive cluster in the target layer; g i Represents the variance of all reduced dimensionality vectors in the i-th positive cluster in the target layer.

[0061] And so on, the weighted variance v2 of the negative samples in the target layer is obtained.

[0062] Finally, the variance v of all dimensionality reduction vectors in the target layer is obtained, v-v1-v2 is recorded as the positive and negative separability of the target layer, and the layer with the largest positive and negative separability in the hierarchical clustering tree is recorded as the best layer.

[0063] It should be noted that the variance of all dimensionality reduction vectors within a category represents the distribution of samples within the category.

[0064] It should be noted that the positive-negative separability represents the inter-class variance between the positive samples and the negative samples in the corresponding layer, reflecting the difference between the corresponding classification results of the positive samples and the negative samples after the corresponding layer in the hierarchical clustering tree classifies the positive samples and the negative samples. Therefore, this embodiment describes the classification effect of the corresponding layer in the hierarchical clustering tree on the positive samples and the negative samples through the positive-negative separability. The greater the positive-negative separability, the better the classification effect of the positive samples and the negative samples in the corresponding layer in the hierarchical clustering tree.

[0065] It should be noted that the hierarchical clustering algorithm obtains a tree-like hierarchical structure by clustering data step by step. Since the hierarchical clustering algorithm is an existing data clustering algorithm, this embodiment does not describe the hierarchical clustering algorithm in detail.

[0066] It should be noted that, when a neural network is used to identify anomalies in the collected vibration data, temperature data and humidity data, the accuracy of the identification result is easily affected by the quality of the training results of the neural network. Therefore, a neural network is used for identification, and the existing Adaboost algorithm cannot effectively process the real-time collected data in real time when classifying the data. Therefore, the purpose of this embodiment is to improve the real-time performance of the Adaboost algorithm, reduce the time required for the training process of the Adaboost algorithm, and accelerate the model convergence speed of the Adaboost algorithm, so as to further improve the timeliness of data anomaly identification under the premise of ensuring the accuracy of the data processing results of the Adaboost algorithm.

[0067] So far, the optimal layer of the hierarchical clustering tree is obtained through the above method.

[0068] Step S003: According to the difference between the dimension reduction vectors of all clusters in the best layer and the samples in each cluster and the dimension reduction vectors of other samples with the same label, the centrality of each sample is obtained, and the weights of the weak classifiers in the Adaboost algorithm are adjusted using the centrality to obtain the adjusted Adaboost algorithm.

[0069] Specifically, in step 3.1, the centrality of each sample is obtained based on the difference between the dimensionality reduction vectors of all clusters in the best layer and the samples in each cluster and the dimensionality reduction vectors of other samples with the same label. The specific acquisition method is:

[0070] First, according to the difference between the dimensionality reduction vectors of samples in all clusters in the optimal layer and the dimensionality reduction vectors of other samples with the same label, the partition centrality of the samples is obtained, specifically including: recording the set consisting of all positive samples in the optimal layer as a positive set, recording the set consisting of all negative samples in the optimal layer as a negative set, obtaining the combination of the dimensionality reduction vector corresponding to any positive sample R in the positive set and the dimensionality reduction vector corresponding to each other positive sample, respectively, and recording them as the first positive sample group of the positive sample R, recording the ratio of the minimum value to the maximum value in the first positive sample group as the first relative ratio of the positive sample R, recording the cumulative sum of all first relative ratios of the positive sample R as the overall relative ratio of the positive sample R, obtaining the ratio of the overall relative ratio of any positive sample R in the positive set to the maximum overall relative ratio of all positive samples, and recording it as the partition centrality of the positive sample R, and so on, to obtain the partition centrality of each negative sample.

[0071] Then, according to the difference between the dimensionality reduction vector of the sample in each cluster in the optimal layer and the dimensionality reduction vector of other samples with the same label, the intra-cluster centrality of the sample is obtained, specifically including: recording the combination formed by the dimensionality reduction vector corresponding to any positive sample R in the optimal layer and the dimensionality reduction vector corresponding to each other sample in the cluster to which it belongs as the second positive sample group of the positive sample R, recording the ratio of the minimum value to the maximum value in the second positive sample group of the positive sample R as the second relative ratio of the positive sample R, recording the cumulative value of all second relative ratios of the positive sample R in the cluster to which it belongs as the intra-cluster similarity of the positive sample R, and recording the ratio of the intra-cluster similarity of the positive sample R to the maximum intra-cluster similarity of all positive samples in the cluster to which the positive sample R belongs as the intra-cluster centrality of the positive sample R.

[0072] Finally, the centrality of the sample is obtained according to the partition centrality and the intra-cluster centrality, specifically including: taking the sum of the partition centrality and the cluster centrality of any positive sample in the best layer as the centrality of the positive sample, obtaining the centrality of each positive sample in the best layer, and so on, obtaining the centrality of each negative sample in the best layer.

[0073] It should be noted that, in this embodiment, the first relative comparison of the dimensionality reduction vectors of the sample and other samples in each layer is obtained to reflect the difference between different samples with the same label. The overall relative comparison of the sample reflects the relative difference between the sample and all other samples with the same label. The partition centrality of the sample further obtained reflects the relative position of the sample in the entire layer to which it belongs. A sample with a higher partition centrality means that the position of the sample is closer to the center among all samples with the same label, that is, the closer the dimensionality reduction vector of the sample is to the average level of the corresponding dimensionality reduction vectors of all samples, the more the sample can be used as the clustering center during hierarchical clustering, and can effectively separate positive and negative samples. Therefore, the greater the centrality of the sample, the more accurate the classification effect of the sample can be provided when the neural network is used as the base classifier of the Adaboost algorithm in the future, thereby improving the recognition accuracy of the cargo transportation status. By calculating the relative ratio between the sample and other samples in the cluster to which it belongs, as well as the relative The cumulative sum of the ratios can be used to obtain the intra-cluster similarity of the samples. Samples with higher intra-cluster similarity mean that the sample is closer to the center within the cluster to which it belongs. Similarly, the greater the intra-cluster centrality of the sample, the more it can cluster other samples with the same label but similar dimensionality reduction vectors into one category, which improves the subsequent classification using multiple neural networks. Samples with the same label and similar dimensionality reduction vectors are classified into one category, which can provide more accurate classification effects when the neural network is used as the base classifier of the Adaboost algorithm in the future, and improve the recognition accuracy of the cargo transportation status; and the centrality of the sample is a comprehensive reflection of the partition centrality and the intra-cluster centrality, that is, a comprehensive evaluation of the sample's position in the entire clustering structure and its position within the cluster to which it belongs. Therefore, by comprehensively considering the partition centrality and the intra-cluster centrality to adjust the parameters of the Adaboost algorithm in the future, the recognition accuracy of the storage cargo transportation status using multiple neural networks can be effectively improved.

[0074] Step 3.2, using the centrality to adjust the weights of the weak classifiers in the Adaboost algorithm to obtain the adjusted Adaboost algorithm, specifically including: using the centrality of the dimensionality reduction vector corresponding to each sample to multiply and adjust the weights of the weak classifiers in the Adaboost algorithm to obtain the adjusted weak classifier weights of the Adaboost algorithm.

[0075] It should be noted that in order to improve the classification effect of all samples after integrating multiple neural networks through the Adaboost algorithm and improve the recognition accuracy of the transportation status of warehouse goods, this embodiment uses the centrality of the sample as a weight to perform weighted adjustment on the weak classifier weight of the Adaboost algorithm to avoid the problem that when the neural network classifies samples with larger centrality, the weight of the corresponding weak classifier is reduced too much, resulting in an unstable classification process and unsatisfactory classification effect, so as to ensure that after all neural networks are combined as base classifiers, the classification ability of the next base classifier and the previous base classifier change stably during the sample classification process.

[0076] At this point, the adjusted Adaboost algorithm is obtained through the above method.

[0077] Step S004: Combining several neural networks with the adjusted Adaboost algorithm to obtain an integrated neural network, and using the integrated neural network to identify the state of the goods transported by the storage and transport vehicle.

[0078] Specifically, in step 4.1, several BP neural networks are constructed, and a set formed by several labeled dimension reduction vectors is used as a data set to train all BP neural networks to obtain trained BP neural networks, and all trained BP neural networks are combined into an integrated neural network in combination with the adjusted Adaboost algorithm.

[0079] Step 4.2, using the integrated neural network to identify the state of the storage and transport vehicle when transporting goods, if the recognition result is 0, the state is normal, if the recognition result is 1, the state is abnormal, the storage and transport vehicle stops running and issues an alarm.

[0080] It should be noted that the specific process of using the set formed by several labeled dimensionality reduction vectors as a data set to train all BP neural networks, obtain the trained BP neural networks, and combine all the trained BP neural networks into an integrated neural network in combination with the adjusted Adaboost algorithm is as follows: the data set is divided into a training set and a validation set to train the BP neural network, and during the training process, the adjusted weak classifier weight corresponding to each sample can be obtained by the adjusted Adaboost algorithm, the output results of the BP neural network are weightedly combined by the adjusted weak classifier weights of the Adaboost algorithm, all BP neural networks are integrated to obtain an integrated neural network.

[0081] like Figure 2 Shown is a flow chart of intelligent identification for the warehousing industry provided by this embodiment.

[0082] It should be noted that both the BP (Back Propagation) neural network and the Adaboost (Adaptive Boosting) algorithm are existing algorithms, and thus they are not described in detail in this embodiment.

[0083] This embodiment implements the AI-based intelligent identification method for the warehousing industry in steps S001 to S004 to monitor the status of goods on the transport vehicle during transportation, achieve more efficient, timely and accurate identification of abnormal cargo status, and improve the reliability and timeliness of logistics transportation.

[0084] At this point, this embodiment is completed.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The AI-based intelligent identification method for the warehousing industry is characterized by: The method comprises the following steps: Obtain the parameter vector formed by several types of data at each moment in the process of transporting goods by the storage and transportation vehicle, and label the parameter vector at each moment according to the transportation status of the goods during the transportation process; Each parameter vector is reduced in dimension to obtain a reduced-dimensional vector of the parameter vector, where the reduced-dimensional vector is a scalar. A reduced-dimensional vector is taken as a sample, and all samples are hierarchically clustered to obtain a hierarchical clustering tree. The best layer in the hierarchical clustering tree is screened out according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree; According to the difference between the dimension reduction vectors of all clusters in the best layer and the samples in each cluster and the dimension reduction vectors of other samples with the same label, the centrality of each sample is obtained, and the weight of the weak classifier in the Adaboost algorithm is adjusted by the centrality to obtain the adjusted Adaboost algorithm; Combine several neural networks with the adjusted Adaboost algorithm to obtain an integrated neural network, and use the integrated neural network to identify the state of goods transported by storage and transportation vehicles; The method of obtaining the centrality of each sample according to the difference between the dimensionality reduction vectors of all clusters in the optimal layer and the samples in each cluster and the dimensionality reduction vectors of other samples with the same label includes: The partition centrality of the sample is obtained based on the difference between the dimensionality reduction vectors of the samples in all clusters in the best layer and the dimensionality reduction vectors of other samples with the same label; According to the difference between the dimension reduction vector of each sample in each cluster in the best layer and the dimension reduction vector of other samples with the same label, the cluster centrality of the sample is obtained; The centrality of the samples is obtained based on the partition centrality and the intra-cluster centrality; The method of obtaining the partition centrality of the samples according to the difference between the dimensionality reduction vectors of the samples in all clusters in the optimal layer and the dimensionality reduction vectors of other samples with the same label includes the following specific methods: The set consisting of all positive samples in the best layer is recorded as the positive set, and the set consisting of all negative samples in the best layer is recorded as the negative set. The combination of the dimensionality reduction vector corresponding to any positive sample R in the positive set and the dimensionality reduction vector corresponding to each other positive sample is obtained, which is recorded as the first positive sample group of the positive sample R. The ratio of the minimum value to the maximum value in the first positive sample group is recorded as the first relative ratio of the positive sample R. The cumulative sum of all the first relative ratios of the positive sample R is recorded as the overall relative ratio of the positive sample R. The ratio of the overall relative ratio of any positive sample R in the positive set to the maximum overall relative ratio of all positive samples is obtained, which is recorded as the partition centrality of the positive sample R. By analogy, the partition centrality of each negative sample is obtained. The method of obtaining the intra-cluster centrality of the sample according to the difference between the dimensionality reduction vector of the sample in each cluster in the optimal layer and the dimensionality reduction vector of other samples with the same label includes the following specific methods: The combination of the dimensionality reduction vector corresponding to any positive sample R in the best layer and the dimensionality reduction vector corresponding to each other sample in the cluster to which it belongs is recorded as the second positive sample group of the positive sample R, the ratio of the minimum value to the maximum value in the second positive sample group of the positive sample R is recorded as the second relative ratio of the positive sample R, the cumulative value of all second relative ratios of the positive sample R in the cluster to which it belongs is recorded as the intra-cluster similarity of the positive sample R, the ratio of the intra-cluster similarity of the positive sample R to the maximum intra-cluster similarity of all positive samples in the cluster to which the positive sample R belongs is recorded as the intra-cluster centrality of the positive sample R, and so on, to obtain the intra-cluster centrality of each negative sample.

2. According to the AI-based intelligent identification method for the warehousing industry according to claim 1, it is characterized in that: The hierarchical clustering tree is obtained by performing hierarchical clustering on all samples, and the best layer in the hierarchical clustering tree is screened out according to the number and distribution of samples with different labels in different clusters in each layer of the hierarchical clustering tree. The specific method includes: The dimensionality reduction vectors at all times are clustered using a hierarchical clustering algorithm, and the clustering results obtained by the hierarchical clustering are recorded as a hierarchical clustering tree; The reduced-dimensional vector corresponding to the parameter vector with label 0 is taken as a positive sample, and the reduced-dimensional vector corresponding to the parameter vector with label 1 is taken as a negative sample; according to the number of samples with different labels in each layer of the hierarchical clustering tree, the clusters in each layer are divided into positive clusters and negative clusters; Record any layer in the hierarchical clustering tree as the target layer, and obtain the weighted variance v1 of the positive samples in the target layer and the weighted variance v2 of the negative samples in the target layer; Get the variance v of all dimensionality reduction vectors in the target layer, record v-v1-v2 as the positive and negative separability of the target layer, and record the layer with the largest positive and negative separability in the hierarchical clustering tree as the best layer.

3. According to claim 2, the AI-based intelligent identification method for the warehousing industry is characterized in that: The method of dividing the clusters in each layer into positive clusters and negative clusters according to the number of samples with different labels in each layer of the hierarchical clustering tree includes: Any cluster in any layer of the hierarchical clustering tree is recorded as the target cluster, and the number of positive samples and the number of negative samples in the target cluster are obtained. When the number of positive samples in the target cluster is greater than the number of negative samples, the target cluster is recorded as a positive cluster. When the number of negative samples in the target cluster is greater than the number of positive samples, the target cluster is recorded as a negative cluster.

4. According to claim 2, the AI-based intelligent identification method for the warehousing industry is characterized in that: The specific method of obtaining the weighted variance v1 of the positive samples in the target layer and the weighted variance v2 of the negative samples in the target layer is as follows: The ratio of the number of positive samples in the positive cluster to the number of all samples in the positive cluster is recorded as the sample purity of the positive cluster. The variance of all dimensionality reduction vectors in each positive cluster is weighted by the sample purity of each positive cluster to obtain the weighted variance v1 of the positive samples in the target layer. Similarly, the weighted variance v2 of the negative samples in the target layer is obtained.

5. According to claim 1, the AI-based intelligent identification method for the warehousing industry is characterized in that: The method of combining several neural networks with the adjusted Adaboost algorithm to obtain an integrated neural network is to use the integrated neural network to identify the state of the goods transported by the storage and transportation vehicle, including the following specific methods: Construct several BP neural networks, use the set formed by several labeled dimension reduction vectors as the data set to train all BP neural networks, obtain the trained BP neural networks, and combine all the trained BP neural networks into an integrated neural network by combining the adjusted Adaboost algorithm; An integrated neural network is used to identify the status of a storage transport vehicle when transporting goods. If the recognition result is 0, the status is normal; if the recognition result is 1, the status is abnormal.

6. The AI-based intelligent identification method for warehousing industry according to claim 1 is characterized in that: The specific method of labeling the parameter vector at each moment according to the transportation status of the goods during transportation is as follows: A label is given to the parameter vector at each moment. When the cargo transported by the transport vehicle is in normal condition at any moment, 0 is used as the label of the parameter vector at the corresponding moment. When the cargo transported by the transport vehicle is damaged at any moment, 1 is used as the label of the parameter vector at the corresponding moment.

7. The AI-based intelligent identification method for warehousing industry according to claim 1 is characterized in that: The specific method of reducing the dimension of each parameter vector to obtain a reduced dimension vector of the parameter vector includes: The PCA algorithm is used to perform dimension reduction processing on the parameter vector at each moment, and the result after the dimension reduction of the parameter vector is recorded as the dimension reduction vector of the parameter vector.

Images