Financial transaction risk assessment method and system based on multilayer neural network

Abstract

The invention relates to the technical field of data processing, and discloses a financial transaction risk assessment method and system based on a multilayer neural network. The method comprises the following steps: acquiring financial multi-dimensional risk indexes to form a feature matrix; converting the feature matrix into a two-dimensional risk map data set; training a CNN-TA model and a multi-layer sensor by using the atlas data to obtain a risk network; and inputting real-time data to calculate a risk level, and generating a hierarchical rating report and an early warning signal. According to the method, the time sequence characteristic and the spatial correlation of the financial data are utilized, a refined multi-level risk rating result is provided, and the method has risk source traceability and risk evolution trend prediction capability, so that omnibearing and dynamic evaluation of financial transaction risks is realized.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a financial transaction risk assessment method and system based on a multi-layer neural network. Background Art

[0002] With the increasing globalization and complexity of financial markets, financial transaction risk assessment has become a core concern for investment institutions and financial regulators. Traditional financial risk assessment methods primarily rely on statistical models, such as value at risk (VaR), expected shortfall (ES), and stress testing. These methods are typically based on statistical analysis of historical data and assume that market conditions follow specific probability distribution models. In recent years, with the advancement of computer science and artificial intelligence, machine learning methods have begun to be applied to financial risk assessment. Among them, neural network models, due to their powerful nonlinear modeling capabilities, have demonstrated significant advantages in processing complex financial data. Currently, mainstream neural network risk assessment methods include risk scoring models based on multilayer perceptrons (MLPs), time series risk prediction systems based on recurrent neural networks (RNNs), and financial anomaly detection methods based on long short-term memory networks (LSTMs). These methods have achieved considerable progress in risk classification, anomaly detection, and risk trend prediction.

[0003] However, existing neural network risk assessment methods still suffer from several technical limitations. First, most models focus on a single type of financial data representation, processing either raw time series data or derived financial indicators, lacking the ability to comprehensively leverage multidimensional financial risk characteristics. Second, existing models typically employ a single neural network architecture, unable to simultaneously capture both the temporal patterns and spatial correlations of financial data, resulting in incomplete risk feature extraction. Third, risk assessment results generally lack interpretability, making it difficult to trace the risk levels output by the models to specific sources of risk, limiting their practical application in decision-making. Fourth, most methods offer only a simple three-level risk classification of high, medium, and low, resulting in a coarse granularity that cannot meet the needs of refined risk management. Fifth, risk assessment results are typically static judgments, lacking the ability to dynamically track and predict risk evolution trends, and are unable to proactively identify potential risk accumulation. These technical limitations severely restrict the practical application of neural network methods in financial risk management. Summary of the Invention

[0004] This application provides a financial transaction risk assessment method and system based on a multi-layer neural network, which is used to utilize the temporal characteristics and spatial correlation of financial data to provide refined multi-level risk rating results, and has the ability to trace risk sources and predict risk evolution trends, thereby realizing a comprehensive and dynamic assessment of financial transaction risks.

[0005] In the first aspect, the present application provides a financial transaction risk assessment method based on a multi-layer neural network, and the financial transaction risk assessment method based on a multi-layer neural network includes: collecting and calculating multi-dimensional risk indicators of financial market data to obtain a financial risk feature matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors; performing time series to image processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map data set suitable for two-dimensional convolution analysis; training a two-dimensional convolutional neural network trading model CNN-TA and a multi-layer perceptron based on the risk feature map data set to obtain a financial transaction risk assessment network; inputting real-time transaction data into the financial transaction risk assessment network to calculate and classify risk levels to obtain a hierarchical structured risk rating report and transaction warning signals.

[0006] In a second aspect, the present application provides a financial transaction risk assessment system based on a multi-layer neural network, the financial transaction risk assessment system based on a multi-layer neural network comprising:

[0007] The acquisition module is used to collect and calculate multi-dimensional risk indicators of financial market data to obtain a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors;

[0008] A conversion module, configured to perform time series to image processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map dataset suitable for two-dimensional convolution analysis;

[0009] A training module, configured to train a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network;

[0010] The classification module is used to input real-time transaction data into the financial transaction risk assessment network to calculate and classify risk levels, and obtain a hierarchical structured risk rating report and transaction early warning signals.

[0011] In a third aspect, a financial transaction risk assessment device based on a multi-layer neural network is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the financial transaction risk assessment device based on the multi-layer neural network executes the above-mentioned financial transaction risk assessment method based on the multi-layer neural network.

[0012] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned financial transaction risk assessment method based on a multi-layer neural network.

[0013] The technical solution provided in this application achieves precise assessment and early warning of financial transaction risks through a combination of technical features: multidimensional risk indicator acquisition and calculation, time series to image processing, fusion training of a two-dimensional convolutional neural network and a multilayer perceptron, and real-time risk level calculation and classification. This approach has significant technical benefits. First, the multidimensional risk indicator acquisition and calculation process transforms complex market information into a structured financial risk feature matrix, comprehensively capturing multi-angle risk signals such as market volatility, liquidity ratios, systemic risk values, and macroeconomic factors. This overcomes the limitations of traditional single-indicator assessment and significantly improves the comprehensiveness and accuracy of risk identification. Second, the time series to image processing technology converts one-dimensional financial data into a two-dimensional risk feature map, creatively transforming the variation patterns of financial time series into spatial distribution features. This enables the subsequent two-dimensional convolutional neural network to effectively extract the spatiotemporal correlation patterns of financial risks. This algorithmic feature fully leverages the advantages of convolutional neural networks in the field of image recognition, successfully transferring them to the field of financial risk assessment, and achieving a deep integration of algorithms and application scenarios. Third, the combined training mechanism of a two-dimensional convolutional neural network (CNN-TA) and a multi-layer perceptron (MLP) enables multimodal feature extraction and fusion. CNN-TA focuses on capturing spatial correlation patterns within risk feature maps, while MLP focuses on learning the complex nonlinear relationships of raw risk indicators. The two complement each other, forming a more comprehensive risk representation capability. This combined design of algorithmic features significantly enhances the model's ability to identify complex financial risk patterns. Fourth, real-time risk level calculation and classification technology transforms risk assessment into practical decision support. Through a nine-level refined risk classification and a three-layer structured risk rating report, it not only provides risk level judgments but also deeply analyzes the sources and evolution trends of risks, significantly enhancing the operability and decision-making value of risk assessment results. It is particularly worth noting that the introduction of the two-dimensional convolutional neural network (CNN-TA) algorithmic features in this solution fully considers the spatiotemporal correlation characteristics of financial risk data. By converting time series features into spatial feature maps, the local receptive field and hierarchical feature extraction capabilities of the convolutional neural network are fully utilized, achieving efficient extraction of financial risk patterns. This algorithmic feature is deeply aligned with the application domain of financial risk assessment and is the key to the solution's excellent results. In summary, this method achieves a significant improvement in the comprehensiveness, accuracy, real-timeness and interpretability of financial transaction risk assessment through precise matching of algorithm features with application requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of a financial transaction risk assessment method based on a multi-layer neural network in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a financial transaction risk assessment system based on a multi-layer neural network in an embodiment of the present application;

[0017] Figure 3 It is a schematic block diagram of the structure of a financial transaction risk assessment device based on a multi-layer neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present application provide a method and system for assessing financial transaction risks based on a multi-layer neural network. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for assessing financial transaction risk based on a multi-layer neural network includes:

[0020] Step S101: Collect and calculate multi-dimensional risk indicators for financial market data to obtain a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors;

[0021] Step S102: Perform time series to image processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map dataset suitable for two-dimensional convolution analysis;

[0022] Step S103: training a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network;

[0023] Step S104: input the real-time transaction data into the financial transaction risk assessment network to calculate and classify the risk level, and obtain a hierarchical structured risk rating report and transaction warning signals.

[0024] It is understandable that the execution subject of this application can be a financial transaction risk assessment system based on a multi-layer neural network, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0025] In this example, historical trading data for major stock indices such as the DJIA, Nasdaq 100, and Nikkei 225 were first collected, covering a market cycle of at least 10 years. This raw financial market data was segmented into three different time scales: daily, weekly, and monthly, to form a multi-time-dimensional financial dataset. Market volatility indicators, liquidity ratios, systemic risk indicators, and macroeconomic factors were then extracted from this dataset. Market volatility indicators include historical volatility and Bollinger Band width; liquidity ratios include trading volume change rate and bid-ask spread ratio; systemic risk indicators include market beta and systemic importance index; and macroeconomic factors include GDP growth rate expectations and inflation rate. These indicators form their own characteristic submatrices, which are then merged to form a comprehensive risk characteristic matrix. This matrix is then normalized to convert all indicator values into a standard normal distribution with a mean of 0 and a standard deviation of 1, ultimately yielding a financial risk characteristic matrix. Based on the financial risk characteristic matrix, the historical trading data is converted from time series to images. Fifteen different risk indicators were selected from the financial risk signature matrix to form a risk indicator subset. Different parameter combinations were set for these indicators to form an indicator parameter matrix. For each risk indicator, 15 consecutive days of time series data were extracted. These time-expanded risk data were reconstructed into a 15×15 two-dimensional matrix to form an initial risk signature map. These initial risk signature maps were numerically normalized to obtain feature images with uniform pixel value distribution. Based on the actual risk results of historical transactions, the feature images were labeled as "high risk," "medium risk," or "low risk," forming a labeled risk signature map. Finally, the labeled risk signature map was split into training, validation, and test sets in a 7:2:1 ratio to obtain the final risk signature map dataset.

[0026] A two-dimensional convolutional neural network trading model (CNN-TA) and a multilayer perceptron (MLP) were trained based on the risk profile dataset. The risk profile dataset was fed into the CNN-TA, which consists of three convolutional layers: 32 5×5 kernels in the first layer, 64 3×3 kernels in the second layer, and 128 3×3 kernels in the third layer. The convolutional layer outputs were subjected to a maximum pooling operation to generate a reduced-dimensional feature map. This map was then transformed through a fully connected layer to obtain the CNN-TA risk feature vector. Simultaneously, the financial risk feature matrix was fed into a multilayer perceptron (MLP) with three fully connected layers, each with 256, 128, and 64 neurons, to obtain the MLP risk feature vector. The CNN-TA risk feature vector was then fused with the MLP risk feature vector to obtain a comprehensive risk representation. This feature representation was mapped to nine risk levels using an output layer consisting of nine neurons: extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk, resulting in a risk probability distribution. Finally, through batch training and parameter optimization, using the cross-entropy loss function and the Adam optimizer with a learning rate of 0.001, a financial transaction risk assessment network was obtained.

[0027] Real-time transaction data is fed into a financial transaction risk assessment network for risk level calculation and classification. Multidimensional risk indicators are extracted from the data to form a real-time risk feature matrix consistent with the training data format. This matrix is then converted into a 15×15 two-dimensional risk feature image to create a real-time risk feature map. This map is then fed into the financial transaction risk assessment network for forward computation, resulting in a risk probability distribution for the real-time transactions. Threshold determination and categorization are performed on the risk probability distribution to determine a nine-level risk rating for the transaction. Based on these results, an overall risk level, risk trend analysis, and key risk indicator analysis are generated to form a three-tiered structured risk rating report. The nine-level risk rating results are compared with pre-set risk thresholds to generate targeted transaction warning signals. Warning signals are recorded and stored on daily, weekly, and monthly timescales to form a multi-period warning signal database. The warning trigger frequency for different risk levels is calculated, and a risk warning heat map and time series trend analysis are generated. Taking a financial institution's foreign exchange transactions as an example, ten years of historical data on the USD / EUR exchange rate are collected, and various risk indicators are extracted to form a 150-dimensional feature matrix. Fifteen key indicators, such as short-term volatility, liquidity indicators, and the political risk index, were selected to construct a 15×15 risk signature map. After training the CNN-TA and multi-layer perceptron, the system assesses the risk level of new foreign exchange transactions in real time. When the system detects a transaction triggering an extremely high risk level, it generates an early warning signal and outputs a detailed risk report, identifying the key risk source as the combined impact of recent declines in liquidity and increases in systemic risk. This helps traders adjust their trading strategies promptly and effectively mitigate potential risks. This entire process requires no human intervention, leveraging the deep learning capabilities of neural networks to automatically and accurately assess and warn of risks.

[0028] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0029] Collect historical trading data of DJIA, Nasdaq 100 and NIKKEI 225 stock indices to obtain raw financial market data with a span of at least 10 years;

[0030] The original financial market data is segmented and processed according to the three time scales of day, week and month to obtain a multi-time dimension financial data set;

[0031] Extract market volatility indicators from multi-time dimension financial data sets to obtain market volatility characteristic submatrix;

[0032] Extract liquidity risk indicators from multi-time dimension financial data sets to obtain liquidity risk characteristic submatrix;

[0033] Extract systemic risk indicators from multi-time dimension financial data sets and obtain systemic risk characteristic submatrix;

[0034] Merge the market volatility characteristic sub-matrix, liquidity risk characteristic sub-matrix and systemic risk characteristic sub-matrix to obtain a comprehensive risk characteristic matrix;

[0035] The comprehensive risk characteristic matrix is standardized to obtain the final financial risk characteristic matrix.

[0036] Specifically, multidimensional risk indicators are collected and calculated from financial market data. This process begins with the collection of historical trading data for major stock indices such as the DJIA, Nasdaq 100, and NIKKEI 225, generating raw financial market data spanning at least 10 years. This data is obtained from professional financial databases and includes basic trading information such as daily opening and closing prices, highs and lows, and trading volume. During data collection, linear interpolation of data from adjacent trading days is used to fill in missing data to ensure data continuity and integrity. After collecting complete raw financial market data, it is segmented and processed according to daily, weekly, and monthly time scales to produce a multi-time financial dataset. Daily data is directly derived from raw trading day data. Weekly data is calculated by aggregating data from five trading days each week, including the weekly opening price (Monday's opening price), weekly closing price (Friday's closing price), weekly high price (the highest price of the week), weekly low price (the lowest price of the week), and weekly trading volume (the total trading volume of the week). Monthly data is derived by similarly aggregating data from approximately 20 trading days each month. This multi-timescale segmentation allows subsequent risk assessments to simultaneously capture short-term, medium-term, and long-term market dynamics.

[0037] Market volatility indicators are extracted from a multi-time-dimensional financial dataset to generate a market volatility characteristic submatrix. Market volatility indicators include historical volatility, Bollinger Band width, average true range, and the CBOE Volatility Index. Historical volatility is calculated as the standard deviation of returns over the past N trading days; Bollinger Band width is the distance between the upper and lower bands divided by the middle band; and the average true range (ATR) is calculated as the N-day moving average of the maximum of the difference between the highest and lowest prices of the day, the difference between the highest price of the day and the previous day's closing price, and the difference between the lowest price of the day and the previous day's closing price. These volatility indicators are calculated separately for different time windows (e.g., 5-day, 10-day, and 20-day windows) to form a market volatility characteristic submatrix encompassing multiple time dimensions. Liquidity risk indicators are also extracted from the multi-time-dimensional financial dataset to generate a liquidity risk characteristic submatrix. Liquidity risk indicators include the trading volume change rate, bid-ask spread ratio, market depth index, and the Amihad illiquidity index. The trading volume change rate is calculated as the ratio of current trading volume to the average trading volume over the past N days; the bid-ask spread ratio is the difference between the ask and bid prices divided by the mid-price; the market depth indicator reflects the number and price distribution of buy and sell orders in the market; and the Amihad illiquidity index is calculated as the absolute value of the daily rate of return divided by the daily trading amount. These liquidity indicators are also calculated over different time windows to form a liquidity risk characteristic submatrix.

[0038] It is also necessary to extract systemic risk indicators from multi-time-dimensional financial datasets to obtain a systemic risk characteristic submatrix. Systemic risk indicators include market beta, systemic importance index, financial stress index, and conditional value at risk (CVaR). Market beta measures the volatility of individual stocks relative to the overall market; the systemic importance index assesses the impact of a specific financial institution or market sector on the overall financial system; the financial stress index comprehensively reflects market stress; and the conditional value at risk (CoVaR) calculates the value at risk of a specific asset under market stress. These systemic risk indicators form a systemic risk characteristic submatrix. The market volatility characteristic submatrix, the liquidity risk characteristic submatrix, and the systemic risk characteristic submatrix are merged to obtain a comprehensive risk characteristic matrix. During this merging process, the submatrices are first aligned row by row (ensuring that each row represents data from the same point in time) and then concatenated column by column, forming a comprehensive matrix with the same number of rows but an increased number of columns. Each row in this matrix represents the complete set of risk characteristics at a specific point in time, and each column represents the value of a specific risk indicator at all time points.

[0039] The comprehensive risk feature matrix is standardized to obtain the final financial risk feature matrix. This standardization uses the Z-score method, subtracting the mean from each feature value and dividing it by the standard deviation, resulting in a distribution with a mean of 0 and a standard deviation of 1. This standardization eliminates dimensional differences between different indicators, making subsequent neural network model training more stable and efficient.

[0040] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0041] Select 15 different risk indicators from the financial risk characteristic matrix to obtain a risk indicator subset;

[0042] Set different parameter combinations for risk indicator subsets to obtain a multi-dimensional indicator parameter matrix;

[0043] Based on the indicator parameter matrix, we extract the 15-day continuous time series data of each indicator to obtain the risk data expanded in the time dimension.

[0044] Reconstruct the risk data expanded in the time dimension into a 15×15 two-dimensional matrix structure to obtain the initial risk feature map;

[0045] Perform numerical normalization on the initial risk feature map to obtain a feature image with uniform pixel value distribution;

[0046] Based on the actual risk results of historical transactions, the feature images are labeled as high risk, medium risk, or low risk to obtain a labeled risk feature map;

[0047] The labeled risk feature map is divided into training set, validation set and test set in a ratio of 7:2:1 to obtain the risk feature map dataset.

[0048] Specifically, the time series data is converted into an image format suitable for processing by a two-dimensional convolutional neural network. First, 15 different risk indicators are selected from the financial risk feature matrix obtained in the previous step to form a risk indicator subset. These 15 indicators are selected based on their importance to risk prediction, determined using a feature importance scoring method. They include market volatility indicators (such as historical volatility and average price-to-earnings ratio), liquidity indicators (such as trading volume change rate and bid-ask spread), systemic risk indicators (such as beta value and financial stress index), and macroeconomic indicators. This selection ensures that the selected indicators can comprehensively reflect the multidimensional risk state of the market while maintaining the manageability of the data dimensions. Different parameter combinations are set for the selected risk indicator subset to obtain a multidimensional indicator parameter matrix. The parameter setting mainly focuses on the time window size. For each risk indicator, values are calculated for three time windows: short-term (5 days), medium-term (10 days), and long-term (20 days). For example, for the historical volatility indicator, the historical volatility for 5 days, 10 days, and 20 days is calculated simultaneously; for the trading volume change rate, the trading volume change rate for 5 days, 10 days, and 20 days is calculated. In this way, the 15 basic indicators are expanded into 45 indicator parameter combinations through different parameter combinations, forming a richer indicator parameter matrix.

[0049] Based on the indicator parameter matrix, 15 consecutive days of time series data for each indicator are extracted to generate time-expanded risk data. Specifically, for each trading day, 15 consecutive days of data, including the current day, are extracted retroactively and arranged in chronological order. This process effectively creates a sliding window of 15 days, sliding along the time axis one day at a time to extract all indicator values within the current window. The time-expanded risk data preserves the time series characteristics of the indicator and captures the patterns and trends of indicator values over time. The time-expanded risk data is reconstructed into a 15×15 two-dimensional matrix to generate the initial risk profile. This step is a key conversion step, converting the one-dimensional time series into a two-dimensional image format. The specific method is to use the 15 different risk indicators as the rows of the matrix and the 15-day time series data for each indicator as the columns, forming a 15×15 two-dimensional matrix. This arrangement ensures that each row of the matrix represents a risk indicator, and each column represents a specific point in time. The entire matrix represents a complete snapshot of the risk status.

[0050] The initial risk signature map is numerically normalized to produce a feature image with uniform pixel value distribution. Normalization maps all values in the matrix to the interval [0, 1], facilitating subsequent convolutional neural network processing. Normalization utilizes a minimum-maximum scaling method: for each element in the matrix, the minimum value of the metric is first subtracted and then divided by the difference between the maximum and minimum values. The normalized matrix values resemble the pixel values of an image and are more evenly distributed, facilitating effective feature extraction by the neural network.

[0051] Based on the actual risk results of historical transactions, feature images are labeled as high, medium, or low risk, resulting in a labeled risk profile. This labeling process is based on actual post-trade risk performance. Specifically, price volatility and maximum drawdown are calculated 1, 5, 10, and 20 days after the trade, and these two metrics are combined with a weighting of 4:6 to create a comprehensive risk index. Based on the distribution of the risk index, samples in the top 25% quantile are labeled "high risk," samples in the bottom 25% quantile are labeled "low risk," and the remaining samples are labeled "medium risk." This labeling method based on historical data provides the label information necessary for supervised learning in the neural network. The labeled risk profile is then split into training, validation, and test sets in a 7:2:1 ratio to generate the final risk profile dataset. This splitting utilizes stratified random sampling to ensure that the proportions of high, medium, and low risk samples in each subset remain consistent with those in the original dataset. The training set is used for model parameter learning, the validation set is used for model selection and hyperparameter tuning, and the test set is used to evaluate the model's generalization ability.

[0052] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0053] The risk feature map dataset was fed into a two-dimensional convolutional neural network (CNN-TA), resulting in a feature extraction structure consisting of three convolutional layers. The first convolutional layer used 32 5×5 convolution kernels, the second convolutional layer used 64 3×3 convolution kernels, and the third convolutional layer used 128 3×3 convolution kernels.

[0054] Perform the maximum pooling operation on the output of the three convolutional layers to obtain the feature map after dimensionality reduction;

[0055] The feature map after dimensionality reduction is transformed through the fully connected layer to obtain the risk feature vector of CNN-TA;

[0056] The financial risk feature matrix is input into a multilayer perceptron, which contains three fully connected layers with 256, 128, and 64 neurons respectively, to obtain the MLP risk feature vector.

[0057] The risk feature vector based on CNN-TA is fused with the risk feature vector of MLP to obtain the comprehensive risk feature representation;

[0058] The comprehensive risk feature representation is mapped to risk levels through the output layer. The output layer contains 9 neurons corresponding to nine levels of risk labels to obtain the risk probability distribution;

[0059] Batch training and parameter optimization are performed on the risk probability distribution. The training uses the cross entropy loss function and Adam optimizer, and the learning rate is set to 0.001 to obtain the financial transaction risk assessment network.

[0060] Specifically, the risk profile dataset was fed into a two-dimensional convolutional neural network (CNN-TA), resulting in a feature extraction architecture consisting of three convolutional layers. CNN-TA, a convolutional neural network designed specifically for time series analysis, effectively captures the spatiotemporal characteristics of financial data. The first convolutional layer uses 32 5×5 convolutional kernels to convolve a 15×15 input image. The convolution stride is set to 1, and the edge padding is set to 'same' to maintain the feature map size. The convolution operation extracts local features by sliding the convolution kernel over the input image and calculating the dot product between the kernel and the corresponding image region. The convolution is followed by a ReLU activation function, which sets negative values to zero and retains positive values, introducing nonlinearity. The second convolutional layer uses 64 3×3 convolutional kernels to process the output feature map of the first layer and extract higher-level feature representations. Convolution with a stride of 1 and the ReLU activation function are also used. The third convolutional layer uses 128 3×3 convolutional kernels to further extract more abstract features, using the same convolution method and activation function. The superposition of three layers of convolution enables the network to learn risk features hierarchically, from simple edge features to complex pattern recognition.

[0061] Max pooling is performed on the outputs of the three convolutional layers to produce a reduced-dimensional feature map. Max pooling is a common method for reducing feature map dimensionality. The maximum value within a 2×2 window is selected as the representative value for that region, with a stride of 2, halving the height and width of the feature map. Max pooling not only reduces data dimensionality and computational complexity but also provides a degree of translation invariance, making feature detection less sensitive to positional changes. After pooling, the dimensionality of the feature map is significantly reduced while retaining the most significant features. The reduced-dimensional feature map is then passed through a fully connected layer for feature transformation to produce the CNN-TA risk feature vector. The fully connected layer first flattens the feature map into a one-dimensional vector and then processes it through a fully connected neural network. The fully connected layer contains 256 neurons, uses the ReLU activation function, and is followed by a dropout layer (with a dropout rate of 0.3) to prevent overfitting. The output of the fully connected layer is a 128-dimensional feature vector, representing the risk features extracted by CNN-TA from the image data.

[0062] The financial risk feature matrix is input into a multilayer perceptron (MLP), which consists of three fully connected layers with 256, 128, and 64 neurons, respectively. The financial risk feature matrix contains raw financial indicator data and retains more direct financial information than the risk feature map. A multilayer perceptron (MLP) is a feedforward neural network that transforms input features through a multi-layer fully connected network. The first layer receives the raw feature matrix and contains 256 neurons; the second layer contains 128 neurons; and the third layer contains 64 neurons. Each layer uses the ReLU activation function to introduce nonlinearity, and a BatchNormalization layer is added after each layer for data normalization. A Dropout layer (with a dropout rate of 0.3) is also added after each layer to prevent overfitting. The output of the MLP is a 64-dimensional MLP risk feature vector, representing the risk features extracted from the raw financial data. The risk feature vector based on the CNN-TA is fused with the MLP risk feature vector to obtain a comprehensive risk feature representation. Feature fusion is achieved through concatenation, concatenating the 128-dimensional CNN-TA feature vector and the 64-dimensional MLP feature vector into a 192-dimensional joint vector. This concatenated vector then passes through a fully connected layer with 128 neurons, using the ReLU activation function, to integrate the information from both features and produce the final comprehensive risk profile. This feature fusion mechanism combines the image feature extraction capabilities of convolutional neural networks with the raw financial data processing capabilities of multilayer perceptrons to form a more comprehensive risk representation.

[0063] The comprehensive risk profile is mapped to risk levels through the output layer, which contains nine neurons corresponding to nine risk labels, resulting in a risk probability distribution. The output layer is a fully connected layer that receives the comprehensive risk profile as input and outputs nine values corresponding to the nine risk labels: extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk. The output layer uses the Softmax activation function to convert the output values into a probability distribution, where each value represents the probability of a sample belonging to the corresponding risk level, and the sum of all probabilities is 1. Batch training and parameter optimization are performed on the risk probability distribution using the cross-entropy loss function and the Adam optimizer with a learning rate of 0.001, resulting in a financial transaction risk assessment network. The cross-entropy loss function is a commonly used loss function for classification problems, measuring the difference between the predicted distribution and the true distribution. Batch training uses mini-batch gradient descent, with each batch containing 64 samples. The gradients for one batch are calculated and the parameters are updated in one iteration. The Adam optimizer combines the advantages of momentum and RMSProp, adaptively adjusting the learning rate for different parameters to accelerate convergence. Model training is performed for 100 epochs, with each epoch traversing the entire training set once. During training, performance is monitored on the validation set, and an early stopping strategy is implemented to prevent overfitting. Training is terminated when the validation loss does not improve for 10 consecutive epochs.

[0064] In a specific embodiment, the process of performing the risk level mapping step on the comprehensive risk feature representation through the output layer may specifically include the following steps:

[0065] The comprehensive risk feature representation is represented by retaining information through the residual connection layer to obtain an enhanced risk feature;

[0066] The enhanced risk features are mapped and transformed through a fully connected layer with 128 neurons to obtain an intermediate feature representation.

[0067] Apply the Leaky ReLU activation function to the intermediate feature representation for nonlinear transformation to obtain the activated feature representation;

[0068] The activated feature representation is randomly deactivated through the Dropout layer, and the Dropout rate is set to 0.3 to obtain an anti-overfitting feature representation;

[0069] The anti-overfitting feature representation is transformed through a fully connected layer of nine output neurons to obtain the original risk score;

[0070] The original risk score is normalized using the Softmax function to obtain the probability value of each risk level. The nine risk labels are: extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk. The risk labels correspond to the combined risk index of price volatility and maximum drawdown rate in different time periods after the transaction.

[0071] The probability value of the risk level is threshold-calibrated, and the risk probability distribution is obtained through the equal bin method and Brier score evaluation.

[0072] Specifically, the comprehensive risk signature representation is passed through a residual connection layer to preserve information, resulting in an enhanced risk signature. Residual connection layers are an effective structure for addressing the vanishing gradient problem in deep neural networks. Their core concept is to create a "shortcut" from input to output, allowing the original feature information to be directly passed to subsequent layers. Specifically, the comprehensive risk signature representation (denoted as X) is directly added to the features processed by a fully connected layer (denoted as F(X)), i.e., X + F(X), to form the final enhanced risk signature. This structure ensures that the original financial risk signature is not lost during the transmission process through the deep network, making it particularly suitable for scenarios in financial risk assessment that require simultaneous consideration of both original indicators and deep features. The enhanced risk signature is then transformed through a fully connected layer with 128 neurons to obtain an intermediate feature representation. A fully connected layer is a fundamental component of a neural network, connecting each input neuron to each output neuron, forming a fully connected structure. Each connection has a weight parameter corresponding to the importance of the input feature. The specific calculation is to multiply the enhanced risk feature vector by the weight matrix and add the bias vector to obtain a 128-dimensional intermediate feature representation. This fully connected layer realizes the mapping from enhanced risk features to a more compact feature representation, reducing feature dimensionality and focusing on the most critical risk information.

[0073] The Leaky ReLU activation function is applied to the intermediate feature representation for a nonlinear transformation, yielding an activated feature representation. Leaky ReLU is an improvement on the traditional ReLU function. For positive inputs, it returns the input value itself, and for negative inputs, it returns the input value multiplied by a small coefficient (typically 0.01). This design avoids the problem of the ReLU function completely "dying" in the negative input region, while preserving the transmission of small amounts of negative information. The Leaky ReLU function introduces nonlinearity to the intermediate feature representation, enhancing the model's ability to represent complex financial risk patterns while mitigating the vanishing gradient problem, making deep networks easier to train. The activated feature representation is then randomly deactivated using a Dropout layer with a dropout rate of 0.3 to yield a feature representation that resists overfitting. Dropout is a regularization technique that randomly "turns off" a subset of neurons during training, preventing them from participating in forward and backward computations. Specifically, it generates a binary mask vector with the same shape as the feature vector, with each element having a 30% probability of being 0 (deactivated) and a 70% probability of being 1 (retained). The activated feature representation is multiplied by this mask vector, resulting in some neurons being reset to zero, i.e., the anti-overfitting feature representation. Dropout technology forces the network to learn more robust feature representations, preventing over-reliance on specific neurons. During training, this is equivalent to integrating multiple sub-networks with different structures, improving the model's generalization capabilities in the face of complex changes in the financial market. The anti-overfitting feature representation is transformed through a fully connected layer of nine output neurons to obtain the original risk score. This fully connected layer maps the reduced anti-overfitting feature representation into nine numerical values, corresponding to nine risk labels. Each output neuron calculates the weighted sum of the anti-overfitting feature representation and adds a bias to obtain the original score for that risk level. These original scores directly reflect the model's assessment of the samples at different risk levels but have not yet been converted into probabilistic form.

[0074] The raw risk scores are normalized using the Softmax function to obtain probability values for each risk level. The Softmax function is the standard output layer for multi-classification problems, converting any real value into a probability value between 0 and 1, where the sum of all probabilities is 1. The calculation involves subtracting the maximum score from each raw score (to prevent overflow), then exponentially dividing it by the sum of all exponentially processed scores. After this processing, the nine output neurons correspond to nine risk labels: extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk. These risk labels correspond to risk indices based on price volatility and maximum drawdown rates over different time periods after trading, reflecting the multidimensional nature of risk assessment. The probability values for each risk level are thresholded and evaluated using the equal binning method and the Brier score to obtain the final risk probability distribution. Probability calibration is a crucial step in ensuring that the probability values output by the model truly reflect the actual risk level. The equal binning method sorts samples by predicted probability and divides them into bins of equal size. The difference between the actual positive rate of samples in each bin and the average predicted probability is calculated. The Brier score is a metric used to evaluate the accuracy of probability predictions. It calculates the mean squared error between the predicted probability and the actual outcome (0 or 1). Based on these evaluation results, the original Softmax output is fine-tuned so that the predicted probability more accurately reflects the actual risk level, resulting in the final risk probability distribution.

[0075] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0076] Extract multi-dimensional risk indicators from real-time transaction data to obtain a real-time risk feature matrix consistent with the training data format;

[0077] The real-time risk feature matrix is converted into a 15×15 two-dimensional risk feature image to obtain a real-time risk feature map;

[0078] Input the real-time risk feature map into the financial transaction risk assessment network for forward calculation to obtain the risk probability distribution of real-time transactions;

[0079] Threshold judgment and level classification are performed on the risk probability distribution to obtain nine-level risk rating results ranging from extremely high risk to extremely low risk;

[0080] Based on the nine-level risk rating results, the overall risk level, risk trend analysis, and key risk indicator analysis are generated to obtain a three-layer structured risk rating report;

[0081] Compare the nine-level risk rating results with the preset risk threshold to obtain targeted transaction warning signals;

[0082] Transaction warning signals are periodically updated and cumulatively analyzed to obtain risk warning heat maps and time series change trends.

[0083] Specifically, after completing model training, the multi-layer neural network-based financial transaction risk assessment method enters the practical application phase, performing risk assessment and early warning on real-time transaction data. First, multi-dimensional risk indicators are extracted from real-time transaction data to obtain a real-time risk feature matrix consistent with the training data format. This process uses the same indicator extraction method as the training phase to calculate various risk indicators based on the market data related to the current transaction, including market volatility indicators (historical volatility, Bollinger Band width, etc.), liquidity risk indicators (trading volume change rate, bid-ask spread ratio, etc.), and systemic risk indicators (market beta value, financial stress index, etc.). Indicator calculation uses a sliding window method, performing real-time analysis of the latest transaction data to ensure the timeliness of the features. The calculated indicators are standardized to ensure the same scale as the training data, forming a real-time risk feature matrix.

[0084] The real-time risk feature matrix is converted into a 15×15 two-dimensional risk feature image to obtain a real-time risk feature map. The conversion process is consistent with the training phase. First, 15 key risk indicators are selected from the real-time risk feature matrix. Then, the continuous data for each indicator over the last 15 days is extracted and reconstructed into a 15×15 two-dimensional matrix. This conversion allows the risk features of real-time trading data to be represented as a two-dimensional image, facilitating subsequent convolutional neural network processing. The converted risk feature map retains the time series characteristics of the risk indicators while incorporating spatial relationships between different indicators, providing a richer feature representation for the model.

[0085] The real-time risk profile is fed into the financial transaction risk assessment network for forward computation, yielding the risk probability distribution of the real-time transaction. Forward computation is the process by which a neural network performs inference, processing data from the input layer through each layer of the network to ultimately produce the output. This process can be expressed as follows:

[0086] P(R|I)=Softmax(f θ (I))

[0087] Where P(R|I) represents the probability distribution of risk level under given input feature I, f θ(I) represents a financial transaction risk assessment network function with parameter θ, and Softmax represents the function that converts the raw output into a probability distribution. The specific calculation process involves first extracting image features from the real-time risk feature map I through the three-layer convolutional structure of the CNN-TA to obtain the CNN-TA risk feature vector. Simultaneously, the original financial risk feature matrix is input into a multi-layer perceptron to obtain the MLP risk feature vector. The two feature vectors are then fused to form a comprehensive risk feature representation. Finally, through a series of processes including residual connections, fully connected layers, leaky ReLU activation, dropout, and an output layer, the probability values of the nine risk levels are calculated to form a risk probability distribution.

[0088] Threshold determination and categorization are performed on the risk probability distribution, resulting in a nine-level risk rating ranging from extremely high risk to extremely low risk. Threshold determination is the process of determining the final risk level based on the probability distribution. The most straightforward approach is to select the level with the highest probability as the risk rating result, but in risk management, more complex judgment strategies are often employed. One common approach is to set a probability threshold. When the probability of a certain level exceeds the preset threshold, the transaction is classified as that risk level. Another approach considers the combination of multiple high-probability levels and determines the final level based on specific rules. The nine risk labels include extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk. This fine-grained risk classification enables more accurate risk assessment.

[0089] Based on the nine-level risk rating results, a three-tiered structured risk rating report is generated, including an overall risk rating, risk trend analysis, and key risk indicator analysis. This three-tiered structure is designed to provide risk information from a macro to micro perspective. The first tier, the overall risk rating, directly displays the nine-level risk rating results, providing a risk overview. The second tier, the risk trend analysis, draws on historical risk rating data to calculate the changing trends and volatility of the risk rating and predict future risk trends. The third tier, the key risk indicator analysis, analyzes the model's internal parameters and feature importance to identify the indicators that contribute most to the current risk rating and explain the sources of risk. This structured report not only provides risk rating information but also provides in-depth analysis of risk causes and trends, providing comprehensive support for trading decisions.

[0090] The nine-level risk rating results are compared with pre-set risk thresholds to generate targeted trading warning signals. Warning signals are a key output of risk management and directly guide trading behavior. Pre-set risk thresholds are determined based on the institution's risk appetite and market environment, with different risk levels corresponding to different warning levels and response strategies. When the risk rating exceeds a specific threshold, a corresponding warning signal, such as "High Risk Warning" or "Medium Risk Caution," is triggered. Warning signals also include targeted risk response recommendations, such as "Increase Hedging Recommendation" or "Reducing Position Recommendation," to directly guide trading strategy adjustments.

[0091] Transaction warning signals are periodically updated and cumulatively analyzed to generate a risk warning heat map and time series trend analysis. Periodic updates of warning signals are a core component of continuous risk monitoring and are typically performed on daily, weekly, and monthly timescales. Historical warning signals are cumulatively analyzed to calculate the trigger frequency of warnings at different risk levels, generating a risk warning heat map that visually displays the distribution of risk. Furthermore, through time series analysis, trend, seasonal, and residual components of warning signals are extracted to predict risk trends and generate risk evolution forecasts. These analytical results provide a basis for long-term risk management strategies and help institutions adjust their overall risk exposure.

[0092] For example, a commodity futures trading company conducted a real-time risk assessment of its copper futures trading. First, multi-dimensional risk indicators for copper futures were extracted from market data. Indicators such as the 20-day historical volatility (28.5%), the 5-day trading volume change rate (1.45), and the bid-ask spread ratio (0.12%) were calculated to form a real-time risk signature matrix. The 15 key indicators in this matrix and their 15-day historical data were reconstructed into a 15×15 risk signature map, which was then fed into a trained financial transaction risk assessment network. The network extracted image features using convolutional layers and processed the raw indicator data using a multi-layer perceptron to generate a comprehensive risk signature. Through residual connections, full connections, activations, and dropout, a nine-level risk probability distribution was calculated: extremely high risk (0.01), high risk (0.03), medium-high risk (0.45), medium-high risk (0.30), and medium risk (0.15). The current transaction risk level was determined to be "medium-high," and a three-layer structured report was generated, including the risk factors (increased market volatility and declining liquidity) and trend forecasts. The system triggered a "medium-to-high risk warning" signal, recommending a 20% reduction in positions and increased hedging strategies. Continuous monitoring showed an increasing risk level in copper futures trading over the past week. The risk warning heat map showed that risks were concentrated in volatility and liquidity, providing effective guidance for the trading team to adjust their strategies in a timely manner.

[0093] In a specific embodiment, the process of performing the steps of periodically updating and cumulatively analyzing the transaction warning signal may specifically include the following steps:

[0094] The transaction warning signals are recorded and stored on a daily, weekly and monthly basis to obtain a multi-period warning signal database;

[0095] Conduct frequency statistical analysis on the warning signals in the multi-period warning signal database to obtain the warning trigger frequency matrix of different risk levels;

[0096] Convert the warning trigger frequency matrix into a color-coded two-dimensional heat map to obtain the risk warning heat map;

[0097] Perform time series decomposition on the multi-period warning signal database to obtain the trend component, seasonal component and residual component of the warning signal;

[0098] Construct a risk evolution curve based on the trend component to obtain a long-term risk change trend chart;

[0099] Conduct correlation analysis between the risk change trend graph and the market benchmark index to obtain quantitative indicators of the correlation between risk and market;

[0100] By matching and comparing the quantitative indicators of risk and market correlation with historical risk events, we can obtain risk evolution prediction results based on historical patterns.

[0101] Specifically, trading warning signals are recorded and stored at daily, weekly, and monthly timescales to create a multi-period warning signal database. Daily warning signals directly record the risk rating and warning level for each trading day. Weekly warning signals are generated by aggregating data from five trading days per week, recording the highest risk level, average risk level, and risk trend within the week. Monthly warning signals aggregate data from approximately 20 trading days to provide an overview of the monthly risk landscape. Warning signals from these three timescales are stored in a relational database, forming a structured multi-period warning signal database. The table structure includes fields such as transaction identifier, assessment date, risk level, warning level, and key risk indicators, and corresponding indexes are established to support efficient query and analysis. Frequency statistical analysis is performed on the warning signals in the multi-period warning signal database to generate a matrix of warning trigger frequencies for different risk levels. Frequency statistical analysis is a fundamental analytical method for risk distribution characteristics, performing statistical calculations across different risk dimensions. The analysis process first determines the statistical dimensions: time (daily, weekly, and monthly), risk level (with nine risk labels), and risk type (market risk, liquidity risk, credit risk, etc.). The number of alert triggers for each dimension combination is then calculated, such as the number of triggers for the "monthly-high-risk-market volatility" combination. The statistical results are organized into a multidimensional frequency matrix, with each element representing the alert trigger frequency for a specific dimension combination. Frequency is calculated by dividing the number of triggers for that dimension combination by the total number of observations, yielding the trigger frequency as a percentage. This frequency matrix comprehensively demonstrates the multidimensional characteristics of risk distribution and provides a data foundation for subsequent analysis.

[0102] The warning trigger frequency matrix is converted into a color-coded two-dimensional heat map, resulting in a risk warning heat map. Heat maps are effective visualization tools for intuitively displaying multidimensional data, mapping numerical values to color depths, making data patterns readily apparent. The conversion process first selects two dimensions for the heat map, typically time as the horizontal axis and risk level as the vertical axis, generating a two-dimensional view of time versus risk level. For frequency matrices of three or greater dimensions, it is necessary to fix the values of the other dimensions or aggregate the data from these dimensions to reduce the dimensionality to two dimensions. A color mapping scheme is then designed. Typically, low frequencies are mapped to cool colors (blue) and high frequencies to warm colors (red), with color transitions reflecting frequency changes. Elements such as the heat map's boundaries, axis labels, and color legend also require careful design to ensure readability. The resulting risk warning heat map intuitively displays the spatiotemporal patterns of risk distribution, facilitating the identification of risk concentration areas and evolving trends. Time series decomposition is performed on the multi-period warning signal database to obtain the trend, seasonal, and residual components of the warning signals. Time series decomposition is a classic method for analyzing the inherent structure of time series data. It decomposes the original sequence into several components with clear interpretations. The specific implementation uses the STL (Seasonal-Trend decomposition using Loess) algorithm, which is suitable for time series data with seasonality. Before decomposition, the risk level is encoded as a numerical value (e.g., very high risk = 9, very low risk = 1), forming a quantifiable risk index sequence. The STL decomposition process consists of an inner loop and an outer loop. The inner loop performs iterative extraction of trend and seasonal components, while the outer loop processes outliers to enhance robustness. The decomposition results include a trend component (reflecting the direction of long-term change), a seasonal component (reflecting cyclical fluctuation patterns), and a residual component (reflecting random fluctuations). This decomposition allows different aspects of risk variation to be separated and analyzed separately, revealing the inherent laws of risk evolution.

[0103] A risk evolution curve is constructed based on the trend component, yielding a long-term risk trend chart. The trend component reflects the direction of long-term change after removing seasonal and random fluctuations and is an important indicator for predicting future risk trends. The process of constructing the risk evolution curve begins by smoothing the trend component data to eliminate residual short-term fluctuations and enhance the long-term trend. This smoothing method uses a median filter with a window size of 7 to perform a secondary smoothing of the trend data. The smoothed trend data is then plotted as a continuous curve, with time plotted on the horizontal axis and the risk index value on the vertical axis. Key turning points, such as local maxima, minima, and inflection points of the risk index, are marked on the curve. These points often correspond to important market events or critical points of risk accumulation. The risk evolution curve clearly illustrates the long-term accumulation and release of risk, providing a macro perspective for risk management. Correlation analysis is performed between the risk trend chart and market benchmark indices to quantify the risk-market correlation. Correlation analysis aims to reveal the relationship between risk changes and market trends and quantify the forward-looking and lagging nature of risk indicators. The analysis begins by selecting an appropriate market benchmark index, such as a major stock market index, volatility index, or industry index. The correlation coefficient between the risk trend and the market index is then calculated at different time lags, typically ranging from -30 to +30 trading days. A negative lag indicates that risk changes are ahead of the market, while a positive lag indicates that risk changes lag behind the market. The correlation coefficient is calculated using the Pearson correlation coefficient method, which measures the degree of linear correlation between two variables. The analysis results form a correlation coefficient sequence, recording the correlation strength at different lags. The maximum correlation coefficient and its corresponding time lag are then identified to quantify the correlation between risk and the market. This metric reflects the predictive power or responsiveness of the risk assessment results to market fluctuations.

[0104] Quantitative indicators of risk-market correlation are matched against historical risk events to generate risk evolution forecasts based on historical patterns. This step compares the current risk status within a historical context, identifying similar patterns and predicting future developments. The matching process begins by establishing a historical risk event database, recording the risk evolution trajectories of major market adjustments, financial crises, or industry shocks. Then, using the Dynamic Time Warping (DTW) algorithm, the current risk trend curve is elastically matched against historical patterns to identify the historical cases with the most similar morphology. The similarity calculation considers multiple factors, including curve shape, risk level, and market correlation, and a comprehensive score is used to determine the most similar historical case. Based on the matching results, the subsequent development trajectories of historical cases are referenced, and the unique factors of the current market environment are incorporated to generate risk evolution forecasts, including risk peak predictions, duration estimates, and impact assessments. This historical pattern-based forecast provides an important reference for risk management decisions.

[0105] For example, an investment company's fixed-income portfolio uses a multi-layer neural network risk assessment method to monitor the risk of its bond investments. Risk ratings are generated daily for the portfolio, with nine risk labels and warning signals recorded to form a daily warning record. Weekly data is aggregated every Friday, with the highest and average risk levels calculated to form a weekly warning record. Monthly data is aggregated at the end of each month to form a monthly risk overview. These multi-period warning signals are stored in a structured database, documenting the risk assessment history for the past two years. Frequency statistical analysis is performed on the database to calculate the triggering frequency of the nine risk labels in various time periods, forming a risk frequency matrix. This matrix is converted into a heat map, with months on the horizontal axis and risk levels on the vertical axis, with colors ranging from dark blue to dark red representing low to high frequency. The heat map shows a recent significant increase in the frequency of high-risk and medium-high-risk events, indicating potential risk accumulation. Time series decomposition of the two-year risk data is performed to extract a long-term upward trend component, a quarterly seasonal component, and short-term random fluctuations. A risk evolution curve is constructed based on the trend component, showing that the risk level has steadily increased from a low level, with recent acceleration. Comparing the risk curve with the credit spread index reveals that risk ratings lead market spreads by approximately 12 trading days, demonstrating the model's robust risk early warning capabilities. A search of the historical database reveals that the current risk evolution pattern closely resembles that preceding a credit risk event three years ago. Based on this historical pattern and current trends, we predict that credit risk will continue to rise over the next month. We recommend preemptive portfolio adjustments, increasing the proportion of high-credit-rated bonds and appropriately increasing liquidity reserves to prepare for potential market adjustments.

[0106] The above describes the financial transaction risk assessment method based on a multi-layer neural network in the embodiment of the present application. The following describes the financial transaction risk assessment system based on a multi-layer neural network in the embodiment of the present application. Figure 2 In one embodiment of the present application, a financial transaction risk assessment system based on a multi-layer neural network includes:

[0107] The acquisition module is used to collect and calculate multi-dimensional risk indicators of financial market data to obtain a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors;

[0108] A conversion module, configured to perform time series to image processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map dataset suitable for two-dimensional convolution analysis;

[0109] A training module, configured to train a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network;

[0110] The classification module is used to input real-time transaction data into the financial transaction risk assessment network to calculate and classify risk levels, and obtain a hierarchical structured risk rating report and transaction early warning signals.

[0111] Through the collaborative efforts of the aforementioned components, the system achieves precise assessment and early warning of financial transaction risks through a combination of technical features: multidimensional risk indicator collection and calculation, time series to image processing, fusion training of a two-dimensional convolutional neural network and a multi-layer perceptron, and real-time risk level calculation and classification. This system demonstrates significant technical effectiveness. First, the multidimensional risk indicator collection and calculation process transforms complex market information into a structured financial risk feature matrix, comprehensively capturing multi-faceted risk signals such as market volatility, liquidity ratios, systemic risk values, and macroeconomic factors. This overcomes the limitations of traditional single-metric assessments and significantly improves the comprehensiveness and accuracy of risk identification. Second, the time series to image processing technology converts one-dimensional financial data into a two-dimensional risk feature map, creatively transforming the variation patterns of financial time series into spatial distribution features. This enables the subsequent two-dimensional convolutional neural network to effectively extract the spatiotemporal correlation patterns of financial risks. This algorithmic feature fully leverages the advantages of convolutional neural networks in image recognition, successfully migrating them to the field of financial risk assessment and achieving a deep integration of algorithms and application scenarios. Third, the combined training mechanism of a two-dimensional convolutional neural network (CNN-TA) and a multi-layer perceptron (MLP) enables multimodal feature extraction and fusion. CNN-TA focuses on capturing spatial correlation patterns within risk feature maps, while MLP focuses on learning the complex nonlinear relationships of raw risk indicators. The two complement each other, forming a more comprehensive risk representation capability. This combined design of algorithmic features significantly enhances the model's ability to identify complex financial risk patterns. Fourth, real-time risk level calculation and classification technology transforms risk assessment into practical decision support. Through a nine-level refined risk classification and a three-layer structured risk rating report, it not only provides risk level judgments but also deeply analyzes the sources and evolution trends of risks, significantly enhancing the operability and decision-making value of risk assessment results. It is particularly worth noting that the introduction of the two-dimensional convolutional neural network (CNN-TA) algorithmic features in this solution fully considers the spatiotemporal correlation characteristics of financial risk data. By converting time series features into spatial feature maps, the local receptive field and hierarchical feature extraction capabilities of the convolutional neural network are fully utilized, achieving efficient extraction of financial risk patterns. This algorithmic feature is deeply aligned with the application domain of financial risk assessment and is the key to the solution's excellent results. In summary, this method achieves a significant improvement in the comprehensiveness, accuracy, real-timeness and interpretability of financial transaction risk assessment through precise matching of algorithm features with application requirements.

[0112] above Figure 2The financial transaction risk assessment system based on a multi-layer neural network in an embodiment of the present invention is described in detail from the perspective of modular functional entities. The financial transaction risk assessment device based on a multi-layer neural network in an embodiment of the present invention is described in detail from the perspective of hardware processing.

[0113] Figure 3 Figure 3 is a schematic diagram of the structure of a multi-layer neural network-based financial transaction risk assessment device provided by an embodiment of the present invention. The multi-layer neural network-based financial transaction risk assessment device 300 may vary significantly depending on configuration or performance. The device may include one or more central processing units (CPUs) 310 (e.g., one or more processors), memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions and operations within the multi-layer neural network-based financial transaction risk assessment device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, executing the series of instructions and operations stored in the storage medium 330 on the multi-layer neural network-based financial transaction risk assessment device 300 to implement the steps of the multi-layer neural network-based financial transaction risk assessment method described above.

[0114] The financial transaction risk assessment device 300 based on a multi-layer neural network may further 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 Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The illustrated structure of the multi-layer neural network-based financial transaction risk assessment device does not limit the multi-layer neural network-based financial transaction risk assessment 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.

[0115] 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. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the financial transaction risk assessment method based on a multi-layer neural network.

[0116] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0117] 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, or the part that contributes to the prior art, or all 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 and includes several instructions for enabling a financial transaction risk assessment device based on a multi-layer neural network (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), magnetic disk or optical disk, etc., various media that can store program code.

[0118] 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 above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can 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 financial transaction risk assessment method based on a multi-layer neural network, characterized in that: The method comprises: Collect and calculate multi-dimensional risk indicators of financial market data to obtain a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors; Based on the financial risk feature matrix, the historical transaction data is converted from time series to images to obtain a risk feature map dataset suitable for two-dimensional convolution analysis; Training a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network; The real-time transaction data is input into the financial transaction risk assessment network to calculate and classify the risk level, and a hierarchical structured risk rating report and transaction early warning signals are obtained.

2. The financial transaction risk assessment method based on a multi-layer neural network according to claim 1, characterized in that: The multi-dimensional risk indicator collection and calculation of financial market data obtains a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors, including: Collect historical trading data of DJIA, Nasdaq 100, and NIKKEI 225 stock indices to obtain raw financial market data covering a span of at least 10 years; Segmenting the original financial market data into three time scales: daily, weekly, and monthly, to obtain a multi-time dimension financial data set; Extracting market volatility indicators from the multi-time dimension financial data set to obtain a market volatility characteristic submatrix; Extracting liquidity risk indicators from the multi-time dimension financial data set to obtain a liquidity risk characteristic submatrix; Extracting systemic risk indicators from the multi-time dimension financial data set to obtain a systemic risk characteristic submatrix; Merging the market volatility characteristic sub-matrix, the liquidity risk characteristic sub-matrix, and the systemic risk characteristic sub-matrix to obtain a comprehensive risk characteristic matrix; The comprehensive risk characteristic matrix is standardized to obtain a final financial risk characteristic matrix.

3. The financial transaction risk assessment method based on a multi-layer neural network according to claim 1, characterized in that: The method of performing time series to image conversion processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map dataset suitable for two-dimensional convolution analysis includes: Selecting 15 different risk indicators from the financial risk characteristic matrix to obtain a risk indicator subset; Setting different parameter combinations for the risk indicator subset to obtain a multi-dimensional indicator parameter matrix; Extracting 15 consecutive days of time series data for each indicator based on the indicator parameter matrix to obtain risk data expanded in the time dimension; Reconstruct the risk data expanded in the time dimension into a 15×15 two-dimensional matrix structure to obtain an initial risk feature map; Performing numerical normalization processing on the initial risk feature map to obtain a feature image with uniform pixel value distribution; Marking the feature image as high risk, medium risk, or low risk based on actual risk results of historical transactions to obtain a labeled risk feature map; The labeled risk signature map is divided into a training set, a validation set, and a test set in a ratio of 7:2:1 to obtain a risk signature map dataset.

4. The financial transaction risk assessment method based on a multi-layer neural network according to claim 1, characterized in that: The method of training a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network includes: The risk feature map dataset is input into a two-dimensional convolutional neural network CNN-TA to obtain a feature extraction structure consisting of three convolutional layers, where the first convolutional layer uses 32 5×5 convolution kernels, the second convolutional layer uses 64 3×3 convolution kernels, and the third convolutional layer uses 128 3×3 convolution kernels; Performing a maximum pooling operation on the outputs of the three convolutional layers to obtain a feature map after dimensionality reduction; The feature map after dimensionality reduction is subjected to feature conversion through a fully connected layer to obtain a risk feature vector of CNN-TA; Input the financial risk feature matrix into a multilayer perceptron, wherein the multilayer perceptron includes three fully connected layers with 256, 128, and 64 neurons, respectively, to obtain an MLP risk feature vector; Perform feature fusion based on the risk feature vector of the CNN-TA and the risk feature vector of the MLP to obtain a comprehensive risk feature representation; The comprehensive risk feature representation is mapped to a risk level through an output layer, where the output layer contains 9 neurons corresponding to nine levels of risk labels, to obtain a risk probability distribution; Batch training and parameter optimization are performed on the risk probability distribution. The training uses the cross entropy loss function and the Adam optimizer with a learning rate set to 0.001 to obtain a financial transaction risk assessment network.

5. The financial transaction risk assessment method based on a multi-layer neural network according to claim 4 is characterized in that: The comprehensive risk feature representation is mapped to a risk level through an output layer, wherein the output layer includes 9 neurons corresponding to nine levels of risk labels, and a risk probability distribution is obtained, including: Retaining information on the comprehensive risk feature representation through a residual connection layer to obtain an enhanced risk feature; The enhanced risk feature is mapped and transformed through a fully connected layer, where the number of neurons in the fully connected layer is 128, to obtain an intermediate feature representation; Applying a Leaky ReLU activation function to the intermediate feature representation to perform a nonlinear transformation to obtain an activated feature representation; Performing a random deactivation operation on the activated feature representation through a Dropout layer, wherein the Dropout rate is set to 0.3, to obtain an anti-overfitting feature representation; Transforming the anti-overfitting feature representation through a fully connected layer of nine output neurons to obtain an original risk score; The raw risk score is normalized using the Softmax function to obtain the probability value of each risk level. The nine risk labels are: extremely high risk, high risk, medium-high risk, medium-high risk, medium risk, medium-low risk, medium-low risk, low risk, and extremely low risk. The risk labels correspond to the combined risk index of price volatility and maximum drawdown rate in different time periods after the transaction. The probability value of the risk level is threshold-calibrated, and the risk probability distribution is obtained through the equal bin method and Brier score evaluation.

6. The financial transaction risk assessment method based on a multi-layer neural network according to claim 1, characterized in that: The real-time transaction data is input into the financial transaction risk assessment network to calculate and classify risk levels, thereby obtaining a hierarchical structured risk rating report and transaction warning signals, including: Extracting multi-dimensional risk indicators from the real-time transaction data to obtain a real-time risk feature matrix consistent with the training data format; Converting the real-time risk feature matrix into a 15×15 two-dimensional risk feature image to obtain a real-time risk feature map; Inputting the real-time risk feature map into the financial transaction risk assessment network for forward calculation to obtain a risk probability distribution of the real-time transaction; Perform threshold judgment and level classification on the risk probability distribution to obtain nine-level risk rating results ranging from extremely high risk to extremely low risk; Based on the nine-level risk rating results, an overall risk level, risk trend analysis, and key risk indicator analysis are generated to obtain a three-layer structured risk rating report; Compare the nine-level risk rating results with the preset risk thresholds to obtain targeted transaction warning signals; The transaction warning signals are periodically updated and cumulatively analyzed to obtain a risk warning heat map and a time series change trend.

7. The financial transaction risk assessment method based on a multi-layer neural network according to claim 6, characterized in that: The transaction warning signal is periodically updated and cumulatively analyzed to obtain a risk warning heat map and a time series change trend, including: Recording and storing the transaction warning signals on a daily, weekly, and monthly basis to obtain a multi-period warning signal database; Performing frequency statistical analysis on the warning signals in the multi-period warning signal database to obtain a warning trigger frequency matrix of different risk levels; Converting the warning trigger frequency matrix into a color-coded two-dimensional heat map to obtain a risk warning heat map; Performing time series decomposition on the multi-period warning signal database to obtain a trend component, a seasonal component, and a residual component of the warning signal; Constructing a risk evolution curve based on the trend component to obtain a long-term risk change trend graph; Conducting correlation analysis on the risk change trend graph and the market benchmark index to obtain a quantitative indicator of the correlation between risk and market; The quantitative indicators of risk and market correlation are matched and compared with historical risk events to obtain risk evolution prediction results based on historical patterns.

8. A financial transaction risk assessment system based on a multi-layer neural network, characterized in that: A method for implementing a financial transaction risk assessment method based on a multi-layer neural network according to any one of claims 1 to 7, wherein the financial transaction risk assessment system based on a multi-layer neural network comprises: The acquisition module is used to collect and calculate multi-dimensional risk indicators of financial market data to obtain a financial risk characteristic matrix including market volatility, liquidity ratio, systemic risk value and macroeconomic factors; A conversion module, configured to perform time series to image processing on historical transaction data based on the financial risk feature matrix to obtain a risk feature map dataset suitable for two-dimensional convolution analysis; A training module, configured to train a two-dimensional convolutional neural network transaction model CNN-TA and a multi-layer perceptron based on the risk feature map dataset to obtain a financial transaction risk assessment network; The classification module is used to input real-time transaction data into the financial transaction risk assessment network to calculate and classify risk levels, and obtain a hierarchical structured risk rating report and transaction early warning signals.

9. A financial transaction risk assessment device based on a multi-layer neural network, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the financial transaction risk assessment method based on a multi-layer neural network according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the financial transaction risk assessment method based on a multi-layer neural network according to any one of claims 1 to 7.

Images