Business object demand quantity prediction method, device and equipment

Abstract

The invention relates to a business object demand quantity prediction method, device and equipment, relates to the technical field of computer intelligent analysis, and improves the accuracy of business object demand quantity prediction. Comprising the following steps: establishing a hierarchical structure formed by each processing flow in a target scene according to the influence of a result obtained by executing a specific processing flow on other different processing flow results; in response to a selection instruction input by a user, selecting a target processing flow from the plurality of processing flows; collecting multi-dimensional related characteristics generated by processing the business object before the current time node by a target processing flow, and forming a historical time node sequence; predicting the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through a single flow prediction model; and inputting the prediction quantity into a harmonic model obtained by training in advance based on the hierarchical structure, and adjusting the prediction quantity by using the speculative quantity through the harmonic model to obtain the harmonic prediction vector.

Description

Technical Field

[0001] The present application relates to the field of computer intelligent analysis technology, and specifically to a method, device and equipment for predicting the quantity of business object demand. Background Art

[0002] Predicting the consumption of business objects in the future is one of the most important means of modern business activities. It can effectively help enterprises make short-term to long-term plans and decisions, such as formulating business object consumption methods, rationally arranging business object inventory, optimizing product structure, etc., to reduce their operating costs and increase corporate profits. The accuracy of the forecast directly affects the rationality and effectiveness of corporate decisions.

[0003] In deep learning, features are one of the most important factors affecting the performance of deep learning methods. Features that affect the prediction results of business objects include: historical processing data of business objects, dates, holidays, weather, etc. The learning method under the existing technology directly inputs the above multi-dimensional features into the model operation, ignoring the differences between different features in the time dimension and the correlation between various features, resulting in poor model results.

[0004] In addition, when executing a process to process business objects, the actual number of business objects consumed is related to the scenario in which the process is executed. The existing technology for predicting the number of objects consumed in the future often ignores the associated constraints between the scenarios in which a process is executed to process business objects, resulting in poor model prediction results. Summary of the invention

[0005] The technical problem to be solved by the present application is to provide a method, device and equipment for predicting the quantity of business object demand, so as to improve the accuracy of predicting the quantity of business object demand.

[0006] In a first aspect, an embodiment provides a method for predicting the quantity of business object requirements, which is applied to electronic devices, and the method includes: establishing a hierarchical structure formed by each processing flow in a target scenario according to the impact of the result achieved by executing a specific processing flow on the results of other different processing flows; the processing flow is the process of running a predetermined program to process the business object; the target scenario contains multiple processing flows; in response to a selection instruction input by a user, selecting a target processing flow among the multiple processing flows; collecting multidimensional related features generated by the target processing flow when processing the business object before the current time node to form a historical time node sequence; the multidimensional related features include but are not limited to the number of business objects, resources consumed in processing the business objects, time information, and scenario environment; inputting the historical time node sequence into a pre-established single-flow prediction model, and predicting the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through the single-flow prediction model as basic prediction information; inputting the predicted number into a reconciliation model pre-trained based on the hierarchical structure, and adjusting the predicted number with the inferred number through the reconciliation model to obtain the reconciliation prediction vector; the inferred number is the number of objects consumed by other processing flows in the hierarchy at future time nodes.

[0007] In one embodiment, the method further includes: using multiple samples to train the reconciliation model based on the hierarchical structure; the process of using a single sample to train the reconciliation model includes: According to the characteristics generated by the target processing flow processing the consumable object at the first time node, predicting the first value of the target processing flow processing the consumable object at the second time node, and collecting the second value of the target processing flow processing the consumable object at the second time node; According to the characteristics of at least one other processing flow processing the consumable object at the first time node, predicting a third value of the consumable object processed by the other processing flow at the second time node, and collecting a fourth value of the consumable object processed by the other processing flow at the second time node; Combining the third value and the first value to form a prediction vector according to the hierarchical structure; Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Resetting the prediction vector using the relationship matrix to obtain a reset prediction vector; the reset prediction vector includes a reset third value and a reset first value; combining the second value and the fourth value to obtain a tag sequence; A loss function is calculated according to the reset prediction vector and the label sequence, and a weight parameter in the relationship matrix is ​​adjusted according to the loss function until the loss function converges.

[0008] In one embodiment, calculating the loss function according to the reset prediction vector and the tag sequence includes: (1) in, is the reset prediction vector, is a tag sequence; is the sequence of the first value after reset corresponding to the second time node, is the sequence of the third value after reset corresponding to the second time node, represents the loss function, Indicates the sequence of samples currently used, is the total number of samples.

[0009] In one embodiment, the reconciliation model is obtained based on the hierarchical structure training; and the process of training the reconciliation model includes: Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Using the relationship matrix as an encoder and the transformation matrix as a decoder to obtain a deep learning framework; The reconciliation model is obtained based on the deep learning framework training.

[0010] In one embodiment, the historical time node sequence is input into a pre-established single-process prediction model, and the single-process prediction model is used to predict the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time, including: Responding to the user's setting instructions, determining the predicted future time node; Setting a basic prediction network, setting a prediction output structure of the basic prediction network according to the future time node, and setting a retrospective output structure of the basic prediction network according to the node information in the historical time node sequence; Inputting the historical time node sequence into any basic prediction network, outputting first prediction information according to the prediction output structure, and outputting first review information according to the review output structure; Inputting the first review information into a basic prediction network corresponding to the next dimension feature, outputting second prediction information and second review information, and inputting the second review information into the next basic prediction network until the outputs of all basic prediction networks are obtained; Combined with the prediction information output by each basic prediction network, the predicted number of business objects processed by the target processing flow at at least one future time node after the current time is obtained.

[0011] In one embodiment, the method further includes: For each basic prediction network, the difference between the output review information and the input is calculated to obtain the residual information filtered by the basic prediction network; Combine the prediction information output by each basic prediction network, including: When the number of basic prediction networks is less than a preset number, the prediction information output by each basic prediction network and the residual information are combined to obtain the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time.

[0012] In one embodiment, the single-process prediction model includes multiple basic prediction networks; the basic prediction network includes a time-related feature extraction network layer and a non-time-related feature extraction network layer; the time-related feature extraction network layer is constructed based on LSTM results, and the non-time-related feature extraction network layer is constructed based on MLP.

[0013] In one embodiment, the historical time node sequence is input into any basic prediction network, the first prediction information is output according to the prediction output structure, and the first review information is output according to the review output structure, including: The historical time node sequence is calculated by the time-related feature extraction network layer to filter the first latent variable feature associated with time in the historical time node sequence; When there are time-independent features in the historical time node sequence, the non-time-related feature extraction network layer is enabled to filter the second latent variable features in the historical time node sequence that are not related to time; the time-independent features refer to features that have no linear relationship with time; Strengthening the first latent variable feature to obtain a first strengthened feature, and strengthening the second latent variable feature to obtain a second strengthened feature; The first prediction information and the first review information are calculated by fusing the first enhanced feature and the second enhanced feature to obtain the feature.

[0014] In a second aspect, an embodiment provides a business object demand quantity prediction device, which is arranged in an electronic device, and the device includes: A hierarchical structure establishment module is used to establish a hierarchical structure formed by various processing flows in a target scenario according to the impact of the results of executing a specific processing flow on the results of other different processing flows; the processing flow is the process of running a predetermined program to process a business object; the target scenario includes multiple processing flows; A process selection module, configured to respond to a selection instruction input by a user and select a target process from the multiple process flows; A feature collection module is used to collect multi-dimensional related features generated by the target processing flow processing the business object before the current time node to form a historical time node sequence; the multi-dimensional related features include but are not limited to the number of business objects, resources consumed in processing business objects, time information, and scene environment; A sequence input module, used to input the historical time node sequence into a pre-established single-process prediction model, and predict the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through the single-process prediction model as basic prediction information; The adjustment module is used to input the predicted quantity into a reconciliation model pre-trained based on the hierarchical structure, and adjust the predicted quantity with the inferred quantity through the reconciliation model to obtain the reconciliation prediction vector; the inferred quantity is the quantity of consumption objects processed by other processing flows in the hierarchical structure at future time nodes.

[0015] The beneficial effects of this application are: The method for predicting the demand quantity of business objects proposed in the embodiment of the present application discovers that the results of different processing flows will affect the results of other processing flows, which is manifested in the sales scenario as follows: the sales quantity of one category affects the sales quantity of another category, or the sales quantity of a certain product in one region is affected by the sales quantity of a certain product in other regions. In view of the above findings, according to the current application scenario of processing business objects, the hierarchical structure of the business object processing flow is mined from aspects such as region, city, store, date, and business object category, and the predicted demand quantity of business objects for different processing flows is optimized with a hierarchical structure to improve the accuracy of the prediction of the demand quantity of business objects. At the same time, the embodiment of the present application also extracts the dependency relationship between features and time, extracts features associated with time and features not associated with time respectively, and strengthens features associated with time and features not associated with time, amplifies the differences between features, clarifies the feature path of deep learning network learning, and thus improves the accuracy of the model output prediction quantity of business objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the steps of executing the method for predicting the quantity of business object demand proposed in the embodiment of the present application; Figure 2 It is a schematic diagram of a hierarchical structure established by an example of this application; Figure 3 It is a schematic diagram of a hierarchical structure established by another example of the present application; Figure 4 It is a schematic diagram of the structure of an exemplary single-process prediction model of the present application; Figure 5 This is a schematic diagram of the structure of a basic prediction network constructed as an example in this application; Figure 6 This is a schematic diagram of the geographical scope in one example of this application; Figure 7 It is a schematic diagram of a learning framework of a reconciliation model of the present application; Figure 8 It is a functional module diagram of the business object demand quantity prediction device proposed in the embodiment of the present application; Fig. 9 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are intended to enable the present application to be better understood. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, and methods. In some cases, some operations related to the present application are not shown or described in the specification, in order to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0018] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0019] The serial numbers assigned to the components herein, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0020] Figure 1 It is a flowchart of the steps of executing the method for predicting the quantity of business object demand proposed in the embodiment of the present application. The execution subject may be an electronic device, and the electronic device may be a server, a computer terminal, a virtual computing system, etc.

[0021] The steps include: S101: Establishing a hierarchical structure formed by various processing flows in a target scenario according to the impact of a result achieved by executing a specific processing flow on results of other different processing flows.

[0022] The processing flow is the process of running a predetermined program to process a business object; the target scenario includes multiple processing flows.

[0023] The applicant discovered that in different scenarios, there is a mutually constrained relationship between the number of different business objects consumed in executing a specific processing flow to process business objects.

[0024] In one example, the target scenario is product processing, which may include multiple smart devices that execute processing flows and mechanical devices that operate on products as instructed by the smart devices, wherein the process of each mechanical device operating on a product as instructed by the smart device is a processing flow.

[0025] In this example, business object a is a part to be processed, and the specific processing flow is the process of the intelligent device running a predetermined program to process the part to be processed. Business object b is a whole part to be assembled, and the specific processing flow is the process of the intelligent device running a predetermined program to assemble the whole part to be assembled. Business object c is still a part to be processed, and the specific processing flow is the process of the intelligent device running a predetermined program to process the part to be processed, where business object b is assembled from business object a and business object c. Then there is a constraint relationship between the number of business objects consumed by executing a specific processing flow: + =N. is the consumption quantity of business object a, is the consumption quantity of business object c, and N is the consumption quantity of business object b.

[0026] Based on the above scenario, the calculation process of the electronic device executing "establishing a hierarchical structure formed by each processing flow in the target scenario according to the influence of the result of executing a specific processing flow on the results of other different processing flows" can be: according to the influence of the number of parts obtained by executing the process of processing parts on the process of assembling workpieces, a hierarchical structure such as Figure 2 As shown, Figure 2 is a schematic diagram of a hierarchical structure established by an example of this application, Figure 2 In the example, business object b corresponds to process flow b which is the root node, business object a corresponds to process flow a, and business object c corresponds to process flow c which are leaf nodes of the root node.

[0027] In one example, the target scenario is selling business objects. The target scenario may include the processes of multiple vending machines each executing a local reservation program. The process of each vending machine executing the local reservation program to sell goods is a processing flow. The process of the server accepting instructions issued by the user through the mini program to control the vending machine to launch goods can also be regarded as a processing flow. Predicting the demand quantity of a business object at a certain time can be understood as the sales quantity of the business object at a certain time. In actual applications, accurately predicting the sales quantity of a business object at a certain time provides a basis for providing the demand for the business object in future time periods, ensuring that the business object will not be in short supply or unsalable.

[0028] In this example, the business object is commodity A to be sold. The computing device installed in the vending equipment 1 is used to receive user instructions to execute a program for delivering commodity A to be sold. The program for delivering commodity A to be sold is a predetermined program for this scenario. The computing device controls the vending machine's pushing mechanism to push commodity A to be sold. The program for delivering commodity B to be sold is a predetermined program for this scenario. The computing device controls the vending machine's pushing mechanism to push commodity B to be sold. Commodity A to be sold and commodity B to be sold are set in different areas. During the execution of the program for delivering commodities to be sold, the computing device of the vending machine can also upload the user request to the server. The server executes a process for supplementing business objects for each vending machine according to the program execution results of the vending machines set in different locations.

[0029] According to the business objects consumed by different vending machines in a certain area in the process of executing the scheduled program to deliver goods, the impact of the robot controlling the execution of the scheduled program on the server in a certain area on the number of vending machines replenishing goods is established, such as Figure 3 As shown, Figure 3 It is a schematic diagram of a hierarchical structure established by another example of the present application. Figure 3 In the figure, the processing flow C corresponding to the commodity C to be sold is the root node, the processing flow B corresponding to the commodity B to be sold, and the processing flow A corresponding to the commodity A to be sold are leaf nodes of the root node.

[0030] Based on the application of step S101 in the scenario shown in the above example, it is obtained that: in the sales scenario, "establishing a hierarchical structure formed by each processing flow in the target scenario based on the impact of the result achieved by executing a specific processing flow on the results of other different processing flows" means: establishing a hierarchical structure of each sales scenario based on the impact of the sales quantity of a business object in a certain sales scenario on the sales quantity of business objects in other sales scenarios.

[0031] Generally, factors affecting the hierarchical structure may include region, city, store, date, etc. Therefore, the electronic device can mine the constraint relationship between factors such as region, city, store, date, business object classification, etc. on the consumption quantity of business objects based on different scenarios of executing specific processing flows, thereby establishing a hierarchical structure.

[0032] S102: In response to a selection instruction input by a user, a target processing flow is selected from the multiple processing flows.

[0033] The objective conditions for executing different processing flows are different, so it is necessary to predict the number of business objects consumed for each processing flow respectively. Specifically, each processing flow can be used as the target processing flow for the current processing, and then the number of business objects consumed can be predicted for the target processing flow.

[0034] The embodiment of the present application sets up a single-process prediction model for each processing flow, and predicts the consumption quantity of business objects in each processing flow based on the single-process prediction model. Before predicting the consumption quantity of business objects in the target processing flow based on the single-process prediction model, features related to business objects in the process of consuming business objects in the execution of the target processing flow are first collected.

[0035] S103: The acquisition target processing flow processes the multi-dimensional related features generated by the business object before the current time node to form a historical time node sequence.

[0036] The multi-dimensional related features include but are not limited to the number of business objects, resources consumed in processing business objects, time information, and scenario environment; each historical time node in the multiple historical time node sequences has corresponding specific dimension related features.

[0037] Taking the business object sales scenario as an example, the number of business objects can be the number of goods sold in a certain period of time, the resources consumed in processing business objects can include promotional activities and advertising investment, and the scenario environment can include holidays, temperature, weather, etc.

[0038] The multi-dimensional related features generated by processing the business object before the current time node can represent: historical data of the business object consumption quantity. For example, the current time is T month TT day, and the historical time nodes before the current time are selected to include: T month TT-1 day, T month TT-2 day, T month TT-3 day, and the multi-dimensional related features generated by processing the business object on T month TT-1 day are collected. , collect multi-dimensional related features generated by processing business objects on TT-2 of T month , collect multi-dimensional related features generated by processing business objects on TT-3rd of T month , get the historical time node sequence: {T month TT-1 day - ,T month TT-2nd- ,T month TT-3rd- }.

[0039] In the actual consumption scenario of business objects, such as the sales scenario of business objects, the features that affect the sales volume of business objects include: endogenous features and exogenous features. Endogenous features include lagged sales history data and features added by merchants to sell business objects, such as promotional activities and advertising investment. Exogenous features reflect external factors such as dates, holidays, and weather. The above factors will affect the sales volume of business objects, and each factor has a different impact on the sales volume, so they can be collected separately based on the features of each dimension.

[0040] For example, assume that the multi-dimensional features include: quantity, Advertising investment, temperature, Weather, then T month TT-1- 、TT-1- 、TT-1- 、TT-1- Each of them can be used as a specific dimension-related feature of the historical time node T month TT-1.

[0041] S104: Inputting the historical time node sequence into a pre-established single-process prediction model, and predicting the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through the single-process prediction model.

[0042] Taking the business object sales scenario as an example, the target processing process is site A for the sales of business objects on a certain street. The historical time node sequence generated by site A is input into the single-process prediction model to predict the number of business objects sold at the site for the sales of business objects at a certain point in the future.

[0043] In one example, if a business object is sold at venue A for the first time, multi-dimensional features of similar products can be collected to obtain a historical time node sequence.

[0044] Figure 4 This is a schematic diagram of the structure of an example single-process prediction model of this application. Figure 4 As shown in Figure 2, the single-process prediction model includes multiple basic prediction networks, which are composed of Figure 4 It can be seen that the output of the basic prediction network includes review information and prediction information, and the review information output by basic prediction network 1 is input into basic prediction network 2. In order to enhance the feature extraction capability, multiple basic prediction networks are stacked to obtain a single-process prediction model, and skip connections are used to prevent the model from degrading due to the increase in the number of layers.

[0045] In the embodiment of the present application, step S104 includes sub-steps: S1041: Respond to the user's setting instruction and determine the predicted future time node.

[0046] If the current time is day T, the user chooses to predict the consumption quantity of business objects within the next Q days, and determines that the future time nodes include {T+1, T+2..., T+Q}.

[0047] The single-process prediction model includes a basic prediction network for each dimension of features.

[0048] S1042: Setting a basic prediction network, setting a prediction output structure of the basic prediction network according to the future time node, and setting a retrospective output structure of the basic prediction network according to the node information in the historical time node sequence.

[0049] For example, if the historical time node sequence is {T+1, T+2..., T+O}, T is the current time, and the future time nodes are {T+1, T+2..., T+Q}, then the predicted output structure can be an O-bit output structure, and the reviewed output structure can be a Q-bit output structure.

[0050] S1043: Input the historical time node sequence into any basic prediction network, output first prediction information according to the prediction output structure, and output first review information according to the review output structure.

[0051] For example, assuming that the future time nodes are {T+1, T+2, ..., T+Q}, the prediction information output based on the prediction output structure is expressed as ,in, is the demand quantity of business objects at the time point (T+1) predicted by the basic forecasting network, is the demand quantity of the business object at the time point (T+2) predicted by the basic forecasting network, It is the demand quantity of the business object at the time point (T+Q) predicted by the basic forecasting network.

[0052] The applicant found that there are both time-dependent and time-independent features in different dimensions. For example, in the sales business object scenario, features such as advertising complaints and temperature are time-independent, while features such as date and sales volume are time-dependent. If a unified network is used to calculate all features, the potential risk is ignored: the interference of time information in mixed feature extraction may lead to performance degradation.

[0053] In view of the above problems, the basic prediction network set up in the embodiment of the present application includes a time-related feature extraction network layer and a non-time-related feature extraction network layer; the time-related feature extraction network layer is constructed based on the LSTM result, and the non-time-related feature extraction network layer is constructed based on MLP.

[0054] Figure 5 This is a schematic diagram of the structure of a basic prediction network constructed as an example in this application. Figure 5 As shown, the time-related feature extraction network layer calculates the historical time node sequence to obtain the first latent variable feature, and then strengthens the first latent variable feature to obtain the strengthened time-related feature sequence, which is then combined with the time-independent feature sequence to first explore the factors affecting the historical consumption quantity of business objects from two aspects, and then integrate them to avoid noise when extracting time-related features and time-independent features respectively, and solve the problem that the interference of time information in feature mixed extraction may cause performance degradation.

[0055] based on Figure 5 In the network shown, executing step S1043 is implemented by running sub-steps S201-S204: S201: Calculating the historical time node sequence through the time-related feature extraction network layer to filter out the first latent variable feature associated with time in the historical time node sequence.

[0056] The temporal correlation feature extraction network layer is built based on the long short-term memory (LSTM) network.

[0057] S202: When there is a time-independent feature in the historical time node sequence, the non-time-related feature extraction network layer is enabled to filter the second latent variable feature that is not related to time in the historical time node sequence. The time-independent feature refers to a feature that has no linear relationship with time.

[0058] The non-time-related feature extraction network layer is built based on the Multi-Layer Perceptron (MLP) mechanism.

[0059] For example, in a sales scenario, features such as promotional activities, advertising investment, and weather can be discrete features. In the actual implementation of the business object demand quantity prediction method, a feature library can be established in advance, or a curve showing the change of features with historical time nodes can be established to determine whether a feature is linearly related to time, thereby determining whether to enable the non-time-related feature extraction network layer.

[0060] In one example of this application, the historical time node sequence T is first divided into time-independent features that are not related to time and time-dependent features (TDFs) that are related to time by feature segmentation. The input historical time node sequence is: a matrix with r-dimensional features and a length of L. The historical time node sequence is sampled using a sliding window to obtain m-dimensional time-related feature TDFs, which are expressed as , the rm dimension is not related to time, time-independent features (TiDFs), expressed as When the sliding window sampling is obtained , it means that there are time-independent features in the historical time node sequence.

[0061] The time-related feature extraction network layer LSTM can calculate the historical time node sequence through formula (2); (2) Represents the first latent variable feature.

[0062] The calculation of the historical time node sequence by the non-time-related feature extraction network layer MLP can be realized by formula (3); (3) Represents the second latent variable feature.

[0063] S203: Strengthen the first latent variable feature to obtain a first strengthened feature, and strengthen the second latent variable feature to obtain a second strengthened feature.

[0064] Strengthening the first latent variable feature can be achieved through formula (4); (4) Indicates the use of sigmoid activation function Perform calculations, Indicates the first reinforcement feature.

[0065] Calculated using the sigmoid activation function , we get the time-dependent feature importance , then Multiply element-wise, filter to get highlight Time-dependent features .

[0066] Strengthening the second latent variable feature can be achieved through formula (5); (5) Represents the second enhanced feature. Calculated using the sigmoid activation function , we get the time-independent feature importance , then Multiply element-wise, filter to get highlight Time-independent features , clarify the connection between time-related features in historical time node sequences, and strengthen the network's learning and memory of time-independent features.

[0067] S204: Calculate the first prediction information and the first review information by fusing the first enhanced feature and the second enhanced feature to obtain the feature.

[0068] The fusion of the first enhanced feature and the second enhanced feature can be achieved by formula (6); (6) Represents element-wise matrix multiplication; Represents the element-wise addition of the matrix, and we get After that, the fully connected layer FC1 is used to calculate the fused information and output the predicted information , use the fully connected layer FC2 to calculate the fused information and output the review information .

[0069] In one possible implementation, the basic prediction networks are stacked to obtain a single-process prediction model. The basic prediction network currently performing calculations outputs first review information and first prediction information. The first review information is used as input to the next basic prediction network performing a calculation task. The next basic prediction network performing a calculation task outputs second review information and second prediction information, and so on.

[0070] S1044: Input the first review information into a basic prediction network corresponding to the next dimension feature, output second prediction information and second review information, and input the second review information into the next basic prediction network until the outputs of all basic prediction networks are obtained.

[0071] The implementation method of inputting the second review information into the basic prediction network corresponding to the next dimension feature includes: performing residual calculation on the second review information and the input of the previous basic prediction network, and using the parameters as all inputs of the next basic prediction network.

[0072] Continue to refer Figure 4 In one example of the present application, the historical time node sequence Input basic prediction network 1, basic prediction network 1 outputs review information and forecast information , calculate the residual of the base prediction network 1 = - , residual As the input of the basic prediction network 2, the output of the basic prediction network 2 and , calculate the residual of the base prediction network 2 = - , residual As the input of the base prediction network 3, and so on, the residual output of the base prediction network k-1 As the input of the base prediction network k, the base prediction network k outputs and .

[0073] S1045: Combining the prediction information output by each basic prediction network, obtaining the predicted number of business objects processed by the target processing flow at at least one future time node after the current time.

[0074] The difference between the retrospective information output and the input of each basic prediction network is calculated to obtain the residual information filtered by the basic prediction network; when the number of basic prediction networks is less than the preset number, the prediction information and residual information output by each basic prediction network are combined to obtain the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time.

[0075] When the number of basic prediction networks stacked is small, the last basic prediction network may still contain more information. Therefore, when the number of basic prediction networks is less than the preset number, the prediction information output by each basic prediction network and the residual information corresponding to the last basic prediction network are fused through formula (7) to obtain the predicted number of consumption objects processed at at least one future time node after the current time.

[0076] (7) It indicates the predicted number of consumption objects to be processed at at least one future time node after the current time, that is, the basic forecast information. In the sales scenario, it can be the predicted number of sales business objects at at least one future time node after the current time. Accurately predicting the sales quantity can provide data support for product preparation to ensure that the prepared goods are within a reasonable range and avoid waste and shortage of goods.

[0077] Indicates that the residual information corresponding to the last basic prediction network is fully connected.

[0078] Indicates The prediction information of the basic prediction network outputs is accumulated for each basic prediction network output, and then combined with the calculation result of the residual information by the fully connected network, the predicted number of consumption objects processed by the target processing flow at least one future time node after the current time is obtained.

[0079] Each processing flow can determine the predicted number of business objects to be processed at at least one future time node after the current time through the above-mentioned embodiments of the present application. The hierarchical structure between the processing flows is mined through the reconciliation model set in the embodiments of the present application to further optimize the predicted number of business object requirements corresponding to each processing flow.

[0080] For example, in the sales scenario, the business data sales area The forecast quantity of the forecast processing business object is , Business Data Sales Area The forecast quantity of the forecast processing business object is , Business Data Sales Area The forecast quantity of the forecast processing business object is , the features generated by a single processing flow may not be comprehensive enough, or there may be special factors that cause prediction errors. In view of this, the embodiment of the present application sets up a reconciliation model that mines the hierarchical constraints between the prediction results of different processing flows and adjusts the prediction results to further optimize the processing results of each processing flow.

[0081] Since the prediction information output by the basic prediction network according to the prediction output structure is expressed as , the prediction information output by each basic prediction network is expressed in the same way.

[0082] S105: Input the predicted quantity into a reconciliation model pre-trained based on the hierarchical structure, adjust the predicted quantity with the inferred quantity through the reconciliation model, and obtain the reconciliation prediction vector.

[0083] The estimated quantity is the quantity of objects consumed by other processing flows in the hierarchy at future time nodes.

[0084] For example, the business data first sales area The forecast quantity of the forecast processing business object is , the second sales area corresponding to the business data The forecast quantity of the forecast processing business object is , Business Data Third Sales Area The forecast quantity for the geographic area including the forecast processing business object is ,analyze , and In the map range data such as Figure 6 As shown, Figure 6 This is a schematic diagram of the geographical scope in an example of this application. Sales quantity of business objects Objective and The sum of Figure 6 This is a location area diagram corresponding to the business object processing flow in an example of this application, and the construction hierarchy is: = + .

[0085] If based on Angle, and As an estimated number, Optimize so that the optimized The number approaches and If based on Angle, and As an estimated number, Optimize so that the optimized The number approaches and difference.

[0086] The method for predicting the demand quantity of business objects proposed in the embodiment of the present application discovers that the results of different processing flows will affect the results of other processing flows, which is manifested in the sales scenario as follows: the sales quantity of one category affects the sales quantity of another category, or the sales quantity of a certain product in one region is affected by the sales quantity of a certain product in other regions. In view of the above findings, according to the current application scenario of processing business objects, the hierarchical structure of the business object processing flow is mined from aspects such as region, city, store, date, and business object category, and the predicted demand quantity of business objects for different processing flows is optimized with a hierarchical structure to improve the accuracy of the prediction of the demand quantity of business objects. At the same time, the embodiment of the present application also extracts the dependency relationship between features and time, extracts features associated with time and features not associated with time respectively, and strengthens features associated with time and features not associated with time, amplifies the differences between features, clarifies the feature path of deep learning network learning, and thus improves the accuracy of the model output prediction quantity of business objects.

[0087] The present application also provides a method for training a reconciliation model. Figure 7 This is a schematic diagram of a learning framework of a reconciliation model in this application, such as Figure 7 As shown, the reconciliation model includes an encoder and a decoder. In this embodiment, the relationship matrix S is used as the encoder and the transformation matrix P is used as the decoder.

[0088] Initial weights are set for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; the relationship matrix is ​​used as an encoder and the transformation matrix is ​​used as a decoder to obtain a deep learning framework; the reconciliation model is obtained by training based on the deep learning framework The transformation matrix P is a matrix of shape M*N, which is used to scale the matrix. Different initial weight parameters are set for different layers, and the initial weight parameters are adjusted during the training process until the loss function of the harmonized model adjusts the predicted quantity and the label converges.

[0089] Training the learning framework to obtain the reconciliation model includes the following steps: S301: predicting a first value of the target processing flow when processing the consumable object at a second time node based on the features generated by the target processing flow when processing the consumable object at a first time node, and collecting a second value of the target processing flow when processing the consumable object at the second time node.

[0090] The first value may be predicted by the method proposed in other embodiments of the present application. The features generated by processing the consumption object include date, weather, temperature, quantity, etc.

[0091] For example, the first time node includes , , , the second time node corresponds to ( )、( )、( ); Features of the collection , Features of the collection , Features of the collection , using the single-process prediction model proposed in other embodiments of the present application, according to the above characteristics { , , }The prediction target processing flow is in ( )、( )、( ) The number of consumed business objects is , , , as the first value; and the acquisition target processing flow is in ( )、( )、( ) The actual number of business objects consumed in time is , , , as the second value.

[0092] S302: Based on the characteristics of at least one other processing flow processing the consumable object at the first time node, predict the third value of the consumable object processed by the other processing flow at the second time node, and collect the fourth value of the consumable object processed by the other processing flow at the second time node.

[0093] The characteristics of acquisition and processing flow 1 include: Features of the collection , Features of the collection , Features of the collection , the characteristics of acquisition and processing flow 2 include: Features of the collection , Features of the collection , Features of the collection According to the above characteristics { , , }Prediction process 1 in ( )、( )、( ) The number of consumed business objects is , , , as the third value corresponding to the processing flow 1, and collect the processing flow 1 at ( )、( )、( ) The number of business objects actually consumed by the time is , , , as the fourth value corresponding to processing flow 1.

[0094] Based on the same method, the third value corresponding to process flow 2 is obtained: , , , obtain the fourth value corresponding to processing flow 2: , , .

[0095] S303: According to the hierarchical structure, combine the third value and the first value to form a prediction vector.

[0096] According to the method of outputting prediction information of the single-process prediction model described in other embodiments of the present application, the output of the single-process prediction model corresponding to the processing flow 1 can be expressed as , ={( )- 、( )- 、( )- }, the output of the single-process prediction model corresponding to process 2 can be expressed as , ={( )- 、( )- 、( )- }, the output of the single-process prediction model corresponding to the target processing process can be expressed as , ={( )- 、( )- 、( )- }.

[0097] For example, suppose the target process is Figure 3 Process C in the case hierarchy shown, the target process is the root node in the hierarchy, and process 1 corresponds to Figure 3 Processing flow A shown in FIG. 1 , processing flow 2 corresponds to Figure 3 Processing flow B shown, refer to Figure 3 In the hierarchical structure shown in FIG. 1 , process 1 and process 2 are leaf nodes of the target process. Therefore, according to the hierarchical structure, the third values ​​predicted by the other process flows are combined with the first value predicted by the target process flow to form a prediction vector, which can be expressed as = .

[0098] S304: Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix.

[0099] The relationship matrix S is used as an encoder, and different weights of attention are used for feature extraction based on the prediction information corresponding to the positions of different hierarchical structures.

[0100] For example, by constructing the initial weights, the control relationship matrix is ​​used to encode the prediction vector. The calculation strategy is determined by the position of the target process in the hierarchy, so as to adjust the weight information in the model training and adjust the relationship matrix The calculation strategy matches the hierarchical constraint relationship among the three.

[0101] S305: Reset the prediction vector using the relationship matrix to obtain a reset prediction vector; the reset prediction vector includes a reset third value and a reset first value.

[0102] References Figure 7 As shown in the learning framework diagram, the process of resetting the prediction vector using the relationship matrix can be expressed by formula (8): (8) in, The third value and the first value form a prediction vector, is the reset prediction vector, is the relationship matrix, is the transformation matrix.

[0103] S306: Combine the second value and the fourth value to obtain a tag sequence.

[0104] The target collection process is as follows: )、( )、( ) The actual number of business objects consumed in time is , , , as the second value, the actual target processing flow of the acquisition is ( )、( )、( ) The actual sequence information of the business data consumed at each time point is expressed as: , process 1 in ( )、( )、( ) The actual sequence information of the business data consumed at each time point is expressed as: , process 1 in ( )、( )、( ) The actual sequence information of the business data consumed at each time point is expressed as: .

[0105] Marker sequence for .

[0106] S307: Calculate a loss function according to the reset prediction vector and the label sequence, and adjust weight parameters in the relationship matrix according to the loss function until the loss function converges.

[0107] This application provides an example of how to calculate the loss function: (1) in, is the reset prediction vector, is a tag sequence; is the sequence of the first value after reset corresponding to the second time node, is the sequence of the third value after reset corresponding to the second time node, represents the loss function, Indicates the sequence of samples currently used, is the total number of samples.

[0108] It can represent the difference between the root node prediction information and the leaf node prediction information. The root node prediction information is the prediction information corresponding to the root node in the hierarchical structure, and the leaf node prediction information is the prediction information corresponding to the leaf node connected to the root node.

[0109] Other examples given in this application are as follows Figure 3 In the case hierarchy shown, the target processing flow, processing flow 1, and processing flow 2 are adjusted using the reconciliation model to adjust the prediction information output by the single process prediction model corresponding to the target processing flow, processing flow 1, and processing flow 2. The prediction information output by the single process prediction model corresponding to the target processing flow is optimized based on the constraints on the number of business objects consumed by processing flow 1 and processing flow 2 in the hierarchy on the number of business objects consumed by the target flow.

[0110] Substitute the prediction vector and label sequence formed by the third value and the first value in the above example into the loss function ,like is the first value after reset. is the third value after reset, i.e. yes ,So yes and .

[0111] Figure 8 is a functional module diagram of the business object demand quantity prediction device proposed in the embodiment of the present application, such as Figure 8 As shown, the business object demand quantity prediction device includes: A hierarchical structure establishment module 81 is used to establish a hierarchical structure formed by various processing flows in a target scenario according to the impact of the result of executing a specific processing flow on the results of other different processing flows; the processing flow is the process of running a predetermined program to process a business object; the target scenario includes multiple processing flows; A process selection module 82, for responding to a selection instruction input by a user and selecting a target process from the multiple process flows; The feature collection module 83 is used to collect multi-dimensional related features generated by the target processing flow in processing the business object before the current time node to form a historical time node sequence; the multi-dimensional related features include but are not limited to the number of business objects, resources consumed in processing the business objects, time information, and scene environment; A sequence input module 84 is used to input the historical time node sequence into a pre-established single-process prediction model, and predict the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through the single-process prediction model as basic prediction information; The adjustment module 85 is used to input the predicted quantity into the reconciliation model pre-trained based on the hierarchical structure, and adjust the predicted quantity with the inferred quantity through the reconciliation model to obtain the reconciliation prediction vector; the inferred quantity is the quantity of consumption objects processed by other processing flows in the hierarchical structure at future time nodes.

[0112] Figure 8 The business object demand quantity prediction provided by the illustrated embodiment can be used to execute the present specification Figures 1 to 7 The technical solution of the method embodiment shown, its implementation principle and technical effect can be further referred to the relevant description in the method embodiment.

[0113] Optionally, the device further comprises a training module, the training module being used to obtain the reconciliation model by training multiple samples based on the hierarchical structure, and specifically being used to predict a first value of the consumption object processed by the target processing flow at a second time node according to the features generated by the consumption object processed by the target processing flow at a first time node, and to collect a second value of the consumption object processed by the target processing flow at the second time node; According to the characteristics of at least one other processing flow processing the consumable object at the first time node, predicting a third value of the consumable object processed by the other processing flow at the second time node, and collecting a fourth value of the consumable object processed by the other processing flow at the second time node; Combining the third value and the first value to form a prediction vector according to the hierarchical structure; Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Resetting the prediction vector using the relationship matrix to obtain a reset prediction vector; the reset prediction vector includes a reset third value and a reset first value; combining the second value and the fourth value to obtain a tag sequence; A loss function is calculated according to the reset prediction vector and the label sequence, and a weight parameter in the relationship matrix is ​​adjusted according to the loss function until the loss function converges.

[0114] Optionally, the training module is further used to set initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Using the relationship matrix as an encoder and the transformation matrix as a decoder to obtain a deep learning framework; The reconciliation model is obtained based on the deep learning framework training.

[0115] Optionally, the sequence input module includes: The node determination submodule is used to respond to the user's setting instructions and determine the predicted future time node; A network setting submodule, used to set a basic prediction network, set a prediction output structure of the basic prediction network according to the future time node, and set a retrospective output structure of the basic prediction network according to the node information in the historical time node sequence; A first output submodule, configured to input the historical time node sequence into any basic prediction network, output first prediction information according to the prediction output structure, and output first review information according to the review output structure; A second output submodule is used to input the first review information into a basic prediction network corresponding to the next dimension feature, output second prediction information and second review information, and input the second review information into the next basic prediction network until the outputs of all basic prediction networks are obtained; The fusion submodule is used to combine the prediction information output by each basic prediction network to obtain the predicted number of business objects processed by the target processing flow at at least one future time node after the current time.

[0116] Optionally, the device further comprises: A residual calculation submodule, used for calculating the difference between the output review information and the input for each basic prediction network, and obtaining the residual information filtered by the basic prediction network; The combining submodule is specifically used to combine the prediction information output by each basic prediction network and the residual information when the number of basic prediction networks is less than a preset number, so as to obtain the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time.

[0117] Optionally, the single-process prediction model includes multiple basic prediction networks; the basic prediction network includes a time-related feature extraction network layer and a non-time-related feature extraction network layer; the time-related feature extraction network layer is constructed based on LSTM results, and the non-time-related feature extraction network layer is constructed based on MLP.

[0118] Optionally, the first output submodule includes: A calculation submodule, used to calculate the historical time node sequence through the time-related feature extraction network layer, and screen the first latent variable feature associated with time in the historical time node sequence; A feature extraction submodule is used to enable a non-time-related feature extraction network layer to filter out a second latent variable feature that is not related to time in the historical time node sequence when there is a time-independent feature in the historical time node sequence; a time-independent feature refers to a feature that has no linear relationship with time; A strengthening submodule, used for strengthening the first latent variable feature to obtain a first strengthened feature, and strengthening the second latent variable feature to obtain a second strengthened feature; A fusion submodule, configured to obtain a feature after fusing the first enhanced feature and the second enhanced feature, and calculate the first prediction information and the first review information. Regarding the various modules / units included in the various devices described in the above embodiments, they can be software modules / units, or hardware modules / units, or they can be partially software modules / units and partially hardware modules / units. For example, for various devices applied to or integrated in a chip, the various modules / units included therein can all be implemented in the form of hardware such as circuits, or at least some of the modules / units can be implemented in the form of software programs, which run on a processor integrated inside the chip, and the remaining modules / units can be implemented in the form of hardware such as circuits; for various devices applied to or integrated in a chip module, the various modules / units included therein can all be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component of the chip module (such as a chip, circuit module, etc.) or in different components, or at least some of the modules / units can be implemented in the form of software programs. The software program may be implemented in the form of a program, which runs on a processor integrated in the chip module, and the remaining modules / units may be implemented in the form of hardware such as circuits; for each device applied to or integrated in the electronic terminal device, each module / unit contained therein may be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (for example, a chip, a circuit module, etc.) or in different components in the electronic terminal device, or at least some modules / units may be implemented in the form of a software program, which runs on a processor integrated in the electronic terminal device, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits.

[0119] Fig. 9 The schematic diagram of the structure of a computing device provided in an embodiment of the present application, the computing device 900 includes a processor 910, a memory 911, a computer program stored in the memory 911 and executable on the processor 910, the processor 910 implements the steps in the aforementioned method embodiment when executing the program, the computing device provided in the embodiment can be used to execute the technical solution of the method embodiment shown above, and its implementation principle and technical effect can be further referred to the method In one embodiment of the present application, a computer storage medium is provided, on which a program is stored. The stored program includes a method that can be loaded by a processor and process any of the above embodiments.

[0120] In one embodiment of the present application, a computer storage medium is provided. The storage medium is a readable storage medium. A program is stored on the storage medium. The stored program includes a method that can be loaded by a processor and process any of the above embodiments.

[0121] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer storage medium, and the storage medium can include: a read-only memory, a random access memory, a disk, an optical disk, a hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, a disk, an optical disk, a flash disk or a mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0122] The above specific examples are used to illustrate the present application, which is only used to help understand the present application and is not intended to limit the present application. For technicians in the technical field to which the present application belongs, they can also make some simple deductions, deformations or substitutions based on the ideas of the present application.

Claims

1. A method for predicting the quantity of business object demand, characterized in that: The method comprises: According to the influence of the result of executing a specific processing flow on the results of other different processing flows, a hierarchical structure formed by each processing flow in the target scenario is established; the processing flow is the process of running a predetermined program to process a business object; the target scenario includes multiple processing flows; In response to a selection instruction input by a user, selecting a target processing flow among the multiple processing flows; The target collection process processes the multi-dimensional related features generated by the business object before the current time node to form a historical time node sequence; the multi-dimensional related features include but are not limited to the number of business objects, resources consumed in processing the business objects, time information, and scene environment; Inputting the historical time node sequence into a pre-established single-process prediction model, and predicting the predicted number of business objects processed by the target processing flow at at least one future time node after the current time by the single-process prediction model as basic prediction information; The predicted quantity is input into a reconciliation model pre-trained based on the hierarchical structure, and the predicted quantity is adjusted by the reconciliation model with the inferred quantity to obtain the reconciliation prediction vector; the inferred quantity is the quantity of consumption objects processed by other processing flows in the hierarchical structure at future time nodes.

2. The method according to claim 1, characterized in that The method further includes: using multiple samples to train the reconciliation model based on the hierarchical structure; and the process of using a single sample to train the reconciliation model includes: According to the characteristics generated by the target processing flow processing the consumable object at the first time node, predicting the first value of the target processing flow processing the consumable object at the second time node, and collecting the second value of the target processing flow processing the consumable object at the second time node; According to the characteristics of at least one other processing flow processing the consumable object at the first time node, predicting a third value of the consumable object processed by the other processing flow at the second time node, and collecting a fourth value of the consumable object processed by the other processing flow at the second time node; Combining the third value and the first value to form a prediction vector according to the hierarchical structure; Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Resetting the prediction vector using the relationship matrix to obtain a reset prediction vector; the reset prediction vector includes a reset third value and a reset first value; combining the second value and the fourth value to obtain a tag sequence; A loss function is calculated according to the reset prediction vector and the label sequence, and a weight parameter in the relationship matrix is ​​adjusted according to the loss function until the loss function converges.

3. The method according to claim 2, characterized in that Calculating the loss function according to the reset prediction vector and the tag sequence includes: （1） in, is the reset prediction vector, is a tag sequence; is the sequence of the first value after reset corresponding to the second time node, is the sequence of the third value after reset corresponding to the second time node, represents the loss function, Indicates the sequence of samples currently used, is the total number of samples.

4. The method according to claim 1, characterized in that: The reconciliation model is obtained based on the hierarchical structure training; The process of training the reconciliation model includes: Setting initial weights for different processing flows and different time nodes in the hierarchical structure to construct a relationship matrix; Using the relationship matrix as an encoder and the transformation matrix as a decoder to obtain a deep learning framework; The reconciliation model is obtained based on the deep learning framework training.

5. The method according to claim 1, characterized in that Inputting the historical time node sequence into a pre-established single-process prediction model, and predicting the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time by the single-process prediction model, including: Responding to the user's setting instructions, determining the predicted future time node; Setting a basic prediction network, setting a prediction output structure of the basic prediction network according to the future time node, and setting a retrospective output structure of the basic prediction network according to the node information in the historical time node sequence; Inputting the historical time node sequence into any basic prediction network, outputting first prediction information according to the prediction output structure, and outputting first review information according to the review output structure; Inputting the first review information into a basic prediction network corresponding to the next dimension feature, outputting second prediction information and second review information, and inputting the second review information into the next basic prediction network until the outputs of all basic prediction networks are obtained; Combined with the prediction information output by each basic prediction network, the predicted number of business objects processed by the target processing flow at at least one future time node after the current time is obtained.

6. The method according to claim 5, characterized in that The method further comprises: For each basic prediction network, the difference between the output review information and the input is calculated to obtain the residual information filtered by the basic prediction network; Combine the prediction information output by each basic prediction network, including: When the number of basic prediction networks is less than a preset number, the prediction information output by each basic prediction network and the residual information are combined to obtain the predicted number of consumption objects processed by the target processing flow at at least one future time node after the current time.

7. The method according to claim 5, characterized in that The single-process prediction model includes multiple basic prediction networks; the basic prediction network includes a time-related feature extraction network layer and a non-time-related feature extraction network layer; the time-related feature extraction network layer is constructed based on LSTM results, and the non-time-related feature extraction network layer is constructed based on MLP.

8. The method according to claim 7, characterized in that Inputting the historical time node sequence into any basic prediction network, outputting first prediction information according to the prediction output structure, and outputting first review information according to the review output structure, including: The historical time node sequence is calculated by the time-related feature extraction network layer to filter the first latent variable feature associated with time in the historical time node sequence; When there are time-independent features in the historical time node sequence, the non-time-related feature extraction network layer is enabled to filter the second latent variable features in the historical time node sequence that are not related to time; the time-independent features refer to features that have no linear relationship with time; Strengthening the first latent variable feature to obtain a first strengthened feature, and strengthening the second latent variable feature to obtain a second strengthened feature; The first prediction information and the first review information are calculated by fusing the first enhanced feature and the second enhanced feature to obtain the feature.

9. A business object demand quantity prediction device, characterized in that: Set in an electronic device, the device includes: A hierarchical structure establishment module is used to establish a hierarchical structure formed by various processing flows in a target scenario according to the impact of the results of executing a specific processing flow on the results of other different processing flows; the processing flow is the process of running a predetermined program to process a business object; the target scenario includes multiple processing flows; A process selection module, configured to respond to a selection instruction input by a user and select a target process from among the multiple process flows; A feature collection module is used to collect multi-dimensional related features generated by the target processing flow processing the business object before the current time node to form a historical time node sequence; the multi-dimensional related features include but are not limited to the number of business objects, resources consumed in processing business objects, time information, and scene environment; A sequence input module, used to input the historical time node sequence into a pre-established single-process prediction model, and predict the predicted number of business objects processed by the target processing flow at at least one future time node after the current time through the single-process prediction model as basic prediction information; The adjustment module is used to input the predicted quantity into a reconciliation model pre-trained based on the hierarchical structure, and adjust the predicted quantity with the inferred quantity through the reconciliation model to obtain the reconciliation prediction vector; the inferred quantity is the quantity of consumption objects processed by other processing flows in the hierarchical structure at future time nodes.

10. A computing device, characterized in that The method comprises at least one processor and at least one memory connected to the processor; the memory stores program instructions for being executed by the processor, and the processor calls the program instructions to execute any method according to claims 1 to 8.

Images