Information mining method and artificial intelligence system based on RPA

Abstract

The present invention provides an information mining method and artificial intelligence system based on RPA, which relates to the field of artificial intelligence technology, realizes the accurate assessment and processing of the abnormal risks of each scheduled node during the process execution process, first obtains the process execution path diagram and the scheduling path diagram of the process to be detected and each scheduled node, and then uses the optimized target RPA abnormal diagnosis model to perform abnormal diagnosis and generate the abnormal risk index of each node. Then, the present invention arranges the nodes in order by risk index, generates risk weights, and then matches the corresponding abnormal processing knowledge base for the process to be detected according to the risk weights, which not only improves the accuracy and efficiency of abnormal detection, but also realizes the accurate matching of abnormal processing knowledge, significantly enhances the stability and reliability of the RPA process, reduces the operation and maintenance costs, and improves the automation level and response speed of business processing.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an information mining method and an artificial intelligence system based on RPA. Background Art

[0002] With the acceleration of digital transformation of enterprises, Robotic Process Automation (RPA) technology has been widely used in various industries because of its ability to efficiently perform highly repetitive tasks with clear rules. RPA technology realizes the automation of business processes by simulating human operations, greatly improving work efficiency and accuracy. However, during the execution of RPA processes, due to various factors such as system errors, data anomalies, process design defects, etc., the process is often interrupted or the execution results are inaccurate, causing unnecessary losses and risks to the enterprise.

[0003] Traditional RPA exception handling methods often rely on manual monitoring and intervention, which is not only inefficient, but also difficult to fully cover all possible exception situations. In addition, with the continuous deepening of RPA applications, the complexity of processes and the amount of data are also increasing, making exception handling more difficult. Therefore, how to effectively mine abnormal information in RPA processes and diagnose and handle them in a timely and accurate manner has become a problem that needs to be solved urgently. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides an information mining method based on RPA, the method comprising:

[0005] Obtaining a first process execution path diagram of the process to be detected and a first scheduling path diagram of each scheduled node of the process to be detected;

[0006] Using the target RPA abnormality diagnosis model, abnormality diagnosis is performed according to the first process execution path diagram and the first scheduling path diagram to generate an abnormality risk index of each scheduled node, wherein the target RPA abnormality diagnosis model is generated by optimizing model parameters according to the second process execution path diagram of the first training process example and the second scheduling path diagram of each sample scheduled node of the first training process example;

[0007] Arranging the scheduled nodes in order based on the abnormal risk index of each scheduled node, and generating a risk weight of each scheduled node;

[0008] According to the risk weight, the exception handling knowledge base corresponding to each scheduled node is matched to the process to be detected.

[0009] For example, in a possible implementation of the first aspect, the step of obtaining a first process execution path graph of the process to be detected and a first scheduling path graph of each scheduled node of the process to be detected includes:

[0010] Determine the scope and boundary of the process to be detected based on a pre-configured process definition document to determine the starting node and the ending node of the process to be detected, as well as all candidate nodes and paths included in the process to be detected;

[0011] Initialize the process execution monitoring system and configure monitoring parameters, including node execution time, node execution status, and data content transmitted between nodes;

[0012] According to the process definition document, a unique identifier is assigned to each node in the process execution monitoring system, and a mapping relationship between each node and the corresponding identifier is established. After the node identifier mapping table is generated, the process to be detected is started, and the data collection function of the process execution monitoring system is turned on to obtain the process execution data collected in real time, and according to the configured monitoring parameters, the execution status, execution time, input data, output data and abnormal information of each node in the process to be detected are recorded in real time according to the real-time collected process execution data to generate a node execution record data set;

[0013] During the process execution, for each node execution completion event, the execution record is associated with the corresponding node according to the node identifier mapping table to generate an associated node execution record, and an initial execution path graph of the process to be detected is constructed in chronological order according to the associated node execution record, wherein the initial execution path graph includes the execution order and connection relationship of all nodes in the process to be detected;

[0014] Verifying the initial execution path graph to check whether there are isolated nodes or incoherent paths in the initial execution path graph and making corrections, outputting the verified process execution path graph, and formatting the verified process execution path graph to generate a first process execution path graph that meets the set format requirements, wherein the first process execution path graph graphically and intuitively displays the relationship between the execution path and nodes of the process to be detected;

[0015] Analyze the first process execution path graph to identify all scheduled nodes, and generate a scheduled node list, wherein the scheduled node list includes all nodes that have been executed in the process to be detected;

[0016] For each scheduled node, extracting a node scheduling information data set of the scheduled node according to the execution record of the scheduled node in the node execution record data set, wherein the node scheduling information data set includes a scheduling time, a scheduling condition, and a scheduling result;

[0017] According to the node scheduling information data set, construct an initial scheduling path graph of each scheduled node, wherein the initial scheduling path graph shows the scheduling process and result of each scheduled node;

[0018] The initial scheduling path diagram is verified and corrected, the verified scheduling path diagram is output, and the verified scheduling path diagram is formatted to generate a first scheduling path diagram that meets the set format requirements. The first scheduling path diagram graphically displays the scheduling path and scheduling results of each scheduled node.

[0019] On the other hand, an embodiment of the present invention further provides an artificial intelligence system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0020] Based on the above aspects, the embodiment of the present application realizes the accurate assessment and processing of the abnormal risk of each scheduled node during the process execution. First, the process execution path diagram and the scheduling path diagram of the process to be detected and each of its scheduled nodes are obtained, and then the abnormal diagnosis is performed using the optimized target RPA abnormal diagnosis model to generate the abnormal risk index of each node. The present invention arranges the nodes in order by risk index, generates risk weights, and then matches the corresponding abnormal handling knowledge base for the process to be detected according to the risk weights, which not only improves the accuracy and efficiency of abnormal detection, but also realizes the accurate matching of abnormal handling knowledge, significantly enhances the stability and reliability of the RPA process, reduces operation and maintenance costs, and improves the automation level and response speed of business processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the execution flow of the RPA-based information mining method provided in an embodiment of the present invention.

[0022] Figure 2 Schematic diagram of the hardware architecture of the artificial intelligence system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The figure is a flow chart of an RPA-based information mining method provided by an embodiment of the present invention. The RPA-based information mining method is introduced in detail below.

[0024] Step S110, obtaining a first process execution path graph of the process to be detected and a first scheduling path graph of each scheduled node of the process to be detected.

[0025] In this embodiment, the server can detect the order processing flow of the enterprise. This order processing flow involves multiple departments and operation links and is a complex business process.

[0026] First, the server determines the scope and boundaries of the process to be tested based on the pre-configured process definition document. In the order processing process, the starting node is the customer placing an order, and the ending node is the order completion and shipment. The candidate nodes included in this process include order receipt, order review, inventory check, payment processing, order sorting, packaging, logistics arrangement, etc. The path is the possible flow sequence between these nodes, such as order review after order receipt, and inventory check after review.

[0027] Then, the server initializes the process execution monitoring system and configures the monitoring parameters. For the node execution time, it is set as the time period from the start of the node processing to the completion of the processing; the node execution status includes normal execution, execution suspension, execution failure, etc.; the data content transmitted between nodes is recorded in detail, such as order information (including product name, quantity, customer address, etc.).

[0028] Next, according to the process definition document, a unique identifier is assigned to each node in the process execution monitoring system. For example, the order receiving node is assigned the identifier "Order_Receive_001", and a mapping relationship between each node and the corresponding identifier is established to generate a node identifier mapping table. Then, the order processing process to be tested is started, and the data collection function of the process execution monitoring system is turned on.

[0029] During the execution of the order processing process, the server collects process execution data in real time. For example, when the order receiving node starts to execute, the server records the execution status (start execution), execution time (accurate to seconds), input data (order details from the customer), output data (order information after format conversion for subsequent processes) and abnormal information (if any, such as a temporary interruption of the network connection causing a delay in receiving data) of the node according to the configured monitoring parameters, and generates a node execution record data set.

[0030] When each node is executed, the execution record is associated with the corresponding node according to the node identifier mapping table. For example, after the order review node is executed, its execution record is associated with the identifier "Order_Review_002" to generate the associated node execution record. The initial execution path diagram of the process to be tested is constructed in chronological order. This initial execution path diagram contains the execution order and connection relationship of all nodes in the order processing process, such as connecting to order review after order receipt, and connecting to inventory inspection after order review is passed.

[0031] After that, the server verifies the initial execution path diagram. Assuming that the inventory check node is found to be isolated in the initial execution path diagram, it may be because some connection relationships are not configured correctly after the process definition document is updated. The server makes corrections, outputs the verified process execution path diagram, and formats it to generate a first process execution path diagram that meets the set format requirements. This first process execution path diagram graphically displays the execution path of the order processing process and the relationship between nodes. For example, it can be represented by a directed graph, with nodes represented by circles and connection relationships represented by arrows.

[0032] Next, the server analyzes the first process execution path diagram to identify all scheduled nodes and generates a scheduled node list. In the order processing process, nodes that have been executed, such as order receipt, order review, etc., are all in the scheduled node list.

[0033] For each scheduled node, extract the node scheduling information dataset of the scheduled node based on the execution record of the scheduled node in the node execution record dataset. Taking the order review node as an example, the scheduling time is a certain moment after the order is received, the scheduling condition is that the order amount does not require additional approval within a certain range, and the scheduling result is approval or failure. Based on these node scheduling information datasets, construct the initial scheduling path diagram of each scheduled node. For example, the initial scheduling path diagram of the order review node shows the process from starting the scheduling review to obtaining the review result.

[0034] Finally, the initial scheduling path diagram is verified and corrected, and the verified scheduling path diagram is output, and the verified scheduling path diagram is formatted to generate a first scheduling path diagram that meets the set format requirements. This first scheduling path diagram graphically displays the scheduling path and scheduling results of each scheduled node. For example, the first scheduling path diagram of the order review node clearly shows the conditions for review and the flow of the review results.

[0035] Step S120, using the target RPA abnormality diagnosis model, perform abnormality diagnosis according to the first process execution path diagram and the first scheduling path diagram, and generate the abnormality risk index of each scheduled node. The target RPA abnormality diagnosis model is generated by optimizing model parameters based on the second process execution path diagram of the first training process sample and the second scheduling path diagram of each sample scheduled node of the first training process sample.

[0036] In this embodiment, it is assumed that the server has built a target RPA anomaly diagnosis model for order processing process anomaly diagnosis. This target RPA anomaly diagnosis model is obtained by training with a large number of first training process examples.

[0037] The target RPA anomaly diagnosis model starts to perform anomaly diagnosis for the first process execution path diagram of the order processing process and the first scheduling path diagram of each scheduled node. Taking the inventory check node as an example, the model will analyze the front and back connection relationship of the inventory check node in the first process execution path diagram, as well as the scheduling conditions, scheduling time and other factors of the node in the first scheduling path diagram.

[0038] When training the target RPA anomaly diagnosis model, assume that the first training process example is some historical order processing process data. The model performs learning and parameter optimization for the second process execution path diagram in these examples and the second scheduling path diagram of the scheduled nodes of each example. For example, in the historical order processing, the inventory check node once had a situation where the inventory quantity did not match the order quantity. The corresponding second process execution path diagram shows the execution status of the entire process when this anomaly occurred, and the second scheduling path diagram of the inventory check node shows the scheduling status at that time, such as whether the database connection for inventory query is normal.

[0039] Back to the current order processing process diagnosis, the target RPA anomaly diagnosis model comprehensively determines the abnormal risk index of the inventory check node based on its position in the first process execution path diagram (such as whether it is on the critical path) and the scheduling situation in the first scheduling path diagram (such as whether the response time for inventory query is too long). If the server where the inventory check node is located has been overloaded recently, causing the inventory query response time to be much longer than normal, the model will determine that the abnormal risk index of the node is high based on similar situations learned previously (from the first training process sample).

[0040] For the order review node, the model will also analyze the relationship between its predecessor nodes (such as whether the data transmitted by the order receiving node is complete and accurate) and successor nodes (such as whether the inventory inspection node depends on the review results) in the first process execution path diagram, as well as the review rules in the first scheduling path diagram (such as determining the review process based on the customer's credit rating and order amount) and other factors. If new fraud methods are found to have appeared recently, which may affect the accuracy of order review, the model will improve the abnormal risk index of the order review node based on the knowledge related to fraud response learned from the first training process examples.

[0041] By performing similar analysis on each scheduled node (such as order receipt, order review, inventory inspection, payment processing, etc.), the target RPA anomaly diagnosis model ultimately generates an anomaly risk index for each scheduled node.

[0042] Step S130: arranging the scheduled nodes in order based on the abnormal risk index of the scheduled nodes to generate risk weights of the scheduled nodes.

[0043] In this embodiment, after obtaining the abnormal risk index of each scheduled node in the order processing flow, the server starts to sort them in order to generate risk weights.

[0044] Assume that the abnormal risk index of the inventory check node is 0.8, the abnormal risk index of the order review node is 0.6, and the abnormal risk index of the order receiving node is 0.4. The server first obtains the cross-call weight of each scheduled node. In the order processing process, the inventory check node may be relied on by multiple subsequent nodes (such as order sorting, packaging, etc.), and its cross-call weight is relatively high, assuming it is 0.9; the result of the order review node mainly affects the inventory check and payment processing, and the cross-call weight is set to 0.7; although the order receiving node is the starting node, it has fewer dependencies on subsequent nodes, and the cross-call weight is 0.5.

[0045] For the inventory inspection node, based on its abnormal risk index of 0.8 and the cross-call weight of 0.9, the risk weight is calculated to be 0.8 * 0.9 = 0.72. For the order review node, the risk weight is 0.6 * 0.7 = 0.42. For the order receiving node, the risk weight is 0.4 * 0.5 = 0.2.

[0046] The scheduled nodes are arranged in order according to the risk weight. The inventory check node has the highest risk weight and is ranked first, followed by the order review node, and finally the order receiving node. In this way, the risk weight of each scheduled node is generated. The higher the risk weight, the greater the impact on the process when an abnormality occurs in the entire order processing process.

[0047] Step S140: Match the exception handling knowledge base corresponding to each scheduled node to the process to be detected according to the risk weight.

[0048] In this embodiment, the corresponding exception handling knowledge base is matched to the order processing flow according to the risk weight of each scheduled node generated previously.

[0049] Taking the inventory check node as an example, since it has the highest risk weight, the server first obtains the node index attribute of the inventory check node. This node index attribute may contain information such as the node name, the module it is in, and the relevant database table. Based on this node index attribute, the server finds the corresponding exception handling knowledge base. This knowledge base may contain processing strategies when the inventory quantity is insufficient (such as transferring goods from other warehouses, notifying customers to delay delivery, etc.), and emergency measures when the inventory query system fails (such as switching to an alternative query system, etc.).

[0050] According to the high risk weight of the inventory check node, the server matches this exception handling knowledge base to the order processing flow to be tested. If an exception occurs in the inventory check node during the subsequent order processing, the corresponding processing solution can be directly obtained from this matched exception handling knowledge base.

[0051] For the order review node, its risk weight is low. The server obtains the node index attribute of the order review node and finds the corresponding exception handling knowledge base, which may include how to ensure a smooth transition between the old and new rules when the review rules are updated, and how to solve problems when the reviewer's authority has problems. Then, according to its risk weight, this knowledge base is matched to the order processing flow.

[0052] Similarly, for order receiving nodes, the exception handling knowledge base (such as the strategy for re-receiving orders when the network connection is interrupted) is found according to its node index attribute, and matched to the order processing flow according to the risk weight. In this way, when any scheduled node in the order processing flow has an exception, an effective processing solution can be quickly obtained from the matching exception handling knowledge base, improving the stability and reliability of the order processing flow.

[0053] Based on the above steps, the embodiment of the present application realizes the accurate assessment and processing of the abnormal risk of each scheduled node during the process execution. First, the process execution path diagram and the scheduling path diagram of the process to be detected and each of its scheduled nodes are obtained, and then the optimized target RPA abnormal diagnosis model is used to perform abnormal diagnosis and generate the abnormal risk index of each node. The present invention arranges the nodes in order by risk index, generates risk weights, and then matches the corresponding abnormal handling knowledge base for the process to be detected according to the risk weights, which not only improves the accuracy and efficiency of abnormal detection, but also realizes the accurate matching of abnormal handling knowledge, significantly enhances the stability and reliability of the RPA process, reduces operation and maintenance costs, and improves the automation level and response speed of business processing.

[0054] For example, in a possible implementation, step S110 includes:

[0055] Step S111 : determining the scope and boundary of the process to be detected based on a pre-configured process definition document, so as to determine the starting node and the ending node of the process to be detected, and all candidate nodes and paths included in the process to be detected.

[0056] Step S112, initializing the process execution monitoring system and configuring monitoring parameters, wherein the monitoring parameters include node execution time, node execution status, and data content transmitted between nodes.

[0057] Step S113, according to the process definition document, assign a unique identifier to each node in the process execution monitoring system, and establish a mapping relationship between each node and the corresponding identifier. After generating a node identifier mapping table, start the process to be detected, and turn on the data collection function of the process execution monitoring system to obtain real-time collected process execution data, and according to the configured monitoring parameters, record the execution status, execution time, input data, output data and abnormal information of each node in the process to be detected in real time according to the real-time collected process execution data to generate a node execution record data set.

[0058] Step S114, during the process execution, for the execution completion event of each node, the execution record is associated with the corresponding node according to the node identifier mapping table to generate an associated node execution record, and based on the associated node execution record, an initial execution path graph of the process to be detected is constructed in chronological order, and the initial execution path graph includes the execution order and connection relationship of all nodes in the process to be detected.

[0059] Step S115, verify the initial execution path graph to check whether there are isolated nodes or incoherent paths in the initial execution path graph and make corrections, output the verified process execution path graph, and format the verified process execution path graph to generate a first process execution path graph that meets the set format requirements. The first process execution path graph graphically displays the relationship between the execution path and nodes of the process to be tested.

[0060] Step S116, analyzing the first process execution path graph to identify all scheduled nodes, and generating a scheduled node list, wherein the scheduled node list includes all nodes that have been executed in the process to be detected.

[0061] Step S117, for each scheduled node, extract the node scheduling information data set of the scheduled node according to the execution record of the scheduled node in the node execution record data set, wherein the node scheduling information data set includes the scheduling time, scheduling conditions, and scheduling results.

[0062] Step S118, constructing an initial scheduling path graph for each scheduled node based on the node scheduling information data set, wherein the initial scheduling path graph shows the scheduling process and results of each scheduled node.

[0063] Step S119, verify and correct the initial scheduling path diagram, output the verified scheduling path diagram, and format the verified scheduling path diagram to generate a first scheduling path diagram that meets the set format requirements. The first scheduling path diagram graphically displays the scheduling path and scheduling results of each scheduled node.

[0064] In this embodiment, taking the supply chain management process of an enterprise as an example, first, the scope and boundaries of the supply chain management process are determined based on the pre-configured process definition document. In this process, the starting node is the initiation of the raw material procurement demand, which may be determined by the production department according to the production plan, and the end node is the delivery of the product to the customer. There are many candidate nodes included in this process, such as supplier selection, purchase order issuance, raw material transportation, warehouse reception, inventory management, production scheduling, production processing, quality inspection, finished product warehousing, order processing, product transportation, etc. The paths between these nodes are also relatively complex. For example, a purchase order can only be issued after the supplier selection is completed, and the warehouse can only receive the raw materials after they are transported to the warehouse.

[0065] Next, the server initializes the process execution monitoring system and configures the monitoring parameters. For the node execution time, the specific time from the start of processing to the completion of processing for each node is accurately recorded, which helps to analyze the efficiency of the process. The node execution status includes a variety of situations. For example, normal execution means that the node operates smoothly as expected. The execution suspension may be due to some external conditions (such as waiting for approval from other departments) that cause the operation to be temporarily interrupted. Failed execution means that the node encountered an insurmountable error during the processing. The content of the data transmitted between nodes is also very critical. In the supply chain management process, the purchase order issuance node transmits purchase order data to suppliers including detailed information such as raw material specifications, quantity, price, and delivery time.

[0066] Then, according to the process definition document, a unique identifier is assigned to each node in the process execution monitoring system. For example, the supplier selection node is assigned the identifier "Supplier_Selection_001", and a mapping relationship between each node and the corresponding identifier is established to generate a node identifier mapping table. After that, the supply chain management process is started, and the data collection function of the process execution monitoring system is enabled. During the process execution, the server collects process execution data in real time. Taking the warehouse receiving node as an example, when it starts to execute, the server records the execution status of the node (such as starting to receive raw materials), execution time (accurate to minutes), input data (transportation list and actual goods situation from the raw material transportation node), output data (inventory update information after warehouse reception) and abnormal information (such as damage to raw materials) in real time according to the configured monitoring parameters, thereby generating a node execution record data set.

[0067] When each node is executed, the execution record is associated with the corresponding node according to the node identifier mapping table to generate an associated node execution record. For example, after the production processing node is executed, its execution record is associated with the identifier "Production_Processing_002". Based on these associated node execution records, the initial execution path diagram of the supply chain management process is constructed in chronological order. This initial execution path diagram covers the execution order and connection relationship of all nodes, such as the connection to supplier selection after the raw material procurement demand is initiated, and the connection to purchase order placement after supplier selection.

[0068] Afterwards, the server verifies the initial execution path diagram. In the supply chain management process, problems with the initial execution path diagram may occur due to process updates or system failures. For example, if it is found that there is a path incoherence in the connection relationship between the production scheduling node and the quality inspection node in the initial execution path diagram, it may be because some process relationships were not correctly updated after the implementation of the new quality control process. The server will correct this situation, output the verified process execution path diagram, and format it to generate a first process execution path diagram that meets the set format requirements. This first process execution path diagram graphically displays the relationship between the execution path and nodes of the supply chain management process, for example, using nodes to represent each process link and arrows to represent the direction of the process.

[0069] Next, the server analyzes the first process execution path graph to identify all scheduled nodes and generate a list of scheduled nodes. In the supply chain management process, nodes that have been executed, such as supplier selection, purchase order placement, etc., will be included in the list of scheduled nodes.

[0070] For each scheduled node, the node scheduling information dataset of the scheduled node is extracted according to the execution record of the scheduled node in the node execution record dataset. Taking the purchase order placement node as an example, its scheduling time is a specific moment after the supplier selection is completed. The scheduling condition is that the supplier's quotation and delivery time meet the enterprise requirements. The scheduling result is that the purchase order is successfully placed or the order fails to be placed due to some reasons (such as the supplier rejecting the order terms). Based on these node scheduling information datasets, the initial scheduling path diagram of each scheduled node is constructed. For example, the initial scheduling path diagram of the purchase order placement node shows the entire scheduling process and results from the beginning of supplier evaluation to the final order placement.

[0071] Finally, the initial scheduling path diagram is verified and corrected. In the initial scheduling path diagram of the purchase order placement node, there may be inaccurate scheduling time records or unclear scheduling condition logic. The server will verify and correct it, output the verified scheduling path diagram, and format the verified scheduling path diagram to generate the first scheduling path diagram that meets the set format requirements. This first scheduling path diagram graphically displays the scheduling path and scheduling results of each scheduled node, making the scheduling status of each scheduled node in the supply chain management process clear and traceable, providing accurate basic data for subsequent abnormal diagnosis and other operations.

[0072] In a possible implementation, the step of training the target RPA abnormality diagnosis model includes:

[0073] Step A110, obtaining the priori abnormal risk index of the first training process sample.

[0074] Step A120, using the initialized RPA anomaly diagnosis model, anomaly diagnosis is performed based on the second process execution path diagram of the first training process sample and the second scheduling path diagram of each sample scheduled node of the first training process sample, and anomaly risk indexes of each sample scheduled node are generated.

[0075] Step A130, using the initialized RPA anomaly diagnosis model, performs a fusion calculation on the anomaly risk index of the scheduled nodes of each sample to generate a third anomaly risk index of the first training process sample.

[0076] Step A140: determining an abnormality diagnosis error parameter of the initialized RPA abnormality diagnosis model according to the prior abnormality risk index and the third abnormality risk index.

[0077] Step A150: Optimize the model parameters of the initialized RPA abnormality diagnosis model according to the abnormality diagnosis error parameter to generate the target RPA abnormality diagnosis model.

[0078] In a possible implementation, the method further includes:

[0079] Step B110, based on the importance distinguishing factor of the initialized RPA abnormal diagnosis model, distinguish the scheduled nodes of each sample to generate a first focus node and a second focus node, wherein the importance distinguishing factor represents the importance degree of each sample scheduled node, and the importance degree of the first focus node is greater than the importance degree of the second focus node.

[0080] Step B120: Obtain a first abnormal risk index of the first training process sample according to the abnormal risk index of the first focus node and each sample scheduled node, and obtain a second abnormal risk index of the first training process sample according to the abnormal risk index of the second focus node and each sample scheduled node.

[0081] Step B130: constructing an abnormal risk difference factor of the initialized RPA abnormal diagnosis model based on the first abnormal risk index and the second abnormal risk index, wherein the abnormal risk difference factor is used to expand the difference between the first abnormal risk index and the second abnormal risk index.

[0082] Step A150 includes: optimizing the model parameters of the initialized RPA abnormal diagnosis model according to the abnormal risk difference factor and the abnormal diagnosis error parameter to generate the target RPA abnormal diagnosis model.

[0083] In a possible implementation, the method further includes:

[0084] Step C110, using the target logic association model, according to the backward focus node of the first training process sample and the scheduled nodes of each sample, determine the fourth abnormal risk index of the first training process sample.

[0085] Step C120, using the fourth abnormality risk index and the priori abnormality risk index, determining a discrete factor of the initialized RPA abnormality diagnosis model.

[0086] The method of optimizing the model parameters of the initialized RPA abnormal diagnosis model based on the abnormal risk difference factor and the abnormal diagnosis error parameter to generate the target RPA abnormal diagnosis model includes: optimizing the model parameters of the initialized RPA abnormal diagnosis model based on the discrete factor, the abnormal risk difference factor and the abnormal diagnosis error parameter to generate the target RPA abnormal diagnosis model.

[0087] In this embodiment, when training the target RPA anomaly diagnosis model, multiple steps are involved, which are described in detail below using the enterprise's supply chain management process as an example.

[0088] The first step is to obtain the prior anomaly risk index of the first training process example. In the training examples of the supply chain management process, the acquisition of the prior anomaly risk index requires a comprehensive consideration of many factors. For the supply chain management process, the prior anomaly risk index may be based on historical data, industry standards, and expert experience. For example, in historical data, the interruption frequency of raw material supply, the probability of transportation delays, and the out-of-stock rate in inventory management can all serve as the basis for the prior anomaly risk index. From a more specific perspective, in the raw material supply link, if past records show that a certain raw material often has supply interruptions due to the singleness of the supply source in a specific season, then the prior anomaly risk index of this link will be relatively high. For the transportation link, if a certain transportation route often causes transportation delays due to geographical environment or traffic congestion, it will also increase the prior anomaly risk index of this link. At the same time, industry standards also have an impact on the prior anomaly risk index. For example, in inventory management, the industry-specified safety stock level, if the enterprise is often below this level, then the prior anomaly risk index of the inventory management link will also increase accordingly. Expert experience cannot be ignored either. For example, in the quality inspection process, experienced quality experts can give a relatively reasonable prior abnormal risk index based on the types and frequencies of previous quality problems.

[0089] Next, the initialization RPA anomaly diagnosis model is used to perform anomaly diagnosis based on the second process execution path diagram of the first training process sample and the second scheduling path diagram of each sample scheduled node of the first training process sample, and generate anomaly risk indexes for each sample scheduled node. In the supply chain management process, the second process execution path diagram shows in detail the execution of the supply chain process in the training sample. Taking the purchase order issuance node as an example, in the second process execution path diagram, all pre-operations before the purchase order is issued, such as the results of supplier selection, the negotiation process with the supplier, etc., as well as the impact after the issuance, such as the start of transportation arrangements, etc., can be seen. For the second scheduling path diagram of this node, the scheduling process of the purchase order issuance is shown in detail, including when to decide to issue the order based on the supplier evaluation results, and the internal approval process followed when the order is issued. The initialization RPA anomaly diagnosis model performs analysis based on this information. If incomplete information is found in the supplier selection link during the purchase order issuance process (for example, the supplier's financial status is not fully evaluated), this may lead to an increase in subsequent supply risks. The model will determine the anomaly risk index of the purchase order issuance node based on this situation and similar situations in historical data. Similarly, for production and processing nodes, the rationality of production scheduling, the timeliness of raw material supply and other factors can be seen in the second process execution path diagram, and the scheduling logic of production and processing task allocation can be seen in the second scheduling path diagram. If the production schedule is too tight, it is easy to cause problems such as equipment overload or personnel fatigue during the production and processing process. The model will determine the abnormal risk index of the production and processing node based on these factors. By performing similar analysis on each sample scheduled node in the supply chain management process, the model generates the abnormal risk index of each sample scheduled node.

[0090] Then, the abnormal risk index of each sample scheduled node is fused and calculated by using the initialized RPA abnormal diagnosis model to generate the third abnormal risk index of the first training process sample. In the training of the supply chain management process, the model adopts a specific fusion calculation method for the abnormal risk index of each sample scheduled node. For example, the abnormal risk index of the purchase order issuance node is 0.3, the abnormal risk index of the production and processing node is 0.4, and the abnormal risk index of the inventory management node is 0.2. The model may perform fusion calculation based on the weight of each node in the entire supply chain process. If the weight of the purchase order issuance node in the entire supply chain is 0.2, the weight of the production and processing node is 0.3, and the weight of the inventory management node is 0.1, then the third abnormal risk index may be obtained by the following calculation: (0.3 * 0.2 + 0.4 * 0.3 + 0.2* 0.1) = 0.2. This third abnormal risk index comprehensively reflects the abnormal risk situation of the entire first training process sample.

[0091] Afterwards, the abnormal diagnosis error parameter of the initialized RPA abnormal diagnosis model is determined based on the prior abnormal risk index and the third abnormal risk index. Assuming that the prior abnormal risk index is 0.25 and the calculated third abnormal risk index is 0.2, the abnormal diagnosis error parameter can be determined by the difference between the two (0.25 - 0.2 = 0.05) or other appropriate calculation methods (such as the ratio of the absolute value of the difference between the two and the prior abnormal risk index). This abnormal diagnosis error parameter reflects the diagnostic error of the initialized RPA abnormal diagnosis model in the current training process sample.

[0092] Next, according to the importance distinguishing factor of the initialized RPA anomaly diagnosis model, the scheduled nodes of each sample are distinguished to generate the first node of interest and the second node of interest. In the supply chain management process, the importance distinguishing factor is determined according to the criticality of the node in the entire supply chain. For example, for the raw material supply node and the finished product transportation node, since they are directly related to the source supply and final delivery of the supply chain, their importance distinguishing factors are high and they may be determined as the first node of interest. Some auxiliary nodes, such as the office supplies procurement node (although it is also part of the supply chain management process, but the impact on the overall supply chain is relatively small), have a lower importance distinguishing factor and are determined as the second node of interest. According to the first node of interest and the abnormal risk index of each sample scheduled node, the first abnormal risk index of the first training process sample is obtained. For example, if the first node of interest includes the raw material supply node and the production and processing node, their abnormal risk indexes are 0.3 and 0.4 respectively, and the first abnormal risk index can be obtained according to a specific calculation method (such as weighted average, etc.). Similarly, according to the abnormal risk index of the second node of interest and the abnormal risk index of each sample scheduled node, the second abnormal risk index of the first training process sample is obtained.

[0093] Then, based on the first abnormal risk index and the second abnormal risk index, the abnormal risk difference factor for initializing the RPA abnormal diagnosis model is constructed. Assuming that the first abnormal risk index is 0.35 and the second abnormal risk index is 0.15, the abnormal risk difference factor may be constructed by the difference between the two (0.35 - 0.15 = 0.2) or other appropriate calculation methods. The role of this difference factor is to expand the difference between the first abnormal risk index and the second abnormal risk index, so as to highlight the impact of important nodes when optimizing the model.

[0094] After that, the target logic association model is used to determine the fourth abnormal risk index of the first training process sample based on the backward focus nodes of the first training process sample and the scheduled nodes of each sample. In the supply chain management process, the backward focus nodes may be those nodes that have a significant impact on the subsequent processes. For example, the quality inspection node is a backward focus node after production and processing, because its results will affect subsequent links such as finished product warehousing and product transportation. The target logic association model analyzes the abnormal risk situations under different circumstances (such as the impact of raw material quality fluctuations on quality inspection results, the impact of production and processing technology stability on quality inspection, etc.) according to the logical relationship between the quality inspection node and other sample scheduled nodes (such as production and processing nodes, raw material supply nodes, etc.), thereby determining the fourth abnormal risk index.

[0095] Next, the fourth abnormal risk index and the prior abnormal risk index are used to determine the discrete factor for initializing the RPA abnormal diagnosis model. Assuming that the fourth abnormal risk index is 0.22 and the prior abnormal risk index is 0.25, the discrete factor can be determined by the difference between the two (0.25 - 0.22 = 0.03) or other appropriate calculation methods.

[0096] Finally, based on the discrete factor, abnormal risk difference factor, and abnormal diagnosis error parameter, the model parameters of the initialized RPA abnormal diagnosis model are optimized to generate the target RPA abnormal diagnosis model. For example, according to a specific optimization algorithm, the discrete factor, abnormal risk difference factor, and abnormal diagnosis error parameter are substituted into the algorithm, and the various parameters in the initialized RPA abnormal diagnosis model (such as node weight parameter, logical judgment threshold, etc.) are adjusted to obtain the optimized target RPA abnormal diagnosis model. This target RPA abnormal diagnosis model can more accurately diagnose the abnormal risk index of the scheduled node in the subsequent supply chain management process abnormal diagnosis.

[0097] In a possible implementation, the method further includes:

[0098] Step D110, using the target logic association model, based on the scheduled nodes of each sample, determine the risk scheduling node of the first training process sample.

[0099] Step D120, using the target logic association model, performs abnormality diagnosis based on the second process execution path diagram and the scheduling path diagram of the risk scheduling node of the first training process sample, and generates a fourth abnormal risk index of the first training process sample.

[0100] Step D130, using the fourth abnormal risk index and the third abnormal risk index, determining a discrete factor of the initialized RPA abnormal diagnosis model.

[0101] Step A150 includes: optimizing the model parameters of the initialized RPA abnormal diagnosis model according to the abnormal diagnosis error parameter and the discrete factor to generate the target RPA abnormal diagnosis model.

[0102] In this embodiment, in the context of the supply chain management process of the enterprise, the target logic association model is used to determine the risk scheduling node of the first training process sample based on the scheduled nodes of each sample. In the supply chain management process, the scheduled nodes include many links, such as supplier selection nodes, purchase order issuance nodes, raw material transportation nodes, warehouse receiving nodes, etc. The target logic association model determines the risk scheduling node according to the role of each node in the entire supply chain and the logical relationship between them. For example, for the raw material transportation node, if it has a close time dependency relationship with multiple subsequent nodes, and there are many uncertain factors in the transportation process, such as the transportation route passes through areas with complex road conditions or the transportation method is susceptible to weather, then this node may be determined as a risk scheduling node. In addition, like the production scheduling node, if its scheduling accuracy is crucial to the smooth progress of the entire production process, once the scheduling is deviated, it will cause subsequent production processing, quality inspection and other links to be seriously affected, and the node will also be regarded as a risk scheduling node.

[0103] Next, the server uses the target logic association model to perform abnormal diagnosis based on the second process execution path diagram and the scheduling path diagram of the risk scheduling node of the first training process sample, and generates the fourth abnormal risk index of the first training process sample. Taking the risk scheduling node of raw material transportation as an example, the second process execution path diagram shows the operation sequence before and after the raw material transportation node in the entire supply chain process and the association with other nodes. Its scheduling path diagram describes the scheduling arrangement of the transportation task in detail, including the selection of transportation tools, the determination of transportation time, and the planning of transportation routes. The target logic association model will analyze this information. If it is found in the scheduling path diagram that the transportation route planning is unreasonable, resulting in too long transportation time, and the second process execution path diagram shows that this situation will affect the normal operation of the warehouse receiving node, thereby affecting inventory management and production scheduling, then the fourth abnormal risk index will be determined based on factors such as the severity of this impact and the probability of occurrence. Assuming that a transportation route planning error may cause the transportation time to be extended by 30% compared to normal, and there is a 50% probability that the inventory replenishment of the warehouse receiving node will be delayed, thereby affecting the production scheduling, according to these data and the algorithm inside the model, the fourth abnormal risk index corresponding to the raw material transportation node can be obtained.

[0104] Then, the server uses the fourth abnormal risk index and the third abnormal risk index to determine the discrete factor for initializing the RPA abnormal diagnosis model. Assume that the third abnormal risk index reflects the comprehensive abnormal risk of the entire supply chain process, with a value of 0.25, and the fourth abnormal risk index of the raw material transportation node is 0.35. The calculation of the discrete factor may be the absolute value of the difference between the two, that is, |0.35 - 0.25| = 0.1. This discrete factor reflects the degree of dispersion between the abnormal risk index of the risk scheduling node and the overall abnormal risk index.

[0105] After that, the model parameters of the initialized RPA abnormal diagnosis model are optimized based on the abnormal diagnosis error parameter and the discrete factor to generate the target RPA abnormal diagnosis model. The abnormal diagnosis error parameter reflects the degree of deviation between the previous diagnosis result of the model and the actual situation, and the discrete factor reflects the difference between a specific node and the whole. When optimizing the initialized RPA abnormal diagnosis model, it is assumed that there are some parameters related to the node weight and the abnormal judgment threshold in the model. If the abnormal diagnosis error parameter is large, it means that the overall diagnostic accuracy of the model needs to be improved, while a large discrete factor indicates that the abnormal situation of the risk scheduling node is significantly different from the whole. For example, for the raw material transportation node, if its discrete factor is large, the weight of this node in the entire model may be increased when optimizing the model parameters, so as to more accurately reflect the impact of its abnormal risk on the entire supply chain in subsequent diagnosis. At the same time, the abnormal judgment threshold is adjusted according to the abnormal diagnosis error parameter, so that the model can more accurately judge the abnormal situation. By comprehensively considering the abnormal diagnosis error parameter and the discrete factor, the various parameters in the model are adjusted to generate the target RPA abnormal diagnosis model, which can more accurately reflect the actual situation in the subsequent abnormal diagnosis of the supply chain management process.

[0106] In a possible implementation, the method further includes:

[0107] Step E110, obtain the actual abnormal risk index of the second training process sample, the third process execution path diagram of the second training process sample, the third scheduling path diagram of each past focus node of the second training process sample, and the backward focus node of the second training process sample.

[0108] Step E120, using the initialized logic association model, based on the third process execution path diagram and the third scheduling path diagram, determine the risk scheduling node of the second training process example.

[0109] Step E130, using the initialized logical association model, anomaly diagnosis is performed based on the third process execution path diagram and the scheduling path diagram of the risk scheduling node of the second training process sample to generate a fifth abnormal risk index of the second training process sample.

[0110] Step E140, optimizing the model parameters of the initialized logical association model according to the error between the actual abnormal risk index and the fifth abnormal risk index to generate the target logical association model.

[0111] In one possible implementation, in the second training process example of the supply chain management process, the actual abnormal risk index may be obtained by statistics and analysis of actual abnormal situations. For example, in a certain period of time, there were actually multiple cases of inventory shortages leading to production interruptions. The actual abnormal risk index was calculated based on the severity of these events, frequency of occurrence and other factors. The third process execution path diagram shows the complete supply chain process execution in this training example, and the third scheduling path diagram of each past focus node describes in detail the scheduling process of these past focus nodes (such as the previously mentioned key supplier selection, production scheduling and other nodes in the supply chain). Backward focus nodes also have an important impact on subsequent supply chain processes. For example, quality inspection nodes are backward focus nodes for product transportation and delivery.

[0112] Next, the initialization logic association model will determine the risk scheduling node in the second training process example based on the logical relationship between the nodes in the supply chain and the importance of each node in the process, just like the analysis of the first training process example. For example, if it is found in the third process execution path diagram that a certain equipment in the production and processing link often fails, and the maintenance scheduling of this equipment is shown to be unreasonable in the third scheduling path diagram, such as the maintenance cycle is too long or the maintenance resources are insufficiently allocated, then the production and processing node may be determined as a risk scheduling node.

[0113] Then, taking the production and processing risk scheduling node as an example, the third process execution path diagram shows the operation sequence before and after the production and processing node and the association with other nodes. Its scheduling path diagram describes in detail the equipment maintenance scheduling, production task allocation scheduling, etc. If it is found in the scheduling path diagram that the equipment maintenance scheduling is unreasonable, resulting in frequent equipment failures, which is shown in the third process execution path diagram to affect the subsequent quality inspection node (because equipment failure may lead to unstable product quality) and the finished product warehousing node (production delays affect the warehousing time), then according to the severity of these impacts, the probability of occurrence and other factors, using the algorithm inside the initialization logic association model, the fifth abnormal risk index corresponding to the production and processing node can be obtained.

[0114] Finally, assuming that the actual abnormal risk index is 0.3, and the calculated fifth abnormal risk index is 0.25, there is a certain error between the two. This error reflects the accuracy of the initialization logic association model in the abnormal diagnosis of the second training process sample. When optimizing the model parameters, if the error is large, it means that the diagnostic ability of the model needs to be improved. For example, there may be parameters related to the weight of the logical relationship between nodes in the model. If the error shows that the model does not accurately judge the logical relationship between the production and processing node and the quality inspection node, the weight of the logical relationship between the two nodes in the model can be adjusted. By adjusting the various parameters in the model, the initialization logic association model can accurately reflect the logical relationship and abnormal risk situation between the nodes in the supply chain management process after multiple such optimization processes, thereby generating a target logic association model. This model can more accurately judge the risk scheduling node and calculate the abnormal risk index in the subsequent abnormal diagnosis of the supply chain management process.

[0115] In a possible implementation, step A110 includes:

[0116] Step A111, obtaining the abnormal thermal coefficient of the backward node scheduling of the first training process sample.

[0117] Step A112, obtaining the backward node guided abnormal thermal coefficient of the first training process example.

[0118] Step A113, determining the a priori abnormal risk index of the first training process example according to the abnormal thermal coefficient of the backward node scheduling and the abnormal thermal coefficient of the backward node guiding.

[0119] In the supply chain management process, the abnormal thermal coefficient of the backward node scheduling reflects the possibility and severity of abnormalities in the scheduling of the backward node. Taking the finished product transportation node as an example, it is the later node in the supply chain. The acquisition of the abnormal thermal coefficient of the backward node scheduling requires comprehensive factors. The server will analyze the data in the transportation scheduling system, such as whether the scheduling arrangement of the transportation vehicle is reasonable and whether there is a situation where the transportation is delayed due to vehicle scheduling conflicts. If it is found in the past transportation records that a certain proportion of transportation tasks failed to depart on time due to vehicle scheduling problems, this part of the data will be used to calculate the abnormal thermal coefficient of the backward node scheduling. At the same time, the transit link in the transportation process will also affect this coefficient. If the cargo loading and unloading scheduling of the transit warehouse is not timely, it often causes the transportation vehicle to wait, which will also increase the value of the abnormal thermal coefficient of the backward node scheduling. In addition, for different transportation destinations, they may face different traffic conditions and logistics infrastructure conditions, and these factors will also be taken into consideration. For example, transportation tasks to remote areas are more likely to have scheduling abnormalities due to poor road conditions and imperfect logistics facilities, which will increase the abnormal thermal coefficient of the backward node scheduling corresponding to the relevant transportation tasks of the destination.

[0120] Next, the server obtains the abnormal thermal coefficient of backward node guidance of the first training process example. In the supply chain management process, the abnormal thermal coefficient of backward node guidance reflects the possibility and severity of abnormal guidance of the backward node to its predecessor node. Taking the finished product transportation node as an example, its operation will have an impact on the predecessor node. If there are frequent delays in the finished product transportation link, this may have a reverse impact on the scheduling of the production link. The server will analyze the relevant data of this reverse impact to determine the abnormal thermal coefficient of backward node guidance. For example, due to transportation delays, the production link has to adjust the production plan and increase inventory storage costs. The frequency of this situation and the degree of impact on the cost of the entire supply chain will be used to calculate the abnormal thermal coefficient of backward node guidance. In addition, the accuracy of goods delivery in the finished product transportation link will also affect this coefficient. If there are many cases of loss or damage of goods during transportation, this will affect customer satisfaction in the order processing link, and then have a chain reaction on the entire supply chain, causing the abnormal thermal coefficient of backward node guidance to increase.

[0121] Finally, the server determines the prior anomaly risk index of the first training process example based on the abnormal thermal coefficient of the backward node scheduling and the abnormal thermal coefficient of the backward node guidance. In the supply chain management process, the prior anomaly risk index is a preliminary assessment of the abnormal risk of the entire process. Assuming that the abnormal thermal coefficient of the backward node scheduling is 0.3 and the abnormal thermal coefficient of the backward node guidance is 0.25, the server may use the weighted summation method to determine the prior anomaly risk index. If the weight of the abnormal thermal coefficient of the backward node scheduling is 0.6 and the weight of the abnormal thermal coefficient of the backward node guidance is 0.4, then the prior anomaly risk index = 0.3×0.6 + 0.25×0.4 = 0.28. This prior anomaly risk index will provide important basic data for subsequent model training and anomaly diagnosis. For example, in the process of initializing the training of the RPA anomaly diagnosis model, the prior anomaly risk index will be compared with the abnormal risk index calculated by the model to evaluate the accuracy of the model and optimize it.

[0122] In a possible implementation manner, each scheduled node includes a target scheduled node whose abnormal risk index meets a target requirement, and step S130 includes:

[0123] From the scheduled nodes, a target scheduled node whose abnormal risk index meets the target requirement is obtained.

[0124] A risk weight of the target scheduled node is determined according to the abnormal risk index of the target scheduled node.

[0125] Step S140 includes:

[0126] Step S141, obtaining the node index attribute of the target scheduled node.

[0127] Step S142: Acquire the exception handling knowledge base of the node index attribute to serve as the exception handling knowledge base corresponding to the target scheduled node.

[0128] Step S143, according to the risk weight of the target scheduled node, matching the exception handling knowledge base of the node index attribute to the process to be detected.

[0129] In this embodiment, each scheduled node covers many links in the process from raw material procurement to product delivery, such as supplier selection node, purchase order issuance node, raw material transportation node, warehouse reception node, inventory management node, production scheduling node, production and processing node, quality inspection node, finished product warehousing node, order processing node, product transportation node, etc.

[0130] In the process of arranging the scheduled nodes in order based on their abnormal risk index to generate the risk weight of each scheduled node, the target scheduled nodes whose abnormal risk index meets the target requirements are first obtained from each scheduled node. Assume that the target requirement is that the abnormal risk index is greater than a certain set value, such as 0.3. The server will traverse all scheduled nodes. In the supply chain management process, such as the raw material supply link, if the supply source of a certain raw material is single and located in a politically unstable area, there is a great deal of uncertainty in its supply. The abnormal risk index of the corresponding supply node may reach 0.4, which meets the target requirements, and this supply node becomes the target scheduled node; for example, for product transportation nodes, if the transportation route passes through areas prone to natural disasters and there is no alternative route, the risk of damage and delay of goods during transportation is high, and its abnormal risk index may be 0.35, which also meets the target requirements and also becomes the target scheduled node.

[0131] Next, the risk weight of the target scheduled node is determined based on the abnormal risk index of the target scheduled node. For each target scheduled node, its abnormal risk index is an important basis for determining the risk weight. Taking the raw material supply node mentioned above as an example, its abnormal risk index is 0.4. The determination of risk weight depends not only on the abnormal risk index, but may also be related to other attributes of the node in the entire supply chain. Assuming that in the entire supply chain, the timeliness of raw material supply has a fundamental impact on all subsequent links, the server can set a higher coefficient (such as 1.5) to calculate its risk weight, then the risk weight of the raw material supply node is 0.4 * 1.5 = 0.6. For the product transportation node, the abnormal risk index is 0.35. If the product transportation link in the supply chain is important, but the impact on the subsequent links is slightly lower than that of the raw material supply, the coefficient is set to 1.2, and its risk weight is 0.35 * 1.2 = 0.42. In this way, the risk weight is determined for each target scheduled node that meets the target requirements.

[0132] Then, according to the risk weight, the exception handling knowledge base corresponding to each scheduled node is matched to the process to be tested. This process first requires obtaining the node index attribute of the target scheduled node. For the raw material supply node, its node index attribute may contain detailed information on the supply source (such as supplier name, location, supply capacity, etc.), type and specification of raw materials, relevant terms of the supply contract, etc. The node index attribute of the product transportation node may include the name of the transportation company, transportation route, transportation mode (land transportation, sea transportation, air transportation, etc.), relevant information of the transportation vehicle or ship, etc.

[0133] Next, obtain the exception handling knowledge base of the node index attribute as the exception handling knowledge base corresponding to the target scheduled node. For the raw material supply node, according to its node index attribute, the corresponding exception handling knowledge base may contain a variety of response strategies. If political turmoil occurs in the location of the supply source and affects the supply, the knowledge base may contain strategies for finding alternative sources of supply, including evaluation criteria for potential alternative suppliers, processes for establishing cooperative relationships with alternative suppliers, etc.; if the supply capacity is insufficient, the knowledge base may contain negotiation strategies for negotiating with suppliers to increase supply, or adjusting production plans to adapt to reduced supply, etc. For product transportation nodes, their exception handling knowledge bases vary according to the node index attribute. If there are operational problems with the transportation company, the knowledge base may contain processes and standards for replacing the transportation company; if the transportation route encounters force majeure such as natural disasters, the knowledge base may contain emergency plans for cargo transshipment, and words for communicating with customers about transportation delays.

[0134] Finally, according to the risk weight of the target scheduled node, the exception handling knowledge base of the node index attribute is matched to the process to be detected. In the supply chain management process, the higher the risk weight of the target scheduled node, the more priority the corresponding exception handling knowledge base needs to be given when matching to the process to be detected. Taking the risk weight of the raw material supply node as 0.6 and the risk weight of the product transportation node as 0.42 as an example, since the risk weight of the raw material supply node is higher, when matching the exception handling knowledge base to the process to be detected, the server will give priority to ensuring that the strategies in the exception handling knowledge base corresponding to the raw material supply node can be effectively deployed and monitored in the process. For example, in the monitoring module of the supply chain management system, indicators related to raw material supply will be checked more frequently, and when an exception occurs, it will be processed according to the strategies in the knowledge base. For the product transportation node, although its exception handling knowledge base is also matched to the process to be detected, it will be relatively lower in priority in monitoring and processing. In this way, by matching the exception handling knowledge base according to the risk weight, the supply chain management process can make targeted use of the existing processing strategies when facing the abnormal situations that may occur in each scheduled node, thereby improving the stability and reliability of the entire supply chain.

[0135] In a possible implementation, step S130 may further include:

[0136] Obtain the cross-call weight of each scheduled node.

[0137] The risk weight of each scheduled node is determined according to the abnormal risk index of each scheduled node and the cross-call weight of each scheduled node.

[0138] In this embodiment, in the supply chain management process, there are many scheduled nodes, including supplier selection, purchase order placement, raw material transportation, warehouse reception, inventory management, production scheduling, production processing, quality inspection, finished product warehousing, order processing, product transportation, etc. For each scheduled node, the server needs to obtain its cross-call weight. The cross-call weight reflects the closeness of the interaction and dependency between a node and other nodes and the degree of influence on other nodes.

[0139] Take the inventory management node as an example. It has a cross-call relationship with multiple nodes. It is closely related to the purchase order placement node because the quantity and time of purchase orders directly affect the inventory level. If too many or too few purchase orders are placed, inventory management will face different risks, such as inventory backlog or out-of-stock. The inventory management node is also related to the production scheduling node. Production scheduling needs to arrange production tasks according to inventory levels. If the inventory is insufficient, it may lead to production interruption, while too much inventory may cause capital occupation. At the same time, inventory management is also related to the sales order processing node. Changes in the order quantity will affect the inventory outbound situation. Based on these complex relationships, the server determines the cross-call weight of the inventory management node according to the pre-set algorithm and historical data statistics. Assuming that the weight of the inventory management node on the purchase order placement node is 0.2, the weight of the production scheduling node is 0.3, and the weight of the sales order processing node is 0.1, then the cross-call weight of the inventory management node is the sum of these weights, that is, 0.6.

[0140] Taking the production and processing node as an example, the determination of its cross-call weight is similar. Production and processing depends on the raw materials provided by the raw material transportation node. If the raw materials are not supplied in time or there are problems with the quality of the raw materials, the production and processing will be affected. Assume that this impact weight is 0.3. The results of production and processing will affect the quality inspection node. The processing quality of the product directly determines the result of the quality inspection. The impact weight is 0.2. In addition, the progress of production and processing will also affect the finished product storage node. If production is delayed, the finished product storage will also be delayed. The impact weight is 0.1. Therefore, the cross-call weight of the production and processing node is 0.6.

[0141] After obtaining the cross-call weight of each scheduled node, the server determines the risk weight of each scheduled node according to the abnormal risk index of each scheduled node and the cross-call weight of each scheduled node.

[0142] For example, the abnormal risk index of the inventory management node is 0.4 after previous analysis and calculation, and its cross-call weight is 0.6, so the risk weight of the inventory management node is 0.4×0.6 = 0.24. For the production and processing node, assuming that its abnormal risk index is 0.3 and the cross-call weight is 0.6, its risk weight is 0.3×0.6 = 0.18. Through this calculation method, the server determines the risk weight for each scheduled node, so that the scheduled nodes can be sorted according to the risk weight. The higher the risk weight of the node, the higher its importance in the entire supply chain management process. It will receive more attention in subsequent process monitoring, resource allocation, and exception handling, which will help enterprises prevent risks in advance, optimize resource allocation, and ensure the efficient and stable operation of the supply chain management process.

[0143] Figure 2 The hardware structure of the artificial intelligence system 100 for implementing the above-mentioned RPA-based information mining method provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the artificial intelligence system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0144] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the artificial intelligence system 100 uses to execute or use to complete the exemplary methods described in the present invention.

[0145] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the RPA-based information mining method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0146] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned artificial intelligence system 100. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0147] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned RPA-based information mining method is implemented.

[0148] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. An information mining method based on RPA, characterized in that: The method comprises: Obtaining a first process execution path diagram of the process to be detected and a first scheduling path diagram of each scheduled node of the process to be detected; Using the target RPA abnormality diagnosis model, abnormality diagnosis is performed according to the first process execution path diagram and the first scheduling path diagram to generate an abnormality risk index of each scheduled node, wherein the target RPA abnormality diagnosis model is generated by optimizing model parameters according to the second process execution path diagram of the first training process example and the second scheduling path diagram of each sample scheduled node of the first training process example; Arranging the scheduled nodes in order based on the abnormal risk index of each scheduled node, and generating a risk weight of each scheduled node; According to the risk weight, matching the exception handling knowledge base corresponding to each scheduled node to the process to be detected; The training steps of the target RPA abnormal diagnosis model include: Obtaining a priori abnormal risk index of the first training process sample; Using the initialized RPA anomaly diagnosis model, anomaly diagnosis is performed according to the second process execution path diagram of the first training process sample and the second scheduling path diagram of each sample scheduled node of the first training process sample, and an anomaly risk index of each sample scheduled node is generated; Using the initialized RPA anomaly diagnosis model, the anomaly risk index of each sample scheduled node is fused and calculated to generate a third anomaly risk index of the first training process sample; Determining an abnormality diagnosis error parameter of the initialized RPA abnormality diagnosis model according to the prior abnormality risk index and the third abnormality risk index; Optimizing the model parameters of the initialized RPA abnormality diagnosis model according to the abnormality diagnosis error parameter to generate the target RPA abnormality diagnosis model; The scheduled nodes include target scheduled nodes whose abnormal risk index meets the target requirement, and the ordering of the scheduled nodes based on the abnormal risk index of the scheduled nodes to generate the risk weight of the scheduled nodes includes: From the scheduled nodes, obtain a target scheduled node whose abnormal risk index meets the target requirement; Determining a risk weight of the target scheduled node according to the abnormal risk index of the target scheduled node; The matching the exception handling knowledge base corresponding to each scheduled node to the process to be detected according to the risk weight includes: Obtaining a node index attribute of the target scheduled node; Acquire the exception handling knowledge base of the node index attribute to serve as the exception handling knowledge base corresponding to the target scheduled node; According to the risk weight of the target scheduled node, the exception handling knowledge base of the node index attribute is matched to the process to be detected.

2. The RPA-based information mining method according to claim 1, characterized in that: The method further comprises: According to the importance distinguishing factor of the initialized RPA abnormal diagnosis model, the scheduled nodes of each sample are distinguished to generate a first focus node and a second focus node, wherein the importance distinguishing factor represents the importance degree of each sample scheduled node, and the importance degree of the first focus node is greater than the importance degree of the second focus node; According to the abnormal risk indexes of the first focus node and the scheduled nodes of the samples, a first abnormal risk index of the first training process sample is obtained; and according to the abnormal risk indexes of the second focus node and the scheduled nodes of the samples, a second abnormal risk index of the first training process sample is obtained; According to the first abnormal risk index and the second abnormal risk index, constructing an abnormal risk difference factor of the initialized RPA abnormal diagnosis model, wherein the abnormal risk difference factor is used to expand the difference between the first abnormal risk index and the second abnormal risk index; The optimizing the model parameters of the initialized RPA abnormality diagnosis model according to the abnormality diagnosis error parameter to generate the target RPA abnormality diagnosis model includes: According to the abnormal risk difference factor and the abnormal diagnosis error parameter, the model parameters of the initialized RPA abnormal diagnosis model are optimized to generate the target RPA abnormal diagnosis model.

3. The RPA-based information mining method according to claim 2, characterized in that: The method further comprises: Determine a fourth abnormal risk index of the first training process sample by using a target logic association model according to the backward focus node of the first training process sample and the scheduled nodes of each sample; Determine a discrete factor of the initialized RPA abnormality diagnosis model by using the fourth abnormality risk index and the prior abnormality risk index; The step of optimizing the model parameters of the initialized RPA abnormal diagnosis model according to the abnormal risk difference factor and the abnormal diagnosis error parameter to generate the target RPA abnormal diagnosis model includes: According to the discrete factor, the abnormal risk difference factor and the abnormal diagnosis error parameter, the model parameters of the initialized RPA abnormal diagnosis model are optimized to generate the target RPA abnormal diagnosis model.

4. The RPA-based information mining method according to claim 1, characterized in that: The method further comprises: Determine the risk scheduling node of the first training process sample based on the scheduled nodes of each sample by using the target logic association model; Using the target logic association model, anomaly diagnosis is performed based on the second process execution path diagram and the scheduling path diagram of the risk scheduling node of the first training process sample to generate a fourth abnormal risk index of the first training process sample; Determine a discrete factor of the initialized RPA abnormality diagnosis model by using the fourth abnormality risk index and the third abnormality risk index; The optimizing the model parameters of the initialized RPA abnormality diagnosis model according to the abnormality diagnosis error parameter to generate the target RPA abnormality diagnosis model includes: According to the abnormality diagnosis error parameter and the discrete factor, the model parameters of the initialized RPA abnormality diagnosis model are optimized to generate the target RPA abnormality diagnosis model.

5. The RPA-based information mining method according to claim 4, characterized in that: The method further comprises: Obtaining an actual abnormal risk index of the second training process sample, a third process execution path diagram of the second training process sample, a third scheduling path diagram of each past focus node of the second training process sample, and a backward focus node of the second training process sample; Determine the risk scheduling node of the second training process example based on the third process execution path diagram and the third scheduling path diagram by using the initialization logic association model; Using the initialized logic association model, anomaly diagnosis is performed according to the third process execution path diagram and the scheduling path diagram of the risk scheduling node of the second training process sample, and a fifth abnormal risk index of the second training process sample is generated; The model parameters of the initialized logical association model are optimized according to the error between the actual abnormal risk index and the fifth abnormal risk index to generate the target logical association model.

6. The RPA-based information mining method according to claim 1, characterized in that: The step of obtaining the priori abnormal risk index of the first training process sample includes: Obtaining the abnormal thermal coefficient of backward node scheduling of the first training process sample; Obtaining the backward node guided abnormal thermal coefficient of the first training process sample; According to the abnormal thermal coefficient of the backward node scheduling and the abnormal thermal coefficient of the backward node guiding, the prior abnormal risk index of the first training process sample is determined.

7. The RPA-based information mining method according to claim 1, characterized in that: The step of arranging the scheduled nodes in order based on the abnormal risk index of each scheduled node to generate the risk weight of each scheduled node includes: Obtaining the cross-call weight of each scheduled node; The risk weight of each scheduled node is determined according to the abnormal risk index of each scheduled node and the cross-call weight of each scheduled node.

8. An artificial intelligence system, characterized in that: The artificial intelligence system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the RPA-based information mining method described in any one of claims 1 to 7.

Images