Operation business risk analysis method and device, equipment and storage medium

Abstract

The invention provides an operation business risk analysis method, device and equipment and a storage medium, and the method comprises the steps: carrying out the multi-modal preprocessing of multi-source heterogeneous data, generating a structured risk feature data set, carrying out the staged training and dynamic optimization of a preset multi-branch deep neural network model, and obtaining a structured risk feature data set; and obtaining a trained risk analysis model. And carrying out distributed deployment and streaming processing configuration on the model, and constructing a real-time risk inference system. And inputting real-time operation business data into the system, and generating a risk analysis result and corresponding feature importance distribution. And through a multi-model integration framework, carrying out comprehensive assessment and interpretability analysis on a risk analysis result and feature importance distribution to form a comprehensive risk assessment report. According to the method, through multi-modal preprocessing of multi-source heterogeneous data, risk features of different data sources are fully mined and fused, and the problem of inaccurate analysis caused by single data and unprocessed heterogeneity in a traditional method is solved.

Description

Technical Field

[0001] The present invention relates to the field of risk analysis, and in particular to an operational business risk analysis method, device, equipment and storage medium. Background Art

[0002] With the rapid development of information technology, the complexity and scale of business operations are increasing, and the risks faced by enterprises in the process of operation are of various types and sources. These risks include market fluctuations, policy changes, natural disasters, supply chain disruptions and other factors. In order to effectively manage and respond to these risks, enterprises are increasingly relying on data-driven risk analysis and prediction models.

[0003] Traditional risk analysis methods usually rely on a single data source or a single analysis model. Although this method can provide certain risk prediction capabilities in certain specific scenarios, it is difficult for traditional methods to comprehensively and accurately analyze and predict risks due to the complex and diverse data sources in modern enterprise operating environments (such as structured data, unstructured data, time series data, etc.). In addition, existing risk analysis models often ignore the heterogeneity and multimodal characteristics between data, resulting in limited performance and accuracy of the model when facing multi-source heterogeneous data. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem that the existing risk analysis models often ignore the heterogeneity and multimodal characteristics between data, resulting in the limited performance and accuracy of the model when facing multi-source heterogeneous data; A first aspect of the present invention provides an operational business risk analysis method, the operational business risk analysis method comprising: Perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and perform phased training and dynamic optimization on a preset multi-branch deep neural network model based on the structured risk feature data set to obtain a trained risk analysis model; Distributed deployment and streaming processing configuration are performed on the risk analysis model to obtain a real-time risk reasoning system; Acquire real-time operational business data, and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; The risk analysis results and the feature importance distribution are subjected to risk assessment synthesis and interpretability analysis through a preset multi-model integration framework to obtain a comprehensive risk assessment report.

[0005] Optionally, in a first implementation of the first aspect of the present invention, multi-source heterogeneous data is subjected to multi-modal preprocessing to obtain a structured risk feature data set, and a preset multi-branch deep neural network model is trained and dynamically optimized in stages according to the structured risk feature data set to obtain a trained risk analysis model, including: Perform data cleaning and format unification processing on multi-source heterogeneous data to obtain a standardized data set, and perform dynamic time window division and multi-scale feature extraction on the standardized data set to obtain a time series feature set; Performing feature engineering processing on the time series feature set and the non-time series data in the standardized data set to obtain label data corresponding to the enhanced feature set; Performing unsupervised pre-training on a preset multi-branch deep neural network model according to the enhanced feature set to obtain a pre-trained model; The pre-trained model is fine-tuned and dynamically optimized in a supervised manner according to the enhanced feature set and label data to obtain a trained risk analysis model.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the pre-trained model is fine-tuned and dynamically optimized according to the enhanced feature set and label data to obtain a trained risk analysis model, including: Performing stratified sampling and data enhancement processing on the enhanced feature set and label data to obtain a balanced training data set; Performing multi-task joint optimization on each branch of the pre-trained model according to the balanced training data set to obtain a preliminary fine-tuning model; Performing adversarial training and noise injection on the preliminary fine-tuning model to obtain a robustness enhanced model; According to the prediction results of the robustness enhancement model, the hard-to-classify samples are identified and focused on learning to obtain a hard-to-classify sample enhancement model; Incremental learning and model distillation techniques are applied to the difficult sample reinforcement model to obtain a trained risk analysis model.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the distributed deployment and streaming processing configuration of the risk analysis model to obtain a real-time risk reasoning system includes: The risk analysis model is compressed and quantified to obtain a lightweight reasoning model, and a microservice architecture is constructed based on the lightweight reasoning model to obtain a distributed reasoning framework; wherein the microservice architecture encapsulates different components of the model into independent service units, and realizes communication between service units through message queues; Perform load balancing and elastic scaling configuration on the distributed reasoning framework to obtain a dynamic resource scheduling system; Design a streaming data processing pipeline according to the dynamic resource scheduling system to obtain a real-time feature extraction module; The real-time feature extraction module and the lightweight reasoning model are integrated and performance optimized to obtain a real-time risk reasoning system.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, acquiring real-time operational business data, and inputting the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distributions includes: Synchronously process the real-time operational business data to obtain an original data stream, and perform real-time data cleaning and anomaly detection on the original data stream to obtain a pre-processed data stream; Performing real-time feature engineering and feature selection on the preprocessed data stream to obtain a feature vector stream; Inputting the feature vector stream into the risk analysis model in the real-time risk reasoning system to obtain a multimodal feature representation, wherein the risk analysis model includes a fully connected layer, a recurrent layer, a convolutional layer and an output layer, the fully connected layer performs nonlinear transformation on the structured data of the feature vector stream, the recurrent layer captures the temporal dependency of the sequence data of the feature vector stream, and the convolutional layer extracts local semantic features from the text data of the feature vector stream; Performing cross-modal attention fusion on the multimodal feature representation to obtain a unified risk representation; The risk score is calculated according to the unified risk representation through the output layer of the risk analysis model to obtain the risk analysis result and the corresponding feature importance distribution.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the risk analysis results and the feature importance distribution are subjected to risk assessment synthesis and interpretability analysis through a preset multi-model integration framework to obtain a comprehensive risk assessment report, including: Performing multi-model fusion on the risk analysis results and feature importance distribution to obtain an integrated risk score, and performing Bayesian calibration based on the integrated risk score and historical risk distribution to obtain a calibrated risk probability; Perform multi-dimensional decomposition on the calibrated risk probability to obtain risk source analysis; wherein the multi-dimensional decomposition includes attributing and quantifying risks according to business type, user group and time period dimensions; Generate decision tree rules according to the risk source analysis and the feature importance distribution to obtain an explainable rule set; The explainability rule set and the calibrated risk probability are comprehensively collated and visualized to obtain a comprehensive risk assessment report.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, after performing multimodal preprocessing on the multi-source heterogeneous data to obtain a structured risk feature data set, and performing phased training and dynamic optimization on a preset multi-branch deep neural network model according to the structured risk feature data set to obtain a trained risk analysis model, the method further includes: Performing a hierarchical freezing process on the risk analysis model to obtain a basic migration model; wherein the hierarchical freezing process selectively fixes model parameters according to the universality of features of different layers; The basic migration model is subjected to domain adaptive training according to the feature distribution of the new business scenario to obtain a domain alignment model; wherein the domain adaptive training minimizes the feature distribution difference between the source domain and the target domain through an adversarial learning method; Performing progressive fine-tuning on the domain alignment model and the new business data to obtain a scenario adaptation model; wherein the progressive fine-tuning adopts a layered learning rate strategy to gradually unfreeze and optimize parameters of different layers; Performing meta-learning parameter optimization according to the prediction results of the scene adaptation model to obtain a fast adaptation model; wherein the meta-learning parameter optimization learns the optimal initialization state of the model parameters by simulating different task distributions; The rapid adaptation model is subjected to knowledge distillation and model compression to obtain a lightweight migration model, and the lightweight migration model is used as a risk analysis model for new business scenarios.

[0011] A second aspect of the present invention provides an operational business risk analysis device, the operational business risk analysis device comprising: A model training module is used to perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and to perform phased training and dynamic optimization on a preset multi-branch deep neural network model according to the structured risk feature data set to obtain a trained risk analysis model; A model deployment module, used for performing distributed deployment and streaming processing configuration of the risk analysis model to obtain a real-time risk reasoning system; A risk analysis module, used to obtain real-time operational business data, and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; The explainability analysis module is used to perform risk assessment synthesis and explainability analysis on the risk analysis results and the feature importance distribution through a preset multi-model integration framework to obtain a comprehensive risk assessment report.

[0012] The third aspect of the present invention provides an operational business risk analysis device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the operational business risk analysis device executes the steps of the above-mentioned operational business risk analysis method.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned operational business risk analysis method.

[0014] The above-mentioned operational business risk analysis method, device, equipment and storage medium generate a structured risk feature data set by multi-modal preprocessing of multi-source heterogeneous data, and perform phased training and dynamic optimization on the preset multi-branch deep neural network model to obtain a trained risk analysis model. The model is distributed deployed and stream processed to build a real-time risk reasoning system. Real-time operational business data is input into the system to generate risk analysis results and corresponding feature importance distributions. Through a multi-model integration framework, the risk analysis results and feature importance distributions are comprehensively evaluated and interpreted to form a comprehensive risk assessment report. This method fully mines and integrates risk features from different data sources by multi-modal preprocessing of multi-source heterogeneous data, thereby overcoming the problem of inaccurate analysis caused by single data and unprocessed heterogeneity in traditional methods.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of a first embodiment of the method for analyzing operational business risks in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of an operating business risk analysis device in an embodiment of the present invention; Figure 3 It is a schematic diagram of an embodiment of an operational business risk analysis device in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0020] To facilitate understanding of this embodiment, a method for analyzing operational business risks disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and perform phased training and dynamic optimization on a preset multi-branch deep neural network model based on the structured risk feature data set to obtain a trained risk analysis model; In one embodiment of the present invention, the multi-source heterogeneous data is subjected to multimodal preprocessing to obtain a structured risk feature data set, and a preset multi-branch deep neural network model is trained and dynamically optimized in stages according to the structured risk feature data set to obtain a trained risk analysis model, including: data cleaning and format unification processing of the multi-source heterogeneous data to obtain a standardized data set, and dynamic time window division and multi-scale feature extraction of the standardized data set to obtain a time series feature set; feature engineering processing is performed on the time series feature set and the non-time series data in the standardized data set to obtain label data corresponding to the enhanced feature set; unsupervised pre-training of the preset multi-branch deep neural network model is performed according to the enhanced feature set to obtain a pre-trained model; supervised fine-tuning and dynamic optimization of the pre-trained model are performed according to the enhanced feature set and the label data to obtain a trained risk analysis model.

[0021] Specifically, first, data cleaning and format unification are performed, which includes removing duplicate data, processing missing values, and correcting outliers. Data sources include user behavior data, transaction records, customer service interaction records, and network usage logs. Through this step, a standardized data set is obtained. Next, the standardized data set is subjected to dynamic time window partitioning and multi-scale feature extraction. Dynamic time window partitioning adaptively adjusts the window size according to the temporal characteristics of different types of data, while multi-scale feature extraction includes the calculation of short-term, medium-term, and long-term features to obtain a time series feature set. Then, feature engineering is performed on the time series feature set and the non-time series data in the standardized data set. This step includes feature combination, feature intersection, feature encoding, and feature selection. Through these operations, an enhanced feature set and its corresponding label data are generated. Feature combination can create new composite features, feature intersection can capture the nonlinear relationship between features, feature encoding converts categorical variables into numerical representations, and feature selection selects the most relevant feature subsets. Next, the enhanced feature set is used to perform unsupervised pre-training on the preset multi-branch deep neural network model. This pre-training process uses a contrastive learning method to learn representations of data of different modalities. Specifically, the model learns to bring the representations of similar samples closer together and push the representations of different samples further apart, thereby learning meaningful feature representations. The output of this step is a pre-trained model that already has a preliminary understanding of the data. Finally, the pre-trained model is supervised fine-tuned and dynamically optimized based on the enhanced feature set and labeled data. Supervised fine-tuning uses a multi-task learning framework to simultaneously optimize multiple risk indicators, such as fraud risk, credit risk, and churn risk. Dynamic optimization includes applying dynamic sampling mechanisms, incremental learning strategies, and adversarial training techniques. The dynamic sampling mechanism adaptively adjusts the sampling probabilities of samples of different categories during training to ensure that the model can fully learn the characteristics of minority class samples. The incremental learning strategy allows the model to adapt to new data distributions without retraining, while the adversarial training technique enhances the robustness of the model by adding carefully designed perturbations. Throughout the training process, different parts of the model are optimized separately. For structured data, embedding layers and tree-structured neural networks are used for feature interaction and nonlinear transformation. For sequence data, recurrent layers are used to capture temporal dependencies. For text data, convolutional layers are used to extract local semantic features. These different branches are fused through a cross-modal attention mechanism to achieve selective integration of information.

[0022] Furthermore, the supervised fine-tuning and dynamic optimization of the pre-trained model according to the enhanced feature set and label data to obtain a trained risk analysis model includes: stratified sampling and data enhancement processing of the enhanced feature set and label data to obtain a balanced training data set; multi-task joint optimization of each branch of the pre-trained model according to the balanced training data set to obtain a preliminary fine-tuning model; adversarial training and noise injection of the preliminary fine-tuning model to obtain a robustness enhanced model; identifying and focusing on learning difficult samples according to the prediction results of the robustness enhanced model to obtain a difficult sample reinforcement model; applying incremental learning and model distillation technology to the difficult sample reinforcement model to obtain a trained risk analysis model.

[0023] Specifically, stratified sampling is performed according to the distribution of different risk types to ensure that the proportion of each type of sample in the training set is balanced. Data augmentation processing includes oversampling minority class samples and generating synthetic samples, such as using the SMOTE (Synthetic Minority Over-sampling Technique) algorithm to generate new minority class samples. The purpose of these steps is to obtain a balanced training data set to solve the class imbalance problem. Next, the balanced training data set is used to perform multi-task joint optimization on each branch of the pre-trained model. Multi-task joint optimization simultaneously considers multiple risk indicators, such as fraud risk, credit risk, and customer churn risk. This process uses a dynamic weight balancing mechanism to automatically adjust the loss contribution of each task according to the difficulty and importance of different tasks. In specific implementation, a soft parameter sharing method can be used to allow different tasks to share some network parameters while retaining task-specific parameters. Through this step, a preliminary fine-tuning model is obtained, which is already able to handle multiple related risk assessment tasks at the same time. Then, adversarial training and noise injection are performed on the preliminary fine-tuning model to enhance the robustness of the model. Adversarial training enhances the generalization ability of the model by generating adversarial samples. Specific methods include algorithms such as FGSM (Fast Gradient Sign Method) or PGD (Projected Gradient Descent). Noise injection adds random perturbations to model parameters and input data during training, such as Gaussian noise or Laplace noise. The application of these techniques makes the model more robust to small input perturbations, thereby obtaining a robustness-enhanced model. Next, the hard-to-distinguish samples are identified and focused on learning based on the prediction results of the robustness-enhanced model. The identification of hard-to-distinguish samples is based on the prediction confidence of the model and the loss value of the sample. For example, samples with prediction confidence close to the decision boundary or high loss value can be selected as hard-to-distinguish samples. When focusing on these hard-to-distinguish samples, the focal loss function is used, which can automatically adjust the weights of different samples so that the model pays more attention to hard-to-distinguish samples. At the same time, a dynamic sampling strategy is used to increase the sampling probability of hard-to-distinguish samples during training. Through these methods, a hard-to-distinguish sample reinforcement model is obtained, which performs better when dealing with complex and boundary situations. Finally, incremental learning and model distillation techniques are applied to the hard sample reinforcement model. Incremental learning allows the model to adapt to new data distributions without retraining. The specific implementation can adopt methods such as Elastic Weight Consolidation to keep the memory of old knowledge while learning new knowledge. Model distillation transfers the knowledge of large and complex models to smaller models to balance performance and computational efficiency. In the distillation process, the large model serves as the teacher model and the small model as the student model, and knowledge is transferred by minimizing the KL divergence of the outputs of the two models.The output of this step is the final trained risk analysis model, which not only has the ability to handle multiple risks, but can also quickly adapt to new data and new scenarios.

[0024] Furthermore, after the multi-source heterogeneous data is subjected to multimodal preprocessing to obtain a structured risk feature data set, and a preset multi-branch deep neural network model is trained and dynamically optimized in stages according to the structured risk feature data set to obtain a trained risk analysis model, it also includes: performing hierarchical freezing processing on the risk analysis model to obtain a basic migration model; wherein the hierarchical freezing processing selectively fixes model parameters according to the feature universality of different layers; performing domain adaptive training on the basic migration model according to the feature distribution of the new business scenario to obtain a domain alignment model; wherein the domain adaptive training is performed through adversarial The learning method minimizes the difference in feature distribution between the source domain and the target domain; the domain alignment model and new business data are progressively fine-tuned to obtain a scenario adaptation model; wherein the progressive fine-tuning adopts a hierarchical learning rate strategy to gradually unfreeze and optimize parameters of different layers; meta-learning parameter optimization is performed according to the prediction results of the scenario adaptation model to obtain a fast adaptation model; wherein the meta-learning parameter optimization learns the optimal initialization state of model parameters by simulating different task distributions; knowledge distillation and model compression are performed on the fast adaptation model to obtain a lightweight migration model, and the lightweight migration model is used as a risk analysis model for new business scenarios.

[0025] Specifically, the trained risk analysis model is subjected to hierarchical freezing. This step selectively fixes model parameters according to the universality of features at different layers. Specifically, the bottom convolutional layer or recurrent layer usually captures more universal features, so it can be completely frozen; while the high-level fully connected layer is more task-specific and needs to remain trainable. The basic migration model obtained by this processing method retains the universal feature extraction capability of the original model, while providing flexibility for subsequent adaptation in new business scenarios. Next, the basic migration model is domain-adaptively trained according to the feature distribution of the new business scenario. This step adopts an adversarial learning method, which aims to minimize the difference in feature distribution between the source domain (original business scenario) and the target domain (new business scenario). The specific implementation can use the Domain-Adversarial Neural Network (DANN). In DANN, in addition to the original feature extractor and task classifier, a domain classifier is also introduced. The goal of the feature extractor is to generate features that are difficult for the domain classifier to distinguish, while the domain classifier attempts to accurately distinguish which domain the feature comes from. Through this adversarial process, the model learns feature representations that are more universal between the source domain and the target domain, thereby obtaining a domain alignment model. Then, the domain alignment model and the new business data are progressively fine-tuned. This process uses a hierarchical learning rate strategy to gradually unfreeze and optimize the parameters of different layers. Specifically, only the last few layers are trained first, and then more layers are gradually unfrozen until the entire network participates in the training. In this process, different layers use different learning rates, usually a smaller learning rate for the bottom layer and a larger learning rate for the upper layer. This strategy not only retains the useful features learned by the model on the original task, but also allows the model to adapt to new business scenarios. Through this step, a scenario adaptation model that better adapts to the new scenario is obtained. Then, meta-learning parameter optimization is performed based on the prediction results of the scenario adaptation model. The goal of meta-learning is to learn the optimal initialization state of the model parameters so that the model can quickly adapt to new tasks. The specific implementation can be the Model-Agnostic Meta-Learning (MAML) algorithm. MAML simulates different task distributions and learns a set of initial parameters so that the model can achieve good performance with only a small number of gradient update steps when facing new tasks. This process involves two layers of optimization: the inner layer optimization is for rapid adaptation of a single task, and the outer layer optimization is to optimize the initial parameters across multiple tasks. In this way, a fast adaptation model is obtained that can quickly adapt to new business scenarios. Finally, the fast adaptation model is subjected to knowledge distillation and model compression to obtain a lightweight migration model. During the knowledge distillation process, the fast adaptation model is used as the teacher model to train a smaller student model. The student model not only learns to match the hard labels of the teacher model (i.e., the final prediction), but also learns to imitate the soft labels of the teacher model (i.e., the category probability distribution).This allows the fine-grained knowledge learned by the teacher model to be transferred. At the same time, model compression is performed, including parameter quantization and structural pruning. Parameter quantization converts model parameters from 32-bit floating point numbers to 8-bit integers, significantly reducing the model size. Structural pruning removes unimportant neurons or connections, reducing computational complexity while maintaining performance. The output of this step is a lightweight migration model that retains the core capabilities of the original model while having higher computational efficiency.

[0026] 102. Distribute and configure the risk analysis model for streaming to obtain a real-time risk reasoning system; In one embodiment of the present invention, the distributed deployment and streaming processing configuration of the risk analysis model to obtain a real-time risk reasoning system includes: compressing and quantizing the risk analysis model to obtain a lightweight reasoning model, and building a microservice architecture based on the lightweight reasoning model to obtain a distributed reasoning framework; wherein the microservice architecture encapsulates different components of the model into independent service units, and realizes communication between service units through message queues; load balancing and elastic scaling configuration of the distributed reasoning framework to obtain a dynamic resource scheduling system; designing a streaming data processing pipeline based on the dynamic resource scheduling system to obtain a real-time feature extraction module; integrating and performance optimizing the real-time feature extraction module and the lightweight reasoning model to obtain a real-time risk reasoning system.

[0027] Specifically, the risk analysis model is first compressed and quantized. Model compression uses knowledge distillation technology to transfer the knowledge of the original large model to a smaller model. Specifically, the original model is used as a teacher model to generate soft labels (category probability distribution), and then a student model with a simpler structure is trained to imitate these soft labels. At the same time, model pruning is performed to remove neuron connections that contribute less to the prediction results. Quantization converts model parameters from 32-bit floating point numbers to 8-bit integer representations, significantly reducing model size and computational complexity. The output of these steps is a lightweight inference model that retains the core functions of the original model but has a smaller size and faster inference speed. Next, a microservice architecture is built based on the lightweight inference model to form a distributed inference framework. In this architecture, different components of the model are encapsulated as independent service units. For example, data preprocessing, feature extraction, risk assessment and other functions are respectively used as independent microservices. These microservices communicate through RESTful API or gRPC, ensuring the modularity and scalability of the system. At the same time, message queues (such as Apache Kafka or RabbitMQ) are introduced to implement asynchronous communication between service units to improve the throughput and fault tolerance of the system. This microservice architecture enables the system to flexibly expand and update individual components without affecting the overall functionality. Then, the distributed inference framework is configured for load balancing and elastic scaling to build a dynamic resource scheduling system. Load balancing uses a consistent hashing algorithm for request distribution to ensure that request remapping is minimized when the number of servers changes. Elastic scaling automatically increases or decreases computing resources by monitoring the system load. The specific implementation can use container orchestration platforms such as Kubernetes to dynamically adjust the number of service instances based on indicators such as CPU usage, memory usage, and request queue length. In this way, the system can automatically adjust resource configuration when traffic fluctuates, which not only ensures service quality but also avoids resource waste. Next, a streaming data processing pipeline is designed based on the dynamic resource scheduling system to build a real-time feature extraction module. This module uses a stream processing framework (such as Apache Flink or Spark Streaming) to process real-time data streams. The data processing pipeline includes data reception, preprocessing, feature calculation, and cache management. The data receiving layer uses message systems such as Kafka to access real-time data streams. The preprocessing stage performs data cleaning and format conversion. The feature calculation stage implements real-time feature engineering, including feature transformation, combination, and encoding. Cache management uses memory databases such as Redis to store intermediate calculation results to improve processing efficiency. The entire streaming processing pipeline supports incremental calculations and sliding window updates to ensure the real-time nature and computational efficiency of features. Finally, the real-time feature extraction module and lightweight reasoning model are integrated and performance optimized to obtain a real-time risk reasoning system.The integration process uses pipeline parallel processing to closely combine feature extraction and model reasoning to reduce data transmission overhead. Performance optimization includes several aspects: First, memory management optimization is performed, and memory pool technology is used to reduce memory allocation and recycling overhead. Second, computational graph optimization is performed, such as operator fusion and constant folding, to reduce unnecessary intermediate calculation steps. Third, hardware acceleration is used, such as using GPU for parallel computing or FPGA for dedicated acceleration. Finally, a batch processing strategy is implemented to package multiple requests for processing while ensuring real-time performance to improve throughput.

[0028] 103. Obtain real-time operational business data, and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; In one embodiment of the present invention, the real-time operational business data is obtained, and the real-time operational business data is input into the real-time risk reasoning system to obtain the risk analysis result and the corresponding feature importance distribution, including: synchronously processing the real-time operational business data to obtain the original data stream, and performing real-time data cleaning and anomaly detection on the original data stream to obtain a preprocessed data stream; performing real-time feature engineering and feature selection on the preprocessed data stream to obtain a feature vector stream; inputting the feature vector stream into the risk analysis model in the real-time risk reasoning system to obtain a multimodal feature representation, wherein the risk analysis model includes a fully connected layer, a loop layer, a convolutional layer and an output layer, the fully connected layer performs a nonlinear transformation on the structured data of the feature vector stream, the loop layer captures the timing dependency of the sequence data of the feature vector stream, and the convolutional layer extracts local semantic features from the text data of the feature vector stream; cross-modal attention fusion is performed on the multimodal feature representation to obtain a unified risk representation; risk score calculation is performed through the output layer of the risk analysis model according to the unified risk representation to obtain the risk analysis result and the corresponding feature importance distribution.

[0029] Specifically, the real-time operational business data is first synchronized. This step uses a distributed time synchronization protocol, such as the Precision Time Protocol (PTP) or the Network Time Protocol (NTP), to ensure that the data timestamps from different systems and devices are consistent. The synchronized data forms a raw data stream, which contains various operational business information, such as user transaction records, network usage logs, customer service interactions, etc. Next, the raw data stream is cleaned and anomaly detected in real time. The data cleaning process includes deduplication, completion, and format conversion. Deduplication uses Bloom filters to quickly identify and delete duplicate data. Data completion uses interpolation algorithms (such as linear interpolation or nearest neighbor interpolation) to fill missing values. Format conversion ensures that all data fields conform to a predefined standard format. Anomaly detection uses a combination of statistical methods and machine learning algorithms. Specifically, the Z-score method is used to identify numerical anomalies, the isolation forest algorithm is used to detect anomalies in the multidimensional feature space, and the rule-based expert system is used to capture known business logic anomalies. The output of these steps is a preprocessed data stream that has been cleaned and preliminarily screened. Then, real-time feature engineering and feature selection are performed on the preprocessed data stream. Feature engineering includes feature transformation, combination and encoding. Numerical features are standardized or normalized to ensure that features of different dimensions are comparable. Categorical features are converted to numerical representations through one-hot encoding or embedding layers. Time-related features extract information such as year, month, day, and hour of the timestamp and calculate the time difference. Feature combination creates high-order features through cross-product operations to capture nonlinear relationships between features. Feature selection uses an online feature screening algorithm based on mutual information to dynamically evaluate the importance of each feature and retain the feature subset with the largest amount of information. The output of this step is a high-dimensional feature vector stream that contains rich risk-related information. Next, the feature vector stream is input into the risk analysis model in the real-time risk reasoning system. The model is a multi-branch deep neural network, including fully connected layers, recurrent layers, convolutional layers, and output layers. The fully connected layer is responsible for processing structured data, such as user attributes, transaction amounts, etc. In specific implementation, a multi-layer perceptron (MLP) structure is used, and batch normalization and ReLU activation functions are added between each layer to accelerate training and introduce nonlinearity. The recurrent layer processes sequence data, such as the user's historical behavior sequence. Long short-term memory networks (LSTM) or gated recurrent units (GRU) are used here to effectively capture long-term dependencies. Convolutional layers are used to process text data, such as customer feedback or service records. A one-dimensional convolutional neural network is used in conjunction with a maximum pooling layer to extract local semantic features. These three branches process different types of input data in parallel, each outputting a feature representation. Then, these multimodal feature representations are fused with cross-modal attention to obtain a unified risk representation. The core idea of ​​the cross-modal attention mechanism is to allow the model to learn the correlation between different modal features. In the specific implementation, the attention weights of each modal feature and other modal features are first calculated.This is done by performing a dot product operation on the features of one modality with the features of the other modalities and then normalizing them with a softmax function. Then, the features of the other modalities are weighted and summed with these weights to obtain the context vector of the current modality. Finally, the original features are concatenated with the context vector and passed through a feedforward neural network to obtain the fused feature representation. This process is repeated for each modality, and finally a unified and informative risk representation is obtained. Finally, the risk score is calculated for the unified risk representation using the output layer of the risk analysis model. The output layer is usually a fully connected layer with the number of neurons corresponding to the number of risk categories. For binary classification problems (such as fraud detection), a single neuron is used with a sigmoid activation function. For multi-classification problems (such as multi-level risk assessment), multiple neurons are used with a softmax activation function. The final output of the model is a risk score, which represents the probability distribution of the sample belonging to each risk category. At the same time, in order to obtain the feature importance distribution, the Integrated Gradients method is used. This method quantifies the contribution of each feature to the final prediction by calculating the gradient integral of the model output relative to the input feature. The specific steps are as follows: First, select a reference input (usually a zero vector or a feature mean). Then, create m intermediate points between the reference input and the actual input. For each intermediate point, calculate the gradient of the model output with respect to the input. Finally, accumulate and average these gradients to obtain the importance score of each feature. This method not only considers the direct impact of the feature, but also captures the interaction effect between features.

[0030] 104. Through the preset multi-model integration framework, the risk analysis results and feature importance distribution are analyzed for risk assessment synthesis and interpretability to obtain a comprehensive risk assessment report.

[0031] In one embodiment of the present invention, the risk analysis results and the feature importance distribution are subjected to risk assessment synthesis and interpretability analysis through a preset multi-model integration framework to obtain a comprehensive risk assessment report, including: multi-model fusion of the risk analysis results and the feature importance distribution to obtain an integrated risk score, and Bayesian calibration based on the integrated risk score and the historical risk distribution to obtain a calibrated risk probability; multi-dimensional decomposition of the calibrated risk probability to obtain a risk source analysis; wherein the multi-dimensional decomposition includes attributing and quantifying risks according to business type, user group and time period dimensions; generating decision tree rules based on the risk source analysis and the feature importance distribution to obtain an interpretability rule set; and comprehensively collating and visualizing the interpretability rule set and the calibrated risk probability to obtain a comprehensive risk assessment report.

[0032] Specifically, first, the risk analysis results and feature importance distribution are multi-model fused. This step uses the Stacking method to take the outputs of multiple basic models as new features and train a meta-learner for secondary prediction. Specifically, first select a variety of different types of models as basic models, such as random forests, gradient boosting decision trees, support vector machines, etc., as well as previously trained deep neural network models. These models predict the input data respectively to generate risk scores and feature importance distributions. Then, the output results of these models and the original features are combined to form a new feature matrix. Next, a meta-learner (usually logistic regression or lightweight gradient boosting tree) is trained to learn on this new feature matrix using cross-validation to obtain the final integrated risk score. This method can combine the advantages of multiple models and improve the stability and accuracy of prediction. After obtaining the integrated risk score, the next step is to perform Bayesian calibration. This process uses historical risk distribution as prior knowledge to calibrate the integrated risk score to obtain a more accurate risk probability estimate. In specific implementation, first build a calibration model, such as Platt scaling or equal binning. Then, the calibration model is trained using real risk events in historical data as labels and integrated risk scores as features. In the application phase, the new integrated risk score is input into the calibration model to obtain the calibrated risk probability. This method can solve the problem of inconsistency between the model output score and the actual risk probability, making the risk assessment results more reliable and meaningful. Next, the calibrated risk probability is decomposed in multiple dimensions to obtain a detailed risk source analysis. This step includes attributing and quantifying risks by dimensions such as business type, user group, and time period. In specific implementation, the classification standards of each dimension are first defined, such as business type can be divided into voice, data, value-added services, etc.; user groups can be divided by age, consumption level, usage habits, etc.; time periods can be divided into different granularities such as day, week, month, etc. Then, each sample is labeled according to these dimensions. Next, the contribution of each dimension to the risk is calculated using the decision tree or SHAP (SHapley Additive exPlanations) value method. For the decision tree method, the importance of each dimension can be quantified by the split point of the tree and the purity change of each node. The SHAP value method determines the importance of each feature by calculating its marginal contribution to the model output. Finally, the contribution of each dimension is summarized to form a multi-level risk source analysis report, which details the risk distribution of different business types, user groups and time periods. Then, decision tree rules are generated based on the risk source analysis and feature importance distribution to obtain an interpretable rule set. This process uses decision tree algorithms, such as C4.5 or CART, to transform complex model decision processes into a series of easy-to-understand if-then rules.The specific steps are as follows: First, the dimensional features and the original feature importance distribution in the risk source analysis are used as input features, and the calibrated risk probability is used as the target variable to train a decision tree model. During the training process, pruning techniques are used to control the complexity of the tree to avoid overfitting. Next, the decision tree is traversed from the root node to the leaf node, and each path is converted into a rule. For example, "If the user's age is <25 and the average monthly consumption is >500 yuan and the data usage increases suddenly at night, the fraud risk probability is >80%". In order to improve the representativeness of the rules, rule extraction algorithms such as RuleFit can be used, which combines the advantages of decision trees and linear models to generate more refined and insightful rule sets. Finally, the generated rules are screened and sorted, and the rules with strong predictive ability and wide coverage are retained to form the final interpretable rule set. The last step is to comprehensively organize and visualize the interpretable rule set and the calibrated risk probability to obtain a comprehensive risk assessment report. This process first classifies and organizes the rule set, and classifies it according to dimensions such as risk type and impact. Then, the calibrated risk probabilities are associated with these rules to display the risk level and confidence interval corresponding to each rule. Next, a series of visualization charts are designed to intuitively present risk information. For example, heat maps are used to display the risk distribution of different business types and user groups; time series diagrams are used to show the trend of risk changes over time; and decision tree visualization tools are used to display key decision paths. At the same time, natural language summaries are generated to provide textual descriptions of high-risk points, abnormal patterns, and risk evolution trends. Finally, all these elements are integrated into a structured report template to form the final comprehensive risk assessment report. This report not only contains quantitative risk scores and probabilities, but also provides qualitative explanations and suggestions, allowing risk managers to fully understand the risk situation and formulate targeted response strategies accordingly.

[0033] In this embodiment, a structured risk feature data set is generated by multi-modal preprocessing of multi-source heterogeneous data, and the preset multi-branch deep neural network model is trained and dynamically optimized in stages to obtain a trained risk analysis model. The model is distributed deployed and configured for streaming processing to build a real-time risk reasoning system. Real-time operational business data is input into the system to generate risk analysis results and corresponding feature importance distributions. Through a multi-model integration framework, the risk analysis results and feature importance distributions are comprehensively evaluated and interpreted to form a comprehensive risk assessment report. This method fully mines and integrates risk features from different data sources through multi-modal preprocessing of multi-source heterogeneous data, thereby overcoming the problem of inaccurate analysis caused by single data and unprocessed heterogeneity in traditional methods.

[0034] The above describes the method for analyzing the operational business risk in the embodiment of the present invention. The following describes the device for analyzing the operational business risk in the embodiment of the present invention. Figure 2 , an embodiment of the operation business risk analysis device in the embodiment of the present invention includes: A model training module 201 is used to perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and to perform phased training and dynamic optimization on a preset multi-branch deep neural network model according to the structured risk feature data set to obtain a trained risk analysis model; A model deployment module 202, configured to perform distributed deployment and stream processing configuration on the risk analysis model to obtain a real-time risk reasoning system; The risk analysis module 203 is used to obtain real-time operational business data and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; The explainability analysis module 204 is used to perform risk assessment synthesis and explainability analysis on the risk analysis results and the feature importance distribution through a preset multi-model integration framework to obtain a comprehensive risk assessment report.

[0035] In an embodiment of the present invention, the operational business risk analysis device runs the above-mentioned operational business risk analysis method. The operational business risk analysis device generates a structured risk feature data set by performing multimodal preprocessing on multi-source heterogeneous data, and performs phased training and dynamic optimization on the preset multi-branch deep neural network model to obtain a trained risk analysis model. The model is distributed deployed and stream processed to build a real-time risk reasoning system. Real-time operational business data is input into the system to generate risk analysis results and corresponding feature importance distributions. Through a multi-model integration framework, the risk analysis results and feature importance distributions are comprehensively evaluated and interpreted to form a comprehensive risk assessment report. This method fully mines and integrates risk features of different data sources by multimodal preprocessing of multi-source heterogeneous data, thereby overcoming the problem of inaccurate analysis caused by single data and unprocessed heterogeneity in traditional methods.

[0036] above Figure 2 The operational business risk analysis device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The operational business risk analysis device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0037] Figure 33 is a schematic diagram of the structure of an operation business risk analysis device provided by an embodiment of the present invention. The operation business risk analysis device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 may be temporary storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the operation business risk analysis device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the operation business risk analysis device 300 to implement the steps of the above-mentioned operation business risk analysis method.

[0038] The operational business risk analysis device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The illustrated operational business risk analysis device structure does not constitute a limitation on the operational business risk analysis device provided by the present invention, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0039] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the operational business risk analysis method.

[0040] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0041] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0042] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing operational business risks, characterized in that: The operational business risk analysis method includes: Perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and perform phased training and dynamic optimization on a preset multi-branch deep neural network model based on the structured risk feature data set to obtain a trained risk analysis model; Distributed deployment and streaming processing configuration are performed on the risk analysis model to obtain a real-time risk reasoning system; Acquire real-time operational business data, and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; The risk analysis results and the feature importance distribution are subjected to risk assessment synthesis and interpretability analysis through a preset multi-model integration framework to obtain a comprehensive risk assessment report.

2. The operational business risk analysis method according to claim 1, characterized in that: The multi-source heterogeneous data is subjected to multi-modal preprocessing to obtain a structured risk feature data set, and a preset multi-branch deep neural network model is trained and dynamically optimized in stages according to the structured risk feature data set to obtain a trained risk analysis model, including: Perform data cleaning and format unification processing on multi-source heterogeneous data to obtain a standardized data set, and perform dynamic time window division and multi-scale feature extraction on the standardized data set to obtain a time series feature set; Performing feature engineering processing on the time series feature set and the non-time series data in the standardized data set to obtain label data corresponding to the enhanced feature set; Performing unsupervised pre-training on a preset multi-branch deep neural network model according to the enhanced feature set to obtain a pre-trained model; The pre-trained model is fine-tuned and dynamically optimized in a supervised manner according to the enhanced feature set and label data to obtain a trained risk analysis model.

3. The method for analyzing operational business risks according to claim 2, characterized in that: The supervised fine-tuning and dynamic optimization of the pre-trained model according to the enhanced feature set and label data to obtain a trained risk analysis model includes: Performing stratified sampling and data enhancement processing on the enhanced feature set and label data to obtain a balanced training data set; Performing multi-task joint optimization on each branch of the pre-trained model according to the balanced training data set to obtain a preliminary fine-tuning model; Performing adversarial training and noise injection on the preliminary fine-tuning model to obtain a robustness enhanced model; According to the prediction results of the robustness enhancement model, the hard-to-classify samples are identified and focused on learning to obtain a hard-to-classify sample enhancement model; Incremental learning and model distillation techniques are applied to the difficult sample reinforcement model to obtain a trained risk analysis model.

4. The method for analyzing operational business risks according to claim 1, characterized in that: The distributed deployment and streaming processing configuration of the risk analysis model to obtain a real-time risk reasoning system includes: The risk analysis model is compressed and quantified to obtain a lightweight reasoning model, and a microservice architecture is constructed based on the lightweight reasoning model to obtain a distributed reasoning framework; wherein the microservice architecture encapsulates different components of the model into independent service units, and realizes communication between service units through message queues; Perform load balancing and elastic scaling configuration on the distributed reasoning framework to obtain a dynamic resource scheduling system; Design a streaming data processing pipeline according to the dynamic resource scheduling system to obtain a real-time feature extraction module; The real-time feature extraction module and the lightweight reasoning model are integrated and performance optimized to obtain a real-time risk reasoning system.

5. The method for analyzing operational business risks according to claim 1, characterized in that: The acquiring of real-time operational business data and inputting the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distributions includes: Synchronously process the real-time operational business data to obtain an original data stream, and perform real-time data cleaning and anomaly detection on the original data stream to obtain a pre-processed data stream; Performing real-time feature engineering and feature selection on the preprocessed data stream to obtain a feature vector stream; Inputting the feature vector stream into the risk analysis model in the real-time risk reasoning system to obtain a multimodal feature representation, wherein the risk analysis model includes a fully connected layer, a recurrent layer, a convolutional layer and an output layer, the fully connected layer performs nonlinear transformation on the structured data of the feature vector stream, the recurrent layer captures the temporal dependency of the sequence data of the feature vector stream, and the convolutional layer extracts local semantic features from the text data of the feature vector stream; Performing cross-modal attention fusion on the multimodal feature representation to obtain a unified risk representation; The risk score is calculated according to the unified risk representation through the output layer of the risk analysis model to obtain the risk analysis result and the corresponding feature importance distribution.

6. The method for analyzing operational business risks according to claim 1, characterized in that: The risk analysis results and the feature importance distribution are subjected to risk assessment synthesis and interpretability analysis through a preset multi-model integration framework to obtain a comprehensive risk assessment report including: Performing multi-model fusion on the risk analysis results and feature importance distribution to obtain an integrated risk score, and performing Bayesian calibration based on the integrated risk score and historical risk distribution to obtain a calibrated risk probability; Performing multi-dimensional decomposition on the calibrated risk probability to obtain risk source analysis; wherein the multi-dimensional decomposition includes attributing and quantifying risks according to business type, user group and time period dimensions; Generate decision tree rules according to the risk source analysis and the feature importance distribution to obtain an explainable rule set; The explainability rule set and the calibrated risk probability are comprehensively collated and visualized to obtain a comprehensive risk assessment report.

7. The method for analyzing operational business risks according to claim 1, characterized in that: After performing multimodal preprocessing on the multi-source heterogeneous data to obtain a structured risk feature data set, and performing phased training and dynamic optimization on a preset multi-branch deep neural network model according to the structured risk feature data set to obtain a trained risk analysis model, the method further includes: Performing a hierarchical freezing process on the risk analysis model to obtain a basic migration model; wherein the hierarchical freezing process selectively fixes model parameters according to the universality of features of different layers; The basic migration model is subjected to domain adaptive training according to the feature distribution of the new business scenario to obtain a domain alignment model; wherein the domain adaptive training minimizes the feature distribution difference between the source domain and the target domain through an adversarial learning method; Performing progressive fine-tuning on the domain alignment model and the new business data to obtain a scenario adaptation model; wherein the progressive fine-tuning adopts a layered learning rate strategy to gradually unfreeze and optimize parameters of different layers; Performing meta-learning parameter optimization according to the prediction results of the scene adaptation model to obtain a fast adaptation model; wherein the meta-learning parameter optimization learns the optimal initialization state of the model parameters by simulating different task distributions; The rapid adaptation model is subjected to knowledge distillation and model compression to obtain a lightweight migration model, and the lightweight migration model is used as a risk analysis model for new business scenarios.

8. An operational business risk analysis device, characterized in that: The operational business risk analysis device comprises: A model training module is used to perform multimodal preprocessing on multi-source heterogeneous data to obtain a structured risk feature data set, and to perform phased training and dynamic optimization on a preset multi-branch deep neural network model according to the structured risk feature data set to obtain a trained risk analysis model; A model deployment module, used for performing distributed deployment and streaming processing configuration of the risk analysis model to obtain a real-time risk reasoning system; A risk analysis module, used to obtain real-time operational business data, and input the real-time operational business data into the real-time risk reasoning system to obtain risk analysis results and corresponding feature importance distribution; The explainability analysis module is used to perform risk assessment synthesis and explainability analysis on the risk analysis results and the feature importance distribution through a preset multi-model integration framework to obtain a comprehensive risk assessment report.

9. An operational business risk analysis device, characterized in that: The operational business risk analysis device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the operational business risk analysis device to perform the steps of the operational business risk analysis method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the operational business risk analysis method as described in any one of claims 1 to 7 are implemented.

Images