Data processing method and system based on multi-source information system integration

Abstract

The invention relates to a data processing method and system based on multi-source information system integration, and belongs to the field of information system integration, and the method comprises the steps: collecting various monitoring data and various fault records of a plurality of information systems of an intelligent system; determining multiple types of fault data according to the monitoring data of the time corresponding to the multiple types of fault records; according to redundancy among various monitoring data, screening an input data type of the fault prediction neural network of each type of faults; according to the type of the input data of the fault prediction neural network of each type of faults, selecting the input data of the fault prediction neural network of each type of faults from the monitoring data, and training to obtain the fault prediction neural network of the corresponding type of faults based on a BP neural network; and performing intelligent system fault prediction by adopting the fault prediction neural network of each type of faults obtained by training. The method can reduce the interference of the monitoring data of the redundant types of the multi-source information system on the prediction accuracy of the corresponding types of faults, and can save the computing resources.

Description

Technical Field

[0001] This application relates to the field of information system integration technology, and in particular to a data processing method and system based on multi-source information system integration. Background Technology

[0002] Intelligent systems in fields such as intelligent transportation, industrial IoT, smart cities, healthcare, energy management, and weather forecasting typically rely on multi-source information system integration for perception and decision-making. Multi-source information system integration involves integrating information systems based on a system architecture that utilizes multiple data sources. This integrates or fuses data from different sources, giving the integrated intelligent system stronger perception and decision-making capabilities. However, the various data collected by the multi-source information systems in intelligent systems may exhibit redundancy due to information overlap. For example, the positive linear correlation between different types of data, such as the real-time average power consumption of intelligent system computing devices and power consumption per unit time, can cause information overlap and data redundancy.

[0003] When predicting faults in intelligent systems, the increasing integration of multi-source information systems leads to the accumulation of redundant data. This redundancy in the input nodes of neurons in artificial neural networks such as BP (Back Propagation) neural networks, as well as weight redundancy between adjacent layers, increases the data processing workload and interferes with the fault prediction data output by the artificial neural networks, causing it to deviate from the true or accurate values. Furthermore, multi-source data processing consumes significant computational resources, is time-consuming, and costly. Therefore, fault prediction in intelligent systems based on integrated data processing from multi-source information systems needs to address the problems of redundant data interference and high computational resource consumption. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a data processing method and system based on the integration of multi-source information systems.

[0005] In a first aspect, this application provides a data processing method based on the integration of multi-source information systems, comprising the following steps: Collect various monitoring data and fault records from multiple information systems within the intelligent system; determine various fault data based on the monitoring data corresponding to the time of the fault records; screen the input data types for the fault prediction neural network for each fault type based on the redundancy among the various monitoring data and the significance test with the null hypothesis that there is no significant difference between the monitoring data and the fault data; select the input data for the fault prediction neural network for each fault type from the monitoring data of the multiple information systems within the intelligent system according to the input data types for the fault prediction neural network for each fault type, and train a corresponding fault prediction neural network based on a BP neural network; use the trained fault prediction neural network for each fault type to perform fault prediction for the intelligent system.

[0006] Specifically, based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are selected, including the following steps: Calculate the absolute value of the correlation coefficient between each type of monitoring data and each type of fault data; identify monitoring data whose absolute value of the correlation coefficient with the fault data is higher than the second correlation threshold as relevant monitoring data for the corresponding type of fault; calculate the absolute value of the correlation coefficient between different types of relevant monitoring data for the same type of fault, and mark the different types of relevant monitoring data whose absolute value of the correlation coefficient is higher than the third correlation threshold as highly redundant relevant monitoring data for the corresponding type of fault; select one of the two types of highly redundant relevant monitoring data and perform a first significance test on the highly redundant relevant monitoring data and the fault data for the corresponding type of fault, the null hypothesis of the first significance test being that there is no significant difference between the highly redundant relevant monitoring data and the fault data for the corresponding type of fault; identify the highly redundant relevant monitoring data with significant differences as the input data for the fault prediction neural network for predicting the corresponding type of fault.

[0007] Specifically, based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are screened, which also includes the following steps: Different types of related monitoring data with absolute values ​​of correlation coefficients below the third correlation threshold are labeled as low-redundancy related monitoring data for the corresponding type of fault. A second significance test is performed on each type of low-redundancy related monitoring data and the corresponding type of fault data. The null hypothesis of the second significance test is that there is no significant difference between the low-redundancy related monitoring data used for testing and the corresponding type of fault data. The low-redundancy related monitoring data with significant differences are determined as the input data for the fault prediction neural network that predicts the corresponding type of fault.

[0008] Specifically, based on the input data type of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system, and a fault prediction neural network based on a BP neural network for the corresponding type of fault is trained, including the following steps: Based on the input data types of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system. Multiple training samples and test samples corresponding to each type of fault are established. Multiple BP neural networks corresponding to the fault classes are initialized. The BP neural networks for the corresponding fault classes are trained and tested using the multiple training samples and test samples corresponding to each type of fault, and the fault prediction neural network for the corresponding fault classes is obtained.

[0009] Specifically, initializing multiple BP neural networks corresponding to fault classes includes the following steps: Based on the number of input data types in the fault prediction neural network for each type of fault, determine the number of neurons in the input layer of the corresponding fault prediction neural network; based on the number of monitoring data types in the fault data for the corresponding fault, determine the number of neurons in the output layer of the corresponding fault prediction neural network; based on the number of training samples, the number of neurons in the input layer, and the number of neurons in the output layer of the corresponding fault prediction neural network, determine the number of neurons in the hidden layer of the corresponding fault prediction neural network; based on the number of neurons in the hidden layer, the number of neurons in the input layer, and the number of neurons in the output layer of the corresponding fault prediction neural network, set the number of neurons in the BP neural network for the corresponding fault; initialize the parameters of the BP neural network; and set the activation functions of the hidden and output layers of the BP neural network.

[0010] Specifically, initializing the BP neural network parameters includes the following steps: Initialize the weights of each path from the input layer neurons to the hidden layer neurons in the fault prediction neural network; initialize the weights of each path from the hidden layer neurons to the output layer neurons in the fault prediction neural network; initialize the bias of each neuron in the input and hidden layers of the fault prediction neural network.

[0011] Specifically, setting the activation functions for the hidden and output layers of a BP neural network includes the following steps: The function with a constant gradient in the positive interval is set as the activation function of the hidden layer of the BP neural network; the Sigmoid function is set as the activation function of the output layer of the BP neural network.

[0012] Specifically, the fault prediction neural network for each type of fault, trained and analyzed, is used to predict faults in the intelligent system, including the following steps: Based on the input data type of the fault prediction neural network for each type of fault, the input data of the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems of the intelligent system, and input into the fault prediction neural network for the corresponding type of fault to predict the corresponding type of fault of the intelligent system.

[0013] Specifically, the data processing method based on multi-source information system integration further includes the following steps: The absolute values ​​of the correlation coefficients between the monitoring data corresponding to the time of different types of fault records are calculated respectively. Different types of faults with absolute values ​​of the correlation coefficients higher than the first correlation threshold are marked as interrelated faults. When using the fault prediction neural network for each type of fault trained, the fault prediction neural network for interrelated faults of the corresponding type of fault and its corresponding input data are also used to perform intelligent system fault prediction.

[0014] Secondly, this application provides a data processing system based on multi-source information system integration, comprising: The input layer data filtering module is used to collect various monitoring data and fault records from multiple information systems of the intelligent system; determine the various types of fault data based on the monitoring data corresponding to the time of the fault records; filter the input data types for the fault prediction neural network of each type of fault based on the redundancy between the various monitoring data and the significance test with the null hypothesis that there is no significant difference between the monitoring data and the fault data; select the input data for the fault prediction neural network of each type of fault from the monitoring data of the multiple information systems of the intelligent system according to the input data types for the fault prediction neural network of each type of fault, and train the corresponding fault prediction neural network based on the BP neural network. The fault prediction module is used to predict faults in intelligent systems by employing a fault prediction neural network trained for each type of fault.

[0015] The beneficial effects of the above-mentioned technical solutions provided in this application include at least the following: The technical solution of this application can reduce the interference of redundant monitoring data of multi-source information systems on the accuracy of corresponding fault prediction, reduce the input of redundant monitoring data into the fault prediction neural network, reduce the number of neural network nodes and the number of weights between input layer and hidden layer nodes caused by redundant monitoring data, reduce the computational complexity of updating weight parameters during backpropagation of BP neural network, reduce the amount of data computation, save computing resources, and make the integration effect of multi-source information systems better.

[0016] Other features and advantages of this application will be set forth in the following description or may be learned by practicing this application.

[0017] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the present application and form part of the specification. They are used together with the embodiments of the present application to explain the application and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a data processing method based on multi-source information system integration, as described in an embodiment of this application. Figure 2 This is a block diagram of a data processing system based on multi-source information system integration, as described in an embodiment of this application. Detailed Implementation

[0019] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0020] To address the problems existing in the prior art, embodiments of this application provide a data processing method and system based on multi-source information system integration.

[0021] This application provides a data processing method based on multi-source information system integration, the process of which is as follows: Figure 1 As shown, it includes the following steps: Step 1: Collect various monitoring data and fault records from multiple information systems of the intelligent system; determine the fault data based on the monitoring data corresponding to the time of the fault records.

[0022] In one specific embodiment, the intelligent system integrates multiple information systems, each including at least one type of monitoring data and fault records for at least one type of fault. The types of monitoring data include real-time equipment utilization, real-time equipment power consumption, cumulative power consumption during equipment startup, operating environment data, equipment temperature data, and data from various interfaces of the ERP system. The types of faults include data leakage, malicious attacks, access hijacking, interface freezing, high-concurrency crashes, service interruptions, equipment hardware failures, and abnormal equipment power supply. The monitoring data corresponding to the fault records for different fault types at different times reflects the monitoring data characteristics of that type of fault. Therefore, by comparing the real-time monitoring data with the monitoring data corresponding to the fault records for different fault types at different times, the probability of the intelligent system experiencing a corresponding type of fault can be predicted.

[0023] Taking data leakage as an example, various monitoring data from previous data leakage events are found in system logs and other log files. Several monitoring data that are highly correlated with data leakage are selected from these monitoring data as fault data for "data leakage" type of faults. Then, according to step 2, the input data types for "data leakage" type faults are selected from various monitoring data and input into the training of a fault prediction neural network for "data leakage" type faults based on BP neural network to obtain a fault prediction neural network for predicting the probability of "data leakage" type faults.

[0024] Among these monitoring data, several types of monitoring data with a high correlation to data leakage are selected. Specifically, based on the time of the failure, data types with a change rate greater than their respective preset thresholds are selected as the types of monitoring data with a high correlation to data leakage. Based on the types of monitoring data with a high correlation to data leakage, several types of monitoring data related to data leakage are selected.

[0025] Step 2: Based on the redundancy among various monitoring data and the significance test with the null hypothesis that there is no significant difference between monitoring data and fault data, screen the input data types for the fault prediction neural network of each type of fault; based on the input data types for the fault prediction neural network of each type of fault, select the input data for the fault prediction neural network of each type of fault from the monitoring data of multiple information systems of the intelligent system, and train the corresponding fault prediction neural network based on the BP neural network.

[0026] To accurately capture data anomalies during fault occurrences, fault data includes various types of monitoring data. Especially for unknown anomalies, for safety, fault data includes all monitoring data at the time of the fault. Therefore, fault data often includes some redundant types of monitoring data. These redundant types of monitoring data are independent of the fault event and cannot reflect whether a fault has occurred. Because a backpropagation (BP) neural network determines the number of input layer nodes based on the dimension of the input data, using the monitoring data types included in the fault data as the input types for the neural network used for fault prediction would cause redundancy in the neuron input nodes and weight redundancy between adjacent layer nodes in the BP neural network. This increases the data processing workload, interferes with the fault prediction data output by the artificial neural network, causing it to deviate from the true or accurate value. Furthermore, multi-source data processing consumes significant computational resources, is time-consuming, and costly. Therefore, to reduce these redundant types of monitoring data, the monitoring data type filtering step described in step 2 is adopted.

[0027] Specifically, based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are selected, including the following steps: Calculate the absolute value of the correlation coefficient for each type of monitoring data and each type of fault data separately. For example, if the monitoring data includes multiple types such as a1, a2, ..., an, and the fault data includes multiple types such as b1, b2, ..., bm according to the fault category, calculate the absolute value of the correlation coefficient between ai and bj, where i is a natural number from 1 to n and j is a natural number from 1 to m. The correlation coefficient can be calculated using formulas such as Pearson correlation coefficient, Spearman Rank correlation coefficient, or Kendall Rank correlation coefficient.

[0028] Monitoring data whose absolute correlation coefficient with fault data is higher than the second correlation threshold are identified as relevant monitoring data for the corresponding fault type. For example, if the monitoring data with an absolute correlation coefficient higher than the second correlation threshold for "data leakage" fault data includes multiple types such as ar1, ar2, ..., arp, where r1 to rp are natural numbers between 1 and n, then the relevant monitoring data for the "data leakage" fault type includes ar1, ar2, ..., arp. By calculating the absolute value of the correlation coefficient with fault data, the types of monitoring data with high correlation to fault data are selected.

[0029] The absolute values ​​of correlation coefficients between different types of related monitoring data for the same type of fault are calculated. Monitoring data with absolute correlation coefficient values ​​higher than a third correlation threshold are marked as highly redundant related monitoring data for the corresponding fault type. For example, to calculate the absolute values ​​of correlation coefficients between the related monitoring data ar1, ar2, ..., arp corresponding to a "data leakage" fault, ar1, ar2, ..., arp are paired, and the absolute values ​​of correlation coefficients between the two types of related monitoring data in each pair are calculated. Monitoring data in each pair with absolute correlation coefficient values ​​higher than the third correlation threshold are marked as highly redundant related monitoring data corresponding to a "data leakage" fault. The selected highly redundant related monitoring data can be used to proceed to the next step by choosing one of two highly redundant related monitoring data. This reduces the types of monitoring data input to the neural network in subsequent steps, thereby reducing the impact of redundant data on prediction accuracy and reducing the computational load of the neural network, saving computational resources.

[0030] One of two sets of highly redundant and correlated monitoring data is selected. A first significance test is performed on the highly redundant and correlated monitoring data and the corresponding fault data. The null hypothesis of the first significance test is that there is no significant difference between the highly redundant and correlated monitoring data and the corresponding fault data. For example, in hypothesis testing, the null hypothesis H0 is that there is no significant difference between the highly redundant and correlated monitoring data and the corresponding fault data, and the alternative hypothesis H1 is that there is a significant difference between the highly redundant and correlated monitoring data and the corresponding fault data. Alternatively, the null hypothesis H0 can be interpreted as the difference between the highly redundant and correlated monitoring data and the corresponding fault data being caused by random factors. The significance level is set to 0.05. Then, based on the data characteristics, an appropriate statistical test method is selected. Based on the selected statistical test method, statistical indicators and p-values ​​are calculated. The p-value is the probability that a more extreme result than the obtained sample observation result will occur when the null hypothesis H0 is true. Common significance tests include the t-test and the chi-square test. The t-test is suitable for normally distributed independent data samples with homogeneous variance, while the chi-square test is suitable for large sample sizes where the expected frequency of each category is not less than 5. If the p-value is below the significance level of 0.05, the null hypothesis H0 is rejected and the alternative hypothesis H1 is accepted. If the p-value is above the significance level of 0.05, the null hypothesis H0 is accepted and the alternative hypothesis H1 is rejected.

[0031] Highly redundant and correlated monitoring data with significant differences were selected as input data for the fault prediction neural network that predicts the corresponding fault types. Significance testing was then conducted to rule out cases where the differences between the highly redundant and correlated monitoring data and the corresponding fault data were due to chance factors, further reducing the types of redundant monitoring data input to the neural network.

[0032] As a supplement, based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are screened, which also includes the following steps: Different types of related monitoring data with absolute values ​​of correlation coefficients below the third correlation threshold are labeled as low-redundancy related monitoring data for the corresponding type of fault. A second significance test is performed on each type of low-redundancy related monitoring data and the corresponding type of fault data. The null hypothesis of the second significance test is that there is no significant difference between the low-redundancy related monitoring data used for testing and the corresponding type of fault data. The low-redundancy related monitoring data with significant differences are determined as the input data for the fault prediction neural network that predicts the corresponding type of fault.

[0033] By conducting significance tests, the possibility that the difference between the low-redundancy correlation monitoring data used for testing and the corresponding fault data is caused by accidental factors is eliminated, further reducing the types of redundant monitoring data input into the neural network.

[0034] Specifically, based on the input data type of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system, and a fault prediction neural network based on a BP neural network for the corresponding type of fault is trained, including the following steps: Based on the input data types of the fault prediction neural network for each type of fault, input data for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system. Multiple training and testing samples are established for each type of fault. Multiple backpropagation (BP) neural networks corresponding to the fault classes are initialized. The BP neural networks for each type of fault are trained and tested using the training and testing samples respectively, resulting in the fault prediction neural network for that type of fault. The BP neural network includes an input layer, a hidden layer, and an output layer; a corresponding BP neural network is set up for each type of fault.

[0035] Specifically, initializing multiple BP neural networks corresponding to fault classes includes the following steps: The number of input data types for each type of fault prediction neural network is determined based on the number of input layer neurons for that fault type. The number of output layer neurons for that fault type is determined based on the number of monitoring data types in the fault data for that fault type. The number of hidden layer neurons for that fault type is determined based on the number of training samples, the number of input layer neurons, and the number of output layer neurons for that fault type. For example, the number of hidden layer neurons can be determined based on... Sure, This represents the number of neurons in the hidden layer of a fault prediction neural network. Indicates the number of training samples. This represents the adjustment constant, which can be selected from natural numbers between 2 and 10. This indicates the number of neurons in the input layer of the fault prediction neural network. The number of neurons in the output layer of the fault prediction neural network is represented, or other empirical formulas are used to determine the number of neurons in the hidden layer; based on the number of neurons in the hidden layer, input layer, and output layer of the fault prediction neural network for the corresponding fault type, the number of neurons in the BP neural network for the corresponding fault type is set; the parameters of the BP neural network are initialized; and the activation functions of the hidden and output layers of the BP neural network are set.

[0036] Specifically, initializing the BP neural network parameters includes the following steps: Initialize the weights of each path from the input layer neurons to the hidden layer neurons in the fault prediction neural network; initialize the weights of each path from the hidden layer neurons to the output layer neurons in the fault prediction neural network; initialize the bias of each neuron in the input and hidden layers of the fault prediction neural network.

[0037] Specifically, setting the activation functions of the hidden and output layers of a BP neural network includes the following steps: First, a function with a constant gradient in the positive interval is set as the activation function of the hidden layer of the BP neural network. Functions with a constant gradient in the positive interval include ReLU (Rectified linear unit), Leaky ReLU, and PReLU (Parametric Rectified linear unit), etc. The advantage of this is that when updating weights using gradient descent, gradient vanishing does not occur, the calculation is simple, and the neural network can converge relatively quickly. Second, the Sigmoid function is set as the activation function of the output layer of the BP neural network. The output value of the Sigmoid function is between 0 and 1, and its output value can be interpreted probabilistically to represent the predicted fault probability.

[0038] Step 3: Use the fault prediction neural network for each type of fault obtained from training to perform fault prediction for the intelligent system.

[0039] Specifically, the fault prediction neural network for each type of fault, trained and analyzed, is used to predict faults in the intelligent system, including the following steps: Based on the input data types of the fault prediction neural network for each type of fault, input data for each type of fault prediction neural network is selected from the monitoring data of multiple information systems within the intelligent system. This input data is then fed into the corresponding fault prediction neural network to predict the corresponding type of fault in the intelligent system. For example, different types of input data are determined for different types of faults, and fault prediction neural networks for predicting different types of faults are trained. Input data of the corresponding type is selected from various types of monitoring data and fed into the corresponding fault prediction neural network, outputting the predicted probability of the corresponding type of fault. This yields the probability of each type of fault occurring in the intelligent system.

[0040] As a supplement, the data processing method based on multi-source information system integration also includes the following steps: The absolute values ​​of correlation coefficients between monitoring data corresponding to different types of fault records at different times are calculated. Faults of different types with absolute correlation coefficient values ​​higher than a first correlation threshold are marked as interrelated faults. When using the trained fault prediction neural network for each type of fault to predict intelligent system faults, the fault prediction neural network for interrelated faults of the corresponding fault type and its corresponding input data are also used to predict intelligent system faults. By using the above method of marking interrelated faults, it is possible to prioritize the prediction of related faults after a fault occurs, without having to predict every other type of fault, thus optimizing the maintenance process.

[0041] The various thresholds, such as the first correlation threshold, the second correlation threshold, and the third correlation threshold, in this application can be obtained or adjusted by assigning initial values ​​and modifying them through a posteriori feedback based on the corresponding effects of the current thresholds.

[0042] The terms "first" and "second" mentioned above do not indicate a sequential order, but rather represent a distinction between different characteristics.

[0043] In the above method of this embodiment, the technical solution of this application can reduce the interference of monitoring data of redundant types in multi-source information systems on the accuracy of corresponding fault prediction, reduce the input of monitoring data of redundant types into the fault prediction neural network, reduce the number of neural network nodes and the number of weights between input layer and hidden layer nodes caused by monitoring data of redundant types, reduce the computational complexity of updating weight parameters during backpropagation of BP neural network, reduce the amount of data computation, save computing resources, and make the integration effect of multi-source information systems better.

[0044] Those skilled in the art can change the above-described order of recording or implementation without departing from the scope of protection of this application.

[0045] Another embodiment of this application provides a data processing system based on multi-source information system integration, including: The input layer data filtering module is used to collect various monitoring data and fault records from multiple information systems of the intelligent system; determine the various types of fault data based on the monitoring data corresponding to the time of the fault records; filter the input data types for the fault prediction neural network of each type of fault based on the redundancy between the various monitoring data and the significance test with the null hypothesis that there is no significant difference between the monitoring data and the fault data; select the input data for the fault prediction neural network of each type of fault from the monitoring data of the multiple information systems of the intelligent system according to the input data types for the fault prediction neural network of each type of fault, and train the corresponding fault prediction neural network based on the BP neural network. The fault prediction module is used to predict faults in intelligent systems by employing a fault prediction neural network trained for each type of fault.

[0046] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0047] In this embodiment, the technical solution of this application can reduce the interference of redundant monitoring data of multi-source information systems on the accuracy of corresponding fault prediction, reduce the input of redundant monitoring data into the fault prediction neural network, reduce the number of neural network nodes and the number of weights between input layer and hidden layer nodes caused by redundant monitoring data, reduce the computational complexity of updating weight parameters during backpropagation of BP neural network, reduce the amount of data computation, save computing resources, and make the integration effect of multi-source information systems better.

[0048] Any modifications, additions, and equivalent substitutions made within the scope of the principles of this application shall still fall within the scope of the patent coverage of this application.

Claims

1. A data processing method based on multi-source information system integration, characterized in that, Includes the following steps: Collect various monitoring data and fault records from multiple information systems within the intelligent system; determine the various fault data based on the monitoring data corresponding to the time of the fault records. Based on the redundancy among various monitoring data and the significance test with the null hypothesis that there is no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network of each type of fault are selected. Based on the input data types of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system, and a corresponding fault prediction neural network based on a BP neural network is trained; the trained fault prediction neural network for each type of fault is used to predict faults in the intelligent system.

2. The data processing method based on multi-source information system integration as described in claim 1, characterized in that, Based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are selected, including the following steps: Calculate the absolute value of the correlation coefficient between each type of monitoring data and each type of fault data; identify monitoring data whose absolute value of the correlation coefficient with the fault data is higher than the second correlation threshold as relevant monitoring data for the corresponding type of fault; calculate the absolute value of the correlation coefficient between different types of relevant monitoring data for the same type of fault, and mark the different types of relevant monitoring data whose absolute value of the correlation coefficient is higher than the third correlation threshold as highly redundant relevant monitoring data for the corresponding type of fault; select one of the two types of highly redundant relevant monitoring data and perform a first significance test on the highly redundant relevant monitoring data and the fault data for the corresponding type of fault, the null hypothesis of the first significance test being that there is no significant difference between the highly redundant relevant monitoring data and the fault data for the corresponding type of fault; identify the highly redundant relevant monitoring data with significant differences as the input data for the fault prediction neural network for predicting the corresponding type of fault.

3. The data processing method based on multi-source information system integration as described in claim 2, characterized in that, Based on the redundancy among various monitoring data and the significance test assuming no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network for each type of fault are screened, which also includes the following steps: Different types of related monitoring data with absolute values ​​of correlation coefficients below the third correlation threshold are labeled as low-redundancy related monitoring data for the corresponding type of fault. A second significance test is performed on each type of low-redundancy related monitoring data and the corresponding type of fault data. The null hypothesis of the second significance test is that there is no significant difference between the low-redundancy related monitoring data used for testing and the corresponding type of fault data. The low-redundancy related monitoring data with significant differences are determined as the input data for the fault prediction neural network that predicts the corresponding type of fault.

4. The data processing method based on multi-source information system integration as described in claim 1, characterized in that, Based on the input data type of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system, and a fault prediction neural network based on a BP neural network for the corresponding type of fault is trained, including the following steps: Based on the input data types of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system. Multiple training samples and test samples corresponding to each type of fault are established. Multiple BP neural networks corresponding to the fault classes are initialized. The BP neural networks for the corresponding fault classes are trained and tested using the multiple training samples and test samples corresponding to each type of fault, and the fault prediction neural network for the corresponding fault classes is obtained.

5. The data processing method based on multi-source information system integration as described in claim 4, characterized in that, Initialize multiple backpropagation (BP) neural networks corresponding to fault classes, including the following steps: Based on the number of input data types in the fault prediction neural network for each type of fault, determine the number of neurons in the input layer of the corresponding fault prediction neural network; based on the number of monitoring data types in the fault data for the corresponding fault, determine the number of neurons in the output layer of the corresponding fault prediction neural network; based on the number of training samples, the number of neurons in the input layer, and the number of neurons in the output layer of the corresponding fault prediction neural network, determine the number of neurons in the hidden layer of the corresponding fault prediction neural network; based on the number of neurons in the hidden layer, the number of neurons in the input layer, and the number of neurons in the output layer of the corresponding fault prediction neural network, set the number of neurons in the BP neural network for the corresponding fault; initialize the parameters of the BP neural network; and set the activation functions of the hidden and output layers of the BP neural network.

6. The data processing method based on multi-source information system integration as described in claim 5, characterized in that, Initializing the parameters of the BP neural network includes the following steps: Initialize the weights of each path from the input layer neurons to the hidden layer neurons in the fault prediction neural network; initialize the weights of each path from the hidden layer neurons to the output layer neurons in the fault prediction neural network; initialize the bias of each neuron in the input and hidden layers of the fault prediction neural network.

7. The data processing method based on multi-source information system integration as described in claim 5, characterized in that, Setting the activation functions for the hidden and output layers of a backpropagation (BP) neural network includes the following steps: The function with a constant gradient in the positive interval is set as the activation function of the hidden layer of the BP neural network; the Sigmoid function is set as the activation function of the output layer of the BP neural network.

8. The data processing method based on multi-source information system integration as described in claim 1, characterized in that, The fault prediction neural network for each type of fault, trained and obtained, is used to predict faults in intelligent systems, including the following steps: Based on the input data type of the fault prediction neural network for each type of fault, the input data of the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems of the intelligent system, and input into the fault prediction neural network for the corresponding type of fault to predict the corresponding type of fault of the intelligent system.

9. The data processing method based on multi-source information system integration as described in claim 8, characterized in that, The data processing method based on multi-source information system integration further includes the following steps: The absolute values ​​of the correlation coefficients between the monitoring data corresponding to the time of different types of fault records are calculated respectively. Different types of faults with absolute values ​​of the correlation coefficients higher than the first correlation threshold are marked as interrelated faults. When using the fault prediction neural network for each type of fault trained, the fault prediction neural network for interrelated faults of the corresponding type of fault and its corresponding input data are also used to perform intelligent system fault prediction.

10. A data processing system based on multi-source information system integration, characterized in that, include: The input layer data filtering module is used to collect various monitoring data and fault records from multiple information systems of the intelligent system; Based on the monitoring data corresponding to the time of the various types of fault records, the various types of fault data are determined; Based on the redundancy among various monitoring data and the significance test with the null hypothesis that there is no significant difference between monitoring data and fault data, the input data types for the fault prediction neural network of each type of fault are selected. Based on the input data types of the fault prediction neural network for each type of fault, input data for the fault prediction neural network for each type of fault is selected from the monitoring data of multiple information systems in the intelligent system, and a fault prediction neural network for the corresponding type of fault based on the BP neural network is trained. The fault prediction module is used to predict faults in intelligent systems by employing a fault prediction neural network trained for each type of fault.

Images