Received data analysis and judgment system for data interface

Abstract

The invention relates to the technical field of data analysis, and discloses a received data analysis and judgment system for a data interface, which comprises a data receiving module, a data preprocessing module, a reinforcement learning analysis module and a result output module. The data receiving module is responsible for receiving data from different data sources; the data preprocessing module is used for cleaning, coding and standardizing the data; the reinforcement learning analysis module constructs a data analysis and judgment model based on reinforcement learning, adapts to data changes through a dynamic adjustment strategy, and improves the accuracy and timeliness of analysis; and the result output module is responsible for outputting a judgment result of the model. According to the method, the data analysis strategy is continuously optimized through the reinforcement learning algorithm, the defects of a traditional system in a complex data environment are effectively overcome, the intelligence and automation level of data processing is improved, and powerful support is provided for enterprise decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a receiving data analysis and judgment system for a data interface. Background Art

[0002] In the era of rapid development of informatization, data has become an indispensable resource for corporate decision-making, scientific research and daily life. With the widespread application of technologies such as the Internet of Things, cloud computing, and big data, the importance of data interfaces as a bridge between different data sources and data processing systems has become increasingly prominent. However, in the face of massive, heterogeneous, and high-speed data, how to efficiently and accurately receive and analyze this data has become an urgent problem to be solved.

[0003] Current data reception and analysis systems often rely on fixed rules and preset algorithms to process data. This static processing method is powerless in the face of complex and changing data environments. On the one hand, it is difficult to adapt to the diversity of data source types and the complexity of data formats, resulting in frequent format mismatches, data loss, or incorrect parsing during data reception. On the other hand, traditional analysis algorithms are often based on manually designed feature extraction and pattern recognition methods, which lack adaptability to unknown or newly emerging data patterns and make it difficult to ensure the accuracy and timeliness of analysis. Summary of the invention

[0004] The object of the present invention is to provide a received data analysis and judgment system for a data interface to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a data receiving data analysis and judgment system for a data interface, the system comprising: A data receiving module, used for receiving data transmitted from data interfaces of different data sources; Data preprocessing module, used to clean, encode and standardize the received data; The reinforcement learning analysis module is used to build a data analysis and judgment model based on reinforcement learning. The module models the data processing process received by the data interface as a sequence decision problem. By analyzing the characteristics, patterns and historical data of the received data, the data analysis strategy and judgment criteria are dynamically adjusted. The steps of building a data analysis and judgment model include: Define the state space, including the data type, data format, data integrity, data timeliness, data quality dimensions, and historical data analysis results received by the data interface; Define the action space, which contains the actions that the model may take in the process of analyzing data, including data preprocessing, feature extraction, pattern recognition, anomaly detection, data classification, data prediction, and triggering early warning mechanisms; Define a reward function to evaluate the results of actions based on the accuracy of data analysis, processing efficiency, model stability, false positive rate, and false negative rate, and give corresponding rewards or penalties; Formulate strategies, select the best actions based on the current state to maximize long-term cumulative rewards, and continuously iterate and optimize strategies through reinforcement learning algorithms; The result output module is used to pass the preprocessed data into the data analysis and judgment model, receive the judgment results of the model and output them.

[0006] Preferably, the data types include structured data, unstructured data and semi-structured data; the data formats include text format, numerical format, image format, audio format and video format; the data integrity includes data missing, redundancy and correlation between data; the data timeliness includes data receiving time, processing time and data updating frequency; the data quality dimensions include data accuracy, consistency, credibility and availability.

[0007] Preferably, the components of the reward function include accuracy reward, processing efficiency reward, stability reward, false positive penalty and false negative penalty: The calculation formula of the accuracy reward function is: R_accuracy = k1 * (number of correct judgments / total number of judgments); where k1 is the accuracy reward coefficient; The calculation formula of the processing efficiency reward function is: R_efficiency = k2 * (1 / average processing time); where k2 is the processing efficiency reward coefficient; The calculation formula of the stability reward function is: R_stability = k3 * (model stability index); where k3 is the stability reward coefficient, and the model stability index is obtained by calculating the performance fluctuation of the model on different data sets; The calculation formula of the false alarm penalty function is: P_false_alarm = k4 * (number of false alarms / total number of judgments); where k4 is the false alarm penalty coefficient; The calculation formula of the missed alarm penalty function is: P_miss_alarm = k5 * (number of missed alarms / actual number of abnormalities); where k5 is the missed alarm penalty coefficient; Combining the above reward and penalty functions, we get the final reward function R: R = R_accuracy + R_efficiency + R_stability - P_false_alarm - P_miss_alarm.

[0008] Preferably, the method for obtaining training data for a data analysis and judgment model includes: receiving actual transmission data from multiple data sources in real time through a data interface; preprocessing and extracting features from the actual data to obtain feature representations of the data; performing simulated analysis and judgment actions on the extracted features in combination with preset analysis rules and judgment criteria, and calculating corresponding immediate rewards; combining the feature representation of the data, the simulated analysis and judgment actions, and the immediate rewards into a feature-action-reward tuple as preliminary training data for the data analysis and judgment model.

[0009] Preferably, historical data records are obtained from a data warehouse or historical data storage, including past data interface transmission data, manually annotated data, and data of known classification or anomalies; the historical data records are subjected to the same preprocessing and feature extraction to obtain a feature representation of the historical data; according to preset analysis rules and judgment criteria, possible analysis and judgment actions are simulated for the feature representation of the historical data, and the corresponding immediate rewards are calculated; the feature representation of the historical data, the simulated analysis and judgment actions, and the immediate rewards are combined into additional feature-action-reward tuples, which are added to the preliminary training data to form a complete data analysis and judgment model training data set.

[0010] Preferably, the reinforcement learning analysis module uses the Deep Q-Network algorithm in the deep reinforcement learning algorithm to train the data analysis and judgment model.

[0011] Preferably, the step of training the data analysis and judgment model includes: S1: Initialize the experience replay memory and the target neural network. The experience replay memory is used to store the experience of data features and analysis and judgment results, and the target neural network is used to provide a stable target value for the training process. At the same time, initialize the main neural network, whose input is the feature representation of the data and output is the Q value estimation of each analysis and judgment action. S2: Set the learning rate α, discount factor γ, exploration probability ε, and training iteration number N. The learning rate α is used to control the learning speed of the model, the discount factor γ is used to calculate the current value of future rewards, the exploration probability ε is used to balance the exploration and utilization behavior of the model, and the training iteration number N determines the overall rounds of training; S3: For each training iteration, perform the following steps: S301: randomly extract a batch of data samples from the training data set, and obtain the current feature representation s of the data through preprocessing and feature extraction; S302: Select an analysis and judgment action a according to the current feature representation s and the ε-greedy strategy; specifically, randomly select an action with a probability of ε, and select the action with the largest Q value output by the current main neural network with a probability of 1-ε; S303: Execute action a, calculate the immediate reward r according to the preset analysis rules and judgment criteria, and observe the next feature representation s'; S304: storing the experience tuple (s, a, r, s') into the experience playback memory; S305: randomly sampling a batch of experience tuples from the experience replay memory for updating the main neural network; S306: For each sampled experience tuple, the target Q value is calculated using the target neural network. The target Q value is calculated as follows: Q_target = r + γ * max_a' Q'(s', a'), where Q' represents the Q value estimate of the target neural network. Then, the parameters of the main neural network are updated according to the loss function of the Deep Q-Network algorithm. S307: Copying the parameters of the main neural network to the target neural network at every preset number of steps; S4: Repeat step S3 until the preset number of training iterations N is reached or the output value of the main neural network converges to a stable state; S5: After the training is completed, the trained main neural network is used as the data analysis and judgment model; for any given data feature representation s, the Q value of each action is calculated through the forward propagation of the main neural network, and the action a with the largest Q value is selected as the optimal analysis and judgment action.

[0012] Preferably, the data preprocessing module performs encoding processing on the data, including: performing one-hot encoding conversion on discrete features.

[0013] Preferably, the data preprocessing module performs standardization processing on the data, including: The Z-score standardization method is used to standardize continuous features. Specifically, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data.

[0014] Preferably, the data preprocessing module cleans the data by: For missing values, linear interpolation algorithm is used to fill them; For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

[0015] Compared with the prior art, the present invention has the following beneficial effects: By introducing the reinforcement learning analysis module, this system can automatically learn and adapt to changes in the types, formats and data characteristics of different data sources, avoiding the processing limitations of traditional systems due to reliance on fixed rules and preset algorithms, allowing the system to receive and analyze massive, heterogeneous and high-speed data more flexibly and efficiently.

[0016] The reinforcement learning analysis module models the data processing process received by the data interface as a sequential decision problem. By dynamically adjusting the data analysis strategy and judgment criteria, it can more accurately capture the characteristics and patterns in the data. At the same time, the system can continuously optimize the analysis strategy based on real-time feedback, ensure the timeliness and accuracy of data analysis, and provide more reliable data support for decision-making.

[0017] By defining the reward function, the system can comprehensively consider multiple dimensions such as data analysis accuracy, processing efficiency, model stability, and false positive and false negative rates, evaluate the analysis actions, and give corresponding rewards or penalties. This mechanism prompts the system to continuously optimize its analysis strategy, thereby reducing the risk of false positives and false negatives and improving the reliability of data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A structural diagram of a data receiving and analyzing system for a data interface according to the present invention; Figure 2 A schematic diagram of the design of the reward function in the data analysis judgment model; Figure 3 This is a training flowchart for the data analysis and judgment model. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1-3 The present invention provides a technical solution: a data interface receiving data analysis and judgment system, the system comprising: Data receiving module: The data receiving module is responsible for receiving data transmitted from data interfaces of different data sources. The module is designed with a flexible data interface that is compatible with a variety of data formats and protocols to ensure smooth data reception. At the same time, the module has a built-in data caching mechanism to cope with the receiving pressure during data peak periods and ensure that data is not lost.

[0021] Data preprocessing module: The data preprocessing module cleans, encodes and standardizes the received data. The cleaning process includes removing invalid data, filling missing values, correcting erroneous data, etc. Encoding converts non-numeric data into numeric data for subsequent analysis; standardization unifies the dimension and distribution of data and improves the generalization ability of the analysis model.

[0022] Reinforcement learning analysis module: The reinforcement learning analysis module is the core part of this system, responsible for building a data analysis and judgment model based on reinforcement learning. The specific implementation steps are as follows: Define the state space, which includes the data type, data format, data integrity, data timeliness, data quality dimension and historical data analysis results received by the data interface. These state information provides a comprehensive data context for the model, which helps the model make accurate judgments. The data types include structured data, unstructured data and semi-structured data; the data formats include text format, numerical format, image format, audio format and video format; the data integrity includes data missing, redundancy and correlation between data; the data timeliness includes data receiving time, processing time and data update frequency; the data quality dimensions include data accuracy, consistency, credibility and availability.

[0023] Define the action space, which contains the actions that the model may take in the process of analyzing data, such as data preprocessing, feature extraction, pattern recognition, anomaly detection, data classification, data prediction, and triggering early warning mechanisms. These actions form the basis of model decision-making.

[0024] Define a reward function, which evaluates the results of actions based on indicators such as data analysis accuracy, processing efficiency, model stability, false positive rate, and false negative rate, and gives corresponding rewards or penalties. Through the design of the reward function, the model is guided to continuously evolve towards the optimization goal.

[0025] Formulating strategies and selecting the best actions based on the current state to maximize the long-term cumulative rewards is the core of the reinforcement learning algorithm. This system uses a deep reinforcement learning algorithm to continuously iterate and optimize strategies so that the model gradually adapts to complex and changing data environments.

[0026] Model training uses historical data and simulation data to train the reinforcement learning model, allowing it to learn and optimize decision-making strategies through continuous trial and error. During the training process, the model performance is gradually improved by adjusting the model parameters and reward function.

[0027] Result output module: The result output module is responsible for transferring the pre-processed data to the data analysis and judgment model and receiving the judgment results of the model. This module can output the results in various forms such as visual reports, real-time alarms, and data interface calls according to actual needs to meet application requirements in different scenarios.

[0028] The present invention will be further described below in conjunction with Examples 1 to 3: Example 1

[0029] The reward function is a key component in the reinforcement learning algorithm, which guides the model to learn and optimize its behavior strategy during the exploration process. The following is a detailed description of the specific implementation of the reward function: The reward function consists of: Accuracy reward (R_accuracy): The accuracy reward is used to measure the correctness of the model's judgment. Its calculation formula is: R_accuracy = k1 * (number of correct judgments / total number of judgments); where k1 is the accuracy reward coefficient, which is an adjustable parameter used to adjust the proportion of the accuracy reward in the total reward. The number of correct judgments refers to the number of data categories or anomalies correctly identified by the model, and the total number of judgments is the total number of all judgments made by the model.

[0030] Processing efficiency bonus (R_efficiency): The processing efficiency reward is used to evaluate the speed at which the model processes data. Its calculation formula is: R_efficiency = k2* (1 / average processing time); where k2 is the processing efficiency reward coefficient, which is also an adjustable parameter. The average processing time refers to the average time required for the model to process a single data sample, which is calculated by measuring the total amount and time consumed by the model over a period of time.

[0031] Stability reward (R_stability): The stability reward is used to reflect the performance stability of the model on different data sets. Its calculation formula is: R_stability = k3 * (model stability index); where k3 is the stability reward coefficient. The model stability index is obtained by calculating the variance or standard deviation of the performance fluctuation (such as accuracy, F1 score, etc.) of the model on different data sets (such as training set, validation set, test set). The lower the value, the more stable the model.

[0032] False alarm penalty (P_false_alarm): The false alarm penalty is used to punish the model for incorrectly marking normal data as abnormal. Its calculation formula is: P_false_alarm = k4 * (number of false alarms / total number of judgments); where k4 is the false alarm penalty coefficient. The number of false alarms refers to the number of normal data that the model incorrectly judges as abnormal.

[0033] Miss alarm penalty (P_miss_alarm): The miss-alarm penalty is used to penalize the model for failing to identify actual abnormal data. The calculation formula is: P_miss_alarm = k5 * (miss-alarm number / actual abnormal number); where k5 is the miss-alarm penalty coefficient. The miss-alarm number refers to the number of actual abnormal data that the model failed to identify, and the actual abnormal number is the total number of abnormal data obtained through manual annotation or other reliable methods.

[0034] Comprehensive reward function: Combining the above reward and penalty functions, we get the final reward function R: R = R_accuracy + R_efficiency + R_stability - P_false_alarm - P_miss_alarm.

[0035] This reward function comprehensively considers the accuracy, processing efficiency, stability, and false positives and negatives of the model's judgment. By adjusting the values ​​of each reward and penalty coefficient, the model can be guided to continuously optimize its behavioral strategy during the exploration process to achieve better data analysis and judgment results.

[0036] Assume that in a data interface receiving data analysis and judgment system, the model processes 1000 data samples in a certain iteration, of which 900 are correctly judged, 50 normal data are falsely reported as abnormal, and 20 actual abnormal data are missed. The average processing time is 0.01 seconds, and the model stability index is 0.05 (assuming that the lower the value, the more stable it is). Set the reward and penalty coefficients as: k1=1.0, k2=0.5, k3=0.2, k4 =0.8, k5 =1.2.

[0037] Calculate the values ​​of each reward and penalty function according to the above formula: R_accuracy=1.0×(900 / 1000)=0.9 R_efficiency=0.5×(1 / 0.01)=50 R_stability=0.2×0.05=0.01 P_false_alarm=0.8×(50 / 1000)=0.04 P_miss_alarm=1.2×(20 / (20+900−50))=0.026 (Note that the actual number of anomalies here should exclude the number of false alarms) The final reward function value is: R=0.9+50+0.01−0.04−0.026=50.844 This value will be used to update the model's strategy to further optimize model performance in subsequent iterations. Example 2

[0038] In order to achieve effective training of the data analysis and judgment model, this embodiment provides a systematic method to collect and construct a training data set, which combines real-time data and historical data to ensure the comprehensiveness and diversity of the training data. The specific implementation steps are as follows: Step 1: Real-time data acquisition and processing Step 1.1: Configure the data interface to receive real-time transmission data from different data sources. These data sources include but are not limited to sensor data, user behavior logs, market transaction records, etc. The data interface should support multiple data formats and transmission protocols to ensure the timeliness and integrity of the data.

[0039] Step 1.2: Preprocess the actual data received. The preprocessing process includes data cleaning (removing invalid, duplicate or erroneous data), data normalization (converting data of different dimensions to a unified scale) and timestamp calibration to ensure the consistency and comparability of the data.

[0040] Step 1.3: Perform feature extraction. According to the goal of data analysis, extract key features from the preprocessed data. Features can include statistics (such as mean, standard deviation), time series patterns, text keywords, etc. These features can reflect the inherent laws and characteristics of the data.

[0041] Step 1.4: Combine the preset analysis rules and judgment criteria to perform simulated analysis and judgment actions on the extracted features. For example, in a fraud detection scenario, the rules may include "if the transaction amount exceeds twice the user's average spending level and occurs at an unconventional time, mark it as suspicious." Based on the simulation results, calculate the instant reward for each action, and the reward value can be set based on factors such as the accuracy and timeliness of the judgment.

[0042] Step 1.5: Combine the feature representation of the data, the simulated analysis and judgment actions, and the immediate rewards into feature-action-reward tuples and store them as preliminary training data.

[0043] Step 2: Historical data integration and utilization Step 2.1: Retrieve historical data records from the data warehouse or historical data storage. These data should include past data interface transmission data, manually annotated data (such as expert judgment results), known classified or abnormal data, etc., to ensure the richness and representativeness of the training data.

[0044] Step 2.2: Perform the same preprocessing and feature extraction process on the historical data records as the real-time data to obtain the feature representation of the historical data.

[0045] Step 2.3: According to the preset analysis rules and judgment criteria, simulate the analysis and judgment of the feature representation of the historical data. Since historical data usually contains known results (such as actual classification labels), the accuracy of the simulated judgment can be evaluated more accurately, and the calculation method of the immediate reward can be adjusted accordingly.

[0046] Step 2.4: Combine the feature representation of historical data, simulated analysis and judgment actions, and immediate rewards into additional feature-action-reward tuples. These tuples are added to the preliminary training data to form a complete data analysis and judgment model training dataset.

[0047] Through iterative training, the model will learn the mapping relationship from data features to analysis and judgment actions. Example 3

[0048] In order to achieve effective training of the data analysis and judgment model based on the Deep Q-Network (DQN) algorithm, this implementation method elaborates on the various steps of model training and its specific implementation method. The following is the specific implementation process: S1: Initialization phase Initialize the experience replay memory, which has a large enough capacity to store the experience tuples of data features and analysis and judgment results, so as to perform experience replay during the training process and improve training efficiency.

[0049] Initialize the target neural network, which has the same structure as the main neural network but with a lower parameter update frequency, to provide a stable target value for the training process and reduce fluctuations during training.

[0050] Initialize the main neural network, whose input layer receives the feature representation of the data, and the output layer is the Q-value estimate of each analysis and judgment action, that is, the value of each possible action is evaluated.

[0051] S2: Parameter setting The learning rate α is set to control the step size of model parameter updates to ensure that the model can learn quickly while avoiding overfitting.

[0052] The discount factor γ is set to calculate the value of future rewards at the current moment, reflecting the importance the model places on future rewards.

[0053] The exploration probability ε is set to balance the model's exploration (trying new actions) and exploitation (selecting known high-value actions) behaviors, ensuring that the model can fully explore the environment and learn the optimal strategy.

[0054] Set the number of training iterations N to determine the overall rounds of training and ensure that the model has enough time to learn and optimize.

[0055] S3: Training iteration process For each training iteration, the following steps are performed: S301: Randomly extract a batch of data samples from the training data set, and obtain the current feature representation s of the data through preprocessing and feature extraction. Preprocessing includes data cleaning and normalization, and feature extraction selects relevant features according to specific tasks.

[0056] S302: Select an analysis and judgment action a based on the current feature representation s and the ε-greedy strategy. Specifically, randomly select an action with a probability of ε to explore new possibilities; select the action with the largest Q value output by the current main neural network with a probability of 1-ε to utilize known high-value actions.

[0057] S303: Execute action a, calculate the immediate reward r according to the preset analysis rules and judgment criteria, and observe the next feature representation s' after executing the action. The immediate reward r is determined according to the accuracy and effect of the action, reflecting the quality of the action.

[0058] S304: Store the experience tuple (s, a, r, s') in the experience replay memory for subsequent experience replay.

[0059] S305: Randomly sample a batch of experience tuples from the experience replay memory for updating the main neural network. Experience replay can break the temporal association between data and improve training stability.

[0060] S306: For each sampled experience tuple, the target Q value is calculated using the target neural network. The target Q value is calculated as follows: Q_target = r + γ * max_a' Q'(s', a'), where Q' represents the Q value estimate of the target neural network. Then, the parameters of the main neural network are updated according to the loss function (such as mean square error) of the DQN algorithm to make the Q value estimate of the main neural network more accurate.

[0061] S307: Every preset number of steps, the parameters of the main neural network are copied to the target neural network to maintain the stability of the target neural network.

[0062] S4: Training termination condition Repeat step S3 until the preset number of training iterations N is reached or the output value of the main neural network (i.e., Q value) converges to a stable state, indicating that the model has learned the optimal strategy.

[0063] S5: Model application After the training is completed, the trained main neural network is used as the data analysis and judgment model. For any given data feature representation s, the Q value of each action is calculated through the forward propagation of the main neural network, and the action a with the largest Q value is selected as the optimal analysis and judgment action. In this way, the model can automatically make the best analysis and judgment based on the input data features.

[0064] In order to achieve efficient data preprocessing and provide accurate and reliable input for the data analysis and judgment model, this embodiment is used to describe the specific implementation method of encoding processing, standardization processing and data cleaning in the data preprocessing module. The following are the specific implementation steps: For discrete features, one-hot encoding is used for conversion. Taking the user gender feature as an example, the gender in the original data may be expressed in text form such as "male" and "female". In the one-hot encoding process, a unique integer index is first assigned to each gender, such as "male" corresponds to 0 and "female" corresponds to 1. Then, a binary vector equal to the number of gender types is created. For each sample, the index position corresponding to its gender is set to 1, and the rest of the positions are set to 0. For example, samples with gender "male" will be converted to [1, 0], and samples with gender "female" will be converted to [0, 1].

[0065] For continuous features, the Z-score standardization method is used to ensure that the data conforms to the standard normal distribution, thereby eliminating the impact of different dimensions on model training. Taking the user's age feature as an example, assume that the mean value of age in the original data set is μ and the standard deviation is σ. For each age data x, its standardized value z will be calculated according to the following formula: z = (x - μ) / σ Among them, x is the original age data, μ is the mean age, σ is the standard deviation of age, and z is the standardized age data. Through this processing, all age data will be converted into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0066] Data cleaning includes: For missing values ​​in the data set, linear interpolation algorithm is used to fill them. Taking time series data as an example, if data is missing at a certain time point, linear interpolation will be performed using two valid data points before and after the time point. Assume that the two data points before and after are (t1, y1) and (t2, y2), and the time of the missing data point is t, then the interpolated data y will be calculated according to the following formula: y = y1 + (y2 - y1) * (t - t1) / (t2 - t1).

[0067] The threshold judgment method is used to handle outliers. First, a reasonable threshold range is set according to the statistical characteristics of the data or the business logic. Then, each data point in the data set is traversed to determine whether it exceeds the threshold range. For data points that exceed the threshold, they are regarded as outliers and replaced with the mean of their neighboring points. Taking the temperature sensor data as an example, assuming that the normal temperature range is set to -20℃ to 50℃, if the temperature value of a data point exceeds this range, it is regarded as an outlier and replaced with the mean of the two normal temperature values ​​before and after it.

[0068] It should be noted that, in this article, relational terms such as first and second, etc. 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.

[0069] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A data receiving data analysis and judgment system for a data interface, characterized in that: The system comprises: A data receiving module, used for receiving data transmitted from data interfaces of different data sources; Data preprocessing module, used to clean, encode and standardize the received data; The reinforcement learning analysis module is used to build a data analysis and judgment model based on reinforcement learning. The module models the data processing process received by the data interface as a sequence decision problem. By analyzing the characteristics, patterns and historical data of the received data, the data analysis strategy and judgment criteria are dynamically adjusted. The steps of building a data analysis and judgment model include: Define the state space, including the data type, data format, data integrity, data timeliness, data quality dimensions, and historical data analysis results received by the data interface; Define the action space, which contains the actions that the model may take in the process of analyzing data, including data preprocessing, feature extraction, pattern recognition, anomaly detection, data classification, data prediction, and triggering early warning mechanisms; Define a reward function to evaluate the results of actions based on the accuracy of data analysis, processing efficiency, model stability, false positive rate, and false negative rate, and give corresponding rewards or penalties; Formulate strategies, select the best actions based on the current state to maximize long-term cumulative rewards, and continuously iterate and optimize strategies through reinforcement learning algorithms; The result output module is used to pass the preprocessed data into the data analysis and judgment model, receive the judgment results of the model and output them.

2. The data receiving data analysis and judgment system for a data interface according to claim 1, characterized in that: The data types include structured data, unstructured data and semi-structured data; the data formats include text format, numerical format, image format, audio format and video format; the data integrity includes data missing, redundancy and correlation between data; the data timeliness includes data receiving time, processing time and data updating frequency; the data quality dimensions include data accuracy, consistency, credibility and availability.

3. A data receiving interface data analysis and judgment system according to claim 1, characterized in that: The components of the reward function include accuracy reward, processing efficiency reward, stability reward, false positive penalty and false negative penalty: The calculation formula of the accuracy reward function is: R_accuracy = k1 * (number of correct judgments / total number of judgments); where k1 is the accuracy reward coefficient; The calculation formula of the processing efficiency reward function is: R_efficiency = k2 * (1 / average processing time); where k2 is the processing efficiency reward coefficient; The calculation formula of the stability reward function is: R_stability = k3 * (model stability index); where k3 is the stability reward coefficient, and the model stability index is obtained by calculating the performance fluctuation of the model on different data sets; The calculation formula of the false alarm penalty function is: P_false_alarm = k4 * (number of false alarms / total number of judgments); where k4 is the false alarm penalty coefficient; The calculation formula of the missed alarm penalty function is: P_miss_alarm = k5 * (number of missed alarms / actual number of abnormalities); where k5 is the missed alarm penalty coefficient; Combining the above reward and penalty functions, we get the final reward function R: R = R_accuracy + R_efficiency + R_stability - P_false_alarm - P_miss_alarm.

4. A data interface receiving data analysis and judgment system according to claim 3, characterized in that: The method for obtaining training data for a data analysis and judgment model includes: receiving actual transmission data from multiple data sources in real time through a data interface; preprocessing and extracting features from the actual data to obtain feature representations of the data; performing simulated analysis and judgment actions on the extracted features in combination with preset analysis rules and judgment criteria, and calculating corresponding immediate rewards; combining the feature representations of the data, simulated analysis and judgment actions, and immediate rewards into feature-action-reward tuples as preliminary training data for the data analysis and judgment model.

5. A data interface receiving data analysis and judgment system according to claim 4, characterized in that: Obtain historical data records from a data warehouse or historical data storage, including past data interface transmission data, manually annotated data, and data of known classification or anomalies; perform the same preprocessing and feature extraction on the historical data records to obtain feature representations of the historical data; simulate possible analysis and judgment actions for the feature representations of the historical data according to preset analysis rules and judgment criteria, and calculate the corresponding immediate rewards; combine the feature representations of the historical data, the simulated analysis and judgment actions, and the immediate rewards into additional feature-action-reward tuples, and add them to the preliminary training data to form a complete data analysis and judgment model training data set.

6. The received data analysis and judgment system for a data interface according to claim 1, characterized in that: The reinforcement learning analysis module uses the Deep Q-Network algorithm in the deep reinforcement learning algorithm to train the data analysis and judgment model.

7. A data interface receiving data analysis and judgment system according to claim 6, characterized in that: The steps of training the data analysis and judgment model include: S1: Initialize the experience replay memory and the target neural network. The experience replay memory is used to store the experience of data features and analysis and judgment results, and the target neural network is used to provide a stable target value for the training process. At the same time, initialize the main neural network, whose input is the feature representation of the data and output is the Q value estimation of each analysis and judgment action. S2: Set the learning rate α, discount factor γ, exploration probability ε, and number of training iterations N; the learning rate α is used to control the learning speed of the model, the discount factor γ is used to calculate the current value of future rewards, the exploration probability ε is used to balance the exploration and utilization behavior of the model, and the number of training iterations N determines the overall rounds of training; S3: For each training iteration, perform the following steps: S301: randomly extract a batch of data samples from the training data set, and obtain the current feature representation s of the data through preprocessing and feature extraction; S302: Select an analysis and judgment action a according to the current feature representation s and the ε-greedy strategy; specifically, randomly select an action with a probability of ε, and select the action with the largest Q value output by the current main neural network with a probability of 1-ε; S303: Execute action a, calculate the immediate reward r according to the preset analysis rules and judgment criteria, and observe the next feature representation s'; S304: storing the experience tuple (s, a, r, s') into the experience playback memory; S305: randomly sampling a batch of experience tuples from the experience replay memory for updating the main neural network; S306: For each sampled experience tuple, the target Q value is calculated using the target neural network; the target Q value is calculated using the formula: Q_target = r + γ * max_a' Q'(s', a'), where Q' represents the Q value estimate of the target neural network, and then the parameters of the main neural network are updated according to the loss function of the Deep Q-Network algorithm; S307: Copying the parameters of the main neural network to the target neural network at every preset number of steps; S4: Repeat step S3 until the preset number of training iterations N is reached or the output value of the main neural network converges to a stable state; S5: After the training is completed, the trained main neural network is used as the data analysis and judgment model; for any given data feature representation s, the Q value of each action is calculated through the forward propagation of the main neural network, and the action a with the largest Q value is selected as the optimal analysis and judgment action.

8. The received data analysis and judgment system for a data interface according to claim 1, characterized in that: The data preprocessing module performs encoding processing on the data, including: performing one-hot encoding conversion on discrete features.

9. The data receiving data analysis and judgment system for a data interface according to claim 1, characterized in that: The data preprocessing module performs standardized processing on the data, including: The Z-score standardization method is used to standardize continuous features. Specifically, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data.

10. The received data analysis and judgment system for a data interface according to claim 1, characterized in that: The data preprocessing module cleans the data in the following manner: For missing values, linear interpolation algorithm is used to fill them; For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

Images