Fusion detection system and fusion detection method of distributed devices in micro-grid

Abstract

A fusion detection system includes the following components: a data classification unit for receiving multiple sensing data from multiple distributed devices and classifying the sensing data into a first type, a second type, a third type, and a fourth type; an anomaly detection unit for performing anomaly detection model calculations based on the first type and second type of sensing data to generate anomaly detection prediction results; a predictive maintenance unit for performing predictive maintenance model calculations based on the first type, second type, third type, and fourth type of sensing data and the anomaly detection prediction results to generate predictive maintenance prediction results; and a cost optimization unit for performing cost optimization model calculations based on the predictive maintenance prediction results to generate cost optimization decisions.

Description

Technical Field

[0001] The present disclosure relates to an anomaly detection mechanism, and more particularly to a fusion detection system and a fusion detection method for distributed devices applied to a microgrid. Prior Art

[0002] With the evolution of emerging energy technologies, various types of energy systems have been developed, including home photovoltaics (PV) systems, PV farms, wind farms, and battery energy storage systems (BESS). When the above different types of energy systems operate, it is necessary to perform anomaly detection and maintenance on the devices (such as distributed devices) in the energy system.

[0003] However, the devices in the energy system include various types. It is a huge challenge for the industry in this field to accurately and instantaneously detect the abnormal states of various types of devices and be able to effectively maintain the abnormal devices.

[0004] In view of the above issues, an improved anomaly detection mechanism is needed that can perform instant anomaly detection and optimized predictive maintenance on the devices in the energy system. Summary of the Invention

[0005] According to one aspect of the present disclosure, a fusion detection system is provided, including the following components. A data processing module for performing preprocessing on a sensed data set, the sensed data set including sensed data of a plurality of distributed devices. A data classification unit of the data processing module for receiving the sensed data and classifying the sensed data into a first type, a second type, a third type, and a fourth type, wherein the sensed data is captured by a data acquisition device. An anomaly detection unit for performing operations of an anomaly detection model according to the sensed data of the first type and the second type to generate an anomaly detection prediction result. A predictive maintenance unit for performing operations of a predictive maintenance model according to the sensed data of the first type, the second type, the third type, and the fourth type and the anomaly detection prediction result to generate a predictive maintenance prediction result, and the predictive maintenance prediction result is used to plan a predictive maintenance strategy. A cost optimization unit for performing operations of a cost optimization model according to the predictive maintenance prediction result to generate a cost optimization decision.

[0006] According to another aspect of the present disclosure, a fusion detection method is provided, including the following steps. A plurality of sensing data of a plurality of distributed devices are captured by a data capture device, and the sensing data forms a sensing data set. Preprocessing is performed on the sensing data set by a data processing module. The sensing data of the plurality of distributed devices is received by the data classification unit of the data processing module, and the sensing data is classified into a first type, a second type, a third type, and a fourth type. An abnormal detection unit performs an operation of an abnormal detection model according to the sensing data of the first type and the second type to generate an abnormal detection prediction result. A prediction maintenance unit performs an operation of a prediction maintenance model according to the sensing data of the first type, the second type, the third type, and the fourth type and the abnormal detection prediction result to generate a prediction maintenance prediction result, and the prediction maintenance prediction result is used to plan a prediction maintenance strategy. A cost optimization unit performs an operation of a cost optimization model according to the prediction maintenance prediction result to generate a cost optimization decision.

[0007] Other aspects and advantages of the present disclosure can be seen by reading the following drawings, detailed descriptions, and claims. Brief Description of the Drawings

[0008] FIG. 1 illustrates the application of the fusion detection system of the present disclosure. FIG. 2 illustrates another example of the application of the fusion detection system of the present disclosure. FIG. 3 illustrates a block diagram of the fusion detection system according to an embodiment of the present disclosure. FIG. 4A illustrates the data structure used when the abnormal detection unit performs model operations. FIGS. 4B and 4C illustrate the operation flow of the abnormal detection unit. FIG. 5A illustrates the data structure used when the prediction maintenance unit performs model operations. FIGS. 5B and 5C illustrate the operation flow of the prediction maintenance unit. Embodiments

[0009] The technical terms in this specification refer to the customary terms in the technical field. If some terms in this specification are explained or defined, the explanations of these terms shall be based on the explanations or definitions in this specification. Each embodiment of the present disclosure has one or more technical features. On the premise of possible implementation, those skilled in the art can selectively implement some or all of the technical features in any embodiment, or selectively combine some or all of the technical features in these embodiments.

[0010] Please refer to FIG. 1, which illustrates the application of the fusion detection system 1000 of the present disclosure. The fusion detection system 1000 can be applied to a microgrid, and the microgrid includes distributed devices 10. The fusion detection system 1000 is used to detect the abnormal state of each of the distributed devices 10 and to plan a predictive maintenance strategy. The distributed devices 10 include, for example, solar panels 11 and wind turbines 12, and so on.

[0011] The fusion detection system 1000 receives a sensing dataset SD, which is provided by a data acquisition device 20. The data acquisition device 20 includes, for example, a smart meter 21, a temperature sensor 22, a camera 23, an infrared sensor 24, a sensor of a robot 25, a sensor of an unmanned aerial vehicle 26, a current sensor 27, a voltage sensor 28, and a weather sensor 29, and so on. In one example, the data acquisition device 20 can be disposed on the distributed device 10, and the data acquisition device 20 is used to acquire the sensing data of the distributed device 10. For example: the sensing data acquired by the camera 23 is a visible image, and the sensing data acquired by the infrared sensor 24 is a thermal image. The multiple sensing data acquired by the data acquisition device 20 can form the sensing dataset SD.

[0012] In operation, the fusion detection system 1000 can be integrated into the Internet of Things (IoT) architecture 2000, and both the distributed device 10 and the data acquisition device 20 are part of the IoT architecture 2000. In addition to the distributed device 10 and the data acquisition device 20, the IoT architecture 2000 further includes an edge gateway 31, an edge server 33, and a cloud computing platform 40. The edge server 33 includes an edge database 32, and the fusion detection system 1000 can be installed or disposed on the edge server 33. For example, the fusion detection system 1000 can be implemented by a hardware component or a software program in the edge server 33. The data acquisition device 20 transmits the sensing dataset SD to the edge server 33 via the edge gateway 31, and the sensing dataset SD can be stored in the edge database 32 of the edge server 33.

[0013] More specifically, the fusion detection system 1000 at least includes a data processing module 150, an anomaly detection unit 400, and a predictive maintenance unit 500. In the embodiment of FIG. 1, the data processing module 150 can receive the sensing dataset SD from the edge gateway 31. And, the sensing dataset SD can be stored in the edge database 32 of the edge server 33 via the data processing module 150.

[0014] The data processing module 150 performs preprocessing on the sensed dataset SD, such as data processing transformation and data mining. Among them, the data processing transformation may include filtering processing, transformation processing, segmentation processing, and compression processing. The preprocessed sensed dataset SD is transmitted to the edge database 32 for storage. The anomaly detection unit 400 performs anomaly detection (AD) based on the preprocessed sensed dataset SD, and the predictive maintenance unit 500 performs predictive maintenance (PM) based on the preprocessed sensed dataset SD to generate device information D_if. In addition, the fusion detection system 1000 further includes a cost optimization unit (not shown in Figure 1) to perform model operations for cost optimization to generate a cost optimization decision CO_d.

[0015] The device information D_if includes the device status, device location, and device type of each of the distributed devices 10. Among them, the device status includes the abnormal status A_s. And the cost optimization decision CO_d includes the decision of anomaly detection and the decision of predictive maintenance. Among them, the decision of predictive maintenance includes the optimal maintenance schedule O_M_s. And the decision of anomaly detection includes warnings and device health indicators (DHI). The device health indicators include, for example, the probabilistic anomaly score (PAS), mean time between failures (MTBF), failure rate (FR), and remaining useful life (RUL) of each of the distributed devices 10.

[0016] The cloud computing platform 40 receives the device information D_if and the cost optimization decision CO_d from the edge server 33. The cloud computing platform 40 includes a message mediation unit 41, a data streaming unit 42, and a cloud database 43. The device information D_if and the cost optimization decision CO_d are stored in the cloud database 43 after being processed by the message mediation unit 41 and the data streaming unit 42. And the device information D_if is transmitted to the terminal device 50.

[0017] The terminal device 50 includes, for example, a head mounted device (HMD) 51, a smart phone 52, and a display 53. The terminal device 50 can display device information D_if to present the device information D_if to the user 60. The user 60 is, for example, an operator of a microgrid.

[0018] Please refer to FIG. 2, which illustrates another example of the application of the fusion detection system 1000 of the present disclosure. The example in FIG. 2 is that the head mounted device 51 of the user 60 captures the sensing data of the distributed device 10, which is different from the example in FIG. 1 (in the example of FIG. 1, the data acquisition device 20 for capturing the sensing data is provided on the distributed device 10).

[0019] More specifically, in this embodiment, each of the distributed devices 10 has a visualized device marker. Correspondingly, the head mounted device 51 is provided with a built-in camera or a built-in infrared sensor. The user 60 observes the respective device markers of the distributed devices 10 via the head mounted device 51, and the built-in camera or the built-in infrared sensor of the head mounted device 51 can capture images of the respective device markers of the distributed devices 10. A plurality of image data obtained by the head mounted device 51 form a sensing data set SD.

[0020] In the embodiment of FIG. 2, the sensing data set SD can be transmitted to the edge database 32 via the edge gateway 31. The fusion detection system 1000 (which is installed or set up on the edge server 33) can obtain the sensing data set SD from the edge database 32, or directly receive the sensing data set SD from the edge gateway 31. The fusion detection system 1000 performs model operations for anomaly detection and predictive maintenance based on the sensing data set SD to generate device information D_if and a cost optimization decision CO_d. The device information D_if is transmitted to the head mounted device 51. Then, the user 60 reads the device information D_if via the head mounted device 51.

[0021] Please refer to FIG. 3, which shows a block diagram of the fusion detection system 1000 according to an embodiment of the present disclosure. The fusion detection system 1000 includes a data processing module 150, an anomaly detection unit 400, a predictive maintenance unit 500, a cost optimization unit 600, a cycle analysis unit 700, and an update unit 800. Moreover, the data processing module 150 includes a data classification unit 100, a signal data processing unit 210, a signal data fusion unit 220, an image data processing unit 310, and an image data fusion unit 320. In one example, the fusion detection system 1000 may be application software installed in the edge server 33 of FIG. 2, and the above-mentioned units may be sub-program modules of the fusion detection system 1000. In another example, the fusion detection system 1000 may be a hardware circuit in the edge server 33 of FIG. 2, such as a microprocessor, an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA), and the above-mentioned units may be sub-circuits of the fusion detection system 1000.

[0022] The data classification unit 100 is communicatively coupled to the edge database 32 of FIG. 2. The data classification unit 100 receives a sensed data set SD from the edge database 32. As described above, the sensed data set SD is composed of the sensed data of each of the distributed devices 10 of the microgrid. The data classification unit 100 classifies these sensed data included in the sensed data set SD into "RDCS" type, "LRDCS" type, "RDI" type, and "DI" type, as shown in Table 1. Table 1 Data acquisition source Type of sensed data Distributed device Microgrid system Environment Fault type Current sensor RDCS ü ü - High current, low current Voltage sensor RDCS ü ü - High voltage, low voltage Temperature sensor RDCS ü - ü Overheat, Heat dissipation Smart meter RDCS ü ü - Voltage sag, high voltage, frequency deviation, power quality monitoring Weather sensor LRDCS - - ü Humidity, Wind speed, Temperature Camera (visible image) RDI ü ü ü Loose and corroded wires, Environmental factors, security vulnerabilities Infrared sensor (infrared image) DI ü ü ü Hot spot

[0023] More specifically, the sensing data of the "RDCS" type is data of real-time, dynamic, and continuous signal response, which can be obtained by the current sensor 27, voltage sensor 28, temperature sensor 22, or smart meter 21 in the data acquisition device 20. The fault type can be determined from the sensing data of the "RDCS" type as excessive current or low current, excessive voltage or low voltage, overheating or heat dissipation, frequency deviation, and so on.

[0024] The sensing data of the "LRDCS" type is data of low frequency, real-time, dynamic, and continuous signal response, which can be obtained by a weather sensor. The fault type can be determined from the sensing data of the "LRDCS" type as humidity, wind speed, or temperature, and so on.

[0025] The sensing data of the "RDI" type is data of real-time and dynamic image, which can be obtained by the camera 23 in the data acquisition device 20. The fault type can be determined from the sensing data of the "RDI" type as loose connections and corroded wires, and so on.

[0026] The sensing data of the "DI" type is data of dynamic image, which can be obtained by the infrared sensor 24 in the data acquisition device 20. The fault type can be determined from the sensing data of the "DI" type as hot spots.

[0027] After classification by the data classification unit 100, the sensing data of the "RDCS" and "LRDCS" types is transmitted to the signal data processing unit 210 for signal processing such as fast Fourier transform, wavelet transform (WT), Kalman filter, or auto-regressive integrated moving average (ARIMA). The signal processing performed by the signal data processing unit 210 can be referred to as "first signal processing". The sensing data of the "RDCS" and "LRDCS" types after signal processing (marked as RDCS' and LRDCS' in Figure 3) is transmitted to the signal data fusion unit 220.

[0028] The signal data fusion unit 220 performs fusion processing such as Intensity-Hue-Saturation (IHS), Principal Component Analysis (PCA), or pyramid algorithm. The fusion processing performed by the signal data fusion unit 220 can be referred to as "first fusion processing". The sensed data of the "RDCS" and "LRDCS" types after the fusion processing (marked as "RDCS" and "LRDCS" in Figure 3) is transmitted to the anomaly detection unit 400 and the predictive maintenance unit 500.

[0029] On the other hand, the classified data of the "RDI" and "DI" types by the data classification unit 100 is transmitted to the image data processing unit 310 for image processing such as image filtering, noise reduction, image normalization, image segmentation, or feature extraction. The image processing performed by the image data processing unit 310 can be referred to as "first image processing". The sensed data of the "RDI" and "DI" types after the image processing (marked as "RDI'" and "DI'" in Figure 3) is transmitted to the image data fusion unit 320.

[0030] The image data fusion unit 320 performs spatial domain fusion processing and transform domain fusion processing such as principal component analysis, weighted average method, Discrete Wavelet Transform (DWT), Laplacian Pyramids (LP), or Gradient Pyramids (GP). The fusion processing performed by the image data fusion unit 320 can be referred to as "second fusion processing". The sensed data of the "RDI" and "DI" types after the fusion processing (marked as "RDI"" and "DI" in Figure 3) is transmitted to the predictive maintenance unit 500.

[0031] The anomaly detection unit 400 performs model operations according to the anomaly detection model M1 to perform operations such as Residual Network (ResNet), Visual Geometry Group Network (VGG_Net), K-Nearest Neighbors (KNN), or Support Vector Machines (SVM) on the fused "RDCS" data and "LRDCS" data (marked as "RDCS" and "LRDCS" in Figure 3) to generate an anomaly detection prediction result A_p. The anomaly detection prediction result A_p is transmitted to the predictive maintenance unit 500 according to the period T1, and the period T1 is, for example, 1 hour.

[0032] The predictive maintenance unit 500 performs model operations according to the predictive maintenance model M2 and performs operations such as Residual Network, KNN, Support Vector Machines, Long Short-Term Memory (LSTM), Q-Learning, or Deep Deterministic Policy Gradient (DDPG) on the fused "RDCS", "LRDCS", "RDI", and "DI" data (marked as "RDCS", "LRDCS", "RDI", and "DI" in Figure 3) according to the anomaly detection prediction result A_p to generate a predictive maintenance prediction result P_p. Then, the predictive maintenance prediction result P_p is transmitted to the cost optimization unit 600.

[0033] The cost optimization unit 600 performs model operations according to the cost optimization model M3 to generate a cost optimization decision CO_d and device information D_if. Then, the cost optimization decision CO_d and the anomaly detection prediction result A_p are transmitted to the period analysis unit 700.

[0034] The period analysis unit 700 is used to analyze the time series periods T1, T2, and T3. Among them, the period T1 is, for example, 1 hour, the period T2 is, for example, 0.1 hour, and the period T3 is, for example, 24 hours.

[0035] The period analysis unit 700 updates the sensing data set SD according to the period T2 in response to the abnormal detection prediction result A_p. For example, an updated sensing data set SD is obtained from the data acquisition device 20 in FIG. 1 at every interval of the period T2 (i.e., 0.1 hour). Further, the period analysis unit 700 controls the update unit 800 according to the period T3 in response to the abnormal detection prediction result A_p to update the sensing data set SD and update the abnormal detection model M1 (i.e., update the abnormal detection model M1 at every interval of the period T3).

[0036] On the other hand, the period analysis unit 700 updates the sensing data set SD according to the period T1 in response to the cost optimization decision CO_d. For example, an updated sensing data set SD is obtained from the data acquisition device 20 in FIG. 1 at every interval of the period T1 (i.e., 1 hour). Further, the period analysis unit 700 controls the update unit 800 according to the period T3 in response to the cost optimization decision CO_d to update the sensing data set SD and update the predictive maintenance model M2 and the cost optimization model M3 (i.e., update the predictive maintenance model M2 and the cost optimization model M3 at every interval of the period T3).

[0037] Details of the model operations performed by the abnormal detection unit 400 and the predictive maintenance unit 500 are described below in conjunction with FIGS. 4A to 4C and FIGS. 5A to 5C. Please first refer to FIG. 4A, which shows the data structure used when the abnormal detection unit 400 performs model operations. The data acquisition source src1 of the abnormal detection unit 400 may be the data acquisition device 20 in FIG. 1. The data acquisition source src1 includes, for example, a smart meter 21, a temperature sensor 22, a camera 23, a current sensor 27, a voltage sensor 28, etc. (the data acquisition source src1 in FIG. 4A is also equivalent to the data acquisition source shown in Table 1). The data acquisition source src1 generates input data in_M1 to be provided to the abnormal detection model M1. The input data in_M1 includes, for example, current i1, voltage i2, power consumption i3, usage time i4, power frequency i5, power quality i6, visible image i7, infrared image i8, and temperature i9, etc.

[0038] The anomaly detection model M1 receives the input data in_M1 to perform model operations. More specifically, the anomaly detection model M1 performs the following operations based on machine learning: feature selection m1, data fusion m2, and anomaly detection m3. Among them, the operation of feature selection m1 is to establish and select useful features according to the input data in_M1, and these useful features can be used as input variables of the anomaly detection model M1. For example, the feature selection m1 can be assisted by a user with professional experience to distinguish, according to experience, the features in the input data in_M1 that affect the output data out_M1 of the anomaly detection model M1, and then select and extract these features. Moreover, the operation of data fusion m2 can be performed by the signal data fusion unit 220 and the image data fusion unit 320 in FIG. 1. Furthermore, the operation of anomaly detection m3 includes model training and model prediction (i.e., the operations in the training stage and the prediction stage of the anomaly detection model M1). After performing feature selection m1, data fusion m2, and anomaly detection m3, the anomaly detection model M1 can generate the output data out_M1. The output data out_M1 is the anomaly detection prediction result A_p in FIG. 3, which includes the anomaly status A_s of the distributed device 10.

[0039] Next, please refer to FIGS. 4B and 4C, which illustrate the operation flow of the anomaly detection unit 400. The operation flow of the anomaly detection unit 400 is generally as follows: Based on data-driven machine learning, continuously monitor the sensing data of the distributed device 10 of the microgrid, and then immediately detect the anomaly status of the distributed device 10. Steps S400 to S408 are the preprocessing of the anomaly detection model M1. First, in step S400, the original input data in_M1 is obtained from the data acquisition source src1 (such as the camera 23 and the current sensor 27, etc.). Then, in step S402, preprocessing is performed on the obtained original input data in_M1 (the preprocessing can be performed by the data processing module 150 in FIGS. 1 and 3).

[0040] Then, in step S404, the operation of feature selection m1 in FIG. 4A is performed. Then, in step S406, the operation of data fusion m2 in FIG. 4A is performed (the data fusion can be performed by the signal data fusion unit 220 and the image data fusion unit 320 in FIG. 3). Then, in step S408, data separation is performed to facilitate subsequent training, verification, and testing of the anomaly detection model M1.

[0041] After step S408, two branch processes are then executed. The first branch process is the training phase of the anomaly detection model M1, including steps S410 to S424. The second branch process is the actual execution phase (i.e., the prediction phase) of the anomaly detection model M1, including steps S426 to S434.

[0042] When the first branch process is executed, first, in step S410, the parameters of the anomaly detection model M1 (i.e., the hyper - parameters) are adjusted. Then, in step S412, training is performed according to the fused input data in_M1. At the same time, in step S414, verification is performed according to the fused input data in_M1. Then, in step S416, the anomaly detection model M1 actually performs anomaly detection (i.e., the operation of the anomaly detection m3 in Figure 4A). Then, in step S418, the anomaly detection model M1 generates an anomaly detection prediction result A_p.

[0043] Then, in step S420, the training results of the anomaly detection model M1 are evaluated. At the same time, in step S422, the verification results of the anomaly detection model M1 are evaluated. Then, in step S424, it is judged whether the anomaly detection model M1 meets the predetermined accuracy according to the evaluation of the training results and the evaluation of the verification results. If the judgment result in step S424 is "no", then return to step S410 or step S416.

[0044] If the judgment result in step S424 is "yes" (indicating that the predetermined accuracy is met), then step S426 of the second branch process is executed: testing is performed according to the fused input data in_M1. Then, in step S428, the anomaly detection model M1 completes training. Then, in step S430, the anomaly detection model M1 actually performs anomaly detection to generate an anomaly detection prediction result A_p (which includes an anomaly state A_s).

[0045] After step S430, steps S432 and S434 can be executed simultaneously. In step S432, the test results of the anomaly detection model M1 are evaluated, and the evaluation criteria include, for example, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE), etc. On the other hand, in step S434, the anomaly detection prediction result A_p is presented to the user 60 via the terminal device 50 in Figure 1.

[0046] Next, refer to FIG. 5A, which illustrates the data structure used when the predictive maintenance unit 500 performs model operations. The data structure shown in FIG. 5A is similar to the data structure of FIG. 4A, with the differences being that: compared to the data acquisition source src1 of FIG. 4A, the data acquisition source src2 of the predictive maintenance unit 500 further includes the anomaly detection prediction result A_p. And, compared to the input data in_M1 of FIG. 4A, the input data in_M2 of the predictive maintenance model M2 further includes the anomaly status A_s. That is, in addition to receiving the sensed data generated by the data acquisition source src2, the predictive maintenance model M2 further receives the anomaly detection prediction result A_p (which includes the anomaly status A_s) generated by the anomaly detection model M1.

[0047] The predictive maintenance model M2 performs feature selection m1b, data fusion m2, and anomaly detection m3b based on the input data in_M2. The difference between the feature selection m1b performed by the predictive maintenance model M2 and the feature selection m1 performed by the anomaly detection model M1 is that: the feature selection m1b performed by the predictive maintenance model M2 further converts the input data in_M2 into a new representation form, and then extracts the basic attributes or features of the sensed data relevant to the target task or target problem. The input data in_M2 can be used as the input variable of a machine learning algorithm or statistical modeling after being converted. And, the anomaly detection m3b performed by the predictive maintenance model M2 is a simplified operation of the anomaly detection m3 in FIG. 4A. The predictive maintenance model M2 generates output data out_M2, which is the predictive maintenance prediction result P_p of FIG. 3 and includes the maintenance schedule M_s for each of the distributed devices 10. According to the maintenance schedule M_s, the user can perform different types of maintenance on the distributed devices 10 of the microgrid. For example, the maintenance schedule M_s includes the following types of maintenance: preventive maintenance (PM), corrective maintenance (CM), replace device (R), and downtime (D). Among them, the "replace device" type of maintenance replaces the old faulty device with a new device. The "downtime" type of maintenance forcibly stops the operation of the faulty device.

[0048] Next, please refer to FIGS. 5B and 5C, which illustrate the operation flow of the predictive maintenance unit 500. The operation flow of the predictive maintenance unit 500 is generally as follows: Based on data-driven machine learning to identify the risks and trends that the distributed devices 10 in the microgrid may face faults or degradation, thereby predicting the maintenance requirements of the distributed devices 10 and proactively providing maintenance strategies. The operation flow of the predictive maintenance unit 500 is generally similar to the operation flow of the anomaly detection unit 400 shown in FIGS. 4B and 4C.

[0049] The preprocessing of the predictive maintenance model M2 includes steps S500 to S508. Step S500 is slightly different from step S400 in FIG. 4B. Step S500 further receives the abnormal state A_s generated by the anomaly detection unit 400.

[0050] Moreover, the training stage of the predictive maintenance model M2 includes steps S510 to S524, which are generally the same as steps S410 to S424 in the training stage of the anomaly detection model M1 in FIG. 4B.

[0051] Furthermore, the actual execution stage (i.e., the prediction stage) of the predictive maintenance model M2 includes steps S526 to S540. The difference from the prediction stage process of the anomaly detection model M1 shown in FIG. 4C is that: after step S530, step S532 is executed to evaluate the test results, and step S534 can also be executed simultaneously: the cost optimization model M3 operates in the actual execution stage. And, in step S536, the abnormal state A_s is provided to the cost optimization model M3.

[0052] After step S534, step S538 is then executed: According to the cost optimization model M3, a cost optimization decision CO_d and device information D_if are generated. The cost optimization decision CO_d includes an optimized maintenance schedule O_M_s. Then, in step S540, the device information D_if and the optimized maintenance schedule O_M_s are presented to the user 60 via the terminal device 50 in FIG. 1.

[0053] The cost optimization model M3 is used to perform model operations for cost optimization to minimize the maintenance cost of the microgrid, enabling the operator of the microgrid to make the most cost-effective maintenance decisions. The objective function of the model operation of the cost optimization model M3 is shown in Equation (1). When the objective function reaches the minimum value, the minimum maintenance cost can be obtained. (1)

[0054] The arguments and parameters of Equation (1) are described as follows. "i" represents the type of the distributed device 10 in the microgrid. The distributed device 10 in the microgrid can also be referred to as distributed energy resources (DER), so "i" also represents the type of distributed energy resources. Among them, i ∈ I, and the set "I" includes: photovoltaic (PV), battery energy storage system (BESS), and wind power conversion (WT), etc. "j" represents the quantity of a specific type of the distributed device 10, j ∈ J. "k" represents the type of maintenance performed, k ∈ K, and the set "K" includes: preventive maintenance, corrective maintenance, device replacement, and downtime. "t" represents the time interval, t ∈ T.

[0055] And, " " is a decision variable, and " " represents the dynamic maintenance cost associated with the distributed device 10 of type i at time t and the selected maintenance type k. Furthermore, " " is as shown in Equation (2). (2)

[0056] In Equation (2), " " is the labor cost. " " and " " are the material cost and the equipment cost. " " is the production loss due to downtime. " " is the contract penalty cost for the failure to fulfill the contract obligation due to downtime. " " is the device replacement cost in the maintenance of the "device replacement" type.

[0057] On the other hand, the constraints of the cost optimization model M3 are as shown in Equations (3-1) to (3-9). More specifically, Equation (3-1) represents the "budget constraint", which is used to limit the total maintenance cost, considering different maintenance types k for each group j of the distributed device 10 of type i at each time t. Among them, "B" represents the maximum maintenance budget in the microgrid. " " represents the maintenance decision variable, which is associated with time t, the distributed device 10 of type i in the j-th group, and the selected maintenance type k. (3-1)

[0058] Equation (3-2) represents the "single maintenance type restriction", which is used to restrict each group j of the distributed device 10 of type i to select only one maintenance type at each time t. (3-2)

[0059] Equation (3-3) represents the "predictive maintenance restriction", which is used to restrict the execution of preventive maintenance only for the distributed device 10 that needs to be maintained according to the maintenance schedule plan. " " represents the predictive maintenance plan, which is associated with the time 𝑡, the distributed device 10 of type 𝑖 in the 𝑗-th group, and the selected maintenance type 𝑘. And, . When = 1, it means the device needs maintenance. On the contrary, when = 0, it means no maintenance is needed. (3-3)

[0060] Equation (3-4) represents the "working asset restriction", which is used to restrict the execution of maintenance on assets that are already in operation. " " represents the state detected for the distributed device 10 of type 𝑖 in the 𝑗-th group at time 𝑡. When = 1, it means the distributed device 10 of this group is operating. On the contrary, when = 0, it means the distributed device 10 of this group is not operating. (3-4)

[0061] Equation (3-5) represents the "decision variable restriction", which is used to restrict the decision variable to be a binary variable. (3-5)

[0062] Equations (3-6) to (3-9) represent the "dynamic maintenance cost equation", which is the dynamic maintenance cost associated with the distributed device 10 of type i and the selected maintenance type k. (3-6) (3-7) (3-8) (3-9)

[0063] In the operation of the cost optimization model M3, the objective function of Equation (1) must satisfy the constraints of Equations (3-1) to (3-9). Under the condition of satisfying the constraints of Equations (3-1) to (3-9), the minimum maintenance cost is calculated according to the objective function with the minimum value. When the minimum maintenance cost is achieved, the cost optimization decision CO_d (which includes the optimized maintenance schedule O_M_s) can be obtained.

[0064] In summary, the disclosed fusion detection system 1000 provides real-time anomaly detection and optimized predictive maintenance for the microgrid. The fusion detection system 1000 has the following solutions and technical effects:

[0065] (1) The fusion detection system 1000 performs a fusion operation on the sensing data of different types of distributed devices 10 of the microgrid (including real-time or low-frequency signals and images, etc.). The fusion detection system 1000 has a data fusion technology for processing different types of sensing data.

[0066] (2) When the fusion detection system 1000 immediately detects the abnormal state A_s of the distributed device 10 of the microgrid, it issues a warning or device health indicator to inform the operator of the microgrid to take immediate action. Moreover, the predictive maintenance model M2 of the fusion detection system 1000 is closely associated with the anomaly detection model M1, so as to provide predictive maintenance services for the immediate maintenance needs of the microgrid.

[0067] (3) The cost optimization model M3 of the fusion detection system 1000 considers the dynamic maintenance cost, labor cost, material and equipment cost, production loss, and penalty cost for delayed or failed maintenance according to the current market price, so as to plan the most cost-effective optimization cost. Moreover, the optimized cost, combined with the predictive maintenance prediction result P_p of the fusion detection system 1000, can obtain the optimized maintenance schedule O_M_s with the lowest cost, so as to provide the maintenance schedule plan with the best cost-effectiveness for the microgrid.

[0068] (4) The fusion detection system 1000 can also cooperate with the cloud computing platform and the edge computing platform (including the edge database 32 and the edge server 33) to combine cloud computing and IoT-based edge computing, and thus can be applied to microgrids of different scales.

[0069] Although the present disclosure has been disclosed in detail with preferred embodiments and examples as above, it is understood that the examples are illustrative rather than restrictive. It is expected that those of ordinary skill in the art can think of various modifications and combinations, and various modifications and combinations fall within the spirit of the present disclosure and the scope of the appended patent application.

[0070] 1000: Fusion Detection System 2000: Internet of Things Architecture 10: Distributed Device 11: Solar Panel 12: Wind Turbine 20: Data Acquisition Device 21: Smart Meter 22: Temperature Sensor 23: Camera 24: Infrared Sensor 25: Robot Sensor 26: Drone Sensor 27: Current Sensor 28: Voltage Sensor 31: Edge Gateway 32: Edge Database 33: Edge Server 40: Cloud Computing Platform 41: Message Intermediation Unit 42: Data Streaming Unit 43: Cloud Database 50: Terminal Device 51: Head-Mounted Device 52: Smart phone 53: Display 60: User 150: Data processing module 100: Data classification unit 210: Signal data processing unit 220: Signal data fusion unit 310: Image data processing unit 320: Image data fusion unit 400: Anomaly detection unit 500: Predictive maintenance unit 600: Cost optimization unit 700: Period analysis unit 800: Update unit SD: Sensing data set M1: Anomaly detection model M2: Predictive maintenance model M3: Cost optimization model RDCS: Sensing data of the "RDCS" type LRDCS: Sensing data of the "LRDCS" type RDI: Sensing data of the "RDI" type DI: Sensing data of the "DI" type RDCS’: Sensing data of the "RDCS" type after signal processing LRDCS’: Sensing data of the "LRDCS" type after signal processing RDI’: Sensing data of the "RDI" type after image processing DI’: Sensing data of the "DI" type after image processing RDCS”: Sensing data of the "RDCS" type after fusion processing LRDCS”: Sensing data of the "LRDCS" type after fusion processing RDI”: Sensing data of the "RDI" type after fusion processing DI”: Sensing data of the "DI" type after fusion processing A_p: Anomaly detection prediction result A_s: Anomaly status P_p: Predictive maintenance prediction result P_s: Maintenance schedule CO_d: Cost optimization decision D_if: Device information src1, src2: Data extraction sources in_M1, in_M2: Input data i1: Current i2: Voltage i3: Electric energy consumption i4: Usage time i5: Power frequency i6: Power quality i7: Visible image i8: Infrared image i9: Temperature m1, m1b: Feature selection m2: Data fusion m3, m3b: Anomaly detection out_M1, out_M2: Output data S400~S434, S500~S540: Steps

Claims

1. A fusion detection system for distributed microgrid devices, comprising: A data processing module is configured to perform preprocessing on a set of sensed data, the set of sensed data including a plurality of sensed data from a plurality of distributed devices. The data processing module includes: a data classification unit for receiving the sensed data and classifying the sensed data into a first type, a second type, a third type, and a fourth type, wherein the sensed data is acquired by a data acquisition device; and an anomaly detection unit for performing an anomaly detection model operation based on the sensed data of the first type and the second type after a first fusion process, to generate an anomaly detection prediction result. A predictive maintenance unit is configured to perform a predictive maintenance model calculation based on the first and second types of sensing data processed by the first fusion process, the third and fourth types of sensing data processed by the second fusion process, and the anomaly detection prediction result, so as to generate a predictive maintenance prediction result, which is used to plan a predictive maintenance strategy; and a cost optimization unit is configured to perform a cost optimization model calculation based on the predictive maintenance prediction result, so as to generate a cost optimization decision.

2. The converged detection system for distributed microgrid devices as described in claim 1, wherein the converged detection system is integrated into an Internet of Things (IoT) architecture, which includes the distributed devices, the data acquisition device, an edge gateway, an edge server, and a cloud computing platform.

3. The fusion detection system for a distributed microgrid device as described in claim 2, wherein the fusion detection system is installed or configured on the edge server, and the edge server includes an edge database.

4. The fusion detection system for a microgrid distributed device as described in claim 3, wherein the data acquisition device transmits the sensing data set to the edge server via the edge gateway, and the sensing data set is stored in the edge database.

5. The fusion detection system for a microgrid distributed device as described in claim 3, wherein the data processing module receives the sensing data set via the edge gateway to perform the preprocessing, the preprocessing including data processing conversion and data exploration.

6. The fusion detection system for distributed microgrid devices as described in claim 5, wherein the anomaly detection unit and the predictive maintenance unit perform an anomaly detection and a predictive maintenance respectively based on the preprocessed sensing data set, thereby generating device information.

7. The fused detection system for distributed microgrid devices as described in claim 6, wherein the cost optimization decision includes the anomaly detection decision and the predictive maintenance decision, the predictive maintenance decision includes an optimized maintenance schedule for each of the distributed devices, and the anomaly detection decision includes an alert and a device health indicator.

8. The fused detection system for distributed microgrid devices as described in claim 7, wherein the optimized maintenance schedule includes preventative maintenance, corrective maintenance, or device replacement.

9. A fusion detection system for distributed microgrid devices as described in claim 7, wherein the device health metrics include a probability anomaly score, mean time between failures, failure rate, and remaining useful life for each of the distributed devices.

10. The fusion detection system for distributed microgrid devices as described in claim 6, wherein the cloud computing platform includes a message broker unit, a data streaming unit and a cloud database, and the cloud computing platform receives device information and cost optimization decisions from the edge server.

11. The fusion detection system for distributed microgrid devices as described in claim 10, wherein after the message broker unit and the data streaming unit process the device information and the cost optimization decision, the cost optimization decision is stored in the cloud database, the device information is displayed on a terminal device and presented to a user, wherein the device information includes an abnormal state of each of the distributed devices.

12. The fusion detection system for a microgrid distributed device as described in claim 11, wherein the anomaly detection prediction result includes the anomaly state.

13. A fusion detection system for a distributed microgrid device as described in claim 1, wherein the first type is a real-time dynamic and continuous signal response (RDCS) type, the second type is a low-frequency real-time dynamic and continuous signal response (LRDCS) type, the third type is a real-time dynamic image (RDI) type, and the fourth type is a dynamic image (DI) type.

14. The fused detection system for a microgrid distributed device as described in claim 1, wherein the data processing module further comprises: A signal data fusion unit is used to perform the first fusion process on the sensing data of the first type and the second type, the first fusion process including intensity-hue-saturation processing, principal component analysis or pyramid calculation processing; and an image data fusion unit is used to perform the second fusion process on the sensing data of the third type and the fourth type, the second fusion process including principal component analysis, weighted average method, discrete wavelet transform, Laplacian pyramid or gradient pyramid processing.

15. The fusion detection system for a distributed microgrid device as described in claim 14, wherein the data processing module further comprises: A signal data processing unit is configured to perform a first signal processing on the sensing data of the first type and the second type prior to the first fusion processing, the first signal processing including fast Fourier transform, wavelet transform, Kalman filter, or autoregressive integrated moving average processing; and an image data processing unit is configured to perform a first image processing on the sensing data of the third type and the fourth type prior to the second fusion processing, the first image processing including image filtering, noise reduction, image normalization, image segmentation, or feature extraction.

16. The fusion detection system for distributed microgrid devices as described in claim 1, wherein the data acquisition device is located in the distributed devices or a terminal device, and the data acquisition device includes a current sensor, a voltage sensor, a temperature sensor, a smart meter, a weather sensor, a camera, or an infrared sensor.

17. The fused detection system for a microgrid distributed device as described in claim 16, wherein the first type of sensing data is signal data acquired by the current sensor, the voltage sensor, the temperature sensor or the smart meter, and the second type of sensing data is signal data acquired by the weather sensor.

18. The fusion detection system for a microgrid distributed device as described in claim 16, wherein the third type of sensing data is visible image data captured by the camera, and the fourth type of sensing data is infrared image data captured by the infrared sensor.

19. The fused detection system for a microgrid distributed device as described in claim 1 further includes: One update unit; And a periodic analysis unit for performing the following operations: updating the sensing data according to a first period in response to the cost optimization decision, and controlling the update unit to update the predictive maintenance model and the cost optimization model according to a third period; and updating the sensing data according to a second period in response to the anomaly detection prediction results, and controlling the update unit to update the anomaly detection model according to the third period.

20. The fusion detection system for distributed microgrid devices as described in claim 1, wherein the cost optimization model is calculated based on an objective function, the dynamic maintenance cost of which includes labor costs, material costs, equipment costs, and device replacement costs.

21. A fusion detection method for distributed microgrid devices, comprising: A data acquisition device acquires multiple sensing data from multiple distributed devices, forming a sensing data set. A data processing module performs preprocessing on the sensing data set. A data classification unit of the data processing module receives the sensing data from the multiple distributed devices and classifies the sensing data into a first type, a second type, a third type, and a fourth type. An anomaly detection unit performs an anomaly detection model calculation based on the sensing data of the first type and the second type after a first fusion process to generate an anomaly detection prediction result. A predictive maintenance unit performs calculations on a predictive maintenance model based on the first and second types of sensing data processed by the first fusion process, the third and fourth types of sensing data processed by the second fusion process, and the anomaly detection prediction result, to generate a predictive maintenance prediction result, which is used to plan a predictive maintenance strategy; and a cost optimization unit performs calculations on a cost optimization model based on the predictive maintenance prediction result, to generate a cost optimization decision.

22. The fusion detection method for distributed microgrid devices as described in claim 21, wherein the fusion detection method operates on an Internet of Things (IoT) architecture, the IoT architecture including the distributed devices, the data acquisition device, an edge gateway, an edge server, and a cloud computing platform.

23. The fusion detection method for a microgrid distributed device as described in claim 22, wherein the fusion detection method is executed by hardware components or software programs located on the edge server, and the edge server includes an edge database.

24. The fusion detection method for a distributed microgrid device as described in claim 23, further comprising, after the step of acquiring the sensing data by the data acquisition device: The sensing data set is transmitted to the edge server via the edge gateway using the data acquisition device. And store the sensing data set in the edge database.

25. The fusion detection method for a distributed microgrid device as described in claim 23, wherein the preprocessing step performed by the data processing module includes: The data processing module receives the sensing data set via the edge gateway to perform the preprocessing, which includes data processing transformation and data exploration.

26. The fusion detection method for distributed microgrid devices as described in claim 25, wherein the steps of performing the calculation of the anomaly detection model by the anomaly detection unit and performing the calculation of the predictive maintenance model by the predictive maintenance unit include: The anomaly detection unit and the predictive maintenance unit perform an anomaly detection and a predictive maintenance respectively based on the preprocessed sensing data set, thereby generating device information.

27. The fusion detection method for distributed microgrid devices as described in claim 26, wherein the cost optimization decision includes the anomaly detection decision and the predictive maintenance decision, the predictive maintenance decision includes an optimized maintenance schedule for each of the distributed devices, and the anomaly detection decision includes an alert and a device health indicator.

28. The fusion detection method for distributed microgrid devices as described in claim 27, wherein the optimized maintenance schedule includes preventative maintenance, corrective maintenance, or device replacement.

29. The fusion detection method for distributed microgrid devices as described in claim 27, wherein the device health indicators include a probability anomaly score, mean time between failures, failure rate, and remaining useful life for each of the distributed devices.

30. The fusion detection method for a microgrid distributed device as described in claim 26, wherein the cloud computing platform includes a message broker unit, a data streaming unit, and a cloud database, and the fusion detection method further includes: The cloud computing platform receives device information and cost optimization decisions from the edge server.

31. The fusion detection method for distributed microgrid devices as described in claim 30 further includes: The device information and cost optimization decision are processed by the message brokering unit and the data streaming unit. The cost optimization decision is stored in the cloud database; and the device information is displayed on a terminal device and presented to a user; wherein the device information includes an abnormal state of each of the distributed devices.

32. The fusion detection method for distributed microgrid devices as described in claim 31, wherein the anomaly detection prediction result includes the anomaly state.

33. The fusion detection method for distributed microgrid devices as described in claim 21, wherein the first type is a real-time dynamic and continuous signal response (RDCS) type, the second type is a low-frequency real-time dynamic and continuous signal response (LRDCS) type, the third type is a real-time dynamic image (RDI) type, and the fourth type is a dynamic image (DI) type.

34. The fusion detection method for distributed microgrid devices as described in claim 21 further includes: The first fusion process, including intensity-hue-saturation processing, principal component analysis, or pyramid calculation, is performed on the sensed data of the first and second types by a signal data fusion unit of the data processing module; and the second fusion process, including principal component analysis, weighted average method, discrete wavelet transform, Laplacian pyramid, or gradient pyramid processing, is performed on the sensed data of the third and fourth types by an image data fusion unit of the data processing module.

35. The fusion detection method for distributed microgrid devices as described in claim 34 further includes: Prior to the first fusion process, a signal data processing unit of the data processing module performs a first signal processing on the sensing data of the first and second types, the first signal processing including fast Fourier transform, wavelet transform, Kalman filtering, or autoregressive integrated moving average processing; and prior to the second fusion process, an image data processing unit of the data processing module performs a first image processing on the sensing data of the third and fourth types, the first image processing including image filtering, noise reduction, image normalization, image segmentation, or feature extraction.

36. The fusion detection method for distributed microgrid devices as described in claim 21, wherein the data acquisition device is disposed in the distributed devices or a terminal device, and the data acquisition device includes a current sensor, a voltage sensor, a temperature sensor, a smart meter, a weather sensor, a camera, or an infrared sensor.

37. The fusion detection method for distributed microgrid devices as described in claim 36, further comprising, prior to the step of receiving the sensing data from the distributed devices: The first type of sensing data is acquired by the current sensor, the voltage sensor, the temperature sensor, or the smart meter; and the second type of sensing data is acquired by the weather sensor.

38. The fusion detection method for distributed microgrid devices as described in claim 36, further comprising, prior to the step of receiving the sensing data from the distributed devices: The camera captures the third type of sensing data; and the infrared sensor captures the fourth type of sensing data.

39. The fusion detection method for distributed microgrid devices as described in claim 21 further includes performing the following operations by a periodic analysis unit: updating the sensing data in response to the cost optimization decision and according to a first period, and controlling an update unit to update the predictive maintenance model and the cost optimization model in response to a third period; and updating the sensing data in response to the anomaly detection prediction result and according to a second period, and controlling the update unit to update the anomaly detection model in response to the third period.

40. The fusion detection method for distributed microgrid devices as described in claim 21, wherein the cost optimization model is calculated based on an objective function, the dynamic maintenance cost of which includes labor costs, material costs, equipment costs, and device replacement costs.

Images