Industrial internet of things running state prediction method and system fusing multi-source heterogeneous data

Abstract

The application discloses a kind of industrial internet of things running state prediction method and system of fusion multi-source heterogeneous data, it is related to industrial internet of things technical field, method includes: obtaining main monitoring period and secondary monitoring period, obtaining heterogeneous state data, obtaining time period to be predicted;Obtain main data sequence, obtain main fluctuation trend index, obtain main correlation coefficient;Obtain secondary data sequence, obtain secondary fluctuation trend index, obtain secondary correlation coefficient;Screen out stable correlation data pair;Causal direction inference is carried out to each stable correlation data pair, determine the causal direction between two heterogeneous state data, build causal diagram;The causal stability index of each stable correlation data pair is calculated, and the causal stable data pair is screened out;According to secondary data sequence and causal diagram, obtain the data change trajectory in prediction time period thereof.The application has the advantages of accurate prediction, self-adaptive adjustment and causal derivation.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) technology, and specifically to an industrial IoT operation status prediction method and system that integrates multi-source heterogeneous data. Background Technology

[0002] In the field of industrial IoT equipment operation status prediction technology, accurately extracting the evolutionary patterns of equipment health status from multi-source heterogeneous sensor data and achieving high-precision prediction is a key technical challenge to ensure the safe and stable operation of equipment. Existing prediction methods typically first acquire status data sequences from multiple consecutive monitoring periods, calculate the fluctuation trend index of each data sequence, and construct a correlation coefficient between any two status data based on the degree of difference in the fluctuation trend index. Then, by comparing the changes in the correlation coefficient in different periods, data pairs with relatively stable correlations are selected. Finally, extrapolation predictions are performed on the data sequences in the most recent period based on these stable correlation data.

[0003] However, such methods can only identify whether there are numerical correlations of co-changing variables, but cannot further distinguish the causal direction between variables. That is, they cannot determine whether the temperature change causes the current change or the current change causes the temperature change, nor can they construct a complete causal transmission path between variables. Although this analysis method based solely on numerical correlation can screen out combinations of co-changing variables, it has significant shortcomings in terms of the interpretability of prediction results and fault tracing capabilities. When the prediction model outputs abnormal equipment status, maintenance personnel cannot determine which abnormal change in the underlying physical quantity triggered the entire fault transmission chain, making it difficult to accurately locate the root cause of the fault and carry out targeted maintenance. In addition, prediction models based on numerical correlation implicitly assume that the correlation between variables has long-term stability. However, in real industrial scenarios, equipment aging, changes in operating conditions, or component replacement may cause structural changes in the causal relationship between variables. When the causal direction is reversed or a new causal path appears, the prediction model that relies solely on the stability of numerical correlation will produce systematic biases, and the prediction accuracy will decrease significantly. Summary of the Invention

[0004] To address the technical problem of how to further identify the causal direction between variables and construct a stable causal graph based on correlation stability screening in existing technologies, thereby achieving interpretable fault tracing while ensuring prediction accuracy, this invention provides an industrial Internet of Things (IoT) operation status prediction method and system that integrates multi-source heterogeneous data.

[0005] A method for predicting the operational status of an industrial Internet of Things (IIoT) system by integrating multi-source heterogeneous data includes: acquiring two consecutive monitoring periods for a target device in the IIoT prior to the current moment, which are designated as the primary and secondary monitoring periods; acquiring various heterogeneous status data during the operation of the target device; and acquiring the time period to be predicted; performing time-aligned preprocessing on the various heterogeneous status data to ensure that all status data have a unified sampling time point; acquiring the primary data sequences of various heterogeneous status data within the primary monitoring period; obtaining the primary fluctuation trend indicators corresponding to various heterogeneous status data based on the primary data sequences; and obtaining the primary correlation coefficient between any two heterogeneous status data based on the primary fluctuation trend indicators; and acquiring the secondary data sequences of various heterogeneous status data within the secondary monitoring period; and obtaining the secondary fluctuation trend indicators corresponding to various heterogeneous status data based on the secondary data sequences. The process involves obtaining the secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend indicators of various heterogeneous state data; selecting two heterogeneous state data pairs whose difference between the primary and secondary correlation coefficients is less than a preset change threshold and using them as stable correlation data pairs; performing causal direction inference on each stable correlation data pair to determine the causal direction between the two heterogeneous state data pairs, and constructing a causal graph based on all causal edges with stable causal directions; calculating the causal stability index of each stable correlation data pair, and selecting heterogeneous state data pairs whose causal stability index is greater than a preset causal threshold as causal stable data pairs; obtaining the data change trajectory of each state data in the causal stable data pair within the prediction time period based on the secondary data sequence and the causal graph; and for any heterogeneous state data, if it changes as a dependent variable, deriving the changes of all directly or indirectly related result variables based on the causal graph.

[0006] Optionally, time-aligned preprocessing for various heterogeneous state data includes: for each type of heterogeneous state data, resampling the data to a unified reference time series using a linear interpolation method based on its original sampling timestamp, so that all state data are synchronized in time.

[0007] Optionally, obtaining the main fluctuation trend indicators corresponding to various heterogeneous state data based on the master data sequences of various heterogeneous state data includes: calculating the difference between two adjacent monitoring values ​​in the master data sequence of the state data in sequence and using it as the main relative change, obtaining the historical theoretical maximum and minimum values ​​corresponding to the state data, and calculating the main absolute change corresponding to the main relative change based on the main relative change, the historical theoretical maximum and the historical theoretical minimum values; and accumulating the main absolute changes in the master data sequence of the state data and using it as the main fluctuation trend indicator of the state data.

[0008] Optionally, obtaining the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend indicators of various heterogeneous state data includes: calculating the main correlation coefficient based on the degree of difference between the two main fluctuation trend indicators and a preset scaling factor, wherein the value of the main correlation coefficient is inversely proportional to the degree of difference and the value range is between 0 and 1.

[0009] Optionally, causal direction inference is performed on each pair of stable associated data to determine the causal direction between the two heterogeneous state data, including: for each pair of stable associated data, constructing a vector autoregressive model with the first type of heterogeneous state data as the dependent variable and the second type of heterogeneous state data as the independent variable, and a vector autoregressive model with the second type of heterogeneous state data as the dependent variable and the first type of heterogeneous state data as the independent variable; calculating the F-statistic and its corresponding p-value of the lagged terms of the independent variables in each model through the F-test; if the p-value of the lagged terms of the independent variables in the first model is... If the p-value of the lagged term in the second model is less than the preset significance level threshold and the p-value of the lagged term in the second model is not less than the threshold, then the causal direction is determined to be from the first heterogeneous state data to the second heterogeneous state data; if the p-value of the lagged term in the second model is less than the threshold and the p-value of the lagged term in the first model is not less than the threshold, then the causal direction is determined to be from the second heterogeneous state data to the first heterogeneous state data; if the p-value in both models is less than the threshold, then bidirectional causality is determined to exist; if the p-value in both models is not less than the threshold, then no causal direction is determined to exist.

[0010] Optionally, the causal stability index of each stable data pair is calculated, and heterogeneous state data pairs with causal stability indices greater than a preset causal threshold are selected as causal stable data pairs. This includes: for each stable data pair, determining its primary causal direction in the main monitoring period and its secondary causal direction in the secondary monitoring period; determining the causal stability index based on the consistency between the primary and secondary causal directions, where the causal stability index reflects the proportion of identical causal directions in the two periods, and a larger value indicates a more stable causal direction; and selecting stable data pairs with causal stability indices greater than a preset causal threshold as causal stable data pairs.

[0011] Optionally, based on the sub-data sequences of each state data in the causal stable data pair and combined with the causal graph, the data change trajectory within the prediction time period is obtained, including: for each state data in the causal stable data pair, if it is a dependent variable, then the future change trajectory is obtained by fitting a prediction model based on the sub-data sequence; if it is an effect variable, then the future change trajectory is derived based on the predicted value of the upstream dependent variable in the causal graph and the causal function relationship between variables; repeat this process until the predicted values ​​of all nodes in the causal graph are obtained.

[0012] A system for predicting the operational status of an industrial IoT device by integrating multi-source heterogeneous data is also provided. The system includes a management platform, a sensor network platform, and an object platform connected sequentially. The management platform includes: a data acquisition module, used to acquire two consecutive monitoring cycles for the target device in the industrial IoT up to the current time, serving as the primary and secondary monitoring cycles, and to acquire various heterogeneous status data during the target device's operation, as well as the time period to be predicted; a data preprocessing module, used to perform time-aligned preprocessing on the various heterogeneous status data to ensure all status data have a unified sampling time point; a primary correlation analysis module, used to acquire the primary data sequences of various heterogeneous status data within the primary monitoring cycle, calculate the primary fluctuation trend indicators corresponding to various heterogeneous status data based on the primary data sequences, and calculate the primary correlation coefficient between any two heterogeneous status data based on the primary fluctuation trend indicators; and a secondary correlation analysis module, used to acquire the secondary data sequences of various heterogeneous status data within the secondary monitoring cycle, and calculate various... The system employs several mechanisms: a secondary fluctuation trend index for heterogeneous state data, and a secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend index; a stable correlation screening module, which calculates the absolute difference between the primary and secondary correlation coefficients of each pair of heterogeneous state data and selects heterogeneous state data pairs with a difference less than a preset change threshold as stable correlation data pairs; a causal inference module, which infers the causal direction of each stable correlation data pair, determines the causal direction between the two heterogeneous state data, and constructs a causal graph based on all causal edges with stable causal directions; a causal stability index for each stable correlation data pair, and selects heterogeneous state data pairs with a causal stability index greater than a preset causal threshold as causal stable data pairs; and a causal prediction module, which obtains the data change trajectory of each state data in the causal stable data pair within the prediction time period based on the secondary data sequence and the causal graph; wherein, for any heterogeneous state data, if it changes as a dependent variable, the changes of all directly or indirectly related result variables are derived based on the causal graph.

[0013] Optionally, the data preprocessing module is also used to: for each type of heterogeneous state data, based on its original sampling timestamp, use a linear interpolation method to resample the data to a unified reference time series, so that all state data are synchronized in time.

[0014] Optionally, the main correlation analysis module is also used to: calculate the difference between two adjacent monitoring values ​​in the main data sequence according to the main data sequence of the state data and use it as the main relative change, obtain the historical theoretical maximum and historical theoretical minimum values ​​corresponding to the state data, and calculate the main absolute change corresponding to the main relative change based on the main relative change, the historical theoretical maximum and the historical theoretical minimum values; and accumulate each main absolute change in the main data sequence of the state data and use it as the main fluctuation trend indicator of the state data.

[0015] The beneficial effects of this invention are reflected in: In the entire industrial IoT operational status prediction method that integrates multi-source heterogeneous data, firstly, based on the correlation stability screening, causal direction inference is further performed on each stable correlation data pair. This accurately identifies the causal relationship between variables, thus upgrading the correlation pairs that can only represent numerical co-changes in traditional methods to causal pairs with clear physical causal meanings. When an abnormal equipment status is detected during subsequent prediction, maintenance personnel can trace back along the constructed causal graph to accurately locate the root cause variable causing the abnormality, providing an intuitive decision-making basis for predictive maintenance. Furthermore, by calculating the causal stability index of each stable correlation data pair and screening out causal stable data pairs whose causal direction remains consistent over two consecutive monitoring periods, false causal pairs that are numerically stable but whose causal direction reverses with changes in operating conditions can be effectively eliminated. This ensures that the variable relationships used for prediction have a long-term reliable physical causal basis. When equipment aging, changes in operating conditions, or component replacement lead to changes in the causal structure, the prediction method can be more effective. When changes occur, this method can promptly detect such changes through real-time monitoring of causal stability indicators, thereby avoiding prediction bias based on outdated causal relationships. Furthermore, this invention performs hierarchical prediction based on the constructed causal graph, independently modeling and predicting the future trajectory of nodes as dependent variables, while deriving the causal function relationship learned from historical data based on the predicted values ​​of upstream dependent variables in the causal graph. This causal transmission prediction method fully preserves the real physical interaction path between variables, avoiding the problem of the separation of the cooperative relationship between variables in traditional multivariate independent prediction, making the prediction results more consistent with the internal mechanism of equipment operation. Furthermore, when the causal stability indicator of a certain causal edge continues to decline and falls below a preset threshold, it can provide an early warning that the causal path may undergo structural changes, prompting maintenance personnel to check whether there are potential fault risks in related components, thereby realizing the shift from single numerical prediction to causal structure monitoring, further enhancing the early warning capability and the level of maintenance decision support. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a schematic diagram illustrating the steps of the industrial IoT operation status prediction method that integrates multi-source heterogeneous data according to the present invention. Figure 2 This is a partial flowchart of steps S1 to S5 in the industrial IoT operation status prediction method that integrates multi-source heterogeneous data according to the present invention. Figure 3 This is a partial flowchart of S5 and S6 in the industrial IoT operation status prediction method that integrates multi-source heterogeneous data of the present invention. Figure 4 This is a schematic diagram of the composition of the industrial IoT operation status prediction system that integrates multi-source heterogeneous data according to the present invention; Figure 5 This is a schematic diagram of the optimized industrial Internet of Things (IoT) involved in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] like Figure 1 , Figure 2 and Figure 3 As shown, an industrial IoT operation status prediction method that integrates multi-source heterogeneous data is provided, including: S1. Obtain two consecutive monitoring cycles for the target device in the Industrial Internet of Things before the current time and use them as the main monitoring cycle and the secondary monitoring cycle. Obtain various heterogeneous state data during the operation of the target device and obtain the time period to be predicted. Perform time alignment preprocessing on the various heterogeneous state data so that all state data have a unified sampling time point.

[0022] In this embodiment, it should be noted that the main purpose of S1 is to establish a time frame and basic data source for subsequent data analysis. Specifically, when implementing prediction, the historical monitoring data stream of the target device is first accessed from the industrial IoT platform. This data stream is collected and uploaded in real time by various types of sensors deployed on the device itself or in the surrounding environment. Two consecutive and adjacent monitoring cycles before the current moment are extracted from the data stream, with the earlier cycle defined as the main monitoring cycle and the later cycle defined as the secondary monitoring cycle. The two cycles are connected end-to-end on the time axis and are both adjacent to the current moment. At the same time, various heterogeneous state data related to the operating status of the target device are extracted from the data stream, including but not limited to pressure, flow rate, rotational speed, displacement, acoustic emission intensity, and chemical component concentration. These data have different physical dimensions and sampling frequencies. Finally, the time period to be predicted is set. This time period is a future time window starting from the current moment, and its length is determined based on the device response characteristics and operation and maintenance requirements.

[0023] S1 also includes: performing time alignment preprocessing on the various heterogeneous state data to ensure that all state data have a unified sampling time point.

[0024] In this embodiment, since different sensors may have different sampling frequencies—for example, a pressure sensor samples every 5 seconds and a flow sensor samples every 10 seconds—directly using the raw timestamps cannot perform point-by-point differential and correlation analysis. Therefore, after S1 and before S2, time-aligned preprocessing is required for the multi-source heterogeneous data. Specifically, a unified reference time series is set, with the time interval being the least common multiple of the sampling periods of all sensors or set according to actual needs (e.g., one time point every 10 seconds). For each type of heterogeneous state data, the value of each reference time point is calculated using a linear interpolation method based on its raw sampling timestamp and value, thus obtaining the time-synchronized data sequence. In this embodiment, a sampling point is uniformly used every 10 minutes to facilitate subsequent calculations.

[0025] S2. Obtain the master data sequence of various heterogeneous state data within the main monitoring period, and obtain the main fluctuation trend index corresponding to various heterogeneous state data based on the master data sequence of various heterogeneous state data, and obtain the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend index of various heterogeneous state data.

[0026] In this embodiment, normalized trend modeling is performed on multi-source heterogeneous state data within the main monitoring period. The core of this approach is to construct a fluctuation representation that can be compared across dimensions. Specifically, for each type of heterogeneous state data, its complete time-series sequence after time alignment within the main monitoring period is extracted as the master data sequence. Point-by-point differencing is performed on adjacent sampling points in this sequence to obtain a signed sequence of principal relative changes. Subsequently, the range between the historical theoretical maximum and minimum values ​​corresponding to the state data is calculated as the normalization benchmark. Each principal relative change is divided by this range to obtain the dimensionless principal absolute change. Finally, all principal absolute changes are algebraically summed to form the principal fluctuation trend index of the state data. Based on this, using the formula Calculate data from any two heterogeneous states and The primary correlation coefficient between them. For the first species and first The principal correlation coefficient between heterogeneous state data is dimensionless and ranges from 1 to 1. The larger the value, the more similar the fluctuation trends of the two data points within the main monitoring period; For the first The main fluctuation trend index of heterogeneous state data within the main monitoring period is dimensionless and is obtained by the sum of normalized differences of adjacent monitoring values ​​in the main data sequence. For the first The main fluctuation trend indicator of heterogeneous state data within the main monitoring period is dimensionless. This is a preset scaling factor, dimensionless, used to adjust the sensitivity of the correlation coefficient to trend differences. It can be set according to the dynamic response characteristics of the equipment (e.g., 2-12); where, Specific values ​​can be obtained through training with historical data: select a period of historical data under stable operating conditions, and calculate different values. Choose the ratio of the correlation coefficient between physical quantities of the same type to the correlation coefficient between physical quantities of different types, and select the value that maximizes this ratio. Value; for equipment with slow response (such as large storage tanks). Typically, a smaller value (2-5) is used; for fast-response devices (such as pneumatic valves). A larger value (8-12) is usually chosen. This represents the absolute value of the difference between two volatility trend indicators.

[0027] S3. Obtain the secondary data sequence of various heterogeneous state data within the secondary monitoring period, and obtain the secondary fluctuation trend index corresponding to various heterogeneous state data based on the secondary data sequence of various heterogeneous state data, and obtain the secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend index of various heterogeneous state data.

[0028] In this embodiment, the same processing procedure as S2 is performed for the secondary monitoring period, but the target data is the latest dynamic data. Specifically, the secondary data sequence of each heterogeneous state data after time alignment within the secondary monitoring period is extracted, and the secondary relative change is obtained by point-by-point difference; normalization is performed based on the same historical theoretical extreme value range to obtain the secondary absolute change; and the results are accumulated to form the secondary fluctuation trend index. ; and then through Calculate the secondary correlation coefficient. Wherein, For the first species and first The secondary correlation coefficient between heterogeneous state data is dimensionless and ranges from 1 to 1. The larger the value, the more similar the fluctuation trends of the two data points are within the monitoring period; For the first The dimensionless indicator of the fluctuation trend of heterogeneous state data within a monitoring period; For the first The dimensionless indicator of the fluctuation trend of heterogeneous state data within a monitoring period; Use the same preset scaling factor as in the main correlation coefficient. S3 ensures that the correlation assessment is based on the immediate response characteristics of the equipment under its current operating conditions.

[0029] S4. Select the two heterogeneous state data corresponding to the difference between the primary correlation coefficient and the secondary correlation coefficient being less than the preset change threshold, and use them as stable correlation data pairs.

[0030] In this implementation, cross-period correlation stability determination is performed. For each pair of heterogeneous state data... Calculate its primary correlation coefficient. Correlation coefficient absolute difference ;like ( If the preset change threshold is used, then the data is determined to be... If the correlation remains stable over two consecutive periods, it is included in the set of stable correlation data pairs; otherwise, it is considered a non-steady-state correlation affected by random disturbances and is removed. The mean and standard deviation of the correlation coefficient difference between the same data under historical normal operating conditions and two consecutive periods can be determined, for example, by taking... This ensures that over 95% of normal fluctuations are acceptable; if there is no historical data, an empirical value of 0.15 can be used first, and then adjusted based on the operational results.

[0031] S5. Perform causal direction inference on each pair of stable data pairs to determine the causal direction between the two heterogeneous state data pairs, and construct a causal graph based on all causal edges with stable causal directions; calculate the causal stability index of each pair of stable data pairs, and select heterogeneous state data pairs with causal stability indices greater than the preset causal threshold as causal stable data pairs.

[0032] In this embodiment, firstly, for each pair of stable correlation data selected by S4... Causal direction inference was performed within both the main and secondary monitoring periods. The causal direction inference employed the time-series-based Granger causality test, with the following specific steps: Taking the data from the main monitoring period as an example, for stable correlation data pairs Construct two vector autoregressive models: Model 1 (test) ): ; Model 2 (test) ): ; in, : No. Heterogeneous state data at time The monitored values. : No. Heterogeneous state data at time The monitored values. Lag order: Represents the number of past time steps included in the model, determined by the AIC or BIC criterion. : Autoregressive coefficient, reflecting the degree of influence of lagged terms on the current value, estimated by the least squares method. : Random error term, representing random fluctuations that the model fails to explain. :right From 1 to The summation symbol.

[0033] For each model, the lagged terms of the independent variables (i.e., those in Model 1) are calculated using the F-test. and in Model 2 The F-statistic and its corresponding p-value are determined. A preset significance level threshold is set. (For example Based on the p-value and The comparison results determine the direction of causality: If the p-value of the lagged term of the independent variable in Model 1 is... And the p-value of the lagged terms of the independent variable in Model 2 Then the direction of causality is determined as follows: ; If the p-value of the lagged term of the independent variable in Model 2 is... And the p-value of the lagged term in Model 1 Then the direction of causality is determined as follows: ; If the p-values ​​of both models are Then it is determined that there is bidirectional causality. ; If the p-values ​​of both models are If the correlation is stable, then it is determined that there is no causal direction (at this time, although the data pair is stable, it lacks a causal basis and should be removed).

[0034] Similarly, the same causal direction inference is performed independently within each monitoring period to obtain the main causal direction. and secondary causal direction .

[0035] Then, a causal graph is constructed based on all causal edges with stable causal directions. Each heterogeneous state data point is treated as a node, and the determined causal directions are treated as directed edges, thus initially constructing a directed graph. Since feedback loops may exist in real-world systems, a loop removal algorithm (such as topological sorting) is used to transform the directed graph into a directed acyclic graph (DAG), which serves as the final causal graph. The causal graph reflects stable causal transmission paths between variables.

[0036] Next, the causal stability index of each stable data pair is calculated. :

[0037] in, : No. species and first A causal stability index among heterogeneous state data, dimensionless, with a value range of [value missing]. The larger the value, the more consistent the causal direction is between the two monitoring periods. : No. species and first This data set represents the primary causal direction within the main monitoring period, containing one or more causal directions (such as...). , , wait). : No. species and first The set of secondary causal directions for each data pair within a monitoring period. The intersection of sets represents the causal direction shared by two periods. The union of sets represents all causal directions that occur in two cycles. The cardinality of a set is the number of elements in the set.

[0038] and Consider it as a set, for example, if the main period is If the next period is also [missing information], then the set contains that direction; Then the size of the intersection is 1, and the size of the union is 1. If the main period is The next cycle is If the intersection is empty, the union size is 2. If the primary period is bidirectional and the secondary period is unidirectional, then the intersection size is 1 and the union size is 2. Set a preset causal threshold. (e.g., 0.8), filter out The data pairs are causally stable. These data pairs are not only numerically stable, but also exhibit long-term consistency in the causal direction, ensuring that the relationships used for prediction have a solid physical causal basis.

[0039] S6. Based on the sub-data sequence of each state data in the causal stable data pair, and combined with the causal graph, obtain the data change trajectory of the data within the prediction time period.

[0040] In this implementation, prediction is based on a causal graph. For each state data point in a causal stable data pair, its role in the causal graph (dependent variable or effect variable) is first determined. For nodes that are dependent variables (i.e., nodes without upstream parent nodes), future values ​​are independently predicted based on the secondary data series. Prediction methods can employ time series models such as linear regression, exponential smoothing, and ARIMA. Specifically, for the prediction of the dependent variable, the stationarity of the secondary data series is first tested (e.g., the ADF test). If the series is stationary, the ARIMA model is used, and the autoregression order and moving average order are determined using ACF and PACF plots. If the series has a trend term, a difference exponential smoothing model (Holt model) or a linear regression model is used. Model parameters are determined using least squares or maximum likelihood estimation, and the model is selected using the AIC or BIC criteria.

[0041] For a node that is the result variable, its future value is influenced not only by its own history but also by the upstream dependent variable. Based on the predicted value of the upstream dependent variable in the causal graph and the causal functional relationship between the variables, the future value of the result variable is derived.

[0042] Causal relationships can be learned from historical data. For example, for nodes... Its upstream parent node set is A regression model can be established:

[0043] in, : No. Heterogeneous state data at time The monitored values ​​(effect variables).

[0044] A causal function represents the mapping relationship between a dependent variable and an effect variable. It can be a linear function, a nonlinear function, or a physical mechanism equation (such as a thermodynamic equation). The causal relationship... The specific form can be determined through the following steps: First, calculate the Pearson correlation coefficient between the upstream dependent variable and the effect variable. If the absolute value of the correlation coefficient is greater than 0.8, then a linear regression model is used. (in, Nodes in a causal graph The set of upstream parent nodes; For the $i$-th heterogeneous state data in the parent node set at time... The monitored values; The intercept term is dimensionless. For the first The regression coefficients of the parent node variables reflect the influence of the parent variable on the result variable. The degree of direct impact; For the historical lag term of the result variable itself, The maximum lag order; These are autoregressive coefficients, reflecting the influence of the historical values ​​of the result variable on the current value; The random error term represents the random fluctuations that the model fails to explain, and is usually assumed to follow a normal distribution with a mean of 0; model parameters The lag order is estimated using the least squares method. (This can be determined using the AIC or BIC criteria). If the absolute value of the correlation coefficient is less than 0.3, it indicates a weak linear relationship and the possibility of a nonlinear relationship. In this case, a nonlinear model such as Support Vector Regression (SVR) or Random Forest can be used, with the model parameters optimized through cross-validation. If the physical mechanism equations (such as thermodynamic equations) are known... The equation can be directly used, and the symbols in the equation are determined by fitting historical data.

[0045] All nodes in the causal graph The heterogeneous state data of the parent node at time The set of monitoring values.

[0046] Nodes in a cause-and-effect graph The set of upstream parent nodes.

[0047] The historical lag term of the result variable itself is used to consider autocorrelation.

[0048] : Random error term, representing random fluctuations that the model fails to explain.

[0049] The model parameters are obtained through training on historical data. During prediction, the predicted values ​​of the upstream dependent variable are substituted into the model to obtain the desired parameters. The predicted values ​​are then propagated downstream from the root node of the causal graph, eventually yielding the predicted trajectories for all nodes.

[0050] It is worth noting that the application of cause-effect graphs not only improves prediction accuracy, but more importantly, provides fault tracing capabilities: when the predicted value of a certain result variable deviates from the normal range, the root cause variable can be traced back along the cause-effect graph to find the root cause variable that caused the anomaly, thereby providing maintenance personnel with accurate fault location information.

[0051] In summary, the entire industrial IoT operational status prediction method integrating multi-source heterogeneous data firstly, based on the correlation stability screening, further performs causal direction inference on each stable correlation data pair. This accurately identifies the causal relationship between variables, thus upgrading the correlation pairs that can only represent numerical co-changes in traditional methods to causal pairs with clear physical causal meanings. When an abnormal equipment status is detected during subsequent prediction, maintenance personnel can trace back along the constructed causal graph to accurately locate the root cause variable causing the abnormality, providing an intuitive decision-making basis for predictive maintenance. Furthermore, by calculating the causal stability index of each stable correlation data pair and screening out causal stable data pairs whose causal direction remains consistent over two consecutive monitoring periods, it is possible to effectively eliminate false causal pairs that, although numerically stable, have causal directions that reverse with changes in operating conditions. This ensures that the variable relationships used for prediction have a long-term reliable physical causal basis. When equipment aging, changes in operating conditions, or component replacement lead to causal inference, the method can effectively predict the causal relationship between variables. When the causal structure changes, this method can promptly detect such changes through real-time monitoring of causal stability indicators, thereby avoiding prediction bias based on outdated causal relationships. Furthermore, this invention performs hierarchical prediction based on the constructed causal graph. Nodes acting as dependent variables are independently modeled and their future trajectories are predicted. For nodes acting as result variables, the predictions are derived based on the predicted values ​​of upstream dependent variables in the causal graph and the causal function relationships learned from historical data. This causal transmission prediction method fully preserves the true physical interaction paths between variables, avoiding the problem of fragmented collaborative relationships between variables in traditional multivariate independent prediction, making the prediction results more consistent with the inherent mechanisms of equipment operation. Furthermore, when the causal stability indicator of a certain causal edge continuously decreases and falls below a preset threshold, an early warning can be issued that the causal path may undergo structural changes, prompting maintenance personnel to check for potential fault risks in related components. This achieves a shift from single numerical prediction to causal structure monitoring, further enhancing early warning capabilities and maintenance decision support levels.

[0052] The specific implementation principle of the present invention will be explained in detail below with reference to specific application scenarios.

[0053] In practical applications of predicting the operating status of chemical reactors, an industrial IoT platform first collects data streams from various types of sensors deployed on the reactor body and its surrounding environment in real time via a communication interface. This includes heterogeneous status data output from devices such as pressure sensors, cooling water flow meters, agitator motor speed encoders, acoustic emission probes, and gas concentration detectors. The processor then extracts the two consecutive and adjacent 30-minute monitoring cycles preceding the current moment from this data stream, using these as the main monitoring cycles. to minutes) and sub-monitoring cycle ( to (minutes), and set the time period to be predicted to be the next 15 minutes (minutes). to (minutes) to complete the data acquisition and time window division in step S1.

[0054] Subsequently, S1-1 data preprocessing is performed: all sensor data are resampled to a 10-minute interval. In this example, the original data is sampled exactly every 10 minutes, so no interpolation is needed, and the original time-point data can be used directly. The historical theoretical range of the pressure sensor is... MPa, or the historical theoretical minimum value Historical theoretical maximum value Extreme The historical theoretical range of cooling water flow meters is: That is, the historical theoretical minimum value Historical theoretical maximum value Extreme .

[0055] In step S2, the processor retrieves the historical theoretical extreme value database of each state data pre-stored in memory. This is applied to the pressure data sequence within the main monitoring period. The principal relative change is obtained by calculating the difference between adjacent sampling points in sequence. MPa; Divide each relative change by the corresponding range MPa, yielding the normalized principal absolute change: The main fluctuation trend indicator that forms pressure after accumulation Rounded to four decimal places, it is approximately .

[0056] Similarly, the cooling water flow rate data sequence is The relative change was calculated. Normalization (divided by) )have to , cumulatively ,Pick Set the scaling factor. Substitute the values ​​into the formula to calculate the main correlation coefficient between pressure and flow rate:

[0057] In step S3, the processor performs the same logic on the latest data within the next monitoring cycle. The stress data sequence for the next monitoring cycle is as follows: The adjacent difference was calculated. Normalization yields , cumulatively ,Pick The cooling water flow data sequence for each monitoring cycle is as follows: The adjacent difference was calculated. Normalization yields , cumulatively ,Pick Substitute the same The correlation coefficient is obtained as follows:

[0058] Proceed to step S4, where the stable association filtering module performs calculations. Less than the preset change threshold The pressure-flow rate pair was determined to be a stable correlation. Simultaneously, the difference in the correlation coefficient between pressure and acoustic emission intensity data was calculated (assuming the acoustic emission data was calculated as follows). Since the value is greater than 0.15, the pressure-acoustic emission pair is eliminated.

[0059] Next, step S5 is executed. First, causal direction inference is performed on the stable pressure-flow correlation pair. Within the main monitoring period, the pressure sequence is taken. and flow sequence (Time alignment has been completed). Perform Granger causality testing and set the lag order. (Due to the limited number of data points), and a significance level threshold was set. Model building and testing :return To obtain the coefficient The p-value is 0.03, which is less than 0.05, indicating that pressure changes have a significant predictive effect on flow rate. Model construction and validation are then performed. :return To obtain the coefficient The p-value was 0.45, greater than 0.05, indicating that flow rate changes had no significant predictive effect on pressure. Therefore, the causal direction was determined to be [blank] within the main monitoring period. (Pressure is the cause, traffic is the effect).

[0060] During this monitoring period, the pressure sequence Flow sequence Similarly, a Granger causality test was performed ( , ):test , The p-value is 0.02 (less than 0.05); test , The p-value was 0.67 (greater than 0.05). Therefore, the causal direction remained unchanged during this monitoring period. .

[0061] Calculate the causal stability index The set of principal period directions is The secondary cycle direction set is also The intersection size is 1, and the union size is 1. Greater than the preset causal threshold Therefore, pressure-flow was selected as a causal stable data pair.

[0062] Similarly, for the pressure-acoustic emission pair (although the correlation is unstable, we still assume we will test its causal direction, which is omitted here). Ultimately, the pressure-flow rate pair is retained as a causally stable data pair, and the causal graph is... .

[0063] Proceed to step S6, where predictions are made based on the causal graph. Stress is the dependent variable; its future trajectory is predicted first. This involves analyzing the stress subsequence data. Perform linear least squares fitting (using the same method as before) to obtain the rate of change. Using the measured value of 1.30 MPa at the current time $t$ as the initial value, and recursively extrapolating in 5-minute increments, the predicted value for the next 15 minutes is obtained: .

[0064] Traffic flow, as a consequence variable, is influenced by pressure. A causal functional model is established: based on historical data (including primary, secondary, and earlier data), a fitting model is created. In this example, due to the limited data, a simple linear relationship can be assumed. Using sub-period data ( ) and main cycle data ( ) Perform parameter estimation to obtain (The numerical values ​​are for illustrative purposes only). When forecasting, substitute the predicted pressure values ​​into the following: (current ), calculated The subsequent predicted values ​​can be obtained by following this pattern.

[0065] The above prediction process fully considers the causal transmission relationship between variables, ensuring that the prediction results conform to the physical mechanism. When an abnormal increase in traffic is finally predicted, it can be traced back to changes in upstream pressure, providing maintenance personnel with clues to the root cause of the fault.

[0066] It should be noted that the preset parameters Method for determining: and As mentioned above; The threshold is generally set between 0.8 and 1.0, but the specific value can be determined based on the frequency of changes in the causal direction in the historical operating data of the equipment. If the equipment mechanism is stable, a higher value can be used (e.g., 0.9); if there are periodic changes, the value can be appropriately reduced (e.g., 0.7). Significance level threshold. The value is usually set to 0.05 or 0.01, according to statistical conventions.

[0067] like Figure 4 As shown, an industrial IoT operation status prediction system integrating multi-source heterogeneous data is also provided. The system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The management platform includes: The data acquisition module is used to acquire two consecutive monitoring cycles for the target device in the industrial Internet of Things before the current time, and use them as the main monitoring cycle and the secondary monitoring cycle. It also acquires various heterogeneous state data during the operation of the target device and acquires the time period to be predicted. The data preprocessing module is used to perform time-aligned preprocessing on the various heterogeneous state data so that all state data have a unified sampling time point; The main correlation analysis module is used to obtain the main data sequence of various heterogeneous state data within the main monitoring period, calculate the main fluctuation trend index corresponding to various heterogeneous state data based on the main data sequence of various heterogeneous state data, and calculate the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend index of various heterogeneous state data. The secondary correlation analysis module is used to obtain the secondary data sequence of various heterogeneous state data within the secondary monitoring period, calculate the secondary fluctuation trend index corresponding to various heterogeneous state data based on the secondary data sequence of various heterogeneous state data, and calculate the secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend index of various heterogeneous state data. The stable association filtering module is used to calculate the absolute difference between the primary association coefficient and the secondary association coefficient of each pair of heterogeneous state data, and filter out heterogeneous state data pairs whose difference is less than a preset change threshold as stable association data pairs. The causal inference module is used to infer the causal direction for each pair of stable data pairs, determine the causal direction between two heterogeneous state data, and construct a causal graph based on all causal edges with stable causal directions; calculate the causal stability index for each pair of stable data pairs, and select heterogeneous state data pairs with causal stability indices greater than a preset causal threshold as causal stable data pairs. The causal prediction module is used to obtain the data change trajectory of each state data in the causal stable data pair within the prediction time period based on the sub-data sequence of each state data in the causal stable data pair and the causal graph. Specifically, for any heterogeneous state data, if it changes as a dependent variable, the changes of all directly or indirectly related result variables are derived based on the causal graph.

[0068] In one implementation, the data preprocessing module is further configured to: for each type of heterogeneous state data, resample the data to a unified reference time series using a linear interpolation method based on its original sampling timestamp, so that all state data are synchronized in time.

[0069] In one implementation, the main correlation analysis module is further configured to: calculate the difference between two adjacent monitoring values ​​in the main data sequence according to the main data sequence of the state data and use it as the main relative change amount; obtain the historical theoretical maximum value and historical theoretical minimum value corresponding to the state data; calculate the main absolute change amount corresponding to the main relative change amount based on the main relative change amount, the historical theoretical maximum value and the historical theoretical minimum value; and accumulate each main absolute change amount in the main data sequence of the state data and use it as the main fluctuation trend indicator of the state data.

[0070] In this embodiment, it should be noted that the specific method of performing the above-mentioned industrial IoT operation status prediction system that integrates multi-source heterogeneous data has been described in detail in the embodiments of the industrial IoT operation status prediction method that integrates multi-source heterogeneous data, and will not be elaborated here.

[0071] It should also be noted that the entire industrial IoT operation status prediction system that integrates multi-source heterogeneous data can be applied to the optimized industrial IoT. Figure 4 This is a schematic diagram of the composition of the industrial Internet of Things (IoT) operation status prediction system that integrates multi-source heterogeneous data according to the present invention. Figure 5 This is a schematic diagram illustrating the optimized industrial Internet of Things (IIoT) involved in this invention. (See diagram below.) Figure 4 and Figure 5 As shown, the optimized Industrial Internet of Things (IIoT) includes a user platform, a service platform, a management platform, a sensor network platform, and an object platform that establish communication in sequence. The user platform is configured to provide front-end services to users; users obtain the necessary perception service information through the user platform, process the perception service information, and transform it into user perception information; users analyze the user perception information and make corresponding decisions based on their own wishes, and transform the user perception information into user control information through the corresponding information system and send it to the service platform, thereby demonstrating the user's corresponding service needs and wishes.

[0072] The physical entities of the user platform include various user terminals, such as mobile phones, computers, and dedicated terminals, which provide user services through integration with user information system software.

[0073] The service platform is configured as an API server or other server used to establish communication between the management platform and the user platform to achieve corresponding functions; the physical entity of the service platform includes various servers.

[0074] The management platform is configured to perform at least one of the following: device operation status monitoring and management, data monitoring and management, device parameter management, and lifecycle management; the management platform is the overall operation platform for the Internet of Things, which may include various management sub-platforms, with different management sub-platforms performing different management tasks; the physical entities of the management platform include various servers.

[0075] The sensor network platform is configured to perform at least one of the following functions: network management, command management, device status management, data protocol management, data parsing, data classification, data transmission monitoring, and data transmission security management. The sensor network platform provides functions such as data communication, transmission, parsing, identification, and classification, avoiding the direct aggregation of data from various object platforms onto the management platform, which would otherwise result in data redundancy and low data processing efficiency. The physical entities of the object platforms include various gateways, edge computing devices, etc.

[0076] The object platform is configured to perform specific production control, detection, measurement and other production tasks; the physical entities of the production objects include various production equipment, sensors and so on.

[0077] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0078] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0079] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for predicting the operational status of an industrial Internet of Things (IoT) by integrating multi-source heterogeneous data, characterized in that, include: The system acquires two consecutive monitoring cycles for the target device in the Industrial Internet of Things up to the current moment, which are used as the main monitoring cycle and the secondary monitoring cycle. It also acquires various heterogeneous state data during the operation of the target device and obtains the time period to be predicted. Time alignment preprocessing is performed on various heterogeneous state data to ensure that all state data have a unified sampling time point; Obtain the master data sequence of various heterogeneous state data within the main monitoring period, and obtain the main fluctuation trend index corresponding to various heterogeneous state data based on the master data sequence of various heterogeneous state data, and obtain the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend index of various heterogeneous state data. Obtain the secondary data sequence of various heterogeneous state data within the secondary monitoring period, and obtain the secondary fluctuation trend index corresponding to various heterogeneous state data based on the secondary data sequence of various heterogeneous state data, and obtain the secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend index of various heterogeneous state data. Select the two heterogeneous state data whose difference between the primary correlation coefficient and the secondary correlation coefficient is less than a preset change threshold and use them as stable correlation data pairs; For each pair of stable data pairs, causal direction inference is performed to determine the causal direction between two heterogeneous state data, and a causal graph is constructed based on all causal edges with stable causal directions. Calculate the causal stability index for each stable data pair, and select heterogeneous state data pairs with causal stability indices greater than a preset causal threshold as causal stable data pairs. Based on the sub-data sequence of each state data in the causal stable data pair, and combined with the causal graph, the data change trajectory within the prediction time period is obtained; wherein, for any heterogeneous state data, if it changes as a dependent variable, the changes of all directly or indirectly related result variables are derived based on the causal graph.

2. The industrial IoT operation status prediction method according to claim 1, characterized in that, The time-aligned preprocessing of multiple heterogeneous state data includes: For each type of heterogeneous state data, a linear interpolation method is used to resample the data to a unified reference time series based on its original sampling timestamp, so that all state data are synchronized in time.

3. The industrial IoT operation status prediction method based on multi-source heterogeneous data according to claim 1, characterized in that, The process of obtaining the main fluctuation trend indicators corresponding to various heterogeneous state data based on the main data sequences of various heterogeneous state data includes: Based on the master data sequence of the status data, the difference between two adjacent monitoring values ​​in the master data sequence is calculated sequentially and used as the main relative change. The historical theoretical maximum and minimum values ​​corresponding to the status data are obtained, and the main absolute change corresponding to the main relative change is calculated based on the main relative change, the historical theoretical maximum and the historical theoretical minimum values. The main absolute changes in the main data sequence of the state data are accumulated and used as the main fluctuation trend indicator of the state data.

4. The industrial IoT operation status prediction method based on multi-source heterogeneous data according to claim 1, characterized in that, The method of obtaining the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend index of various heterogeneous state data includes: The primary correlation coefficient is calculated based on the degree of difference between two primary fluctuation trend indicators and a preset scaling factor. The value of the primary correlation coefficient is inversely proportional to the degree of difference and ranges from 0 to 1.

5. The industrial IoT operation status prediction method based on multi-source heterogeneous data according to claim 1, characterized in that, The step of inferring the causal direction for each pair of stable correlated data to determine the causal direction between two heterogeneous state data includes: For each pair of stable associated data, construct a vector autoregression model with the first type of heterogeneous state data as the dependent variable and the second type of heterogeneous state data as the independent variable, and construct a vector autoregression model with the second type of heterogeneous state data as the dependent variable and the first type of heterogeneous state data as the independent variable. Calculate the F-statistic and its corresponding p-value for the lagged terms of the independent variables in each model using the F-test; If the p-value of the lagged term of the independent variable in the first model is less than the preset significance level threshold and the p-value of the lagged term of the independent variable in the second model is not less than the significance level threshold, then the causal direction is determined to be from the first heterogeneous state data to the second heterogeneous state data. If the p-value of the lagged term of the independent variable in the second model is less than the threshold and the p-value of the lagged term of the independent variable in the first model is not less than the significance level threshold, then the causal direction is determined to be from the second heterogeneous state data to the first heterogeneous state data. If the p-value in both models is less than the threshold, then bidirectional causality is determined to exist; if the p-value in both models is not less than the threshold, then no causal direction is determined to exist.

6. The industrial IoT operation status prediction method according to claim 1, characterized in that, The calculation of the causal stability index for each stable data pair, and the selection of heterogeneous state data pairs with causal stability indices greater than a preset causal threshold as causal stable data pairs, includes: For each stable data pair, determine its primary causal direction in the main monitoring period and its secondary causal direction in the secondary monitoring period. The causal stability index is determined based on the degree of consistency between the primary causal direction and the secondary causal direction. The causal stability index reflects the proportion of causal directions that are the same in two periods. The larger the value, the more stable the causal direction. Stable data pairs with causal stability indices greater than a preset causal threshold are selected as causal stable data pairs.

7. The industrial IoT operation status prediction method based on multi-source heterogeneous data according to claim 1, characterized in that, Based on the sub-data sequences of each state data in the causal stable data pair, and combined with the causal graph, the data change trajectory within the prediction time period is obtained, including: For each state data in a causal stable data pair, if it is used as the dependent variable, its future trajectory can be obtained by fitting a prediction model based on the second data sequence. If it is treated as an effect variable, its future trajectory can be deduced based on the predicted value of the upstream dependent variable in the causal graph and the causal functional relationship between the variables. Repeat this process until the predicted values ​​for all nodes in the causal graph are obtained.

8. An industrial IoT operation status prediction system that integrates multi-source heterogeneous data, characterized in that, The system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The management platform includes: The data acquisition module is used to acquire two consecutive monitoring cycles for the target device in the industrial Internet of Things before the current time, and use them as the main monitoring cycle and the secondary monitoring cycle. It also acquires various heterogeneous state data during the operation of the target device and acquires the time period to be predicted. The data preprocessing module is used to perform time-aligned preprocessing on the various heterogeneous state data so that all state data have a unified sampling time point; The main correlation analysis module is used to obtain the main data sequence of various heterogeneous state data within the main monitoring period, calculate the main fluctuation trend index corresponding to various heterogeneous state data based on the main data sequence of various heterogeneous state data, and calculate the main correlation coefficient between any two heterogeneous state data based on the main fluctuation trend index of various heterogeneous state data. The secondary correlation analysis module is used to obtain the secondary data sequence of various heterogeneous state data within the secondary monitoring period, calculate the secondary fluctuation trend index corresponding to various heterogeneous state data based on the secondary data sequence of various heterogeneous state data, and calculate the secondary correlation coefficient between any two heterogeneous state data based on the secondary fluctuation trend index of various heterogeneous state data. The stable association filtering module is used to calculate the absolute difference between the primary association coefficient and the secondary association coefficient of each pair of heterogeneous state data, and filter out heterogeneous state data pairs whose difference is less than a preset change threshold as stable association data pairs. The causal inference module is used to infer the causal direction for each pair of stable data pairs, determine the causal direction between two heterogeneous state data, and construct a causal graph based on all causal edges with stable causal directions; calculate the causal stability index for each pair of stable data pairs, and select heterogeneous state data pairs with causal stability indices greater than a preset causal threshold as causal stable data pairs. The causal prediction module is used to obtain the data change trajectory of each state data in the causal stable data pair within the prediction time period based on the sub-data sequence of each state data in the causal stable data pair and the causal graph. Specifically, for any heterogeneous state data, if it changes as a dependent variable, the changes of all directly or indirectly related result variables are derived based on the causal graph.

9. The industrial IoT operation status prediction system that integrates multi-source heterogeneous data according to claim 8, characterized in that, The data preprocessing module is also used for: For each type of heterogeneous state data, a linear interpolation method is used to resample the data to a unified reference time series based on its original sampling timestamp, so that all state data are synchronized in time.

10. The industrial IoT operation status prediction system that integrates multi-source heterogeneous data according to claim 8, characterized in that, The master association analysis module is also used for: Based on the master data sequence of the status data, the difference between two adjacent monitoring values ​​in the master data sequence is calculated sequentially and used as the main relative change. The historical theoretical maximum and minimum values ​​corresponding to the status data are obtained, and the main absolute change corresponding to the main relative change is calculated based on the main relative change, the historical theoretical maximum and the historical theoretical minimum values. The main absolute changes in the main data sequence of the state data are accumulated and used as the main fluctuation trend indicator of the state data.

Images