Overall multidirectional anomaly and scene transition detection for fluidic networks

Abstract

A data-driven method for continuously monitoring a fluid in a fluid network and triggering an alarm in the event of an unexpected characteristic of the fluid, and a system for implementing the method, are disclosed. The method comprises the following steps: a) respectively measuring one or more physical quantities in three or more measurement units, said physical quantities relating to the fluid transport, said measurement units being respectively located at three or more different locations in said fluid network, each measurement unit comprising one or more sensors, a determination means for determining one or more measured values of the physical quantity, the physical quantity being time-dependent and each of the measured values being continuously determined within a predetermined measurement time interval; the measurement values are transmitted to a data processing unit, c) an observation time sequence of the measurement values of each physical quantity is determined for each measurement unit by using the data processing unit, and d) one or more description steps, one or more estimation steps and one or more evaluation steps are processed by using the data processing unit.

Description

Technical Field

[0001] The present invention relates to a data-driven method for continuously monitoring a fluid in a fluid network and triggering an alarm in the case of an unexpected characteristic of the fluid, and a system for performing the method. Background Art

[0002] Leaks in networks such as water distribution networks having a fluid result in losses of both resources and energy. Thus, the loss of fluid in the form of treated / purified water is a loss of revenue for the utilities of the water industry. Losses of other forms of fluid such as wastewater can also lead to environmental degradation.

[0003] Fluid losses can have multiple causes, such as losses caused by leaking pipes and other components of the network infrastructure, incorrect or missing measurement data, or unexpected and unrecorded consumption. Incorrect measurement data caused by sensor failures means a loss of information and is accompanied by out-of-control situations. For example, unexpected water consumption can burden the network and its infrastructure, and may also be a sign of unauthorized consumption, and thus cause revenue losses.

[0004] Common methods for detecting unexpected consumption include mass balance of a pipe network or area, such as district metering areas (DMA), monitoring minimum night flow, simulating a pipe network using a hydraulic model, and using artificial neural networks for prediction to estimate the expected consumption at key points.

[0005] These methods generally require a detailed understanding of each distribution network, such as individual tables, timestamps, and locations. Currently available data-driven solutions mainly rely on the application of user-specific thresholds that have to be entered manually. Summary of the Invention

[0006] The object of the present invention is to provide a data-driven method for monitoring a fluid in a fluid network from an overall perspective and detecting anomalies in the characteristics of the fluid in the fluid network.

[0007] This object is achieved by the method specified in independent claim 1 and the system for performing the method described in claim 11.

[0008] Methodologically, this object is solved by the following steps:

[0009] a) Measuring one or more physical quantities at each of three or more measurement units, the physical quantities being related to fluid conveyance, the measurement units being located at three or more different positions in the fluid network respectively, each of the measurement units including one or more sensors for determining one or more measurement values of the physical quantity, wherein the physical quantity is time-dependent and each of the measurement values is continuously determined in a predetermined measurement time interval,

[0010] b) Transmit the measurement values to the data processing unit,

[0011] c) Use the data processing unit to determine, for each measurement unit, the observed time series of the measurement values of each physical quantity,

[0012] d) And use the data processing unit to process:

[0013] One or more description steps, including:

[0014] – Decompose each observed time series,

[0015] – Fit each observed time series using a time series model,

[0016] – Perform one or more statistical hypothesis tests on the time series model,

[0017] – And measure one or more statistical distances between every two different observed time series,

[0018] One or more estimation steps, including:

[0019] – Determine a time-sensitive neural network, which is trained based on the historical observed time series of one or more physical quantities from all measurement units,

[0020] – And use the time-sensitive neural network to process the estimation: For each measurement unit, estimate the estimated time series of each physical quantity, and use the time-sensitive neural network to process the estimation: For each measurement unit (MU), estimate the estimated time series of each physical quantity, where, for each estimation, all measurement units (MU) are divided into the measurement unit (MU) being estimated and the remaining measurement units (MU), and where the observed time series of each physical quantity from the remaining measurement units (MU) are used to estimate the estimated time series of each corresponding physical quantity from the measurement unit (MU) being estimated, and the estimated time series includes the estimated values of all measurement values of each corresponding physical quantity,

[0021] And one or more evaluation steps, including:

[0022] – For each measurement unit, determine the deviation between each measurement value of each physical quantity and each of the estimated values,

[0023] – If one or more predetermined limit values of the deviation are exceeded, collect the statistical distances determined from one or more description steps,

[0024] – Quantify the impact of the interference causing the deviation in the affected measurement units, and rate the proximity of the measurement units to the interference source,

[0025] – If more than one predetermined limit value of the statistical distance is exceeded, retrain the time-sensitive neural network.

[0026] – If at least one but not all of the predetermined limit values of the statistical distance are exceeded, output an alarm.

[0027] Advantageously, one or more physical quantities include at least one of the following: flow rate, temperature, level, pressure. One or more physical quantities indicate the characteristics of the fluid.

[0028] In one embodiment, the data processing unit is located at a control center at a distance.

[0029] In one embodiment, the time series model includes at least one of the following: autoregressive (AR) model, autoregressive moving average (ARMA) model, autoregressive integrated moving average (ARIMA) model, and autoregressive fractionally integrated moving average (ARFIMA) model.

[0030] In one embodiment, one or more statistical hypothesis tests include the Dickey - Fuller test.

[0031] In one embodiment, one or more statistical distances include the Kullback - Leibler divergence.

[0032] In one embodiment, the time-sensitive neural network includes at least one of: recurrent neural network (RNN), long short-term memory network (LSTM), and transformer.

[0033] In one embodiment, the estimation includes at least one of the following: sequence method and sequence-to-sequence method.

[0034] In one embodiment, the evaluation step includes performing a cluster analysis on the measured values of the observed time series and / or the phase shift of the observed time series if one or more predetermined limit values of the deviation are exceeded.

[0035] In one embodiment, the alarm relates to at least one of the following: abnormal characteristics of the fluid within a specific measurement time range, abnormal characteristics of the fluid located between at least two measurement units, and sensor failure.

[0036] This object is also achieved by a system for performing the above method, the system comprising:

[0037] a) a fluid network

[0038] b) Three or more measurement units, each of the three or more measurement units being located at three or more different positions in the fluid network, each measurement unit including one or more sensors for determining one or more measurement values of one or more physical quantities, the physical quantities being related to fluid conveyance and being time - related, and each of the measurement values being continuously determined in a predetermined measurement time interval,

[0039] c) And a data processing unit configured to receive the measurement values, wherein the data processing unit is configured to determine, for each measurement unit, an observed time series of the measurement values of each physical quantity, and the data processing unit is configured to process:

[0040] One or more descriptive steps, including:

[0041] – Decompose each observed time series,

[0042] – Fit each observed time series using a time - series model,

[0043] – Perform one or more statistical hypothesis tests on the time - series model,

[0044] – And measure one or more statistical distances between every two different observed time series,

[0045] One or more estimation steps, including:

[0046] – Determine a time - sensitive neural network trained based on historical observed time series of one or more physical quantities from all measurement units,

[0047] – And use the time - sensitive neural network to process the estimation: For each measurement unit, estimate an estimated time series of each physical quantity, and use the time - sensitive neural network to process the estimation: For each measurement unit (MU), estimate an estimated time series of each physical quantity, wherein, for each estimation, all measurement units (MU) are divided into the measurement unit (MU) being estimated and the remaining measurement units (MU), and wherein the observed time series of each physical quantity from the remaining measurement units (MU) is used to estimate the estimated time series of each corresponding physical quantity from the measurement unit (MU) being estimated, the estimated time series including estimated values of all measurement values of each corresponding physical quantity,

[0048] And one or more evaluation steps, including:

[0049] – For each measurement unit, determine the deviation between each measurement value of each physical quantity and each of the estimated values,

[0050] – If one or more predetermined limit values of the deviation are exceeded, collect the statistical distances determined from one or more description steps,

[0051] – Quantify the effect of the interference that causes the deviation of the affected measurement unit and rate the proximity of the measurement unit to the interference source,

[0052] – If more than one predetermined limit value of the statistical distance is exceeded, retrain the time-sensitive neural network,

[0053] – If at least one but not all of the predetermined limit values of the statistical distance are exceeded, output an alarm. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] This is explained in more detail with reference to the following figures.

[0055] Figure 1 Embodiments of a fluid network and a data processing unit are shown.

[0056] Figure 2 Components of the fluid network are shown.

[0057] Figure 3 Embodiments of a time-sensitive neural network are shown.

[0058] In the figures, the same features are labeled with the same reference numerals. DETAILED DESCRIPTION

[0059] See Figure 1 , the fluid network 1 communicates with the data processing unit 2. The communication between the fluid network 1 and the data processing unit 2 can be wired or wireless. The fluid network 1 can be a water distribution network or an oil and gas transportation network. The data processing unit 2 can be an edge computer or a cloud data service.

[0060] In Figure 2 , the fluid network 1 includes an inlet where fluid enters the network - described as IN and an outlet where fluid leaves the network - described as OUT. The fluid network 1 includes six measurement units (MU1, MU2, MU3, MU4, MU5, MU6) and two consumers 4, all located at different positions.

[0061] Each measurement unit (MU) contains three sensors 3, such as a flow meter, a pressure sensor, and a temperature sensor. The three sensors 3 continuously measure three physical quantities, such as flow rate, pressure, and temperature, respectively, at a predetermined measurement time interval. The flow rate, pressure, and temperature indicate water leakage in the water network.

[0062] Each MU communicates with the data processing unit 2 (see Figure 1), the measured values of the flow rate, pressure, and temperature from each MU are transmitted to the data processing unit 2. D-Out describes the data interface with the logic module of the present invention (such as the data processing unit 2). Therefore, the measured values are transmitted to the data processing unit 2 through the data interface - D-Out.

[0063] See Figure 3 , the data processing unit 2 receives the measured values from each measurement unit (MU). The data processing unit 2 is configured to respectively determine three observed time series of the measured values of the flow rate, pressure, and temperature for each measurement unit 3. Figure 3 In, PHM represents parameter health monitoring, which includes all the measured values that the MU can provide. In Figure 3 On the left side (actual measured values), the box named "Mu1 X PHM X t" represents all the measured values that the measurement unit 1 (MU1) can provide and their measurement times.

[0064] The data processing unit 2 is configured to process a descriptive time series analysis including, for example, four descriptive steps.

[0065] The four descriptive steps include decomposing each observed time series into multiple component series, such as a trend component, a seasonal component, and a noise component. The data processing unit 2 fits each observed time series using a time series model (such as an autoregressive integrated moving average model (ARIMA)). Then, a statistical hypothesis test, such as the Dickey–Fuller test, is performed on each time series model. And the statistical distance between every two different observed time series is measured. This statistical distance can be the Kullback-Leibler divergence.

[0066] In the estimation step, the data processing unit 2 first determines a time-sensitive neural network, which is trained based on the historical observed time series of the flow rate, pressure, and temperature from all measurement units (MU). Then, the data processing unit 2 uses the time-sensitive neural network to process the prediction / estimation: for each MU, three prediction / estimation time series are estimated, one for the flow rate, one for the pressure, and one for the temperature. When estimating the estimated time series, the estimated time series of the physical quantity is only estimated by the observed time series of the corresponding physical quantity, so there is no situation where the third physical quantity is estimated by the first physical quantity and the second physical quantity. In addition, the data processing unit 2 further divides all MUs into the MUs being estimated and the remaining MUs, where the observed time series of the physical quantity from the remaining MUs are used to estimate the estimated time series of the corresponding physical quantity from the MUs being estimated ( Figure 3The predicted measured values therein). For example, the estimated flow time series at MU1 is estimated based on all the observed flow time series recorded at MU2, MU3, MU4, MU5, and MU6 (rather than MU1).

[0067] The time-sensitive neural network is not used to estimate new physical quantities, but rather physical quantities that already exist in other and the same measurement units (MUs), so that it can provide proxy information for the incorrect measurements of the individual sensors in the network.

[0068] The estimation step is followed by an evaluation (∆) that includes one or more evaluation steps. It determines the deviation between each measured value and each estimated value of each physical quantity for each MU. In the evaluation, if one or more of the pre-determined limits of the deviation are exceeded, the data processing unit 2 then collects the statistical distances determined according to the description step. The data processing unit 2 quantifies the impact of the interference causing the deviation in the affected MU and rates the proximity of the MU to the interference source. If more than one of the pre-determined limits of the said statistical distances are exceeded, the data processing unit 2 retrains the time-sensitive neural network. If the estimation result is considered to be accurate enough, an expert (e.g., the user or operator of the facility) can stop the retraining. And if at least one but not all of the pre-determined limits of the statistical distances are exceeded, the data processing unit 2 outputs an alarm.

[0069] List of reference numerals

[0070] 1 Fluid network

[0071] 2 Data processing unit

[0072] 3 Sensor

[0073] 4 Consumer

[0074] MU2 Measurement unit 2

Claims

1. A data-driven method for continuously monitoring a fluid in a fluid network (1) and triggering an alarm in the event of an unexpected characteristic in the fluid, comprising: a) Measuring one or more physical quantities, each related to fluid conveyance, at each of three or more measurement units (MU), the measurement units (MU) being located at three or more different positions in the fluid network respectively, each of the measurement units (MU) including one or more sensors (3) for determining one or more measured values of the physical quantity, wherein the physical quantity is time-dependent and each of the measured values is continuously determined in a predetermined measurement time interval; b) Transmitting the measured values to a data processing unit (2); c) Using the data processing unit (2) to determine, for each measurement unit (MU), an observed time series of the measured values of each physical quantity; d) And using the data processing unit (2) to process: i) One or more description steps, including: – Decomposing each observed time series; – Fitting each observed time series with a time series model; – Performing one or more statistical hypothesis tests on the time series model; – And measuring one or more statistical distances between every two different observed time series; ii) One or more estimation steps, including: – Determining a time-sensitive neural network trained based on historical observed time series of one or more physical quantities from all measurement units (MU); – And using the time-sensitive neural network to process the estimation: for each measurement unit (MU), estimating an estimated time series of each physical quantity, wherein for each estimation, all measurement units (MU) are divided into the measurement unit (MU) being estimated and the remaining measurement units (MU), and the observed time series of each physical quantity from the remaining measurement units (MU) is used to estimate the estimated time series of each corresponding physical quantity from the measurement unit (MU) being estimated, the estimated time series including estimated values of all measured values of each corresponding physical quantity; iii) And one or more evaluation steps, including: – For each measurement unit (MU), determining the deviation between each measured value of each physical quantity and each of the estimated values; – If one or more predetermined limit values of the deviation are exceeded, collecting the statistical distances determined from the one or more description steps; – Quantifying the impact of the disturbance causing the deviation of the affected measurement unit (MU) and rating the proximity of the measurement unit to the disturbance source; – If more than one predetermined limit value of the statistical distance is exceeded, retraining the time-sensitive neural network; – If at least one but not all of the predetermined limit values of the statistical distance are exceeded, outputting an alarm.

2. The method according to claim 1, characterized in that the one or more physical quantities include at least one of the following: flow rate, temperature, level, pressure.

3. The method according to at least one of claims 1 to 2, characterized in that the data processing unit (2) is located in a control center at a distance.

4. The method according to at least one of claims 1 to 3, characterized in that the time series model includes at least one of the following: autoregressive (AR) model, autoregressive moving average (ARMA) model, autoregressive integrated moving average (ARIMA) model, and autoregressive fractionally integrated moving average (ARFIMA) model.

5. The method according to at least one of claims 1 to 4, characterized in that the one or more statistical hypothesis tests include the Dickey - Fuller test.

6. The method according to at least one of claims 1 to 5, characterized in that the one or more statistical distances include the Kullback - Leibler divergence.

7. The method according to at least one of claims 1 to 6, characterized in that the time - sensitive neural network includes at least one of the following: recurrent neural network (RNN), long short - term memory network (LSTM), and transformer.

8. The method according to at least one of claims 1 to 7, characterized in that the estimation includes at least one of the following: sequence method and sequence - to - sequence method.

9. The method according to at least one of claims 1 to 8, characterized in that the evaluation step includes performing a cluster analysis on the measured values of the observed time series and / or the phase shift of the observed time series if one or more predetermined limit values of the deviation are exceeded.

10. The method according to at least one of claims 1 to 9, characterized in that the alarm relates to at least one of the following: abnormal characteristics of the fluid within a specific measurement time range, abnormal characteristics of the fluid between at least two measurement units (MU), and sensor failures.

11. A system for implementing the method according to at least one of claims 1 to 10, comprising: a) a fluid network (1) b) three or more measurement units (MU), the three or more measurement units (MU) being respectively located at three or more different positions in the fluid network (1), each of the measurement units (MU) including one or more sensors (3) for determining one or more measured values of one or more physical quantities related to fluid conveyance, the physical quantities being time - related, and each of the measured values being continuously determined at a predetermined measurement time interval, c) and a data processing unit (2) for receiving the measured values, wherein the data processing unit (2) is configured to determine an observed time series of the measured values of each physical quantity for each measurement unit (MU), and the data processing unit (2) is configured to process: i) one or more description steps, including: – decomposing each observed time series, – fitting each observed time series using a time series model, – perform one or more statistical hypothesis tests on the time series model, – and measure one or more statistical distances between each pair of different observed time series, ii) one or more estimation steps, including: – determine a time-sensitive neural network, the time-sensitive neural network being trained based on historical observed time series of one or more physical quantities from all measurement units (MUs), – and use the time-sensitive neural network to process the estimation: for each measurement unit (MU), estimate an estimated time series of each physical quantity, and use the time-sensitive neural network to process the estimation: for each measurement unit (MU), estimate an estimated time series of each physical quantity, wherein, for each estimation, all measurement units (MUs) are divided into the measurement unit (MU) being estimated and the remaining measurement units (MUs), and wherein the observed time series of each physical quantity from the remaining measurement units (MUs) is used to estimate the estimated time series of each corresponding physical quantity from the measurement unit (MU) being estimated, the estimated time series including estimated values of all measured values of each corresponding physical quantity, iii) and one or more evaluation steps, including: – for each measurement unit (MU), determine the deviation between each measured value of each physical quantity and each of the estimated values, – if one or more predetermined limit values of the deviation are exceeded, collect the statistical distances determined from the one or more description steps, – quantify the impact of the interference causing the deviation of the affected measurement unit (MU), and rate the proximity of the measurement unit to the interference source, – if more than one predetermined limit value of the statistical distance is exceeded, retrain the time-sensitive neural network, – if at least one but not all of the predetermined limit values of the statistical distance are exceeded, output an alert.

Images