Multi-modal fusion power plant financial risk assessment method, device, equipment and medium

Abstract

The invention provides a multi-modal fusion power plant financial risk assessment method and device, equipment and a medium, and relates to the field of risk assessment, and the method comprises the steps: obtaining time sequence numerical data, text data and image data, and obtaining dynamic features and static features through feature extraction; copying the static features and aligning the static features with the dynamic features at the corresponding time; calculating the attention weight of the dynamic feature of each time slice to the static feature, and splicing the static feature and the dynamic feature according to the attention weight to obtain a fusion feature; constructing a prediction model according to the fusion features and prior risk information; and obtaining to-be-evaluated data, and inputting the to-be-evaluated data into the prediction model to obtain a prediction result. According to the method, attention of dynamic data to different static data is calculated by integrating multiple data types such as time sequence numerical values, texts and images, so that data of different modalities are effectively fused, and key factors influencing financial risks of power generation enterprises are accurately captured.

Description

Technical Field

[0001] The present invention relates to the field of risk assessment, and in particular to a multi-modal fusion power plant financial risk assessment method, device, equipment and medium. Background Art

[0002] As the market environment faced by power generation companies becomes increasingly complex, the importance of financial risk assessment continues to increase. Traditional financial risk assessment methods mainly rely on historical financial statement data (such as balance sheets, income statements, etc.), and judge the financial health of enterprises by calculating financial indicators (such as debt-to-asset ratio, current ratio, etc.). However, this method has obvious limitations: it ignores the impact of external factors (such as power market price fluctuations, policy and regulatory changes, equipment operating status, etc.) on financial risks, and is difficult to adapt to the rapidly changing market environment. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-modal fusion power plant financial risk assessment method, device, equipment and medium to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows: In a first aspect, the present application provides a multi-modal fusion power plant financial risk assessment method, comprising: Acquire time series numerical data, text data and image data, and obtain dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; The static features are copied and aligned with the dynamic features at the corresponding time; Calculate the attention weight of the dynamic features of each time slice to the static features, and concatenate the static features with the dynamic features according to the attention weight to obtain the fused features; constructing a prediction model based on the fusion features and prior risk information; Obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

[0004] In a second aspect, the present application provides a multi-modal fusion power plant financial risk assessment system, comprising: The first module is used to obtain time series numerical data, text data and image data, and obtain dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; The second module is used to copy the static features and align them with the dynamic features at the corresponding time; The third module is used to calculate the attention weight of the dynamic features of each time slice to the static features, and concatenate the static features with the dynamic features according to the attention weight to obtain the fused features; The fourth module is used to build a prediction model based on the fusion features and prior risk information; The fifth module is used to obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

[0005] In the third aspect, the present application also provides a multimodal fusion power plant financial risk assessment device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor executes the computer program, the steps of the multimodal fusion power plant financial risk assessment method as described above are implemented.

[0006] In a fourth aspect, the present application further provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the multi-modal fusion power plant financial risk assessment method as described above are implemented.

[0007] The beneficial effects of the present invention are: The present invention integrates multiple data types such as time series values, texts and images, calculates the attention of dynamic data to different static data, and thus effectively integrates data of different modes to accurately capture the key factors affecting the financial risks of power generation enterprises. Compared with the traditional single data source method, this multimodal fusion can more accurately reflect the risk status in a complex market environment, thereby improving the comprehensiveness and accuracy of the assessment.

[0008] Other features and advantages of the present invention will be set forth in the following description, and in part will become apparent from the description, or may be understood by practicing embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 This is a flow chart of the multi-modal fusion power plant financial risk assessment method according to an embodiment of the present application; Figure 2 This is a structural diagram of a multi-modal fusion power plant financial risk assessment system according to an embodiment of the present application; Figure 3 This is a structural diagram of the multi-modal fusion power plant financial risk assessment equipment in an embodiment of the present application.

[0011] Markings in the figure: 100 - first module; 110 - first unit; 120 - second unit; 130 - third unit; 140 - fourth unit; 200 - second module; 300 - third module; 400 - fourth module; 500 - fifth module; 800 - multi-modal fusion power plant financial risk assessment equipment; 801 - processor; 802 - memory; 803 - multimedia component; 804 - I / O interface; 805 - communication component. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0013] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0014] At present, some energy companies use semi-automatic financial risk monitoring systems, which have certain data collection and preliminary analysis capabilities. The system is mainly composed of data collection modules, simple analysis modules and basic display modules.

[0015] Data collection module: mainly obtains financial data from the company's internal financial system, such as balance sheet, income statement data, etc. For external data, only some public industry statistics can be manually collected, and the update frequency is low. For example, at the end of each month, financial personnel manually download the monthly energy market supply and demand data released by the industry association.

[0016] Simple analysis module: Use fixed financial indicator formulas for calculation and analysis, such as calculating asset-liability ratio and current ratio to evaluate debt repayment ability. However, the analysis method is relatively simple and lacks in-depth data mining. For example, when evaluating the debt repayment risk of an enterprise, the indicators are calculated only based on historical data, without considering the potential impact of energy market fluctuations on the company's future debt repayment ability.

[0017] Basic display module: presents analysis results in tables and simple charts (such as bar charts and line charts). The display content is limited to the changing trends of basic financial indicators and cannot intuitively reflect complex risk relationships. Example 1

[0018] This embodiment provides a multi-modal fusion power plant financial risk assessment method, including steps S100, S200, S300, S400 and S500; S100, acquiring time series numerical data, text data and image data, and obtaining dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; In this step, the time series numerical data includes electricity price data, raw material price data, power generation equipment operation data, enterprise financial data, etc.; the text data includes text information such as policies and regulations, enterprise announcements, and news reports; the image data includes equipment monitoring images (such as infrared thermal imaging) and power grid layout diagrams.

[0019] Dynamic features are extracted from time series numerical data, which are features that change dynamically over time. The method for extracting dynamic features is as follows: S110, performing multi-scale decomposition on the time series numerical data by using wavelet transform to obtain wavelet coefficients of different scales; The discrete wavelet transform formula is: ;

[0020] in, is the wavelet coefficient, indicating the signal at scale and location The wavelet component on ; is the input signal; is the scale level (number of decomposition levels, controlling the scaling of the wavelet); is the translation parameter, controlling the position of the wavelet; is the discrete wavelet basis function; n represents the sampling point of the signal.

[0021] S120, perform Fourier transform on the wavelet coefficients of each scale to obtain a transformation result. The transformation formula is: ;

[0022] in, The scale is The representation of wavelet coefficients in the frequency domain; is the frequency; is an imaginary unit; S130, extracting low-frequency components, medium-frequency components, and high-frequency components of the time series numerical data based on the transformation result, and calculating the importance of each frequency component based on an energy ratio formula to obtain a feature weight of each frequency component; First, we need to define the dividing point between low-frequency components, medium-frequency components, and high-frequency components, that is, to divide the frequency range; let the total frequency range be F and the low-frequency range be F1 , the intermediate frequency range is F 2 , the high frequency range is F 3 , F m is one of the frequency ranges, m=1, 2, 3; The importance of each frequency component is calculated according to the energy ratio formula, that is, the ratio of the energy of each frequency band to the total energy: ;

[0023] in, The frequency range F m The energy proportion of is the characteristic weight of each frequency component; S140, weighting each frequency component using a corresponding feature weight to obtain dynamic features of the time series numerical data.

[0024] In the dynamic characteristics, low-frequency components can reflect the long-term trend of data, medium-frequency components can reflect the periodic fluctuations of data, and high-frequency components can reflect the random fluctuations of data; The extracted static features include text features and image features, among which the image features include equipment status features and power grid node features.

[0025] Text features are extracted using the BERT model; BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model based on the Transformer architecture. It can capture the semantic information of words in different contexts by learning from large-scale text data. When a piece of text is input into the BERT model, BERT generates a high-dimensional vector representation for each word or subword in the text. This vector contains the semantic information of the word in the current context. In this step, BERT can extract and understand key information in policies and regulations, corporate announcements, or news reports.

[0026] The equipment status features are extracted using the ResNet50 (Residual Network with 50 layers) deep convolutional neural network. The input of ResNet50 can be the infrared thermal imaging of the equipment. By analyzing the image features of the equipment, it can be determined whether the equipment's operating efficiency or temperature distribution is normal.

[0027] The original surveillance image is resized to a uniform size (e.g. 224x224 pixels) and normalized. If the image is a color image (RGB), it is converted to a format suitable for input to ResNet50. ResNet50 performs convolution operations on the image layer by layer, gradually extracting low-level features (such as edges, textures) and high-level features (such as shapes, patterns). At the output of the last convolutional layer of ResNet50, a high-dimensional feature vector (usually 2048 dimensions) is generated. This vector represents the high-level semantic features of the image.

[0028] In order to obtain the characteristics of power grid nodes, we first need to obtain power grid information and build a power grid topology map based on the power grid information. The nodes in the power grid topology map are key equipment in the power grid, such as power plants, substations, load points, etc. The edges represent the connection relationship between the equipment. According to the working parameters of the equipment in the power grid, the initial node vector of the power grid topology graph is generated, and the edge vector is generated according to whether the equipment is connected to each other; The initial node vector can be represented by the performance of each device such as the power generation capacity and efficiency. For example, the initial node vector of the substation is [80,60], which represents its capacity and load capacity. The edge vector describes the connection relationship between nodes. For example, if there is a connection between power plant A and substation B, the edge vector between the two is 1, and if there is no connection, it is 0; Based on the graph convolutional network, the power grid topology graph is convolved, the initial node vector is updated, and the power grid node features are obtained.

[0029] This application convolves the power grid topology graph through GCN (Graph Convolutional Network), which is a deep learning model specifically designed to process graph structured data. It generates feature representations for each node by aggregating information about the node and its neighbors. After completing the last layer of convolution, the power grid node features of each node can be obtained, which contain information about itself and nearby nodes. By analyzing the power grid topology graph, the possible propagation paths of power grid faults can be predicted, and the operating status of the equipment can be evaluated by combining its physical properties and topological relationships.

[0030] S200, copying the static features and aligning them with the dynamic features at the corresponding time; Directly concatenating dynamic features (such as time series data) and static features (such as text and images) will lead to feature dimension mismatch or information loss. Therefore, it is necessary to first align the static features with the time series features in the time dimension.

[0031] Assume that a time series data has T time slices, and the text feature within the corresponding time is F Text , then copy it T times to form T F Text Image feature FImage Similarly, after replication, T F Image . Each static feature is aligned with the time slice and associated.

[0032] S300, calculating the attention weight of the dynamic feature of each time slice to the static feature, and concatenating the static feature with the dynamic feature according to the attention weight to obtain a fusion feature; Since the importance of the same static feature may be different at different time slices, the attention mechanism can be used to dynamically adjust the weights of static features at different time slices.

[0033] This step uses the Multi-Head Attention mechanism to calculate the dynamic weights of static features: dynamic feature FTS, text feature F Text and image features F Image Perform linear transformations on each of them to generate query, key, and value vectors. Separate the query, key, and value into multiple heads, calculate the attention score on each head, concatenate the outputs of all heads, and obtain the final output through linear transformation.

[0034] The text feature taking into account the attention weight is denoted as FText_d, and the image feature taking into account the attention weight is denoted as FImage_d.

[0035] After splicing, we get the fusion feature F Fusion : F Fusion =Concat(FTS,FText_d,FImage_d).

[0036] S400, constructing a prediction model according to the fusion features and prior risk information; When training the model, this embodiment not only uses fusion features but also introduces interaction features, so as to capture the potential relationship between modal data.

[0037] Firstly, principal component analysis is used to filter and process the time series numerical features. Principal component analysis (PCA) projects high-dimensional data into low-dimensional space through linear transformation while retaining the main features of the data. At the same time, recursive feature elimination is used to filter and process semantic features, device status features and power grid node features to obtain the target feature subset, which contains the filtered and retained time series numerical features, semantic features, device status features and power grid node features. Constructing an interactive feature according to the correlation of each feature in the target feature subset, and splicing the fusion feature and the interactive feature to obtain a first processing result; the step of constructing the interactive feature specifically includes: Inputting the semantic features into a classification model to obtain a category corresponding to the text data; In this embodiment, the text data can be classified into the following categories: Energy subsidies: content related to energy-related subsidy policies provided by the government; Environmental protection requirements: content related to environmental protection regulations or emission standards; Market dynamics: Content related to electricity market price fluctuations or industry trends.

[0038] First, you need to pre-train the classification model and collect and annotate the training data. For example, extract text from policies and regulations, corporate announcements, or news reports, and manually annotate the subject categories to which they belong. The data format is usually a table with two columns: text content and corresponding subject labels.

[0039] Determine a time window according to the category of the text data, and determine a time window weight according to the semantic feature; In the previous step, the semantic features output by the BERT model are high-dimensional vectors, which need to be converted into scalar values. The high-dimensional vectors can be mapped to scalar weights through linear mapping; or the high-dimensional semantic features can be input into a multi-layer perceptron to map them to scalar weights.

[0040] The release time point of the text data is obtained, and the electricity price fluctuation data is obtained near the release time point based on the time window and its weight, so as to obtain the first interactive feature.

[0041] When the time window is determined to be K days based on the text data, that is, the abnormally fluctuating electricity price data is extracted K days before and after the release time point, the first interactive feature is obtained based on the data, and the first interactive feature is weighted according to the time window weight.

[0042] In addition, other interactive features can be constructed, such as: Extract policy-related equipment status change features, such as photovoltaic equipment status changes under new energy policies, and extract policy-related grid layout change features, such as new transmission lines and substation construction; It is also possible to use time series numerical features, text features, and image features as nodes to construct a relationship graph, and use a graph convolutional network to perform feature learning on the constructed relationship graph to obtain interactive features between multiple modalities.

[0043] The interactive features constructed above can be concatenated with the fusion features and used together as the input of the training model.

[0044] A prediction model is constructed based on the first processing result and prior risk information.

[0045] The first processing result is the result after multimodal fusion, which is used as the model input and the corresponding prior risk information is used as the output to train the model; The long short-term memory network model architecture is adopted. The hyperparameter range of the prediction model is determined based on the random search method. The target hyperparameter is determined by searching within the hyperparameter range based on the grid search method. In order to avoid overfitting, the early stopping method is introduced in the training process. Specifically, the performance on the validation set is continuously monitored. If the performance of the validation set does not improve in 10 consecutive iterations, the training is stopped and the current best model is retained.

[0046] The model is trained based on the loss function and the target hyperparameters, and the learning rate of the model parameters is adjusted by an adaptive moment estimation (Adam) optimizer to obtain a target prediction model. The loss function is a cross entropy loss function.

[0047] S500: Obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

[0048] The data to be evaluated are the current time series numerical data, text data and image data. This application uses ETL tools (Extract-Transform-Load) to refresh the data regularly, so that the model can always make predictions based on the latest data, thereby improving accuracy. The constructed prediction model is used to dynamically and in real time conduct financial risk assessments, which not only includes the function of traditional financial risk systems to predict financial risks based on financial statements, but also can monitor and predict in real time the impact of energy policies and market fluctuations on corporate finances. Example 2

[0049] This embodiment provides a multi-modal fusion power plant financial risk assessment system, including: The first module 100 is used to obtain time series numerical data, text data and image data, and obtain dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; The second module 200 is used to copy the static features and align them with the dynamic features at the corresponding time; The third module 300 is used to calculate the attention weight of the dynamic feature of each time slice to the static feature, and splice the static feature with the dynamic feature according to the attention weight to obtain a fusion feature; The fourth module 400 is used to build a prediction model based on the fusion features and prior risk information; The fifth module 500 is used to obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

[0050] As an optional implementation manner, the first module 100 includes: The first unit 110 is used to perform multi-scale decomposition on the time series numerical data by using wavelet transform to obtain wavelet coefficients of different scales; The second unit 120 is used to perform Fourier transform on the wavelet coefficients of each scale to obtain a transform result; The third unit 130 is used to extract the low-frequency component, the medium-frequency component and the high-frequency component of the time series numerical data based on the transformation result, and calculate the importance of each frequency component based on the energy ratio formula to obtain the feature weight of each frequency component; The fourth unit 140 is used to weight each frequency component using a corresponding feature weight to obtain a dynamic feature of the time series numerical data. Example 3

[0051] Corresponding to the above method embodiment, a multimodal fusion power plant financial risk assessment device is also provided in this embodiment. The multimodal fusion power plant financial risk assessment device described below and the multimodal fusion power plant financial risk assessment method described above can be referenced to each other.

[0052] Figure 3 FIG. 8 is a block diagram of a multi-modal fusion power plant financial risk assessment device 800 according to an exemplary embodiment. Figure 3As shown, the multimodal fusion power plant financial risk assessment device 800 includes a processor 801 and a memory 802. The multimodal fusion power plant financial risk assessment device 800 may also include a multimedia component 803, an input / output (I / O) interface 804, and one or more of a communication component 805. Among them, the processor 801 is used to control the overall operation of the multimodal fusion power plant financial risk assessment device 800 to complete all or part of the steps in the multimodal fusion power plant financial risk assessment method described above. The memory 802 is used to store various types of data to support the operation of the multimodal fusion power plant financial risk assessment device 800, which may include, for example, commands for any application or method used to operate on the multimodal fusion power plant financial risk assessment device 800, and application-related data, such as contact data, messages sent and received, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0053] The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, and the microphone is used to receive external audio signals.

[0054] The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the multimodal fusion power plant financial risk assessment device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0055] In an exemplary embodiment, the device 800 for mutual signing and verification of digital files can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned multi-modal fusion power plant financial risk assessment method. Example 4

[0056] Corresponding to the above multimodal fusion power plant financial risk assessment method embodiment, this embodiment also provides a readable storage medium, and the readable storage medium described below and the multimodal fusion power plant financial risk assessment method described above can be referenced to each other.

[0057] A readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the embodiment of the multi-modal fusion power plant financial risk assessment method.

[0058] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0059] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0060] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A multi-modal fusion power plant financial risk assessment method, characterized in that: include: Acquire time series numerical data, text data and image data, and obtain dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; The static features are copied and aligned with the dynamic features at the corresponding time; Calculate the attention weight of the dynamic features of each time slice to the static features, and concatenate the static features with the dynamic features according to the attention weight to obtain the fused features; constructing a prediction model based on the fusion features and prior risk information; Obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

2. The multi-modal fusion power plant financial risk assessment method according to claim 1 is characterized in that: Acquire time series numerical data, text data and image data, and extract dynamic and static features through feature extraction; including: Using wavelet transform to perform multi-scale decomposition on the time series numerical data to obtain wavelet coefficients of different scales; Perform Fourier transform on the wavelet coefficients of each scale to obtain the transformation result; Extracting low-frequency components, medium-frequency components and high-frequency components of the time series numerical data based on the transformation results, and calculating the importance of each frequency component based on an energy ratio formula to obtain a feature weight of each frequency component; Each frequency component is weighted using the corresponding feature weight to obtain the dynamic characteristics of the time series numerical data.

3. The multi-modal fusion power plant financial risk assessment method according to claim 1 is characterized in that: The static characteristics include grid node characteristics; Acquire time series numerical data, text data and image data, and extract dynamic and static features through feature extraction; including: Obtain power grid information and build a power grid topology map based on the power grid information; According to the working parameters of the equipment in the power grid, the initial node vector of the power grid topology graph is generated, and the edge vector is generated according to whether the equipment is connected to each other; Based on the graph convolutional network, the power grid topology graph is convolved, the initial node vector is updated, and the power grid node characteristics are obtained.

4. The multi-modal fusion power plant financial risk assessment method according to claim 1 is characterized in that: The static features include semantic features, device status features and power grid node features; The constructing of a prediction model according to the fusion features and prior risk information includes: The principal component analysis is used to filter the time series numerical features, and the recursive feature elimination is used to filter the semantic features, equipment status features and power grid node features to obtain the target feature subset. Constructing interactive features according to the correlation of each feature in the target feature subset, and splicing the fusion feature and the interactive feature to obtain a first processing result; A prediction model is constructed based on the first processing result and prior risk information.

5. The multi-modal fusion power plant financial risk assessment method according to claim 4 is characterized in that: The time series numerical data includes electricity price data; the interactive features are constructed according to the correlation of each feature in the target feature subset, including: Inputting the semantic features into a classification model to obtain a category corresponding to the text data; Determine a time window according to the category of the text data, and determine a time window weight according to the semantic feature; The release time point of the text data is obtained, and the electricity price fluctuation data is obtained near the release time point based on the time window and its weight, so as to obtain the first interactive feature.

6. The multi-modal fusion power plant financial risk assessment method according to claim 1 is characterized in that: A prediction model is constructed based on the fusion features and prior risk information, including: Determine a hyperparameter range of a prediction model based on a random search method, and search within the hyperparameter range based on a grid search method to determine a target hyperparameter; The model is trained based on the loss function and the target hyperparameters, and the learning rate of the model parameters is adjusted by an adaptive moment estimation optimizer to obtain a target prediction model, wherein the loss function is a cross entropy loss function.

7. A multi-modal fusion power plant financial risk assessment system, characterized in that: include: The first module is used to obtain time series numerical data, text data and image data, and obtain dynamic features and static features through feature extraction; the dynamic features include time series numerical features of at least one time slice; The second module is used to copy the static features and align them with the dynamic features at the corresponding time; The third module is used to calculate the attention weight of the dynamic features of each time slice to the static features, and concatenate the static features with the dynamic features according to the attention weight to obtain the fused features; The fourth module is used to build a prediction model based on the fusion features and prior risk information; The fifth module is used to obtain the data to be evaluated, input the data to be evaluated into the prediction model, and obtain the prediction result.

8. The multi-modal fusion power plant financial risk assessment system according to claim 7 is characterized in that: The first module includes: The first unit is used to perform multi-scale decomposition on the time series numerical data by wavelet transform to obtain wavelet coefficients of different scales; The second unit is used to perform Fourier transform on the wavelet coefficients of each scale to obtain the transformation result; The third unit is used to extract the low-frequency component, the medium-frequency component and the high-frequency component of the time series numerical data based on the transformation result, and calculate the importance of each frequency component based on the energy ratio formula to obtain the feature weight of each frequency component; The fourth unit is used to weight each frequency component using a corresponding feature weight to obtain a dynamic feature of the time series numerical data.

9. A multi-modal fusion power plant financial risk assessment device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the multimodal fusion power plant financial risk assessment method as claimed in any one of claims 1 to 6 when executing the computer program.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the multi-modal fusion power plant financial risk assessment method according to any one of claims 1 to 6 are implemented.

Images