Methods, apparatus, equipment, storage media, and products for predicting the risks of operating instructions.
By extracting features from the operation instructions of the middle platform operation and maintenance management system and training a risk prediction model, the problem of low risk prediction accuracy in existing technologies has been solved, and higher risk prediction accuracy and identification of complex correlations have been achieved.
Patent Information
- Application Number
- CN202410340702.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-22
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-03-22
AI Technical Summary
Existing middleware operation and maintenance management early warning methods have the problem of low accuracy in risk prediction when identifying risks in operation and maintenance instructions, and are difficult to adapt to system changes and capture the complex correlations and interactions between multiple indicators.
By extracting features from the original operation instructions to be executed in the middle platform operation and maintenance management system, the first target feature vector is obtained, which includes the degree of correlation between characters and the degree of influence of the execution result. The risk prediction model is used for prediction, and the target neural network model, language representation model and horizontal federated learning are used for feature extraction and model training.
It improves the accuracy of risk prediction for operation and maintenance instructions, enabling more accurate identification and prediction of risks associated with these instructions, and enhances the ability to capture complex correlations between multiple indicators.
Smart Images

Figure CN118211153B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a method, apparatus, device, storage medium and product for predicting the risk of operation instructions. Background Technology
[0002] Currently, the early warning methods in middleware operation and maintenance management are mainly used to identify risks in the execution of operation and maintenance instructions and to take corresponding measures for early warning and handling. Among them, there are three commonly used early warning methods in middleware operation and maintenance management: rule-based and threshold-based early warning methods, statistical analysis-based early warning methods, and expert system-based early warning methods. However, rule-based and threshold-based early warning methods are usually set in the early stages of system design or implementation, making it difficult to adapt to changes in the system and environment, resulting in false alarms or missed alarms and an inability to deal with newly emerging problems. Statistical analysis-based early warning methods usually have high data requirements, while expert system-based early warning methods have knowledge limitations and cannot cover all possible problems. Furthermore, both statistical analysis-based and expert system-based early warning methods struggle to capture the complex correlations and interactions between multiple indicators.
[0003] In summary, the current mid-platform operation and maintenance management early warning method has a technical problem of low risk prediction accuracy when identifying risks in operation and maintenance instructions. Summary of the Invention
[0004] The main objective of this application is to provide a method, apparatus, device, storage medium, and product for predicting the risks of operational instructions. The aim is to provide a risk prediction strategy for operational instructions to improve the accuracy of risk prediction for maintenance operational instructions.
[0005] To achieve the above objectives, this application provides a method for predicting the risk of operating instructions, the method comprising the following steps:
[0006] Feature extraction is performed on the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector of the original operation instructions. The first target feature vector represents the degree of correlation between each character in the original operation instructions except for the execution object and the degree of influence of each string on the execution result of the original operation instructions.
[0007] The first target feature vector is input into the risk prediction model to obtain the risk prediction result of the original operation instruction.
[0008] Optionally, the step of extracting features from the original operation instructions to be executed in the middleware operation and maintenance management system to obtain the first target feature vector of the original operation instructions includes:
[0009] Determine the parameter feature vector of the original operation instructions to be executed in the middle platform operation and maintenance management system;
[0010] The parameter feature vector is input into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object;
[0011] The first target feature vector of the original operation instruction is determined based on the parameter feature vector and the basic feature vector.
[0012] Optionally, the step of determining the parameter feature vector of the original operation instruction to be executed in the middleware operation and maintenance management system includes:
[0013] The original operation instructions to be executed in the middle platform operation and maintenance management system are filtered to obtain the remaining operation instructions;
[0014] Based on preset rules, each character in the remaining operation instructions is converted into a numerical form to obtain the parameter feature vector of the original operation instructions.
[0015] Optionally, the step of determining the first target feature vector of the original operation instruction based on the parameter feature vector and the basic feature vector includes:
[0016] The intermediate vector is determined based on the parameter feature vector and the basic feature vector;
[0017] The intermediate vector is input into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and an execution result association task;
[0018] The initial feature vector is subjected to dimensionality reduction processing to obtain the first target feature vector of the original operation instruction.
[0019] Optionally, the step of performing dimensionality reduction processing on the initial feature vector to obtain the first target feature vector of the original operation instruction includes:
[0020] The initial feature vector is reduced in dimensionality based on a preset encoder to obtain the first target feature vector of the original operation instruction. The preset encoder is trained using an edge likelihood loss function.
[0021] Optionally, before the step of inputting the intermediate vector into the target language representation model to obtain the first feature vector of the original operation instruction, the method further includes:
[0022] Obtain a sample dataset, wherein the sample dataset includes multiple sample data, and each sample data includes a second target feature vector of a historical operation instruction, the instruction type of the historical operation instruction, the actual execution result of the historical operation instruction, and the degree of correlation between the historical operation instruction and the actual execution result;
[0023] When the training task to be performed is a preset instruction type recognition task, each of the second target feature vectors is used as model input data, and the label data constructed based on each of the real instruction types is used as model training labels to fine-tune the language representation model to be trained; and,
[0024] When the training task to be performed is a preset execution result association task, each of the second target feature vectors and each of the real execution results are used as model input data, and the label data constructed based on each of the association degrees is used as model training labels to fine-tune the language representation model to be trained.
[0025] The fine-tuned language representation model was determined as the target language representation model for instruction feature extraction.
[0026] Optionally, before the step of inputting the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction, the method further includes:
[0027] The global base model of the coordinator participating in the horizontal federated learning is used as the global model to be trained. The global model to be trained is distributed as the local model to be trained to multiple participants in the horizontal federated learning, so that each participant can perform parameter update operation on the received local model to be trained based on the local sample dataset to obtain parameter update information.
[0028] The parameter update information returned by multiple participating parties is federated and merged to obtain global update information;
[0029] The global model to be trained is updated based on the global update information to obtain the updated model. The updated model is then used as the new global model to be trained, and the process of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning is returned until the global model to be trained converges, thus obtaining a risk prediction model for instruction risk prediction.
[0030] Furthermore, to achieve the above objectives, this application also provides a risk prediction device for operation instructions, the risk prediction device for operation instructions comprising the following steps:
[0031] The feature extraction module is used to extract features from the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector of the original operation instructions. The first target feature vector represents the degree of correlation between each character in the original operation instructions other than the execution object and the degree of influence of each string on the execution result of the original operation instructions.
[0032] The risk prediction module is used to input the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction, wherein the risk prediction model is trained through horizontal federated learning.
[0033] In addition, to achieve the above objectives, this application also provides a risk prediction device for operation instructions. The risk prediction device for operation instructions includes: a memory, a processor, and a risk prediction program for operation instructions stored in the memory and executable on the processor. When the risk prediction program for operation instructions of the risk prediction device is executed by the processor, it implements the steps of the risk prediction method for operation instructions as described above.
[0034] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a risk prediction program for operation instructions, wherein the risk prediction program for operation instructions, when executed by a processor, implements the steps of the risk prediction method for operation instructions as described above.
[0035] In addition, to achieve the above objectives, the present invention also provides a computer program product, the computer program product including an operation instruction risk prediction program, which, when executed by a processor, implements the steps of the operation instruction risk prediction method as described above.
[0036] This application embodiment extracts features from the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain a first target feature vector in the original operation instructions. It should be noted that the first target feature vector represents the degree of correlation between the characters in the original operation instructions other than the execution object, as well as the degree of influence of each string in the original operation instructions other than the execution object on the execution result of the original operation instructions. Then, the first target feature vector is input into the risk prediction model to obtain the risk prediction result of the original operation instructions.
[0037] Thus, compared to traditional, less accurate, platform operation and maintenance management early warning methods, the first target feature vector extracted in this application not only contains information about the sequence of operation instructions but also information about the execution results of the operation instructions, thereby improving the prediction accuracy of the risk prediction model based on the first target feature vector for instruction risk prediction. Attached Figure Description
[0038] Figure 1 This is a schematic diagram of the hardware operating environment of the risk prediction device for operation instructions involved in the embodiments of this application;
[0039] Figure 2 This is a flowchart illustrating the steps of the first embodiment of the risk prediction method for operation instructions in this application.
[0040] Figure 3 This is a schematic diagram of the instruction risk prediction process according to an embodiment of the risk prediction method for operation instructions in this application;
[0041] Figure 4 This is a schematic diagram of the functional modules of a risk prediction device for operation instructions according to an embodiment of this application.
[0042] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0043] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.
[0044] Reference Figure 1 , Figure 1 This is a schematic diagram of the hardware operating environment of the risk prediction device for operation instructions involved in the embodiments of this application.
[0045] It should be noted that the embodiments of this application relate to a risk prediction device for operation instructions in the field of artificial intelligence technology. Specifically, the risk prediction device for operation instructions can be a smartphone, PC (Personal Computer), tablet computer, portable computer, etc.
[0046] like Figure 1As shown, the risk prediction device for this operation instruction may include: a processor 1001, such as a CPU; a communication bus 1002; a user interface 1003; a network interface 1004; and a memory 1005. The communication bus 1002 is used to establish communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard. Optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM or a stable non-volatile memory, such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001. For ease of description, the execution entity of the method steps in each embodiment is omitted below.
[0047] Those skilled in the art will understand that Figure 1 The structure of the risk prediction device for the operation instructions shown does not constitute a limitation on the risk prediction device for the operation instructions. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0048] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a risk prediction program for operating instructions.
[0049] exist Figure 1 In the terminal shown, network interface 1004 is mainly used to connect to the backend server and communicate data with it; user interface 1003 is mainly used to connect to the client and communicate data with it; and processor 1001 can be used to call the risk prediction program of operation instructions stored in memory 1005 and perform the following operations:
[0050] Feature extraction is performed on the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector of the original operation instructions. The first target feature vector represents the degree of correlation between each character in the original operation instructions except for the execution object and the degree of influence of each string on the execution result of the original operation instructions.
[0051] The first target feature vector is input into the risk prediction model to obtain the risk prediction result of the original operation instruction.
[0052] Furthermore, the operation of extracting features from the original operation instructions to be executed in the middleware operation and maintenance management system to obtain the first target feature vector of the original operation instructions includes:
[0053] Determine the parameter feature vector of the original operation instructions to be executed in the middle platform operation and maintenance management system;
[0054] The parameter feature vector is input into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object;
[0055] The first target feature vector of the original operation instruction is determined based on the parameter feature vector and the basic feature vector.
[0056] Furthermore, the operation of determining the parameter feature vector of the original operation instruction to be executed in the middleware operation and maintenance management system includes:
[0057] The original operation instructions to be executed in the middle platform operation and maintenance management system are filtered to obtain the remaining operation instructions;
[0058] Based on preset rules, each character in the remaining operation instructions is converted into a numerical form to obtain the parameter feature vector of the original operation instructions.
[0059] Further, the operation of determining the first target feature vector of the original operation instruction based on the parameter feature vector and the basic feature vector includes:
[0060] The intermediate vector is determined based on the parameter feature vector and the basic feature vector;
[0061] The intermediate vector is input into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and an execution result association task;
[0062] The initial feature vector is subjected to dimensionality reduction processing to obtain the first target feature vector of the original operation instruction.
[0063] Furthermore, the operation of reducing the dimensionality of the initial feature vector to obtain the first target feature vector of the original operation instruction includes:
[0064] The initial feature vector is reduced in dimensionality based on a preset encoder to obtain the first target feature vector of the original operation instruction. The preset encoder is trained using an edge likelihood loss function.
[0065] Furthermore, before the step of inputting the intermediate vector into the target language representation model to obtain the first feature vector of the original operation instruction, the processor 1001 can also be used to call the risk prediction program of the operation instruction stored in the memory 1005 to perform the following operations:
[0066] Obtain a sample dataset, wherein the sample dataset includes multiple sample data, and each sample data includes a second target feature vector of a historical operation instruction, the instruction type of the historical operation instruction, the actual execution result of the historical operation instruction, and the degree of correlation between the historical operation instruction and the actual execution result;
[0067] When the training task to be performed is a preset instruction type recognition task, each of the second target feature vectors is used as model input data, and the label data constructed based on each of the real instruction types is used as model training labels to fine-tune the language representation model to be trained; and,
[0068] When the training task to be performed is a preset execution result association task, each of the second target feature vectors and each of the real execution results are used as model input data, and the label data constructed based on each of the association degrees is used as model training labels to fine-tune the language representation model to be trained.
[0069] The fine-tuned language representation model was determined as the target language representation model for instruction feature extraction.
[0070] Furthermore, before the step of inputting the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction, the processor 1001 can also be used to call the risk prediction program of the operation instruction stored in the memory 1005 to perform the following operations:
[0071] The global base model of the coordinator participating in the horizontal federated learning is used as the global model to be trained. The global model to be trained is distributed as the local model to be trained to multiple participants in the horizontal federated learning, so that each participant can perform parameter update operation on the received local model to be trained based on the local sample dataset to obtain parameter update information.
[0072] The parameter update information returned by multiple participating parties is federated and merged to obtain global update information;
[0073] The global model to be trained is updated based on the global update information to obtain the updated model. The updated model is then used as the new global model to be trained, and the process of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning is returned until the global model to be trained converges, thus obtaining a risk prediction model for instruction risk prediction.
[0074] It should be noted that operations and maintenance management refers to the management activities of operating and maintaining an enterprise's or organization's information technology infrastructure. A middle platform typically refers to a shared service platform situated between the front-end and back-end, providing unified and reusable technical capabilities and services to support rapid innovation and iteration of front-end business. Middle platform operations and maintenance management refers to the management activities of operating and maintaining the middle platform.
[0075] Current early warning solutions for middleware platform operation and maintenance management mainly fall into three categories: rule-based and threshold-based early warning, statistical analysis-based early warning, and expert system-based early warning. Rule-based and threshold-based early warning primarily monitor various indicators and performance parameters of the middleware platform system by setting rules and thresholds. For example, when an indicator or parameter exceeds a set rule or threshold, the system triggers an early warning notification. Statistical analysis-based early warning primarily analyzes and models the data of the middleware platform system using statistical methods to identify potential problems and trends. Expert system-based early warning utilizes expert domain knowledge and rules to construct an expert system for middleware platform operation and maintenance management early warning. However, all three commonly used early warning methods have their shortcomings. Specifically, rule-based and threshold-based early warning methods are difficult to adapt to changes in the system and environment because rules and thresholds are usually set in the early stages of system design or implementation. If the rules and thresholds are set too strictly or too leniently, it may lead to false alarms or missed alarms. Moreover, rule-based and threshold-based early warning schemes can only identify abnormal situations within the already set rules and threshold ranges, and cannot effectively deal with unknown abnormalities or newly emerging problems. In statistical analysis-based early warning methods, statistical analysis methods usually have high requirements for data and need to meet certain assumptions. If the data does not meet these assumptions, it may cause false alarms or missed alarms. In addition, statistical analysis methods are usually based on single or a few indicators, making it difficult to capture the complex correlations and interactive effects between multiple indicators. In expert system-based early warning methods, early warning schemes based on expert systems rely on experts' domain knowledge and experience to build knowledge bases and rules, and require manual intervention to update, improve, and verify the rules. This increases labor and time costs, and experts' knowledge may be limited and unable to fully cover all situations and problems. Moreover, the middle platform system is usually complex, involving multiple components and interrelated parts, and the early warning solution based on the expert system may be unable to handle this complexity and capture the complex relationships and interactions between multiple parts.
[0076] Based on the above structure, various embodiments of the risk prediction method for operation instructions are proposed.
[0077] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the risk prediction method for operation instructions according to this application. It should be noted that although the logical order is shown in the flowchart, in some cases, the risk prediction method for operation instructions may execute the steps shown or described in a different order. In this embodiment, the execution subject of the risk prediction method for operation instructions can be a personal computer, smartphone, or other device; this is not a limitation. The risk prediction method for operation instructions includes:
[0078] Step S10: Extract features from the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector of the original operation instructions. The first target feature vector represents the degree of correlation between each character in the original operation instructions other than the execution object and the degree of influence of each string on the execution result of the original operation instructions.
[0079] It should be noted that operation and maintenance (O&M) instructions refer to specific instructions or commands triggered by O&M personnel within the middleware O&M management system, which involve configuring, adjusting, maintaining, or optimizing IT (Information Technology) infrastructure, systems, or services. Before execution, O&M instructions awaiting execution in the middleware (hereinafter referred to as original operation instructions for distinction) require a risk assessment. For example, the execution risks of operation instructions may include risks of misoperation, design risks, environmental dependencies, security risks, and data integrity risks. Specifically, misoperation risks include: O&M personnel may make mistakes when executing operation instructions, such as accidentally deleting important configuration files or database tables, leading to system malfunction or data loss. This risk is usually caused by careless operation, lack of necessary review mechanisms, or unclear operation instruction design. Security risks include: If the execution of operation instructions involves sensitive operations, such as modifying permissions or configuring security policies, unauthorized or malicious operation instructions may lead to system attacks or data leaks. Furthermore, executing operational instructions may trigger security vulnerabilities or known attack vectors; data integrity risks: certain operational instructions may affect data integrity, such as data migration, backup and recovery operations. If operational instructions are executed improperly or data verification is insufficient, it may lead to problems such as data corruption, loss, or inconsistency.
[0080] In this embodiment, when the operation and maintenance personnel trigger the original operation command in the middle platform operation and maintenance management system, the original operation command is feature extracted to obtain the feature vector of the original operation command (hereinafter referred to as the first target feature vector for distinction), wherein the first target feature vector is the input of the command risk prediction model.
[0081] In one feasible implementation, the original operation instructions are preprocessed. Specifically, sensitive data in the original operation instructions is de-identified or encrypted to protect data privacy and security. Then, the de-identified or encrypted original operation instructions are subjected to execution object filtering, retaining the remaining part of the original operation instructions excluding the execution object. For example, if the original operation instruction is ps-aux|grep... The original operation command contains six characters: ps, -, aux, |, grep, and python. python is the execution object of this original operation command. The execution object python is filtered out, and the remaining part, ps-aux|grep, is retained. According to a preset rule, each character in the remaining operation command (i.e., parameter features, such as the number, type, and keywords of parameters in the remaining operation command) is mapped into numerical form to obtain a parameter feature vector, such as ps:n, where n is the index of ps in the word list formed after deduplication of words in the training corpus. The parameter feature vector is input into a pre-trained model (hereinafter referred to as the target neural network model for distinction) to obtain a basic feature vector representing the degree of correlation between the characters in the remaining operation command. An intermediate vector is determined based on the parameter feature vector and the basic feature vector, and this intermediate vector is input into a pre-trained BERT (Bidirectional Encoder Representations from...) model. Transformers (a bidirectional encoder from Transformers) model (hereinafter referred to as the target language representation model for distinction) obtains the initial feature vector of the original operation instructions; the initial feature vector output by the BERT model is reduced in dimensionality using a pre-defined encoder to obtain the final feature vector used as input to the subsequent federated reinforcement learning model.
[0082] In this embodiment, step S10 includes:
[0083] Step S101: Determine the parameter feature vector of the original operation instruction to be executed in the middle platform operation and maintenance management system;
[0084] It should be noted that the mapping rules between each character in the operation instruction and the parameter features are predefined, i.e., the preset rules.
[0085] In one feasible implementation, after performing object filtering on the original operation instruction, the remaining operation instructions are obtained, and the parameter features corresponding to each character in the remaining operation instructions are determined according to preset rules. Then, the parameter feature vector of the original operation instruction is determined based on each parameter feature.
[0086] In this embodiment, step S101 includes:
[0087] Step S1011: Perform execution object filtering on the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the remaining operation instructions;
[0088] In this embodiment, the original operation instructions to be executed in the middle platform operation and maintenance management system are subjected to execution object filtering processing, that is, the execution object part of the instruction is removed to obtain the remaining operation instructions.
[0089] In a specific implementation, the original operation instructions to be executed in the current middleware operation and maintenance management system are preprocessed. Specifically, sensitive data in the original operation instructions is desensitized or encrypted to protect data privacy and security. Then, the execution object is filtered for the desensitized or encrypted original operation instructions, retaining the remaining part of the original operation instructions except for the execution object. For example, in the original operation instruction ps-aux|grep python, the execution object python in the instruction is filtered out, and the remaining part, i.e., ps-aux|grep, is retained.
[0090] Step S1012: Based on preset rules, convert each character in the remaining operation instructions into a numerical form to obtain the parameter feature vector of the original operation instructions.
[0091] It should be noted that the operation command data of operation and maintenance personnel, including command text and corresponding operation results or execution status, is obtained from operation and maintenance logs, operation records, or other relevant data sources. When collecting operation and maintenance command data, sensitive data is anonymized or encrypted to protect data privacy and security. The collected data undergoes data cleaning to remove redundant noise, ensuring data reliability and validity. Data is also labeled according to security standards, historical records, or expert knowledge, indicating data attributes, including high-risk commands and normal commands, and labeling the command type, such as query, modification, deletion, restart, etc., to ensure a relative balance between the number of high-risk and normal command samples. Then, all remaining characters in the training corpus are statistically analyzed to construct a mapping table (i.e., the aforementioned preset rules). Specifically, in one feasible implementation, all characters in the commands can be mapped to numerical form, such as ps:n, where n is the index of ps in the word list formed after deduplication of words in the training corpus.
[0092] In this embodiment, based on preset rules, each character in the remaining operation instruction after filtering out the execution object from the original operation instruction is converted into a numerical form to obtain the parameter feature vector of the original operation instruction.
[0093] In a specific implementation, assuming the original operation instruction is ps-aux|grep python, after filtering out the execution object, the remaining operation instruction is ps-aux|grep. This remaining operation instruction has a total of 5 characters. Based on the preset rules, "ps" is defined as the number 1, "-" is defined as the number 2, "aux" is defined as the number 3, "|" is defined as the number 4, and "grep" is defined as the number 5. Thus, the parameter feature vector of the original operation instruction is [1,2,3,4,5].
[0094] Step S102: Input the parameter feature vector into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object;
[0095] In this embodiment, the parameter feature vector obtained through mapping transformation is used as the model input data and input into a pre-trained target neural network model. This target neural network model is a word2vec model trained on a preset operation instruction dataset, using skip-grams. The target neural network model outputs a representation of the degree of association between characters in the remaining operation instructions based on the model input data. For example, the target neural network model can predict the degree of association between 1 and 2 in the parameter feature vector and assign weights to the degree of association between each character. It should be noted that the preset operation instruction dataset includes parameter feature vectors for multiple historical operation instructions.
[0096] Step S103: Determine the first target feature vector of the original operation instruction based on the parameter feature vector and the basic feature vector.
[0097] In this embodiment, the intermediate vector of the original operation instruction is determined based on the parameter feature vector and the basic feature vector of the original operation instruction. The intermediate vector is input into the pre-trained language representation model to obtain the feature vector output by the language representation model based on the intermediate vector (hereinafter referred to as the initial feature vector for distinction). Then, the initial feature vector is subjected to dimensionality reduction processing to obtain the first target feature vector.
[0098] In one feasible implementation, step S103 includes:
[0099] Step S1031: Determine an intermediate vector based on the parameter feature vector and the basic feature vector. In this embodiment, the parameter feature vector and the basic feature vector are fused to obtain the intermediate vector. In a specific implementation, the parameter feature vector and the basic feature vector are directly concatenated to obtain the intermediate vector.
[0100] Step S1032: Input the intermediate vector into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and execution result association task;
[0101] It should be noted that BERT, as a pre-trained model, takes a vector as input. Since computers cannot directly understand text, it needs to be converted into numerical form. This involves converting operation instructions into a sequence of numbers based on predefined rules, resulting in the aforementioned parameter feature vector. The word2vec model then uses this parameter feature vector as input to learn the relationships between characters in the operation instructions, outputting a feature vector containing these relationships—the aforementioned basic feature vector. The parameter feature vector and the basic feature vector are added together to form an intermediate vector, which serves as the input to BERT, providing basic information about the operation sequence. This information only represents string-level information. This application defines two training tasks: an instruction type recognition task and an execution result association task. The instruction type recognition task is a multi-class classification task, while the execution result association task is a binary classification task. During the classification task, the vector undergoes nonlinear transformation through a multi-layer neural network, resulting in vectors containing various deep semantic information. Finally, different activation functions are used to perform inference and prediction for different tasks. In other words, BERT learns a feature vector containing deep semantic relationships through pre-training tasks.
[0102] In this embodiment, the intermediate vector is input into a pre-trained language representation model to obtain an initial feature vector that characterizes the degree of association between characters other than the execution object in the original operation instruction and the degree of influence of each string on the execution result of the original operation instruction.
[0103] In a specific implementation, for the instruction type identification task, the intermediate vector corresponding to the original operation instruction is input into the BERT model, and the type of the operation instruction is output. For example, the instruction type corresponding to `ps-aux|grep python` is a query instruction. For the execution result association task, similar to the NSP (Next Sentence Prediction) task of the original BERT fine-tuning task, it is determined whether the input historical operation instruction and the corresponding instruction execution result are related, that is, whether the execution of the historical operation instruction can obtain the instruction execution result. The fine-tuned BERT can be applied to the operation instruction feature extraction task, which can extract the vector features in the operation instruction, and the fine-tuned BERT model is used to perform the final feature extraction of the operation instruction.
[0104] Step S1033: Perform dimensionality reduction processing on the initial feature vector to obtain the first target feature vector of the original operation instruction.
[0105] In this embodiment, after obtaining the initial feature vector output by the language representation model, the initial feature vector is subjected to dimensionality reduction processing to obtain the first target feature vector of the original operation instruction.
[0106] It should be noted that since the first target feature vector is the input data for the subsequent risk prediction model, dimensionality reduction of the initial feature vector can reduce the complexity of subsequent model training.
[0107] In one feasible implementation, step S1033 includes:
[0108] Step A10: The initial feature vector is reduced in dimensionality based on a preset encoder to obtain the first target feature vector of the original operation instruction. The preset encoder is trained using an edge likelihood loss function.
[0109] It should be noted that the encoder neural network is improved beforehand. Specifically, unlike the mean squared error (MSE) loss function of traditional autoencoders, this application uses the marginal likelihood loss function as the pre-defined encoder's loss function. The initial feature vector of the operation command is used as input, and the weights and biases of the autoencoder are continuously adjusted through backpropagation and gradient descent optimization algorithms to minimize the error between the dimensionality-reduced vector and the original vector. After training, the encoder part of the improved autoencoder is used to reduce the dimensionality of the initial feature vector of the original operation command.
[0110] In this embodiment, the initial feature vector is reduced in dimensionality based on a preset encoder to obtain the first target feature vector of the original operation instruction. The preset encoder is trained using the edge likelihood loss function.
[0111] Step S20: Input the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction, wherein the risk prediction model is trained by horizontal federated learning.
[0112] It should be noted that this application employs a federated reinforcement learning approach to identify and warn of risks in middleware operation commands. Federated Reinforcement Learning (FRL) combines federated learning with customized reinforcement learning. The process involves defining the participants and environment configuration. Participants in the horizontal federated learning are responsible for downloading new models from the coordinator's server, training the models using local data, and uploading encrypted gradients. The environment is configured as a distributed environment shared by the participants, including network traffic and log records. A federated learning environment is constructed to enable all participants to share and exchange information. A customized DQN (Deep Q-network) reinforcement learning network is built, defining the target feature vector of the operation command and the model prediction result as the state space, defining the model's judgment behavior as the action space, and defining the correctness of the model's judgment as feedback.
[0113] In this embodiment, the first target feature vector is input into a pre-trained risk prediction model to obtain the risk prediction result of the original operation instruction. The risk prediction model is trained by horizontal federated learning, and the risk prediction result is either that there is a risk or that there is no risk.
[0114] This application extracts features from the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector in the original operation instructions. It should be noted that the first target feature vector represents the degree of correlation between the characters in the original operation instructions other than the execution object, as well as the degree of influence of each string in the original operation instructions other than the execution object on the execution result of the original operation instructions. Then, the first target feature vector is input into a pre-trained risk prediction model to obtain the risk prediction result of the original operation instructions.
[0115] Thus, compared to traditional, less accurate, platform operation and maintenance management early warning methods, the first target feature vector extracted in this application not only contains information about the sequence of operation instructions but also information about the execution results of the operation instructions, thereby improving the prediction accuracy of the risk prediction model based on the first target feature vector for instruction risk prediction.
[0116] Furthermore, based on the first embodiment of the risk prediction method for operation instructions described above, a second embodiment of the risk prediction method for operation instructions described above is proposed.
[0117] In this embodiment, before step S1032, the following steps are also included:
[0118] Step B10: Obtain a sample dataset, wherein the sample dataset includes multiple sample data, and each sample data includes a second target feature vector of a historical operation instruction, the instruction type of the historical operation instruction, the actual execution result of the historical operation instruction, and the degree of correlation between the historical operation instruction and the actual execution result;
[0119] In this embodiment, operation command data from operation and maintenance personnel, including command text and corresponding operation results or execution status, is obtained from operation and maintenance logs, operation records, or other relevant data sources. Sensitive data is anonymized or encrypted during data collection to protect data privacy and security. The collected data is cleaned to remove redundant noise, ensuring data reliability and validity. Data is labeled according to security standards, historical records, or expert knowledge, indicating data attributes, including high-risk and normal commands, and the command type (e.g., query, modify, delete, restart) to ensure a relative balance between the number of high-risk and normal command samples. Sample data is extracted from the preprocessed data. Each sample data includes a target feature vector of a historical operation command (hereinafter referred to as the second target feature vector for distinction), the command type of the historical operation command, the actual execution result of the historical operation command, and the degree of correlation between the historical operation command and the actual execution result. It should be noted that in this embodiment, the degree of correlation between the historical operation command and the actual execution result can be correlated or not correlated. A sample dataset is constructed based on multiple sample data for training the language representation model to be trained.
[0120] Step B20: When the training task to be performed is a preset instruction type recognition task, each of the second target feature vectors is used as model input data, and the label data constructed based on each of the real instruction types is used as model training labels to fine-tune the language representation model to be trained; and Step B30: When the training task to be performed is a preset execution result association task, each of the second target feature vectors and each of the real execution results are used as model input data, and the label data constructed based on each of the association degrees is used as model training labels to fine-tune the language representation model to be trained;
[0121] In this embodiment, when the training task to be performed by the language representation model to be trained is a preset instruction type recognition task, each second target feature vector is used as the model input data, and the label data constructed based on each real instruction type is used as the model training label to train the language representation model. Conversely, when the training task to be performed is a preset execution result association task, each second target feature vector and each real execution result are used as the model input data, and the label data constructed based on each degree of association is used as the model training label to train the language representation model. It should be noted that this application does not limit the execution order of the above two training tasks. Step B40: Determine the fine-tuned language representation model as the target language representation model for instruction feature extraction.
[0122] In this embodiment, the language representation model to be trained is continuously fine-tuned during the training task, and the language representation model obtained after completing the above two training tasks is determined to be the target language representation model for instruction feature extraction.
[0123] Thus, this application involves pre-designing the instruction type identification task and the execution result association task, fine-tuning the BERT pre-trained model, and using the fine-tuned BERT to extract feature vectors from the operation instructions. This results in the vectors containing not only information about the operation instruction sequence but also information about the operation instruction execution result, thereby improving the accuracy of subsequent models.
[0124] Furthermore, based on the first and / or second embodiments of the risk prediction method for operation instructions described above, a third embodiment of the risk prediction method for operation instructions described above is proposed.
[0125] In this embodiment, before step S20 above, the following steps are also included:
[0126] Step C10: The global base model of the coordinator participating in the horizontal federated learning is used as the global model to be trained. The global model to be trained is distributed as the local model to be trained to multiple participants in the horizontal federated learning, so that the multiple participants can perform parameter update operations on the received local model to be trained based on their local sample datasets to obtain parameter update information.
[0127] In this embodiment, the global base model in the coordinator participating in the horizontal federated learning is used as the global model to be trained, and the global model to be trained is distributed to each participant as a local model to be trained. After receiving the local model to be trained, each participant performs parameter update operations on the local model to be trained based on its own local sample dataset to obtain parameter update information. The local sample dataset includes multiple sample data, each of which includes the target feature vector of a historical operation instruction and the actual execution risk level of the historical operation instruction (whether there is risk or not). When a participant trains the local model to be trained based on its local sample dataset, the target feature vector of the historical operation instruction is used as the model input data, and the label data constructed based on the above-mentioned actual execution risk level is used as the model training label.
[0128] Step C20: Federate and fuse the parameter update information returned by multiple participating parties to obtain global update information;
[0129] In this embodiment, after multiple participants complete a round of training, the final parameter update information is returned to the coordinator. The coordinator performs federated fusion of the received multiple sets of parameter update information to obtain global update information, which is used to update the coordinator's global model to be trained.
[0130] Step C30: Update the global model to be trained based on the global update information to obtain the updated model. Use the updated model as the new global model to be trained, and return to execute the step of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning until the global model to be trained converges to obtain a risk prediction model for instruction risk prediction.
[0131] In this embodiment, the coordinator updates the global model to be trained based on global update information, obtaining the updated model. It then determines whether the updated model has converged. If not, the updated model is used as the new global model to be trained, and the process returns to the step of distributing the global model to be trained as a local model to multiple participants in the horizontal federated learning so that they can continue training the received model. If yes, the current model is used as the risk prediction model for instruction risk prediction. In a specific implementation, the aforementioned global base model is a DQN reinforcement learning network. Under the federated learning architecture, each participant interacts with the customized DQN reinforcement learning network according to its local environment, collecting experience data, including observations, actions, and behavioral feedback, and optimizing its local strategy based on this experience. Participants periodically share the parameters and gradients of their local models with the coordinator of the federated learning for updating the global model DQN network. After the DQN network update is complete, the coordinator sends it to the participants, who then use the updated DQN model in the next round of interaction and training. This model update iteration is repeated until the maximum number of training rounds is reached or the model converges, at which point the model is saved.
[0132] For example, such as Figure 3The diagram illustrates the instruction risk prediction process. First, operation command data from maintenance personnel is obtained from maintenance logs, operation records, or other relevant data sources. Then, the collected data undergoes preprocessing, including data anonymization and data cleaning. Next, the parameter feature vectors for each historical operation instruction are determined according to preset rules. These parameter feature vectors are then used as input data for a word2vec model to be trained. The correlation between characters in each parameter feature vector is used as the training labels for the word2vec model. The word2vec model is then trained to obtain a target neural network model for extracting the basic feature vectors of historical operation instructions. Finally, based on the parameter feature vectors and the basic feature vectors, the... The BERT model is used for customized fine-tuning tasks to obtain a fine-tuned BERT, which is then used to extract initial feature vectors from operation instructions. The encoder is improved and trained to obtain an improved encoder, and the initial feature vectors are then dimensionality-reduced based on the improved encoder to obtain the target feature vectors for each historical operation instruction. In a horizontal federated learning architecture, a global base model is distributed to all participants through a central server of the coordinator. Each participant periodically shares its local model parameters, gradients, and other parameter updates with the coordinator, allowing the coordinator to update the global base model. After the DQN network update is completed, the coordinator sends it back to the participants, who then use the updated DQN model in the next round of interaction and training. This model update iteration is repeated until the maximum number of training rounds is reached or the model converges. The current model is saved as a risk prediction model, and the risk level of the actual operation instructions is predicted based on this risk prediction model.
[0133] Thus, by employing federated reinforcement learning, this application utilizes a customized reinforcement learning network to train a risk prediction and early warning model for middle platform operation instructions through continuous interaction with the environment and updates to the global model. This effectively avoids the reliance of the risk prediction and early warning model for middle platform operation instructions on rules and human expertise, enabling it to adapt to complex relationships and changing system and environment, and accurately identify less common operation instructions.
[0134] In addition, embodiments of this application also provide a risk prediction device for operation instructions.
[0135] Please refer to Figure 4 , Figure 4 This is a functional module diagram of an embodiment of the risk prediction device for operation instructions in this application, as shown below. Figure 4 As shown, the risk prediction device for the operation instructions of this application includes:
[0136] The feature extraction module 10 is used to extract features from the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the first target feature vector of the original operation instructions. The first target feature vector represents the degree of correlation between each character in the original operation instructions except for the execution object and the degree of influence of each string on the execution result of the original operation instructions.
[0137] The risk prediction module 20 is used to input the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction.
[0138] Furthermore, the feature extraction module 10 includes:
[0139] The first feature extraction unit is used to determine the parameter feature vector of the original operation instruction to be executed in the middle platform operation and maintenance management system;
[0140] The second feature extraction unit is used to input the parameter feature vector into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object;
[0141] The determining unit is used to determine the first target feature vector of the original operation instruction based on the parameter feature vector and the basic feature vector.
[0142] Furthermore, the first feature extraction unit is also used to perform execution object filtering on the original operation instructions to be executed in the middle platform operation and maintenance management system to obtain the remaining operation instructions; and to convert each character in the remaining operation instructions into a numerical form based on preset rules to obtain the parameter feature vector of the original operation instructions.
[0143] Further, the unit is defined, including:
[0144] Determine a subunit, used to determine an intermediate vector based on the parameter feature vector and the basic feature vector;
[0145] The feature extraction subunit is used to input the intermediate vector into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and an execution result association task;
[0146] The dimensionality reduction processing subunit is used to perform dimensionality reduction processing on the initial feature vector to obtain the first target feature vector of the original operation instruction.
[0147] Furthermore, the dimensionality reduction subunit is used to perform dimensionality reduction processing on the initial feature vector based on a preset encoder to obtain the first target feature vector of the original operation instruction, wherein the preset encoder is trained using an edge likelihood loss function.
[0148] Furthermore, the risk prediction device for the operation instructions in this application also includes:
[0149] The acquisition module is used to acquire a sample dataset, wherein the sample dataset includes multiple sample data, and each sample data includes a second target feature vector of a historical operation instruction, the instruction type of the historical operation instruction, the actual execution result of the historical operation instruction, and the degree of correlation between the historical operation instruction and the actual execution result;
[0150] The first training module is used to fine-tune the language representation model to be trained by using the second target feature vectors as model input data and the label data constructed based on the real instruction types as model training labels when the training task to be executed is a preset instruction type recognition task; and,
[0151] The second training module is used to fine-tune the language representation model to be trained when the training task to be executed is a preset execution result association task, taking each second target feature vector and each of the real execution results as model input data, and taking the label data constructed based on each of the association degrees as model training labels.
[0152] The determination module is used to determine the fine-tuned language representation model as the target language representation model for instruction feature extraction.
[0153] Furthermore, the risk prediction device for the operation instructions in this application also includes:
[0154] The distribution module is used to take the global base model of the coordinator participating in the horizontal federated learning as the global model to be trained, and distribute the global model to be trained as the local model to be trained to multiple participants participating in the horizontal federated learning, so that the multiple participants can perform parameter update operations on the received local model to be trained based on their local sample datasets to obtain parameter update information.
[0155] The federated fusion module is used to federate and fuse the parameter update information returned by multiple participants to obtain global update information;
[0156] The update module is used to update the global model to be trained based on the global update information to obtain the updated model, use the updated model as the new global model to be trained, and return to execute the step of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning until the global model to be trained converges to obtain a risk prediction model for instruction risk prediction.
[0157] This application also provides a computer storage medium storing a risk prediction program for operation instructions, which, when executed by a processor, implements the steps of the risk prediction program method for operation instructions as described in any of the above embodiments.
[0158] The specific embodiments of the computer storage medium in this application are basically the same as the embodiments of the risk prediction program method of the operation instructions in this application described above, and will not be repeated here.
[0159] This application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the risk prediction method for the operation instructions of this application as described in any of the above embodiments, which will not be elaborated here.
[0160] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0161] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a risk prediction device (which may be a TWS earphone, etc.) to execute the methods described in the various embodiments of this application.
[0163] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A method for predicting the risk of operational instructions, characterized in that, The risk prediction method for the operation instructions includes the following steps: Determine the parameter feature vector of the original operation instructions to be executed in the middle platform operation and maintenance management system; The parameter feature vector is input into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object; The intermediate vector is determined based on the parameter feature vector and the basic feature vector; The intermediate vector is input into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and an execution result association task; The initial feature vector is subjected to dimensionality reduction processing to obtain the first target feature vector of the original operation instruction, wherein the first target feature vector characterizes the degree of correlation between each character in the original operation instruction except for the execution object and the degree of influence of each string on the execution result of the original operation instruction; The first target feature vector is input into the risk prediction model to obtain the risk prediction result of the original operation instruction; Before the step of inputting the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction, the method further includes: The global base model of the coordinator participating in the horizontal federated learning is used as the global model to be trained. The global model to be trained is distributed as the local model to be trained to multiple participants in the horizontal federated learning, so that each participant can perform parameter update operation on the received local model to be trained based on the local sample dataset to obtain parameter update information. The parameter update information returned by multiple participating parties is federated and merged to obtain global update information; The global model to be trained is updated based on the global update information to obtain the updated model. The updated model is then used as the new global model to be trained, and the process of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning is returned until the global model to be trained converges, thus obtaining a risk prediction model for instruction risk prediction.
2. The risk prediction method for operation instructions as described in claim 1, characterized in that, The step of determining the parameter feature vector of the original operation instruction to be executed in the middle platform operation and maintenance management system includes: The original operation instructions to be executed in the middle platform operation and maintenance management system are filtered to obtain the remaining operation instructions; Based on preset rules, each character in the remaining operation instructions is converted into a numerical form to obtain the parameter feature vector of the original operation instructions.
3. The risk prediction method for operation instructions as described in claim 1, characterized in that, The step of performing dimensionality reduction processing on the initial feature vector to obtain the first target feature vector of the original operation instruction includes: The initial feature vector is reduced in dimensionality based on a preset encoder to obtain the first target feature vector of the original operation instruction. The preset encoder is trained using an edge likelihood loss function.
4. The risk prediction method for operation instructions as described in claim 1, characterized in that, Before the step of inputting the intermediate vector into the target language representation model to obtain the first feature vector of the original operation instruction, the method further includes: Obtain a sample dataset, wherein the sample dataset includes multiple sample data, and each sample data includes a second target feature vector of a historical operation instruction, the instruction type of the historical operation instruction, the actual execution result of the historical operation instruction, and the degree of correlation between the historical operation instruction and the actual execution result; When the training task to be performed is a preset instruction type recognition task, each of the second target feature vectors is used as model input data, and the label data constructed based on the instruction types of each of the historical operation instructions is used as model training labels to fine-tune the language representation model to be trained; and, When the training task to be performed is a preset execution result association task, each of the second target feature vectors and each of the real execution results are used as model input data, and the label data constructed based on each of the association degrees is used as model training labels to fine-tune the language representation model to be trained. The fine-tuned language representation model was determined as the target language representation model for instruction feature extraction.
5. A risk prediction device for operating instructions, characterized in that, The risk prediction device for the operation command includes the following steps: A feature extraction module is used to determine the parameter feature vector of the original operation instruction to be executed in the middleware operation and maintenance management system; input the parameter feature vector into the target neural network model to obtain the basic feature vector of the original operation instruction, wherein the basic feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object; determine an intermediate vector based on the parameter feature vector and the basic feature vector; input the intermediate vector into the target language representation model to obtain the initial feature vector of the original operation instruction, wherein the target language representation model is trained by performing a preset instruction type recognition task and an execution result association task; perform dimensionality reduction processing on the initial feature vector to obtain the first target feature vector of the original operation instruction, wherein the first target feature vector represents the degree of correlation between each character in the original operation instruction except for the execution object and the degree of influence of each string on the execution result of the original operation instruction; The risk prediction module is used to input the first target feature vector into the risk prediction model to obtain the risk prediction result of the original operation instruction; The risk prediction device for the operation command also includes: The distribution module is used to take the global base model of the coordinator participating in the horizontal federated learning as the global model to be trained, and distribute the global model to be trained as the local model to be trained to multiple participants participating in the horizontal federated learning, so that the multiple participants can perform parameter update operations on the received local model to be trained based on their local sample datasets to obtain parameter update information. The federated fusion module is used to federate and fuse the parameter update information returned by multiple participants to obtain global update information; The update module is used to update the global model to be trained based on the global update information to obtain the updated model, use the updated model as the new global model to be trained, and return to execute the step of distributing the global model to be trained as the local model to be trained to multiple participants in the horizontal federated learning until the global model to be trained converges to obtain a risk prediction model for instruction risk prediction.
6. A risk prediction device for operating instructions, characterized in that, The risk prediction device for the operation instructions includes a memory and a processor. The memory stores a risk prediction program for the operation instructions that can be run on the processor. When the risk prediction program for the operation instructions is executed by the processor, it implements the steps of the risk prediction method for the operation instructions as described in any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a risk prediction program for operation instructions, which, when executed by a processor, implements the steps of the risk prediction method for operation instructions as described in any one of claims 1 to 4.
8. A computer program product, characterized in that, The computer program product includes a risk prediction program for operation instructions, which, when executed by a processor, implements the steps of the risk prediction method for operation instructions as described in any one of claims 1 to 4.
Citation Information
Patent Citations
Risk prediction method and device, storage medium and electronic equipment
CN113362168A
Media data correction method and device, computer, storage medium and program product
CN114611493A
