A method and system for monitoring abnormal fluctuation risk in farmers' markets based on cloud computing

Abstract

The present invention discloses a method and system for monitoring the risk of abnormal fluctuations in farmers' markets based on cloud computing, and relates to the field of monitoring farmers' markets. The method comprises: extracting time series features of crowd emotions and body movements; fusing the extracted time series features of crowd emotions and body movements with farmers' market data, and applying a deep learning algorithm to predict abnormal fluctuations in farmers' markets; based on the predicted abnormal fluctuations in farmers' markets, analyzing the degree of influence of abnormal fluctuations on the stability of farmers' markets, and constructing an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations; feeding back the fluctuation risk assessment results to the farmers' market management end in real time through the cloud, and executing an adaptive control mechanism based on the feedback data. The present invention not only provides a rich data basis for subsequent abnormal fluctuation predictions by collecting multimedia data from farmers' markets and using a cloud computing platform, but also greatly improves the efficiency and accuracy of data processing.

Description

Technical Field

[0001] This invention relates to the field of agricultural market monitoring, and more specifically, to a cloud computing-based method and system for monitoring the risk of abnormal fluctuations in agricultural markets. Background Technology

[0002] With the acceleration of agricultural modernization and the continuous improvement of living standards for urban and rural residents, farmers' markets, as an indispensable part of the agricultural product circulation system, play an irreplaceable role in ensuring the supply of agricultural products, stabilizing market prices, and promoting farmers' income. As a bridge connecting producers and consumers, farmers' markets not only provide urban residents with fresh and safe agricultural products, but also play a crucial role in promoting local economic development, maintaining food safety, and ensuring social stability.

[0003] However, in actual operation, agricultural markets often face abnormal fluctuations due to various internal and external factors, such as drastic price fluctuations and supply-demand imbalances. These abnormal fluctuations not only disrupt the normal operation of the market but may also adversely affect the rights and interests of consumers and the business operations of merchants, even impacting the entire industry chain, affecting the stability of farmers' income and their enthusiasm for agricultural production. Therefore, establishing a sound and effective monitoring mechanism is particularly important.

[0004] Traditional agricultural market monitoring methods primarily focus on analyzing structured information such as transaction data and price fluctuations. While these indicators are important for understanding the market, they often overlook unstructured information generated by human behavior, such as changes in market sentiment and evolution of consumer behavior patterns. This leads to incomplete risk identification, difficulty in detecting potential early warning signals, and thus limits managers' ability to gain a deeper understanding of market dynamics and make precise adjustments. This not only reduces the efficiency of responding to emergencies but also makes market management inflexible.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] In response to the problems in related technologies, this invention proposes a cloud computing-based method and system for monitoring the risk of abnormal fluctuations in agricultural markets, in order to overcome the aforementioned technical problems existing in the existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for monitoring the risk of abnormal fluctuations in agricultural markets based on cloud computing is provided, the method comprising the following steps: S1. Collect multimedia data from farmers' markets and process the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements. S2. The extracted time-series features of crowd emotions and body movements are integrated with agricultural market data to establish a multimodal time-series dataset. Deep learning algorithms are then used to detect abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time-series dataset in order to predict abnormal fluctuations in agricultural markets. S3. Based on the predicted abnormal fluctuations in farmers' markets, analyze the degree of impact of abnormal fluctuations on the stability of farmers' markets, and construct an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations. S4. The volatility risk assessment results are fed back to the agricultural market management terminal in real time via the cloud, and an adaptive control mechanism is implemented based on the feedback data; The extracted time-series features of crowd emotions and body language are fused with agricultural market data to establish a multimodal time-series dataset. Deep learning algorithms are then used to detect anomalous co-fluctuation patterns and anomalous emotion diffusion characteristics in the multimodal time-series dataset to predict anomalous fluctuations in agricultural markets. The steps include: S21. Collect data from farmers' markets and combine it with the extracted time-series features of crowd emotions and body movements. Align the data according to timestamps to establish a multimodal time-series dataset. S22. Based on the deep learning algorithm, construct a multimodal deep learning model and train the multimodal deep learning model using a pre-collected historical multimodal time series dataset. S23. Input the established multimodal time series dataset into the trained multimodal deep learning model to identify abnormal cooperative fluctuation patterns between different data sources; S24. Using the self-excited point process algorithm, model the propagation pattern of crowd emotional fluctuations, analyze the correlation between emotional fluctuations and body movements, and combine the isolated forest algorithm to detect the characteristics of abnormal emotional diffusion. S25. Analyze the dynamic correlation between abnormal collaborative fluctuation patterns and abnormal sentiment diffusion characteristics to predict abnormal fluctuations in agricultural markets.

[0008] In one embodiment, collecting multimedia data from a farmers' market and processing the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements includes the following steps: S11. Using cloud technology, collect multimedia data from farmers' markets and preprocess the collected multimedia data, which includes video and audio data. S12. Based on deep learning algorithms, analyze facial expressions in preprocessed video data and integrate preprocessed audio data to identify the emotional characteristics of people in farmers' markets. S13. Using a pose estimation algorithm, extract the limb movement data of the crowd in the preprocessed video data, and capture the limb movement features of the crowd from the limb movement data through a pre-built behavior detection model. S14. Using the distributed computing framework of the cloud computing platform, process the emotional and physical characteristics of the crowd to extract the time-series features of the crowd's emotions and physical movements.

[0009] In one embodiment, analyzing facial expressions in preprocessed video data using a deep learning algorithm and fusing it with preprocessed audio data to identify the emotional characteristics of people in a farmers' market includes the following steps: S121. Construct a facial expression recognition model based on deep learning algorithms, and train the facial expression recognition model using historical multimedia data. S122. Input the preprocessed video data into the trained facial expression recognition model and output the expression and emotion features corresponding to each face. S123. Using the Long Short-Term Memory Network algorithm, analyze the preprocessed audio data and output the corresponding audio emotional features. S124. Based on a weighted algorithm, facial expression emotion features and audio emotion features are fused to identify the emotion features of people in farmers' markets.

[0010] In one embodiment, the self-excited point process algorithm is used to model the propagation pattern of crowd emotional fluctuations, analyze the correlation between emotional fluctuations and body movements, and combine the isolated forest algorithm to detect abnormal emotional diffusion characteristics, including the following steps: S241. Based on a multimodal time series dataset, the emotion propagation function is calculated using the self-excitation point process algorithm, and the propagation pattern of emotion fluctuations in the population is modeled using the emotion propagation function. S242. Based on the completed modeling of the population's emotional fluctuation patterns, use the Bayesian network algorithm to analyze the correlation between emotional fluctuations and physical movements in the multimodal time series dataset. S243. For each time step, combine the emotional features and body movement features in the multimodal time series dataset to establish a comprehensive feature vector; S244. Using the isolated forest algorithm, construct an anomaly detection model and train the anomaly detection model using comprehensive feature vectors. S245. Calculate the anomaly score for the comprehensive feature vector at each time step using the trained isolated forest model, and identify the characteristics of abnormal emotion diffusion based on the preset anomaly threshold.

[0011] In one embodiment, the formula for the emotion propagation function is: ; In the formula, or ( t ) indicates time t The intensity of emotional fluctuations at the location; t Indicates the duration of the current emotional fluctuation; r Indicates the foundation strength; t i Indicates the period of past emotional fluctuations; s Indicates the trigger coefficient; l This indicates the attenuation rate.

[0012] In one embodiment, based on the modeled patterns of emotional fluctuations in the population, analyzing the correlation between emotional fluctuations and physical actions in a multimodal time series dataset using a Bayesian network algorithm includes the following steps: S2421. Based on the emotional fluctuation pattern of the population after modeling, select emotional parameters and body movement parameters from the multimodal time series dataset as nodes to construct the Bayesian network structure. S2422. Construct an initial Bayesian network structure using the maximum support tree algorithm, and process the initial Bayesian network structure using the hill-climbing algorithm to generate a Bayesian network structure as the initial population. S2423. Calculate the fitness value of each Bayesian network structure in the initial population, sort them according to the fitness value, and select several Bayesian network structures with high fitness values ​​as the new population. S2424. Based on the Bayesian network structure of each individual in the new population, recalculate the corresponding fitness value, identify the individuals with the best and worst fitness in the current population, and update the individual positions based on the individuals with the best and worst fitness. S2425. Repeat steps S2421 to S2424 until the preset number of iterations is reached to obtain the optimal Bayesian network structure. S2426. Using the optimal Bayesian network structure, analyze the correlation between emotional fluctuations and physical actions.

[0013] In one embodiment, based on predicted abnormal fluctuations in farmers' markets, the impact of these abnormal fluctuations on the stability of the farmers' markets is analyzed, and an abnormal behavior risk assessment model is constructed. The assessment of abnormal fluctuation risk includes the following steps: S31. Based on the predicted abnormal fluctuations in the farmers' market, use the local anomaly factor algorithm to calculate the abnormal sentiment index and determine the range of abnormal sentiment distribution. S32. Combining the determined abnormal sentiment distribution range with the conditional value at risk algorithm, analyze the degree of impact of abnormal fluctuations on the stability of farmers' markets. S33. Based on the degree of impact of abnormal fluctuations on the stability of agricultural markets, construct an abnormal behavior risk assessment model using an ensemble learning algorithm, and assess the risk of abnormal fluctuations through the abnormal behavior risk assessment model.

[0014] In one embodiment, analyzing the impact of abnormal fluctuations on the stability of farmers' markets, by combining a determined abnormal sentiment distribution range with a conditional value at risk algorithm, includes the following steps: S321. Based on the determined abnormal emotion distribution interval, construct a risk value model based on different quantiles, and use historical abnormal emotion characteristics to train the risk value model to obtain the risk value under each abnormal emotion distribution interval. S322. Combining value at risk and conditional variables, construct a conditional value at risk model, and train the conditional value at risk model using historical abnormal sentiment characteristics to obtain the conditional value at risk for each abnormal sentiment distribution interval. S323. Compare the value of risk and conditional value of risk under different emotional states, assess the impact of emotional changes on the value of risk, and analyze the degree of impact of abnormal fluctuations on the stability of the agricultural market based on the assessment results.

[0015] In one embodiment, the formula for the conditional value at risk model is: ; In the formula, CoVaR ( q ) indicates the quantile q Conditional risk value; q Indicates quantiles; α Represents the intercept term; β The coefficient representing the impact of Value at Risk; VaR ( q ) indicates the quantile q The risk value; c This indicates the stability value of the farmers' market; M Value indicators representing the risks of farmers' markets.

[0016] According to another aspect of the present invention, a cloud computing-based agricultural market abnormal fluctuation risk monitoring system is also provided. The system includes a data acquisition and processing module, a data analysis and anomaly detection module, a fluctuation analysis and risk assessment module, and a real-time feedback and adaptive control module. The data acquisition and processing module, the data analysis and anomaly detection module, the fluctuation analysis and risk assessment module, and the real-time feedback and adaptive control module are connected sequentially. The data acquisition and processing module is used to collect multimedia data from farmers' markets and process the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements. The data analysis and anomaly detection module is used to integrate the extracted time-series features of crowd emotions and body movements with agricultural market data to establish a multimodal time-series dataset. It also uses deep learning algorithms to detect abnormal co-fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time-series dataset in order to predict abnormal fluctuations in agricultural markets. The volatility analysis and risk assessment module is used to analyze the impact of predicted abnormal fluctuations on the stability of farmers' markets, and to construct an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations. The real-time feedback and adaptive control module is used to feed the fluctuation risk assessment results back to the agricultural market management terminal in real time via the cloud, and to execute an adaptive control mechanism based on the feedback data.

[0017] The beneficial effects of this invention are as follows: 1. This invention collects multimedia data from farmers' markets and utilizes the efficient processing capabilities of cloud computing platforms to extract time-series features of crowd emotions and physical movements in real time and accurately. This not only provides a rich data foundation for subsequent prediction of abnormal fluctuations but also greatly improves the efficiency and accuracy of data processing.

[0018] 2. By integrating multiple data sources, this invention constructs a multimodal time series dataset, enabling a more comprehensive and in-depth understanding of the dynamic changes in agricultural markets. At the same time, by employing deep learning algorithms, it can accurately identify abnormal collaborative fluctuation patterns and sentiment diffusion characteristics, providing early warnings of abnormal market fluctuations and offering a powerful tool for timely risk prevention.

[0019] 3. This invention analyzes abnormal fluctuations and their impact on market stability, and constructs an abnormal behavior risk assessment model. This not only quantifies the risk level, but also provides a solid basis for formulating scientific risk management strategies. As a result, it can more accurately assess the risk status of agricultural markets and take more effective measures to ensure the stable operation of the market. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a cloud computing-based method for monitoring the risk of abnormal fluctuations in agricultural markets according to an embodiment of the present invention; Figure 2This is a schematic diagram of a cloud computing-based agricultural market abnormal fluctuation risk monitoring system according to an embodiment of the present invention.

[0022] In the picture: 1. Data acquisition and processing module; 2. Data analysis and anomaly detection module; 3. Fluctuation analysis and risk assessment module; 4. Real-time feedback and adaptive control module. Detailed Implementation

[0023] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0024] According to an embodiment of the present invention, a method and system for monitoring the risk of abnormal fluctuations in agricultural markets based on cloud computing are provided.

[0025] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown in the figure, the cloud computing-based risk monitoring method for agricultural markets according to an embodiment of the present invention includes the following steps: S1. Collect multimedia data from farmers' markets and process the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements.

[0026] In this optional embodiment, collecting multimedia data from farmers' markets and processing the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements includes the following steps: S11. Using cloud technology, collect multimedia data from farmers' markets and preprocess the collected multimedia data, which includes video and audio data.

[0027] It should be further explained that the use of cloud technology to collect multimedia data from farmers' markets, and the preprocessing of the collected multimedia data, including video and audio data, are implemented in the following specific examples: Step 1: Data Collection Video data acquisition: Use high-definition cameras or webcams to set up multiple monitoring points in the farmers' market to collect video data in real time; consider using a frame capture device to capture video footage and convert it into digital format.

[0028] Audio data acquisition: Use high-sensitivity microphones or audio recording equipment to collect ambient and transaction sounds within the farmers' market; ensure that the audio equipment can cover every corner of the market to obtain comprehensive audio data.

[0029] Convert video and audio data into formats suitable for cloud storage and processing; upload the converted data to cloud storage systems such as Alibaba Cloud and Tencent Cloud; and use cloud storage services such as object storage and block storage to store the collected video and audio data.

[0030] Step 2: Data Preprocessing (1) Video data preprocessing Video shot segmentation: Video is segmented into shots using pixel difference methods, histogram-based methods, block matching methods, or motion-based methods; by comparing the differences between consecutive frames, the boundaries of shots are detected, and the video is divided into multiple shots.

[0031] Keyframe extraction: Extract keyframes from each shot, representing the most important and representative images in the shot; keyframe extraction can be performed using the first and last frame method, the middle frame method, or a feature extraction-based method.

[0032] Feature extraction: Extract features from keyframes, such as color features, texture features, and shape features; these features can be used for subsequent image analysis, object detection, and other tasks.

[0033] (2) Audio data preprocessing Audio noise reduction: Using filtering algorithms or noise reduction software, the acquired audio data is processed to reduce noise; environmental noise and background noise are removed to improve the clarity of the audio data.

[0034] Audio segmentation: Based on the characteristics of audio data, such as volume and frequency, audio is divided into multiple segments; these segments can represent different sound events, such as transaction sounds, conversation sounds, etc.

[0035] Feature extraction: Extract features from audio segments, such as Mel-frequency cepstral coefficients (MFCC) and spectral features; these features can be used for subsequent tasks such as speech recognition and sound classification.

[0036] S12. Based on deep learning algorithms, analyze facial expressions in the preprocessed video data and integrate the preprocessed audio data to identify the emotional characteristics of people in the farmers' market.

[0037] In this optional embodiment, the analysis of facial expressions in the preprocessed video data based on a deep learning algorithm, and the fusion of preprocessed audio data, to identify the emotional characteristics of people in a farmers' market includes the following steps: S121. Construct a facial expression recognition model based on deep learning algorithms, and train the facial expression recognition model using historical multimedia data. S122. Input the preprocessed video data into the trained facial expression recognition model and output the expression and emotion features corresponding to each face. S123. Using the Long Short-Term Memory Network algorithm, analyze the preprocessed audio data and output the corresponding audio emotional features. S124. Based on a weighted algorithm, facial expression emotion features and audio emotion features are fused to identify the emotion features of people in farmers' markets.

[0038] It should be further explained that, based on deep learning algorithms, the facial expressions in the preprocessed video data are analyzed and fused with the preprocessed audio data to identify the emotional characteristics of people in the farmers' market. A specific implementation example is shown below: Step 1: Build and train a facial expression recognition model Collect historical multimedia data, including facial images and video data, which should cover features such as different ages, genders, races, expressions, and lighting conditions to improve the robustness of the model; use data cleaning tools to check the image quality and annotation accuracy in the dataset, and remove data that does not meet the requirements; use data annotation tools to add labels to each facial image, and the labels can be expression categories, such as happy, sad, angry, etc.; generate new data samples through data augmentation techniques (such as rotation, scaling, flipping, etc.) to increase the diversity and scale of the dataset.

[0039] Convolutional Neural Networks (CNNs) are chosen as the basic architecture for the facial expression recognition model. Pre-trained weights, such as VGG and ResNet, are introduced into the CNN to accelerate the training process and improve model performance. The last layer of the network is modified according to the needs to adapt to the expression classification task, such as adjusting the number of neurons in the output layer to correspond to the expression category.

[0040] The model is trained using labeled face image data. Appropriate training strategies, optimization algorithms, and loss functions are selected to improve the model's performance. Regularization, dropout, and other techniques are used during training to prevent overfitting. The model performance is evaluated using a validation set, and hyperparameters are adjusted based on the evaluation results.

[0041] Step 2: Identify facial expressions and emotional features in the video. Keyframes or consecutive frames are extracted from the preprocessed video data. Face detection and cropping are performed on the frame images to extract face regions. The face regions are then preprocessed with grayscale conversion, binarization, edge detection, and other steps. The detected face regions are then input into a trained CNN model to output the emotional features of each face, such as happiness, sadness, and anger.

[0042] Step 3: Analyze the emotional features in the audio data Extract audio features (such as Mel-frequency cepstral coefficients, MFCC); construct a Long Short-Term Memory (LSTM) network model to process sequential audio feature data. LSTM can capture temporal dependencies in audio data; train the LSTM model using a labeled audio dataset (containing emotion labels) to enable it to recognize emotional features in audio; input the preprocessed audio data into the trained LSTM model to output corresponding audio emotion features, such as excitement, relaxation, and frustration.

[0043] Step 4: Integrate facial expression emotion features with audio emotion features A weighted algorithm is used to fuse facial expression emotion features with audio emotion features. The weighting algorithm can assign weights based on the importance or reliability of the features. For example, if the video data is of high quality and the facial expression recognition model performs well, higher weights can be given to facial expression emotion features; conversely, if the audio data contains more emotional information, higher weights can be given to audio emotion features.

[0044] Based on the fused features, a classifier (such as Support Vector Machine (SVM), Random Forest, or Deep Learning Model) is used to identify the emotional characteristics of people in the farmers' market; the final emotion recognition result is obtained, which is either the emotion category (such as happiness, sadness, etc.) or the emotion intensity (such as the degree of joy, anger, etc.).

[0045] S13. Using a pose estimation algorithm, extract the limb movement data of the crowd from the preprocessed video data, and capture the limb movement features of the crowd from the limb movement data through a pre-built behavior detection model.

[0046] It should be noted that the following is a specific implementation example of using pose estimation algorithms to extract the limb movement data of the crowd from the preprocessed video data, and capturing the limb movement features of the crowd from the limb movement data using a pre-built behavior detection model: Step 1: Attitude Estimation The preprocessed video data is processed frame by frame using pose estimation algorithms such as OpenPose or MediaPipe to extract the joint position information of the human body in each frame, including key parts such as the head, neck, shoulders, elbows, wrists, hips, knees and ankles; the joint position information is converted into relative coordinates or normalized coordinates to facilitate subsequent processing and analysis.

[0047] Step 2: Feature Extraction Based on the extracted joint position information, the posture features of the human body, such as joint angles and limb length ratios, are calculated. The joint position information of consecutive frames is tracked and analyzed to extract dynamic features such as the human body's motion trajectory and speed. The posture features and dynamic features are then fused to form complete limb motion data.

[0048] Step 3: Building the Behavior Detection Model Based on the specific behavior detection task (such as recognizing running, jumping, waving, etc.), the corresponding training dataset is collected and labeled, and a behavior detection model is built using deep learning frameworks, such as 3D convolutional neural networks (3D CNN) and long short-term memory networks (LSTM). The model is then trained and optimized using the training dataset to enable it to accurately recognize the target behavior.

[0049] Step 4: Behavior Capture and Recognition The preprocessed video data is input into the trained behavior detection model. The model captures the limb movement features of the crowd based on the input limb movement data and outputs the corresponding behavior recognition results. The recognition results are then post-processed and analyzed, such as noise removal, smoothing, and behavior classification.

[0050] S14. Using the distributed computing framework of the cloud computing platform, process the emotional and physical characteristics of the crowd to extract the time-series features of the crowd's emotions and physical movements.

[0051] It should be further explained that, through the distributed computing framework of the cloud computing platform, a specific example of processing the emotional and physical characteristics of a crowd to extract the time-series features of their emotions and physical movements is as follows: Configure an Apache Spark cluster on a cloud computing platform, including a master node and multiple slave nodes; upload the pre-processed video data to cloud storage, and write Spark jobs to process the data in parallel.

[0052] In the Spark job, emotion recognition models and body motion recognition models are used to extract features from video frames. For each video frame, emotion features (such as facial expressions and voice features) and body motion features (such as joint angles and movement trajectories) are extracted. The extracted features are then organized according to time series to form a time series feature dataset.

[0053] The time series feature dataset was analyzed using the time series analysis library statsmodels, including trend analysis, periodic analysis, and correlation analysis, to extract key time series features that reflect changes in people's emotions and physical movements, such as the trend of emotional intensity changes and the frequency and amplitude of physical movements.

[0054] For example, a week's worth of surveillance video data from a farmers' market was selected as experimental data, covering different time periods and crowd densities. After preprocessing the video data, approximately 100,000 video frames were obtained, containing emotional and physical characteristics.

[0055] The time-series characteristics of crowd emotions and body movements were extracted. Through time-series analysis, it was found that the intensity of crowd emotions showed obvious fluctuations in certain time periods, which were related to factors such as market promotion activities and crowd density. At the same time, body movement characteristics also showed certain periodic changes, such as people's activities being more frequent and body movement characteristics being more obvious on weekends.

[0056] S2. The extracted time-series features of crowd emotions and body movements are integrated with agricultural market data to establish a multimodal time-series dataset. Deep learning algorithms are then used to detect abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time-series dataset in order to predict abnormal fluctuations in agricultural markets.

[0057] The process involves fusing extracted time-series features of crowd emotions and body language with agricultural market data to create a multimodal time-series dataset. Deep learning algorithms are then used to detect anomalous co-fluctuation patterns and anomalous emotion diffusion characteristics within this dataset, in order to predict anomalous fluctuations in agricultural markets. The steps include: S21. Collect data from farmers' markets and combine it with the extracted time-series features of crowd emotions and body movements. Align the data according to timestamps to establish a multimodal time-series dataset.

[0058] It should be added that data on farmers' markets, such as transaction details, stall layout, and foot traffic, were collected through methods such as observation and interviews.

[0059] In a specific embodiment, assume a data collection activity was conducted at a farmers' market, and the information is as follows: Time: [Date] (9:00 AM to 5:00 PM).

[0060] Location: A large farmers' market in a certain city.

[0061] Step 1: Data Collection Farmers' market data: Through interviews with stall owners and observations of the market, hourly transaction volume, number of customers, and stall usage were recorded.

[0062] Crowd emotion data: Using a facial expression recognition model, the emotions of customers were identified, and the proportion of people with positive emotions (such as happiness and satisfaction) and negative emotions (such as dissatisfaction and anxiety) was recorded every hour.

[0063] Body movement data: Through a behavior detection model, the body movements of customers and vendors were captured, and hourly body movement features, such as movement frequency and amplitude, were extracted.

[0064] Step 2: Data Processing Data cleaning: Removes abnormal data caused by equipment failure or human error.

[0065] Feature extraction: Time series statistical features such as mean and variance of transaction volume, number of customers and stall utilization rate were extracted from the behavior detection model; the proportion of positive and negative emotions was extracted from the crowd emotion data; and the time series features of movement frequency and movement amplitude were extracted from the body movement data.

[0066] Timestamp alignment: All data was aligned according to hourly timestamps to ensure that the data at each point in time is complete.

[0067] S22. Based on the deep learning algorithm, construct a multimodal deep learning model and train the multimodal deep learning model using a pre-collected historical multimodal time series dataset.

[0068] It should be noted that choosing a multimodal deep learning model based on the Transformer architecture can effectively capture long-range dependencies in time-series data and is suitable for the fusion and collaborative learning of multimodal data.

[0069] S23. Input the established multimodal time series dataset into the trained multimodal deep learning model to identify abnormal cooperative fluctuation patterns between different data sources.

[0070] It should be further explained that the preprocessed multimodal time series dataset is input into the trained multimodal deep learning model. The model extracts effective features from the data of each modality through the individual feature learning stage; through the information fusion stage, the features of different modalities are fused to form a shared representation that can comprehensively reflect the information of multiple modal data; in the testing stage, the model performs anomaly detection on the input data. When the model finds that the data features at a certain time point are significantly different from the normal data features, it is considered that there is an abnormal co-fluctuation pattern at that time point.

[0071] Suppose that at a certain point in time, the transaction volume in a farmers' market suddenly increases, while the transaction value and the price of agricultural products remain unchanged or decrease slightly. At the same time, this point in time happens to be a rainy day and not a holiday. Based on these characteristics, the model judges that there is an abnormal coordinated fluctuation pattern at this point in time, which may be due to abnormal consumer purchasing behavior caused by weather.

[0072] Based on the model's judgment results, the reasons for the abnormal fluctuations can be further analyzed. For example, rainy days may cause inconvenience for consumers, thus reducing the frequency of their purchases at farmers' markets. At the same time, due to the relatively stable supply of agricultural products, prices have not fluctuated significantly. These factors work together to lead to the abnormal phenomenon of increased transaction volume but unchanged transaction value and prices.

[0073] S24. Using the self-excited point process algorithm, model the propagation pattern of crowd emotional fluctuations, analyze the correlation between emotional fluctuations and body movements, and combine the isolated forest algorithm to detect the characteristics of abnormal emotional diffusion.

[0074] In this optional embodiment, the propagation pattern of crowd emotional fluctuations is modeled using the self-excited point process algorithm, the correlation between emotional fluctuations and body movements is analyzed, and the abnormal emotional diffusion characteristics are detected by combining the isolated forest algorithm, including the following steps: S241. Based on a multimodal time series dataset, the emotion propagation function is calculated using the self-excitation point process algorithm, and the propagation pattern of emotion fluctuations in the population is modeled using the emotion propagation function.

[0075] In this optional embodiment, the formula for the emotion propagation function is: ; In the formula, or ( t ) indicates time t The intensity of emotional fluctuations at the location; t Indicates the duration of the current emotional fluctuation; r Indicates the foundation strength; t i Indicates the period of past emotional fluctuations; s Indicates the trigger coefficient; l This indicates the attenuation rate.

[0076] S242. Based on the completed modeling of the population's emotional fluctuation patterns, use the Bayesian network algorithm to analyze the correlation between emotional fluctuations and physical movements in the multimodal time series dataset.

[0077] In this optional embodiment, based on the modeled crowd emotion fluctuation patterns, analyzing the correlation between emotion fluctuations and body movements in a multimodal time series dataset using a Bayesian network algorithm includes the following steps: S2421. Based on the emotional fluctuation pattern of the population after modeling, select emotional parameters and body movement parameters from the multimodal time series dataset as nodes to construct the Bayesian network structure. S2422. Construct an initial Bayesian network structure using the maximum support tree algorithm, and process the initial Bayesian network structure using the hill-climbing algorithm to generate a Bayesian network structure as the initial population. S2423. Calculate the fitness value of each Bayesian network structure in the initial population, sort them according to the fitness value, and select several Bayesian network structures with high fitness values ​​as the new population. S2424. Based on the Bayesian network structure of each individual in the new population, recalculate the corresponding fitness value, identify the individuals with the best and worst fitness in the current population, and update the individual positions based on the individuals with the best and worst fitness. S2425. Repeat steps S2421 to S2424 until the preset number of iterations is reached to obtain the optimal Bayesian network structure. S2426. Using the optimal Bayesian network structure, analyze the correlation between emotional fluctuations and physical actions.

[0078] S243. For each time step, combine the emotional features and body movement features in the multimodal time series dataset to establish a comprehensive feature vector.

[0079] S244. Using the isolated forest algorithm, construct an anomaly detection model and train the anomaly detection model using comprehensive feature vectors.

[0080] S245. Calculate the anomaly score for the comprehensive feature vector at each time step using the trained isolated forest model, and identify the characteristics of abnormal emotion diffusion based on the preset anomaly threshold.

[0081] It should be further explained that, using the trained Isolation Forest model, an anomaly score is calculated for the comprehensive feature vector at each time step. This score reflects the degree of deviation of the data at that time step from the normal data. Based on a preset anomaly threshold, abnormal emotion diffusion characteristics are identified, which may include diffusion speed, diffusion range, diffusion intensity, etc. When the anomaly score at a certain time step exceeds the threshold, it is considered that there are abnormal emotion diffusion characteristics at that time step. The identified abnormal emotion diffusion characteristics are analyzed to explore their possible causes and effects.

[0082] S25. Analyze the dynamic correlation between abnormal collaborative fluctuation patterns and abnormal sentiment diffusion characteristics to predict abnormal fluctuations in agricultural markets.

[0083] It should be further explained that, based on the extracted abnormal co-fluctuation patterns and abnormal sentiment diffusion characteristics, a prediction model is constructed. The model can be a multiple regression model, a machine learning model (such as support vector machine, random forest, etc.), or a deep learning model (such as neural network). The input features of the model can include statistics of abnormal co-fluctuation patterns, quantitative indicators of abnormal sentiment diffusion characteristics, etc., and the output of the model can be the predicted value of abnormal fluctuations in agricultural market prices.

[0084] S3. Based on the predicted abnormal fluctuations in the farmers' market, analyze the degree of impact of abnormal fluctuations on the stability of the farmers' market, and construct an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations.

[0085] In this optional embodiment, based on the predicted abnormal fluctuations in the farmers' market, the impact of these abnormal fluctuations on the stability of the farmers' market is analyzed, and an abnormal behavior risk assessment model is constructed. The assessment of the abnormal fluctuation risk includes the following steps: S31. Based on the predicted abnormal fluctuations in the farmers' market, use the local anomaly factor algorithm to calculate the abnormal sentiment index and determine the range of abnormal sentiment distribution.

[0086] It should be further explained that, based on the predicted abnormal fluctuations in the farmers' market, the following is a specific implementation example of using the local anomaly factor algorithm to calculate the abnormal sentiment index and determine the interval of the abnormal sentiment distribution: Step 1: Local Anomaly Factor Algorithm (1) Define distance metric Before calculating the local anomaly factor, define a distance metric between sample points. For example, Euclidean distance can be used to measure the distance between sample points of different emotions.

[0087] (2) Calculate the k-distance and k-distance neighborhood. For each emotion sample point, calculate its distance to the k-th nearest neighbor sample point, i.e., the k-distance; determine the k-distance neighborhood of each emotion sample point, i.e., the region containing the k nearest neighbor sample points.

[0088] (3) Calculate the k-reachable distance For each emotion sample point, calculate its k-reachable distance to every sample point in its k-distance neighborhood; the k-reachable distance is the larger value of the sample point's k-distance and its distance to the sample points in that neighborhood.

[0089] (4) Calculate the locally reachable density Based on the k-reachability distance, calculate the local reachability density of each emotion sample point; the local reachability density is the reciprocal of the sum of the k-reachability distances of all sample points in the k-distance neighborhood, divided by the number of sample points in the k-distance neighborhood.

[0090] (5) Calculate the local anomaly factor For each sentiment sample point, calculate its local anomaly factor; the local anomaly factor is the average of the ratio of the local reachability density of all sample points in its k-distance neighborhood to the local reachability density of that sample point.

[0091] Step 2: Determine the indicators and distribution range of abnormal emotions. (1) Set threshold Based on the calculation results of local abnormal factors, a threshold is set to distinguish between normal emotions and abnormal emotions. Sample points with local abnormal factors greater than a certain threshold are considered abnormal points.

[0092] (2) Calculate abnormal emotion indicators For each emotion sample point, if its local abnormality factor is greater than the set threshold, it is marked as an abnormal emotion. Statistical indicators such as the proportion and number of abnormal emotions can be calculated to quantify the intensity of abnormal emotions.

[0093] (3) Determine the distribution range of abnormal emotions Based on the timestamps of abnormal emotion sample points, the distribution range of abnormal emotions over time can be determined; time series plots or histograms can be drawn to visualize the distribution of abnormal emotions.

[0094] S32. Combining the determined abnormal sentiment distribution range with the conditional value at risk algorithm, analyze the degree of impact of abnormal fluctuations on the stability of farmers' markets.

[0095] In this optional embodiment, the analysis of the impact of abnormal fluctuations on the stability of farmers' markets, combining the determined abnormal sentiment distribution range with the conditional value at risk algorithm, includes the following steps: S321. Based on the determined abnormal emotion distribution interval, construct a risk value model based on different quantiles, and use historical abnormal emotion characteristics to train the risk value model to obtain the risk value under each abnormal emotion distribution interval.

[0096] S322. Combining value at risk and conditional variables, construct a conditional value at risk model, and train the conditional value at risk model using historical abnormal sentiment characteristics to obtain the conditional value at risk for each abnormal sentiment distribution interval.

[0097] In this optional embodiment, the formula for the conditional value-at-risk model is: ; In the formula, CoVaR ( q ) indicates the quantile q Conditional risk value; q Indicates quantiles; α Represents the intercept term; β The coefficient representing the impact of Value at Risk; VaR ( q ) indicates the quantile q The risk value; c This indicates the stability value of the farmers' market; M Value indicators representing the risks of farmers' markets.

[0098] S323. Compare the value of risk and conditional value of risk under different emotional states, assess the impact of emotional changes on the value of risk, and analyze the degree of impact of abnormal fluctuations on the stability of the agricultural market based on the assessment results.

[0099] It should be further explained that, in the specific embodiment, it is assumed that three abnormal emotion distribution intervals are determined: low emotion interval, medium emotion interval, and high emotion interval. Each interval has corresponding emotion characteristic data, such as emotion index and emotion duration. Market risk data of farmers' markets at different time points are collected, such as price fluctuations and changes in transaction volume. The data should match the timestamps of the abnormal emotion characteristic data.

[0100] Step 1: Construct a Value at Risk (VaR) model We choose the 5% and 1% quantiles to define VaR.

[0101] For each abnormal sentiment distribution interval, a regression model is constructed using historical abnormal sentiment feature data and market risk data. The goal of the model is to predict the maximum possible loss at a given confidence level (e.g., 95% or 99%) within a specific future time (e.g., one day). The model is trained using historical data to learn the mapping relationship between abnormal sentiment features and market risk data. The trained model can predict the corresponding VaR value based on different abnormal sentiment distribution intervals.

[0102] Step 2: Construct the Conditional Value at Risk (CoVaR) model. When constructing the CVaR model, VaR is introduced as a condition variable into the model, while other factors that may affect market risk are also considered, such as market trends and policy changes (in this embodiment, these factors are simplified to abnormal sentiment characteristics).

[0103] For each abnormal sentiment distribution interval, a conditional regression model is constructed using historical abnormal sentiment feature data, market risk data, and corresponding VaR values. The goal of the model is to predict the average loss of the portfolio (i.e., CVaR) when the loss exceeds VaR. The model is trained using historical data to learn the complex relationship between abnormal sentiment features, VaR, and market risk data. The trained model can predict the corresponding CVaR value based on different abnormal sentiment distribution intervals.

[0104] Step 3: Comparison and Analysis Using the trained VaR and CVaR models, VaR and CVaR values ​​were calculated for different abnormal emotion distribution intervals.

[0105] By comparing VaR and CVaR values ​​under different abnormal sentiment distribution intervals, the impact of sentiment changes on value at risk is analyzed. As the level of abnormal sentiment increases, VaR and CVaR values ​​also increase accordingly, indicating that market risk increases.

[0106] Based on abnormal market fluctuations (such as price volatility and changes in trading volume), the impact of VaR and CVaR values ​​on market stability under abnormal sentiment distribution intervals is analyzed. If VaR and CVaR values ​​increase significantly in high sentiment intervals, and market volatility is also large, then it can be considered that abnormal fluctuations have a significant impact on the stability of the agricultural market, as shown in Table 1. Table 1: Data Example S33. Based on the degree of impact of abnormal fluctuations on the stability of agricultural markets, construct an abnormal behavior risk assessment model using an ensemble learning algorithm, and assess the risk of abnormal fluctuations through the abnormal behavior risk assessment model.

[0107] It should be further noted that selecting an ensemble learning algorithm suitable for handling classification problems, such as random forest or gradient boosting tree (GBDT), can combine the results of multiple weak classifiers to improve the accuracy and robustness of the model. The ensemble learning model can be trained using the impact of historical abnormal fluctuations on the stability of farmers' markets. In-depth analysis of the impact can be conducted to assess the degree of influence of abnormal fluctuation risks on the stability of farmers' markets.

[0108] S4. The volatility risk assessment results are fed back to the agricultural market management terminal in real time via the cloud, and an adaptive control mechanism is implemented based on the feedback data.

[0109] It should be added that the cloud server transmits the volatility risk assessment results to the farmers' market management terminal in real time. The farmers' market management terminal software displays the volatility risk assessment results in the form of charts, alarm information and other forms, so that managers can intuitively understand the market situation.

[0110] Based on the volatility risk assessment results, the agricultural market management software automatically calculates and generates control instructions. These instructions are displayed to management personnel through the management interface and can be adjusted and confirmed according to their needs. Once confirmed, the control instructions are transmitted via network to the execution equipment at the agricultural market site, such as price adjustment systems and transaction volume control systems, to achieve adaptive control.

[0111] like Figure 2 As shown, according to another embodiment of the present invention, a cloud computing-based agricultural market abnormal fluctuation risk monitoring system is also provided. The system includes a data acquisition and processing module 1, a data analysis and anomaly detection module 2, a fluctuation analysis and risk assessment module 3, and a real-time feedback and adaptive control module 4. The data acquisition and processing module 1, the data analysis and anomaly detection module 2, the fluctuation analysis and risk assessment module 3, and the real-time feedback and adaptive control module 4 are connected in sequence. Data acquisition and processing module 1 is used to collect multimedia data from farmers' markets and process the multimedia data through a cloud computing platform to extract time-series features of crowd emotions and body movements. The data analysis and anomaly detection module 2 is used to integrate the extracted time-series features of crowd emotions and body movements with agricultural market data to establish a multimodal time-series dataset. It also uses deep learning algorithms to detect abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time-series dataset in order to predict abnormal fluctuations in the agricultural market. The volatility analysis and risk assessment module 3 is used to analyze the impact of predicted abnormal fluctuations on the stability of farmers' markets, and to construct an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations. The real-time feedback and adaptive control module 4 is used to feed the fluctuation risk assessment results back to the agricultural market management terminal in real time via the cloud, and to execute an adaptive control mechanism based on the feedback data.

[0112] In summary, by utilizing the technical solution of this invention, and through the collection of multimedia data from farmers' markets and the efficient processing capabilities of cloud computing platforms, the time-series characteristics of crowd emotions and physical movements can be extracted in real time and accurately. This not only provides a rich data foundation for subsequent prediction of abnormal fluctuations but also significantly improves the efficiency and accuracy of data processing. By integrating multiple data sources and constructing a multimodal time-series dataset, a more comprehensive and in-depth understanding of the dynamic changes in farmers' markets can be achieved. Simultaneously, the application of deep learning algorithms can accurately identify abnormal collaborative fluctuation patterns and emotion diffusion characteristics, providing early warnings of abnormal market fluctuations and offering a powerful tool for timely risk prevention. By analyzing abnormal fluctuations and their impact on market stability and constructing an abnormal behavior risk assessment model, not only is the risk level quantified but also a solid basis for formulating scientific risk management strategies. This allows for a more accurate assessment of the risk status of farmers' markets, enabling more effective measures to ensure their stable operation.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring abnormal fluctuation risk in farmers' markets based on cloud computing, characterized in that: The method comprises the following steps: S1. Collect multimedia data from farmers' markets and process them through cloud computing platforms to extract time series features of crowd emotions and body movements. S2. The extracted time series features of crowd emotions and body movements are integrated with the farmers' market data to establish a multimodal time series dataset. The deep learning algorithm is then used to detect the abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time series dataset to predict abnormal fluctuations in the farmers' market. S3. Based on the predicted abnormal fluctuations in the farmers' market, analyze the impact of abnormal fluctuations on the stability of the farmers' market, and build an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations; S4. Feedback the volatility risk assessment results to the farmers' market management end in real time through the cloud, and implement an adaptive control mechanism based on the feedback data; The method of fusing the extracted time series features of crowd emotions and body movements with the farmers' market data to establish a multimodal time series data set, and using a deep learning algorithm to detect abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time series data set to predict abnormal fluctuations in the farmers' market includes the following steps: S21. Collect farmer's market data, combine it with the extracted time series features of crowd emotions and body movements, align them according to timestamps, and establish a multimodal time series dataset; S22. Construct a multimodal deep learning model according to the deep learning algorithm, and train the multimodal deep learning model using the pre-collected historical multimodal time series data set; S23, inputting the established multimodal time series data set into the trained multimodal deep learning model to identify abnormal collaborative fluctuation patterns between different data sources; S24. Use the self-excitation point process algorithm to model the propagation pattern of crowd emotion fluctuations, analyze the correlation between emotion fluctuations and body movements, and combine the isolation forest algorithm to detect abnormal emotion diffusion characteristics; S25. Analyze the dynamic correlation between abnormal co-fluctuation patterns and abnormal sentiment diffusion characteristics to predict abnormal fluctuations in farmers' markets.

2. According to the cloud computing-based method for monitoring abnormal fluctuation risks in farmers' markets according to claim 1, it is characterized in that: The method of collecting multimedia data from the farmers' market, processing the multimedia data through a cloud computing platform, and extracting time series features of crowd emotions and body movements includes the following steps: S11. Using cloud technology to collect multimedia data of the farmers' market, and preprocessing the collected multimedia data, wherein the multimedia data includes video data and audio data; S12. Analyze the facial expressions in the preprocessed video data according to the deep learning algorithm, and integrate the preprocessed audio data to identify the emotional characteristics of the crowd in the farmers' market; S13, using a posture estimation algorithm to extract body movement data of the crowd from the pre-processed video data, and using a pre-built behavior detection model to capture body movement features of the crowd from the body movement data; S14. Through the distributed computing framework of the cloud computing platform, the emotional characteristics and body movement characteristics of the crowd are processed to extract the time series characteristics of the crowd's emotions and body movements.

3. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 2 is characterized in that: The method of analyzing facial expressions in the preprocessed video data according to the deep learning algorithm and fusing the preprocessed audio data to identify the emotional characteristics of people in the farmers' market includes the following steps: S121. Constructing a facial expression recognition model according to a deep learning algorithm, and training the facial expression recognition model using historical multimedia data; S122, inputting the pre-processed video data into the trained facial expression recognition model, and outputting the expression and emotion features corresponding to each face; S123, using a long short-term memory network algorithm to analyze the preprocessed audio data and output corresponding audio emotion features; S124. Based on a weighted algorithm, the facial expression emotion features and the audio emotion features are integrated to identify the emotion characteristics of people in the farmers' market.

4. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 1, characterized in that: The method of using the self-excitation point process algorithm to model the propagation mode of crowd emotional fluctuations, analyzing the correlation between emotional fluctuations and body movements, and combining the isolation forest algorithm to detect abnormal emotional diffusion characteristics includes the following steps: S241. Based on the multimodal time series data set, the emotion propagation function is calculated using the self-excitation point process algorithm, and the propagation pattern of the emotional fluctuation of the crowd is modeled through the emotion propagation function; S242. Based on the crowd emotion fluctuation pattern after modeling, the Bayesian network algorithm is used to analyze the correlation between emotion fluctuation and body movements in the multimodal time series data set; S243, for each time step, combining the emotion features and body movement features in the multimodal time series data set to establish a comprehensive feature vector; S244, using the isolation forest algorithm, constructing an anomaly detection model, and using the comprehensive feature vector to train the anomaly detection model; S245. Calculate the anomaly score for the comprehensive feature vector of each time step using the trained isolation forest model, and identify abnormal emotion diffusion characteristics based on a preset anomaly threshold.

5. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 4 is characterized in that: The formula of the emotion propagation function is: ； In the formula, η ( t ) indicates the time t The intensity of emotional fluctuations; t Indicates the current emotional fluctuation time; ρ Indicates basic strength; t i Indicates the time of emotional fluctuations in the past; σ Indicates the trigger coefficient; λ Represents the decay rate.

6. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 5, characterized in that: The method of analyzing the correlation between emotional fluctuations and body movements in a multimodal time series data set using a Bayesian network algorithm according to the emotional fluctuation pattern of the crowd after modeling is completed includes the following steps: S2421. According to the emotion fluctuation pattern of the crowd after modeling, select emotion parameters and body movement parameters from the multimodal time series data set as nodes for constructing the Bayesian network structure; S2422, constructing an initial Bayesian network structure using a maximum spanning tree algorithm, and processing the initial Bayesian network structure using a hill climbing algorithm to generate a Bayesian network structure as an initial population; S2423, calculating the fitness value of each Bayesian network structure in the initial population, sorting them according to the fitness value, and selecting several Bayesian network structures with high fitness values ​​as new populations; S2424, based on each Bayesian network structure in the new population, recalculate the corresponding fitness value, identify the individuals with the best and worst fitness in the current population, and update the individual positions based on the individuals with the best and worst fitness; S2425, repeatedly executing steps S2421 to S2424 until a preset number of iterations is reached to obtain an optimal Bayesian network structure; S2426. Use the optimal Bayesian network structure to analyze the correlation between emotional fluctuations and body movements.

7. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 1, characterized in that: Based on the predicted abnormal fluctuations of the farmers' market, the influence of the abnormal fluctuations on the stability of the farmers' market is analyzed, and an abnormal behavior risk assessment model is constructed. The assessment of the abnormal fluctuation risk includes the following steps: S31. Based on the predicted abnormal fluctuations of the farmers' market, the abnormal sentiment index is calculated using the local abnormal factor algorithm, and the interval of abnormal sentiment distribution is determined; S32. Combine the determined abnormal sentiment distribution interval with the conditional value at risk algorithm to analyze the impact of abnormal fluctuations on the stability of the farmers' market; S33. According to the impact of abnormal fluctuations on the stability of farmers' markets, an abnormal behavior risk assessment model is constructed using an integrated learning algorithm, and the abnormal fluctuation risk is assessed through the abnormal behavior risk assessment model.

8. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 7, characterized in that: The analysis of the influence of abnormal fluctuations on the stability of the farmers' market by combining the determined abnormal sentiment distribution interval with the conditional value at risk algorithm includes the following steps: S321. According to the determined abnormal emotion distribution interval, construct a risk value model based on different quantiles, and train the risk value model using historical abnormal emotion characteristics to obtain the risk value under each abnormal emotion distribution interval; S322. Combining the risk value and the conditional variable, constructing a conditional risk value model, and using the historical abnormal sentiment characteristics to train the conditional risk value model to obtain the conditional risk value under each abnormal sentiment distribution interval; S323. Compare the risk value and conditional risk value under different emotional states, evaluate the impact of emotional changes on the risk value, and analyze the impact of abnormal fluctuations on the stability of the farmers' market based on the evaluation results.

9. The method for monitoring abnormal fluctuation risk of farmers' markets based on cloud computing according to claim 8, characterized in that: The formula of the conditional value-at-risk model is: ； In the formula, CoVaR ( q ) indicates the quantile q The conditional value at risk of q represents quantile; α represents the intercept term; β The impact coefficient representing the risk value; VaR ( q ) indicates the quantile q Value at risk; γ Indicates the stability value of the farmers' market; M A value indicator representing the risk of farmers' markets.

10. A cloud computing-based farmer's market abnormal fluctuation risk monitoring system, used to implement the cloud computing-based farmer's market abnormal fluctuation risk monitoring method described in any one of claims 1 to 9, characterized in that: The system includes a data acquisition and processing module, a data analysis and anomaly detection module, a fluctuation analysis and risk assessment module, and a real-time feedback and adaptive control module; The data acquisition and processing module, the data analysis and anomaly detection module, the fluctuation analysis and risk assessment module, and the real-time feedback and adaptive control module are connected in sequence; The data collection and processing module is used to collect multimedia data from the farmers' market and process the multimedia data through the cloud computing platform to extract the time series characteristics of crowd emotions and body movements; The data analysis and anomaly detection module is used to integrate the extracted time series features of crowd emotions and body movements with the farmers' market data to establish a multimodal time series data set, and use a deep learning algorithm to detect abnormal collaborative fluctuation patterns and abnormal emotion diffusion characteristics in the multimodal time series data set to predict abnormal fluctuations in the farmers' market; The fluctuation analysis and risk assessment module is used to analyze the impact of abnormal fluctuations on the stability of the farmers' market based on the predicted abnormal fluctuations of the farmers' market, and to build an abnormal behavior risk assessment model to assess the risk of abnormal fluctuations; The real-time feedback and adaptive control module is used to feed back the volatility risk assessment results to the farmers' market management end in real time through the cloud, and execute an adaptive control mechanism based on the feedback data.

Images