DETERMINING THE FLUID DENSITY IN AN ELECTRICAL DEVICE

Description

[0001] The invention relates to a method for determining the fluid density of a fluid in an encapsulated electrical device.

[0002] Many electrical devices are at least partially filled with a fluid, i.e., a gas or liquid, as an insulating medium. This is done, for example, to increase the dielectric strength, extinguish arcs, and / or cool the device. Gas-insulated switchgear, for instance, is filled with a pressurized insulating gas such as sulfur hexafluoride or air, and transformers are often filled with transformer oil. The fluid density within the electrical device is often a crucial and critical factor, particularly for its operational reliability. For example, the dielectric strength of an insulating gas in a gas-insulated switchgear depends critically on the density of that gas.Furthermore, a leak of insulating gas such as sulfur hexafluoride from a gas-insulated switchgear can lead to monetary penalties for the operator of the switchgear due to the environmentally damaging effect of the insulating gas.

[0003] The fluid density in an electrical device is therefore typically monitored to detect fluid leakage and, if necessary, to counteract changes in fluid density or, for safety reasons, to shut down the device in the event of a critical change in fluid density. The measurement accuracy of fluid density in an electrical device is subject to fluctuations, which can be caused in particular by external weather influences. This limits the detection of fluid density changes, especially over short periods. The influence of changing weather conditions on fluid density measurements is generally only recorded and evaluated over long observation periods (several weeks or months), which allows for the calculation of long-term trends.Document DE 696 04 842 T2 describes a device with a microprocessor configured to calculate the density of a gas using equations of state. Document US 6 205 846 B1 deals with determining the leakage rate of a gas-insulated, encapsulated device.

[0004] The invention is based on the objective of improving the determination of the fluid density of a fluid in an encapsulated electrical device, in particular to reliably detect changes in fluid density even in the short term.

[0005] The problem is solved according to the invention by a method with the features of claim 1, a computer program with the features of claim 14 and an electrical device with the features of claim 15.

[0006] Advantageous embodiments of the invention are the subject of the dependent claims.

[0007] In the inventive method for determining the fluid density of a fluid in an encapsulated electrical device, measurement data is acquired using a sensor unit, from which measured values ​​for the fluid density are derived, and weather data on the weather conditions in the vicinity of the electrical device are collected. A digital model for the influence of the weather conditions on a measurement deviation of a measured value from the correct fluid density is generated using machine learning. A correction value for measured values ​​is calculated using the digital model as a function of the weather data, and a measured value is corrected with this correction value.

[0008] The method according to the invention thus enables a correction of fluid density measurements that takes into account the influence of weather conditions in the environment of the electrical device. For this purpose, a digital model is trained using machine learning, which models the influence of weather conditions on measurement deviations. This allows, in particular, short-term influences of weather conditions on the fluid density measurement to be considered, thereby significantly improving the accuracy of the fluid density measurement and enabling the rapid and reliable detection of changes in fluid density, for example, due to fluid losses from leaks in the electrical device. In particular, this increases the operational reliability of the electrical device and improves its maintainability.

[0009] In one embodiment of the method according to the invention, the digital model comprises an artificial neural network with multiple layers of interconnected artificial neurons. Preferably, the artificial neural network is a recurrent artificial neural network and / or includes at least one memory cell. A recurrent artificial neural network is understood to be a feedback-based artificial neural network whose neurons are interconnected in sub-layers with connections from neurons of one layer not only to neurons of a deeper layer but also to neurons of the same or a higher layer. A memory cell is understood to be a cell with so-called Long Short-Term Memory (LSTM), that is, a cell with a type of long-lasting short-term memory. Such cells are also called LSTM cells, and an artificial neural network with LSTM cells is also called an artificial neural LSTM network.Artificial neural networks with many layers are extremely capable of learning. LSTM technology ensures the reliable and stable functioning of such artificial neural networks, and the recurrent interconnection of neurons enables, among other things, the retrieval and evaluation of sequenced information within the data processed by the artificial neural network. Recurrent LSTM artificial neural networks are already successfully used in many fields, for example, for handwriting recognition, speech recognition, and machine translation of texts into different languages. The method according to the invention advantageously utilizes the capabilities of such artificial neural networks for the analysis and correction of fluid density measurements as a function of weather data.

[0010] In a further embodiment of the method according to the invention, the measurement data and / or the measured values ​​are transferred to a data cloud and / or the correction value is calculated using the digital model in a data cloud. By transferring the measurement data and / or measured values ​​to a data cloud, this data can be made available and evaluated regardless of location or user. Calculating the correction values ​​using the digital model in a data cloud advantageously allows the use of the high computing capacity that can be provided by a data cloud.

[0011] In a further embodiment of the method according to the invention, additional training values ​​for measured values ​​and / or weather data are generated from measured values ​​and / or weather data to train the digital model. For example, training values ​​are generated by shifting weather data relative to measured values ​​in time, scaling measured values ​​and / or weather data, and / or shifting the range of measured values. Furthermore, training values ​​for simulated fluid losses can be generated to train the digital model, for example, by adding values ​​from simulated trend lines to measured values. These embodiments of the method according to the invention are particularly advantageous if there are initially not enough suitable measured values ​​and / or weather data available to train the digital model.The designs provide for the generation of suitable additional training values ​​from the (few) available measured values ​​and / or weather data through so-called data augmentation and / or from simulated fluid losses.

[0012] In a further embodiment of the method according to the invention, a calculation period, for example a period of 24 hours, is specified, and the correction value for measured values ​​acquired within the calculation period is calculated using the digital model. This allows the consideration of the influence of weather conditions on the measured fluid density to be advantageously kept up-to-date and compared with the results for previous calculation periods, for example, to identify trends in these results.

[0013] In a further embodiment of the method according to the invention, the weather data includes temperature, wind speed, precipitation, humidity, and / or air pressure in the vicinity of the electrical device. This embodiment of the method according to the invention advantageously takes into account weather conditions that primarily influence the measurement of fluid density.

[0014] In a further embodiment of the method according to the invention, the digital model is generated specifically for one electrical device. Alternatively, the digital model is generated for different but similar electrical devices. Generating a digital model for only one specific electrical device allows for the consideration of special properties of the device. Generating a digital model for several similar electrical devices advantageously enables the use of the model for an entire class of electrical devices and thereby reduces the development effort and development costs for the measurement correction according to the invention for these electrical devices compared to generating a separate digital model for each of these devices.

[0015] In a further embodiment of the method according to the invention, only measured values ​​and weather data are supplied to the digital model as input variables. This embodiment of the method according to the invention is particularly suitable for generating a digital model specifically for a particular electrical device. Alternatively, measured values, weather data, and additional data generated from the measured values ​​and weather data are supplied to the digital model as input variables. This alternative embodiment of the method according to the invention is particularly suitable for generating a digital model for several similar electrical devices. The additional data is used to account for differences between the electrical devices.Additional data includes, for example, derivations of measured values ​​from weather data, which describe, for example, changes in measured values ​​depending on a temperature or air pressure in the vicinity of an electrical device.

[0016] A computer program according to the invention comprises instructions which, when the computer program is executed by a control unit or in a data cloud, cause it to execute the digital model of a method according to the invention.

[0017] An electrical device according to the invention comprises a control unit on which a computer program according to the invention is executed, or a connection to a data cloud in which a computer program according to the invention is stored.

[0018] The properties, features and advantages of this invention described above, as well as the manner in which they are achieved, will become clearer and more easily understood in connection with the following description of exemplary embodiments, which are explained in more detail in conjunction with the drawing.

[0019] The single figure shows a structural diagram of an embodiment of the inventive method for determining the fluid density of a fluid in an encapsulated electrical device 1.

[0020] In the illustrated embodiment, a sensor unit 3 acquires measurement data 5, and a processing unit 7 derives measured values ​​9 for the fluid density from these measurement data 5. Furthermore, weather data 13 concerning weather conditions in the vicinity of the electrical device 1, provided by a weather data source 11, are collected. A digital model 15 is generated using machine learning to represent the influence of weather conditions on a measurement deviation 9 from the correct fluid density. Using the digital model 15, a correction value 17 is calculated for the measured values ​​9 based on the weather data 13, and a measured value 9 is corrected by this correction value 17.

[0021] For example, the electrical device 1 is a gas-insulated switchgear and the fluid is a pressurized insulating gas in the gas-insulated switchgear. Alternatively, the electrical device 1 is, for example, an oil-filled transformer and the fluid is transformer oil in the transformer. The invention is not limited to one type or kind of electrical device.

[0022] The sensor unit 3 is configured to acquire measurement data 5 from which measured values ​​9 for the fluid density can be derived. For example, the sensor unit 3 includes sensors configured to acquire the fluid pressure and temperature as measurement data 5. Alternatively, the sensor unit 3 includes, for example, two quartz crystals, one of which operates in a controlled reference environment and the other of which operates in the fluid, and the sensor unit 3 acquires the resonance frequencies of the two quartz crystals as measurement data 5. The invention is not limited to any one type or design of sensor unit 3.

[0023] The processing unit 7 determines a measured value 9 for the fluid density from the measured data 5. For example, if the measured data 5 include a fluid pressure and a fluid temperature, the processing unit 7 calculates a measured value 9 for the fluid density from the fluid pressure and the fluid temperature. If, for example, the measured data 5 include resonance frequencies of two quartz crystals of a sensor unit 3 described above, the processing unit 7 calculates a measured value 9 for the fluid density from the difference between the resonance frequency in the fluid and the resonance frequency in the reference environment. The invention is not limited to any one type or design of the processing unit 7.

[0024] The weather data source 11 is, for example, a weather station that records the weather data 13. Alternatively, the weather data source 11 is a weather database, for example in a data cloud, that provides the weather data 13. The weather data source 11 can also comprise a weather station and such a weather database. The weather data 13 includes, for example, temperature, wind speed, precipitation, humidity, and / or air pressure in the vicinity of the electrical device 1. The invention is not limited to any one type or kind of weather data source 11.

[0025] The digital model 15 features an artificial neural network 19 with multiple layers 21, 22, 23 of interconnected artificial neurons 25 and memory cells 27 (LSTM cells). The neural network 19 is configured as a recurrent neural LSTM network. An arrow from one neuron 25 to another neuron 25 or to a memory cell 27 symbolizes that an output value of neuron 25 is passed to the other neuron 25 or the memory cell 27 as an input value. Similarly, an arrow from a memory cell 27 to a neuron 25 symbolizes that an output value of the memory cell 27 is passed to a neuron 25 as an input value. The neural network 19 is shown here only schematically with an input layer 21, an intermediate layer 22, an output layer 23, and a memory cell 27.In an actual implementation, the neural network 19 has significantly more intermediate layers 22, neurons 25 and memory cells 27 than are shown in the figure.

[0026] Optionally, in addition to the measured values ​​9 and the weather data 13, supplementary data 29 are supplied to the digital model 15 as input variables. These supplementary data are generated by the processing unit 7 from the measured values ​​9 and the weather data 13. The supplementary data 29 are, for example, derivations of measured values ​​9 from weather data 13, which describe, for instance, changes in measured values ​​9 as a function of temperature or air pressure in the vicinity of an electrical device 1. Such supplementary data 29 are supplied to the digital model 15 as input variables, in particular, when the digital model 15 is generated not only specifically for a particular electrical device 1 but for different (but similar) electrical devices 1.

[0027] For example, the measurement data 5 and / or the measured values ​​9 are transferred to a data cloud and / or the correction value 17 is calculated with the digital model 15 in a data cloud.

[0028] To train the digital model 15, further training values ​​for measurements 9 and / or weather data 13 can be generated from measurements 9 and / or weather data 13 by means of so-called data augmentation, particularly if there are initially not enough suitable measurements 9 and / or weather data 13 available for training the digital model 15. For example, such training values ​​are generated by shifting weather data 13 in time relative to measurements 9, scaling measurements 9 and / or weather data 13, and / or shifting the value range of measurements 9. Furthermore, training values ​​for simulated fluid losses can be generated to train the digital model 15.

[0029] For example, the digital model calculates a correction value 17 for each measurement value 9 that is recorded within a specified calculation period, for example a period of 24 hours.

Claims

1. A method for determining a fluid density of a fluid in an encapsulated electrical equipment (1), wherein - measuring data (5), from which measured values (9) for the fluid density are derived, are acquired with a sensor unit (3), - weather data (13) on weather conditions in an environment of the electrical equipment (1) are collected, - a digital model (15) for the influence of the weather conditions on a measurement deviation of a measured value (9) from the correct fluid density is created by machine learning, a correction value (17) for measured values (9) is calculated as a function of the weather data (13) with the digital model (15), and - a measured value (9) is corrected with the correction value (17).

2. The method according to claim 1, wherein the digital model (15) has an artificial neural network (19) comprising a plurality of layers (21, 22, 23) of crosslinked artificial neurons (25).

3. The method according to claim 2, wherein the artificial neural network (19) is a recurrent artificial neural network (19).

4. The method according to claim 2 or 3, wherein the artificial neural network (19) has at least one retentive cell (27).

5. The method according to any one of the preceding claims, wherein the measuring data (5) and / or the measured values (9) are transmitted into a data cloud and / or the correction value (17) is calculated with the digital model (15) in a data cloud.

6. The method according to any one of the preceding claims, wherein further training values for measured values (9) and / or weather data (13) are generated from measured values (9) and / or weather data (13) for training the digital model (15).

7. The method according to claim 6, wherein training values are generated by temporally shifting weather data (13) relative to measured values (9), scaling measured values (9) and / or weather data (13), and / or shifts in the value range of the measured values (9).

8. The method according to any one of the preceding claims, wherein training values for simulated fluid losses are generated for training the digital model (15).

9. The method according to any one of the preceding claims, wherein a calculation period, such as a period of 24 hours, is predefined, and the correction value (17) for measured values (9) acquired within the calculation period is calculated with the digital model (15).

10. The method according to any one of the preceding claims, wherein the weather data (13) comprises a temperature, wind speed, precipitation, humidity, and / or atmospheric pressure in the environment of the electrical equipment (1).

11. The method according to any one of the preceding claims, wherein the digital model (15) is created specifically for an electrical equipment (1) or for pieces of electrical equipment (1) that are different from one another.

12. The method according to any one of the preceding claims, wherein only measured values (9) and weather data (13) are supplied to the digital model (15) as input variables.

13. The method according to any one of claims 1 to 11, wherein measured values (9), weather data (13), and additional data (29) generated from the measured values (9) and weather data (13) are supplied to the digital model (15) as input variables.

14. A computer program comprising instructions which, when the computer program is executed by a control unit or in a data cloud, cause it to execute the digital model (15) of a method according to any one of the preceding claims.

15. An encapsulated electrical equipment (1) comprising a control unit on which a computer program according to claim 14 is executed, or a system from an encapsulated electrical equipment (1) and an associated data cloud, in which a computer program according to claim 14 is executed.