Machine Learning-Based Order Demand Prediction Method, System, and Electronic Device

Abstract

This application relates to the field of demand forecasting. Specifically, it discloses a method, system, and electronic device for order demand forecasting based on machine learning. First, it collects a plurality of customer portraits containing multiple portrait tags and their weights obtained from a database, and historical order data obtained from the database, where the historical order data includes the order quantity, date, time, and product type. Then, using deep learning technology, it performs feature extraction and correlation analysis on the two. Finally, it obtains a classification result through a classifier to judge the purchase tendency of customers, further predict the order demand for the next month, and then optimize the enterprise production plan and inventory management plan, reduce the situation of overstock or out-of-stock, thereby improving the operation efficiency and business growth.

Description

Technical Field

[0001] The present application relates to the field of demand forecasting, and more specifically, to a method, system, and electronic device for order demand forecasting based on machine learning. Background Art

[0002] Order demand forecasting plays a crucial role in business operations. By accurately predicting future order demands, enterprises can make more informed decisions, optimize supply chain management, improve customer satisfaction, and minimize inventory costs and production costs to the greatest extent.

[0003] In traditional business models, enterprises usually make order and production plans based on past sales data and experience. However, this static method cannot cope with the dynamic changes and uncertainties in the market. Making decisions solely based on past data and experience, a single artificial factor judgment, may lead to problems such as imbalance between supply and demand, inventory backlog, or out-of-stock situations.

[0004] Therefore, there is a need for a method, system, and electronic device for order demand forecasting based on machine learning. Summary of the Invention

[0005] To solve the above technical problems, the present application is proposed. Embodiments of the present application provide a method, system, and electronic device for order demand forecasting based on machine learning. First, it collects multiple customer portraits containing multiple portrait tags and their weights obtained from a database and historical order data obtained from the database. The historical order data includes order quantity, date, time, and product type. Then, using deep learning technology, feature extraction and correlation analysis are performed on both, and finally, a classification result is obtained through a classifier to judge the purchase tendency of customers, further predict the order demand of the next month, and then optimize the enterprise production plan and inventory management plan, reduce the situation of overstock or out-of-stock, thereby improving operational efficiency and business growth.

[0006] According to one aspect of the present application, there is provided a method for order demand forecasting based on machine learning, which includes:

[0007] Collect multiple customer portraits containing multiple portrait tags and their weights obtained from a database and historical order data obtained from the database, where the historical order data includes order quantity, date, time, and product type;

[0008] Extract customer portrait correlation feature vectors and historical order correlation feature vectors from the multiple customer portraits containing multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database;

[0009] Based on the customer portrait associated feature vector and the historical order associated feature vector, determine the purchase tendency of the customer.

[0010] According to another aspect of the present application, there is provided an order demand prediction system based on machine learning, which includes:

[0011] An order data collection module, configured to collect a plurality of customer portraits including a plurality of portrait tags and their weights obtained from a database and historical order data obtained from the database, wherein the historical order data includes the order quantity, date, time, and product type;

[0012] An order data feature extraction module, configured to extract a customer portrait associated feature vector and a historical order associated feature vector from the plurality of customer portraits including a plurality of portrait tags and their weights obtained from the database and the historical order data obtained from the database;

[0013] A purchase tendency judgment generation module, configured to determine the purchase tendency of the customer based on the customer portrait associated feature vector and the historical order associated feature vector.

[0014] According to still another aspect of the present application, there is provided an electronic device, including: a processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are run by the processor, the processor is caused to execute the above-mentioned order demand prediction method based on machine learning.

[0015] Compared with the prior art, an order demand prediction method, system, and electronic device provided by the present application first collect a plurality of customer portraits including a plurality of portrait tags and their weights obtained from a database and historical order data obtained from the database, wherein the historical order data includes the order quantity, date, time, and product type, then use deep learning technology to perform feature extraction and correlation analysis on the two, and finally obtain a classification result through a classifier to determine the purchase tendency of the customer, so as to further predict the order demand of the next month, and further optimize the enterprise production plan and inventory management plan, reduce the situation of overstock or out-of-stock, thereby improving the operation efficiency and business growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1Flow chart of a machine learning-based order demand prediction method according to an embodiment of the present application.

[0018] Figure 2 Flow chart for semantic encoding of multiple customer portraits containing multiple portrait tags and their weights obtained from a database and historical order data obtained from the database in a machine learning-based order demand prediction method according to an embodiment of the present application.

[0019] Figure 3 Flow chart for feature extraction of the multiple portrait tag feature vectors in a machine learning-based order demand prediction method according to an embodiment of the present application.

[0020] Figure 4 Block diagram of a machine learning-based order demand prediction system according to an embodiment of the present application.

[0021] Figure 5 Block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0022] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0023] Figure 1 Flow chart of a machine learning-based order demand prediction method according to an embodiment of the present application. As Figure 1 shown, a machine learning-based order demand prediction method according to an embodiment of the present application includes: S110, collecting multiple customer portraits containing multiple portrait tags and their weights obtained from a database and historical order data obtained from the database, where the historical order data includes order quantity, date, time, and product type; S120, extracting a customer portrait association feature vector and a historical order association feature vector from the multiple customer portraits containing multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database; S130, judging the purchase tendency of the customer based on the customer portrait association feature vector and the historical order association feature vector.

[0024] In the above machine learning-based order demand prediction method, in step S110, multiple customer portraits including multiple portrait tags and their weights obtained from the database and historical order data obtained from the database are collected. Among them, the historical order data includes the order quantity, date, time, and product type. It should be understood that the customer portraits and historical order data obtained from the database provide a detailed description of customers and their purchase histories, and rich feature information can be collected for analyzing customers' purchase behaviors and trends. Specifically, the customer portrait contains multiple portrait tags and their weights, and these tags can be keywords or attributes describing customer characteristics, such as age, gender, geographical location, hobbies, etc. These tags and their weights can provide personalized characteristics of customers to help understand customers' purchase preferences and behaviors. The historical order data contains information such as order quantity, date, time, and product type. This information can be used to analyze customers' purchase frequencies, purchase time patterns, and purchase product type preferences. By analyzing the historical order data, potential purchase patterns and trends can be discovered, thus better predicting customers' future purchase tendencies.

[0025] In the above machine learning-based order demand prediction method, in step S120, customer portrait-related feature vectors and historical order-related feature vectors are extracted from the multiple customer portraits including multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database. It should be understood that the customer portraits and historical order data obtained from the database provide rich information, but there may be some problems in directly using the original data for prediction. Therefore, it is necessary to extract customer portrait-related feature vectors and historical order-related feature vectors from these data to convert the original data into a numerical representation that can be processed by machine learning algorithms, and then better represent the relevant information of customer portraits and historical orders, so as to more accurately predict customers' purchase tendencies. Specifically, for customer portraits, feature vectors can be constructed according to tags and weights. For example, each tag can be used as the dimension of a feature, and the weight as the value of that dimension. In this way, each customer portrait can be represented as a vector, where each dimension corresponds to a tag and its weight. For historical order data, relevant features can be extracted according to information such as order quantity, date, time, and product type. For example, the average order quantity of each customer, the time interval since the last purchase, and the preference for the product type purchased can be calculated. These features can be used to represent customers' purchase behaviors and trends. In this way, these feature vectors can be used to train and construct an order demand prediction model to achieve accurate prediction of customers' purchase tendencies.

[0026] In a specific embodiment of the present application, the step S120 includes: performing semantic encoding on the multiple customer portraits containing multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database to obtain multiple portrait tag feature vectors and multiple historical order feature vectors; extracting features from the multiple portrait tag feature vectors to obtain the customer portrait association feature vector; extracting features from the multiple historical order feature vectors to obtain the historical order association feature vector. It should be understood that semantic encoding is a technology for converting text or data into a continuous vector representation. Through semantic encoding, the semantic similarity and relevance between data can be captured, so as to better represent the characteristics of the data. For customer portraits, semantic encoding can convert each portrait tag and its weight into a feature vector. This feature vector can be generated by word embedding techniques (such as Word2Vec or GloVe) or deep learning models (such as BERT or GPT). In this way, each portrait tag will be represented as a vector, which can capture the semantic meaning and weight information of the tag. For historical order data, semantic encoding can convert information such as order quantity, date, time, and product type into a feature vector. Similarly, word embedding techniques or deep learning models can be used to generate this feature vector. Through semantic encoding, these order features can be converted into a continuous vector representation, so as to better represent the semantic association and trend between orders. By performing semantic encoding on customer portraits and historical order data, multiple portrait tag feature vectors and multiple historical order feature vectors can be obtained to convert the original data into a feature vector representation with more semantic information. These feature vectors have more semantic information and can be better used for training and constructing an order demand prediction model.

[0027] Furthermore, in order to extract more informative and expressive features from the original features, features are extracted from the multiple portrait tag feature vectors to construct a more accurate customer portrait association feature vector. It should be understood that feature extraction is a process of extracting more representative and discriminative features from the original features. It can be achieved through various statistical, mathematical, and machine learning methods. In the case of customer portraits, feature extraction can help extract the most representative and discriminative features from multiple portrait tag feature vectors. Some commonly used feature extraction methods include principal component analysis (PCA), linear discriminant analysis (LDA), information gain, etc. These methods can select and extract the most relevant and discriminative features. By extracting features from the multiple portrait tag feature vectors, a more compact and representative customer portrait association feature vector can be obtained.

[0028] Furthermore, feature extraction from multiple historical order feature vectors can extract more informative and expressive features from the original features to better represent the correlation between historical orders. It should be understood that historical order data may contain information in multiple dimensions, such as order quantity, date, time, product type, etc. Feature extraction methods can include statistical feature extraction, frequency feature extraction, time series feature extraction, etc. For example, the average value, standard deviation, maximum value, and minimum value of the order quantity can be calculated as statistical features, the order frequency of different product types can be calculated as frequency features, and the seasonal and trend features of the order date and time can be extracted as time series features. Through feature extraction, the original historical order feature vectors can be converted into more representative and relevant historical order correlation feature vectors, thereby better representing the correlation and patterns between orders.

[0029] Figure 2 Flowchart for semantic encoding of the multiple customer portraits containing multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database in the machine learning-based order demand prediction method according to an embodiment of the present application. As Figure 2As shown, in a specific embodiment of the present application, semantic encoding is performed on the multiple customer portraits containing multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database to obtain multiple portrait tag feature vectors and multiple historical order feature vectors, including: S210, the multiple customer portraits containing multiple portrait tags and their weights obtained from the database are passed through a portrait tag feature extractor based on a context encoder to convert each portrait tag of each customer portrait in the multiple customer portraits into a feature vector to obtain the multiple portrait tag feature vectors; S220, the historical order data obtained from the database is passed through a context-based encoder model including an embedding layer to obtain the multiple historical order feature vectors. It should be understood that in a customer portrait, each portrait tag represents a specific attribute or feature, such as age, gender, geographical location, etc. These tags usually exist in the database in text form. However, machine learning models usually need to represent the input as numerical feature vectors for calculation and prediction. The portrait tag feature extractor based on a context encoder can convert the portrait tags in text form into numerical feature vectors. Specifically, the portrait tag feature extractor based on a context encoder can map words or phrases in the text to a low-dimensional continuous vector space, so that words with similar meanings are closer in the vector space, to convert each portrait tag into a feature vector. More specifically, the embedding layer of the portrait tag feature extractor based on the context encoder is used to convert the multiple customer portraits containing multiple portrait tags and their weights obtained from the database into embedding vectors to obtain a sequence of customer portrait embedding vectors; the transformer of the portrait tag feature extractor based on the context encoder is used to perform global encoding on the sequence of customer portrait embedding vectors to obtain the multiple portrait tag feature vectors.

[0030] As mentioned above, the various information usually contained in historical order data is also stored in a database in a structured or semi-structured form. Machine learning models usually need to represent the input as a numerical feature vector for calculation and prediction. Therefore, a context-based encoder model is used to convert historical order data into feature vectors. Further, the embedding layer is part of the encoder model, which can map discrete order data (such as product numbers) to a low-dimensional continuous vector representation. A common way to do this is to use an embedding matrix, where each unique order data value is mapped to a vector. By inputting the historical order data into the context-based encoder model, each order can be converted into a feature vector. This feature vector captures the key information and features of the order, such as purchase behavior, product preferences, etc. Specifically, the historical order data obtained from the database is tokenized to obtain a historical order word sequence; the embedding layer of the context-based encoder model containing the embedding layer is used to map each word in the historical order word sequence to a historical order word embedding vector respectively to obtain a sequence of historical order word embedding vectors; the Transformer-based Bert model of the context-based encoder model containing the embedding layer is used to perform global context semantic encoding on the sequence of historical order word embedding vectors to obtain the multiple historical order feature vectors.

[0031] Figure 3 A flowchart for feature extraction of the multiple portrait label feature vectors in the machine learning-based order demand prediction method according to an embodiment of the present application. As Figure 3As shown, in a specific embodiment of the present application, feature extraction is performed on the multiple portrait label feature vectors to obtain the customer portrait association feature vector, including: S310, multiplying each portrait label feature vector corresponding to each customer portrait in the multiple portrait label feature vectors by the weights corresponding to each portrait label in each customer portrait to obtain multiple portrait label weighted feature vectors; S320, performing feature encoding on the multiple portrait label weighted feature vectors to obtain the customer portrait association feature vector. It should be understood that multiplying each portrait label feature vector corresponding to each customer portrait in the multiple portrait label feature vectors by the weights corresponding to each portrait label in each customer portrait is to assign different importance or weights to different portrait labels. A customer portrait is a comprehensive overview of customer characteristics and behaviors, usually composed of multiple portrait labels, and each portrait label represents a feature or behavior of the customer. For example, portrait labels may include age, gender, geographical location, purchase history, hobbies, etc. When representing the features of a customer portrait, considering that different portrait labels may have different importance for the customer's purchase tendency, in the technical solution of the present application, a weight is defined for each portrait label. These weights can be determined according to business requirements, domain knowledge, or through data analysis and model training. By multiplying each portrait label feature vector corresponding to each customer portrait by the corresponding weight, the features of different portrait labels can be weighted. In this way, important portrait label features can be emphasized and secondary portrait label features can be weakened, so as to better reflect the customer's characteristics and purchase tendency. The generation of weighted feature vectors can be achieved through simple element multiplication, multiplying the elements in each portrait label feature vector by the corresponding weights. The finally obtained weighted feature vectors will more accurately represent the portrait features of the customer.

[0032] Furthermore, performing feature encoding on multiple portrait label weighted feature vectors can integrate the multiple portrait label features of the customer into a comprehensive feature vector, thereby representing the overall features of the customer. It should be understood that the customer portrait association feature vector is a numerical representation of the customer portrait. Encoding the multiple portrait label features of the customer can facilitate the processing and analysis of these features by machine learning models. Specifically, feature encoding is the process of converting the original features into numerical representations that can be processed by machine learning models. Common feature encoding methods include one-hot encoding, label encoding, embedding encoding, etc. In this case, when performing feature encoding on multiple portrait label weighted feature vectors, an appropriate encoding method can be used to convert these features into numerical representations, so as to capture the overall features and behaviors of the customer.

[0033] In a specific embodiment of the present application, performing feature encoding on the weighted feature vectors of the multiple portrait tags to obtain the customer portrait association feature vector includes: concatenating the weighted feature vectors of the multiple portrait tags to concatenate the weighted feature vectors of each portrait tag of each customer portrait in the multiple customer portraits into multiple customer portrait feature vectors; passing the multiple customer portrait feature vectors through a customer portrait feature extractor based on a convolutional neural network to obtain the customer portrait association feature vector. It should be understood that concatenating the weighted feature vectors of each portrait tag of each customer portrait in the multiple customer portraits into multiple customer portrait feature vectors can integrate the features of different customers to form a more comprehensive customer portrait representation. Specifically, in the order demand prediction task, there are usually portrait data of multiple customers, and the portrait of each customer consists of multiple weighted feature vectors of portrait tags, representing the weights of the customer on different features. In order to integrate the features of these customers, the weighted feature vectors of each customer can be concatenated to form a multi-dimensional customer portrait feature vector. The concatenation operation connects multiple vectors in a certain order to form a more comprehensive vector, which contains the feature information of all customers.

[0034] Furthermore, passing the multiple customer portrait feature vectors through a customer portrait feature extractor based on a convolutional neural network can obtain the customer portrait association feature vector. It should be understood that a convolutional neural network can capture local and global context information in the data. For customer portrait data, a convolutional neural network can identify the associations and dependencies between different features, thereby better understanding the feature representation of the customer. By inputting the multiple customer portrait feature vectors into the convolutional neural network, the context modeling ability of the convolutional neural network can be utilized to obtain a customer portrait association feature vector with more semantic information, thereby extracting more meaningful feature representations, achieving dimension matching, and capturing the context relationship between customer features. Specifically, each layer of the customer portrait feature extractor based on the convolutional neural network performs convolutional processing, pooling processing, and non-linear activation processing on the input data respectively during the forward pass of the layer, and the customer portrait association feature vector is output by the last layer of the customer portrait feature extractor based on the convolutional neural network, where the input of the customer portrait feature extractor based on the convolutional neural network is the multiple customer portrait feature vectors.

[0035] In a specific embodiment of the present application, feature extraction is performed on the multiple historical order feature vectors to obtain the historical order association feature vector, including: arranging the multiple historical order feature vectors in a two-dimensional manner to obtain a historical order feature matrix; passing the historical order feature matrix through a historical order feature extractor based on a convolutional neural network to obtain the historical order association feature vector. It should be understood that the historical order feature vector usually contains multiple feature dimensions, and each dimension represents an order feature. By arranging these feature vectors, they can be organized into the form of a two-dimensional matrix, where each row represents an order and each column represents a feature. Arranging the multiple historical order feature vectors in a two-dimensional manner can better organize and represent the order data, provide richer information for the model, and thus improve the prediction effect.

[0036] Furthermore, the convolutional neural network can automatically learn the local and global features in the matrix and combine them through a hierarchical structure to obtain a more expressive feature vector. Such a feature vector can better represent the association and feature information between historical orders and is used for tasks such as order demand prediction. Specifically, it is further used to process the historical order feature matrix with the following convolution formula using the historical order feature extractor based on the convolutional neural network to obtain the historical order association feature vector;

[0037] where the convolution formula is: ; where is the input of the historical order feature extractor based on the convolutional neural network at the th layer, is the output of the historical order feature extractor based on the convolutional neural network at the th layer, is the global transformation matrix of the historical order feature extractor based on the convolutional neural network at the th layer, and is the bias vector of the historical order feature extractor based on the convolutional neural network at the th layer, represents a non-linear activation function.

[0038] In the above machine learning-based order demand prediction method, in step S130, based on the customer profile associated feature vector and the historical order associated feature vector, the purchase tendency of the customer is determined. It should be understood that the customer profile associated feature vector and the historical order associated feature vector can capture the association between customer characteristics and purchase behaviors. The customer profile associated feature vector reflects the personal attributes and preferences of the customer, while the historical order associated feature vector reflects the past purchase behaviors of the customer. By combining these two feature vectors, the relationship between the customer's purchase behavior and their personal attributes and preferences can be discovered, thereby better predicting the customer's purchase tendency.

[0039] In a specific embodiment of the present application, step S130 includes: fusing the customer profile associated feature vector and the historical order associated feature vector to obtain an order prediction feature vector; strengthening the distribution characteristics of the order prediction feature vector based on perceptual analysis to obtain an optimized order prediction feature vector; passing the optimized order prediction feature vector through a classifier to obtain a classification result, and the classification result is used to determine the purchase tendency of the customer. It should be understood that fusing the customer profile associated feature vector and the historical order associated feature vector can comprehensively consider the personal attributes, preferences and past purchase behaviors of the customer, capture the association between the features, and thus determine the purchase tendency of the customer. Further, passing the order prediction feature vector through a classifier to obtain a classification result to establish an association between the order prediction feature vector and different purchase tendencies, and classify the purchase tendency of the customer. The classifier can learn the importance of different features for the purchase tendency and classify the purchase tendency of the customer according to the prediction feature vector, such as high tendency and low tendency. This can help enterprises better understand the purchase willingness and behaviors of customers.

[0040] Further, perform distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector to obtain an optimized order prediction feature vector, including: multiplying the order prediction feature vector by the classification weight matrix of the classifier to obtain an intermediate feature vector; concatenating the order prediction feature vector and the intermediate feature vector to obtain a concatenated feature vector; multiplying the first weight matrix by the concatenated feature vector and then adding the first bias vector to obtain an encoded concatenated feature vector; activating the encoded concatenated feature vector through a sigmoid function to obtain an activation value; multiplying the subtraction of the activation value from 1 by the intermediate feature vector element-wise to obtain a weighted intermediate feature vector; multiplying the activation value by the order prediction feature vector element-wise to obtain a weighted order prediction feature vector; subtracting the order prediction feature vector from the intermediate feature vector element-wise to obtain a difference feature vector; adding the weighted intermediate feature vector and the weighted order prediction feature vector and then dividing by the difference feature vector to obtain a second intermediate feature vector; passing the second intermediate feature vector through a ReLU function to obtain an optimized order prediction feature vector.

[0041] Specifically, in the technical solution of this application, considering that the order prediction feature vector involves the association of customer portrait features and historical order features, this complex feature association may lead to a very complex relationship between data points in the feature space, resulting in poor monotonicity of the overall feature distribution flow. Extracting the order prediction feature vector requires comprehensive consideration of the relationship between customer portrait features and historical order features, which may increase the complexity of feature extraction. If the monotonicity of the overall feature distribution flow is poor, it may be difficult for the feature extractor to accurately capture the association information between data, thereby affecting the classification ability of the classifier. Due to the poor monotonicity of the overall feature distribution flow of the order prediction feature vector, the classifier may not be able to fully utilize the feature information of the data during classification, resulting in insufficient classification regression constraint of the classifier. This will affect the accuracy of the classification result and make it difficult for the classifier to accurately judge the purchase tendency of customers. To improve this situation, in the technical solution of this application, perform distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector to obtain an optimized order prediction feature vector.

[0042] Among them, performing distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector to obtain an optimized order prediction feature vector includes: performing distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector with the following optimization formula, where the optimization formula is:

[0043] ; where represents the order prediction feature vector, is the classification weight matrix of the classifier, represents the concatenation function, represents the first weight matrix, represents the first bias vector, represents matrix multiplication, represents the sigmoid activation function, represents the activation value, represents the rectified linear unit function, represents the optimized order prediction feature vector.

[0044] In the technical solution of the present application, the manifold monotonicity of the overall feature distribution of the order prediction feature vector is poor, resulting in insufficient classification regression constraint with respect to the overall feature distribution of the classifier when it is classified and regressed by the classifier, affecting the accuracy of the classification result.

[0045] Therefore, in the technical solution of the present application, distribution characteristic enhancement based on perceptual analysis is performed on the order prediction feature vector. It performs auxiliary analysis and description of the geometric characteristics of the feature manifold of the order prediction feature vector by using the classification weight matrix of the classifier, and uses the difference between the auxiliary description analysis result and the feature manifold of the original order prediction feature vector as a differential factor to support the description of the improved order prediction feature vector for the predetermined classification result of the classifier. Finally, an activation mechanism is added for activation to maintain the enhancement of the distribution dependence with positive description. In this way, the manifold monotonicity of the overall feature layout of the order prediction feature vector is significantly improved, thereby enhancing its classification regression constraint with respect to the overall feature layout of the classifier and improving the accuracy of the classification result.

[0046] Furthermore, in a specific embodiment of the present application, the classifier is used to process the optimized order prediction feature vector with the following classification formula to obtain the classification result; wherein, the classification formula is: ; wherein, to is the weight matrix, to is the bias vector, is the optimized order prediction feature vector, represents the function,

[0047] In summary, in the embodiments of the present application, multiple customer portraits including multiple portrait tags and their weights obtained from a database and historical order data obtained from the database are first collected. The historical order data includes the order quantity, date, time, and product type. Then, deep learning technology is used to perform feature extraction and correlation analysis on the two. Finally, a classification result is obtained through a classifier to judge the purchase tendency of customers, so as to further predict the order demand of the next month, and then optimize the enterprise production plan and inventory management plan, reduce the situation of overstock or out-of-stock, and thus improve the operation efficiency and business growth.

[0048] Figure 4 FIG. is a block diagram of an order demand prediction system based on machine learning according to an embodiment of the present application. As Figure 4 shown, the order demand prediction system 100 based on machine learning according to an embodiment of the present application includes: an order data collection module 110, configured to collect multiple customer portraits including multiple portrait tags and their weights obtained from a database and historical order data obtained from the database, where the historical order data includes the order quantity, date, time, and product type; an order data feature extraction module 120, configured to extract a customer portrait association feature vector and a historical order association feature vector from the multiple customer portraits including multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database; and a purchase tendency judgment generation module 130, configured to judge the purchase tendency of a customer based on the customer portrait association feature vector and the historical order association feature vector.

[0049] As described above, the order demand prediction system 100 based on machine learning according to an embodiment of the present application can be implemented in various terminal devices, such as a server deployed with an order demand prediction control algorithm based on machine learning. In one example, the order demand prediction system 100 based on machine learning can be integrated into a terminal device as a software module and / or a hardware module. For example, the order demand prediction system 100 based on machine learning can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the order demand prediction system 100 based on machine learning can also be one of many hardware modules of the terminal device.

[0050] Alternatively, in another example, the order demand prediction system 100 based on machine learning and the terminal device can also be separate devices, and the order demand prediction system 100 based on machine learning can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0051] Here, those skilled in the art can understand that the specific operations of each step in the above machine learning-based order demand prediction system have been described in detail in the description of the machine learning-based order demand prediction method with reference to Figures 1 to 3 and thus, the repeated description thereof will be omitted.

[0052] The embodiments of the present application provide an electronic device. Figure 5 It is a block diagram of the electronic device according to the embodiments of the present application.

[0053] As Figure 5 shown, the electronic device 10 includes an input device 11, an input interface 12, a central processing unit 13, a memory 14, an output interface 15, an output device 16, and a bus 17. Among them, the input interface 12, the central processing unit 13, the memory 14, and the output interface 15 are connected to each other through the bus 17, and the input device 11 and the output device 16 are respectively connected to the bus 17 through the input interface 12 and the output interface 15, and then connected to other components of the electronic device 10.

[0054] Specifically, the input device 11 receives external input information and transmits the input information to the central processing unit 13 through the input interface 12; the central processing unit 13 processes the input information based on the computer-executable instructions stored in the memory 14 to generate output information, temporarily or permanently stores the output information in the memory 14, and then transmits the output information to the output device 16 through the output interface 15; the output device 16 outputs the output information to the outside of the electronic device 10 for user use.

[0055] In one embodiment, Figure 5 the electronic device 10 shown can be implemented as a network device, and the network device may include: a memory configured to store a program; a processor configured to run the program stored in the memory to execute any one of the image-based device working power control methods described in the above embodiments.

[0056] Those of ordinary skill in the art will understand that all or some of the steps in the methods disclosed above, and the functional modules / units in systems and devices, can be implemented as software, firmware, hardware, or a suitable combination thereof. In a hardware implementation, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be executed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory, or other memory technologies, CD-ROM, digital versatile disks (DVDs), or other optical disk storage, magnetic cassettes, tapes, magnetic disk storage, or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically contains computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0057] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principles of the present application, but the present application is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present application, and these modifications and improvements are also regarded as the protection scope of the present application.

Claims

1. A method for predicting order demand based on machine learning, characterized in that: include: Collect multiple customer portraits including multiple portrait tags and their weights obtained from a database and historical order data obtained from the database, wherein the historical order data includes order quantity, date, time, and product type; Extracting a customer portrait-related feature vector and a historical order-related feature vector from the multiple customer portraits obtained from the database and including multiple portrait tags and their weights and the historical order data obtained from the database; Judging the customer's purchasing tendency based on the customer portrait-related feature vector and the historical order-related feature vector; The step of judging the customer's purchasing tendency based on the customer portrait-related feature vector and the historical order-related feature vector includes: Fusion of the customer portrait-related feature vector and the historical order-related feature vector to obtain an order prediction feature vector; Performing distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector to obtain an optimized order prediction feature vector; Passing the optimized order prediction feature vector through a classifier to obtain a classification result, wherein the classification result is used to judge the customer's purchasing tendency; The order prediction feature vector is enhanced based on the perceptual analysis to obtain an optimized order prediction feature vector, including: Multiplying the order prediction feature vector by the classification weight matrix of the classifier to obtain an intermediate feature vector; Concatenating the order prediction feature vector with the intermediate feature vector to obtain a concatenated feature vector; The first weight matrix is ​​multiplied by the concatenated feature vector and then the first bias vector is added to obtain a coded concatenated feature vector; Activating the encoded concatenated feature vector through a sigmoid function to obtain an activation value; Subtract the activation value from one and multiply it by position with the intermediate feature vector to obtain the weighted intermediate feature vector; Multiplying the activation value by the order prediction feature vector by position to obtain a weighted order prediction feature vector; Subtracting the intermediate feature vector from the order prediction feature vector by position to obtain a difference feature vector; Adding the weighted intermediate feature vector and the weighted order prediction feature vector and then dividing by the difference feature vector to obtain a second intermediate feature vector; The second intermediate feature vector is passed through the ReLU function to obtain an optimized order prediction feature vector.

2. The order demand forecasting method based on machine learning according to claim 1, characterized in that: Extracting a customer portrait-related feature vector and a historical order-related feature vector from the multiple customer portraits including multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database includes: Performing semantic coding on the multiple customer portraits obtained from the database and including multiple portrait tags and their weights and the historical order data obtained from the database to obtain multiple portrait tag feature vectors and multiple historical order feature vectors; Performing feature extraction on the multiple portrait label feature vectors to obtain the customer portrait associated feature vector; Feature extraction is performed on the multiple historical order feature vectors to obtain the historical order associated feature vector.

3. The order demand forecasting method based on machine learning according to claim 2, characterized in that: Semantically encoding the multiple customer portraits obtained from the database and including multiple portrait tags and their weights and the historical order data obtained from the database to obtain multiple portrait tag feature vectors and multiple historical order feature vectors includes: The plurality of customer portraits obtained from the database and including the plurality of portrait labels and their weights are passed through a portrait label feature extractor based on a context encoder to convert each portrait label of each customer portrait in the plurality of customer portraits into a feature vector to obtain the plurality of portrait label feature vectors; The historical order data obtained from the database is passed through a context-based encoder model including an embedding layer to obtain the multiple historical order feature vectors.

4. The order demand forecasting method based on machine learning according to claim 3 is characterized in that: Extracting features from the multiple portrait label feature vectors to obtain the customer portrait associated feature vector includes: Multiplying each portrait label feature vector corresponding to each customer portrait in the multiple portrait label feature vectors by the weight corresponding to each portrait label in each customer portrait to obtain multiple portrait label weighted feature vectors; The plurality of portrait label weighted feature vectors are feature encoded to obtain the customer portrait associated feature vector.

5. The order demand forecasting method based on machine learning according to claim 4 is characterized in that: Feature encoding the multiple portrait label weighted feature vectors to obtain the customer portrait associated feature vector includes: Cascading the multiple portrait label weighted feature vectors to cascade each portrait label weighted feature vector of each customer portrait in the multiple customer portraits into multiple customer portrait feature vectors; The multiple customer portrait feature vectors are passed through a customer portrait feature extractor based on a convolutional neural network to obtain the customer portrait associated feature vector.

6. The order demand forecasting method based on machine learning according to claim 5, characterized in that: Extracting features from the plurality of historical order feature vectors to obtain the historical order associated feature vector includes: Arrange the plurality of historical order feature vectors in two dimensions to obtain a historical order feature matrix; The historical order feature matrix is ​​passed through a historical order feature extractor based on a convolutional neural network to obtain the historical order associated feature vector.

7. An order demand forecasting system based on machine learning, characterized in that: include: An order data collection module, used to collect multiple customer portraits including multiple portrait tags and their weights obtained from the database and historical order data obtained from the database, wherein the historical order data includes order quantity, date, time and product type; An order data feature extraction module, used to extract a customer portrait-related feature vector and a historical order-related feature vector from the multiple customer portraits including multiple portrait tags and their weights obtained from the database and the historical order data obtained from the database; A purchase tendency judgment generation module, used to judge the customer's purchase tendency based on the customer portrait associated feature vector and the historical order associated feature vector; The step of judging the customer's purchasing tendency based on the customer portrait-related feature vector and the historical order-related feature vector includes: Fusion of the customer portrait-related feature vector and the historical order-related feature vector to obtain an order prediction feature vector; Performing distribution characteristic enhancement based on perceptual analysis on the order prediction feature vector to obtain an optimized order prediction feature vector; Passing the optimized order prediction feature vector through a classifier to obtain a classification result, wherein the classification result is used to judge the customer's purchasing tendency; The order prediction feature vector is enhanced based on the perceptual analysis to obtain an optimized order prediction feature vector, including: Multiplying the order prediction feature vector by the classification weight matrix of the classifier to obtain an intermediate feature vector; Concatenating the order prediction feature vector with the intermediate feature vector to obtain a concatenated feature vector; The first weight matrix is ​​multiplied by the concatenated feature vector and then the first bias vector is added to obtain a coded concatenated feature vector; Activating the encoded concatenated feature vector through a sigmoid function to obtain an activation value; Subtract the activation value from one and multiply it by position with the intermediate feature vector to obtain the weighted intermediate feature vector; Multiplying the activation value by the order prediction feature vector by position to obtain a weighted order prediction feature vector; Subtracting the intermediate feature vector from the order prediction feature vector by position to obtain a difference feature vector; Adding the weighted intermediate feature vector and the weighted order prediction feature vector and then dividing by the difference feature vector to obtain a second intermediate feature vector; The second intermediate feature vector is passed through the ReLU function to obtain an optimized order prediction feature vector.

8. An electronic device, characterized in that: include: at least one processor; And, a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the machine learning-based order demand forecasting method as described in any one of claims 1-6.

Images