-
公开(公告)号：EP4354346A1
公开(公告)日：2024-04-17
申请号：EP22200674.4
申请日：2022-10-10
申请人： Hitachi Energy Ltd
CPC分类号： G06N3/0455 , G06N3/09 , G06N20/00 , G01R19/2513 , H02J13/00002
摘要： One component of time domain protection in a power system is estimating the location of a fault. In an embodiment, a multi-objective problem is formulated that comprises a non-smoothness penalization function that drives the primary objective function for fault location estimation towards a solution that respects smoothness between the inputs and outputs of a machine-learning model. This technique improves the accuracy, blind zone, and speed of state-of-the-art techniques, in the context of time domain protection, as well as for other regression tasks. In an additional or alternative embodiment that is specific to time domain protection, the multi-objective problem may comprise a phasor-deviation penalization function that drives the primary objective function towards a solution that minimizes deviations in phasor values. The trained machine-learning model may be executed in a line protection system to determine whether or not to trip a circuit breaker of a power line.
-
2.发明公开
公开(公告)号：EP4435397A1
公开(公告)日：2024-09-25
申请号：EP23162860.3
申请日：2023-03-20
申请人： Hitachi Energy Ltd
摘要： The present invention relates to a method for estimating a physical quantity of a static electric induction device assembly (10). The static electric induction device assembly (10) comprises an enclosure (14), a static electric induction device (12) and a liquid (18) whereby the enclosure (14) accommodates the static electric induction device (12) and the liquid (18) such that the static electric induction device (12) is at least partially, preferably fully, submerged into the liquid (18). The method comprising using measured temperature data obtained from a measurement assembly (20). The measured temperature data comprises a temperature (Ttrue(x, t)) in each one of a plurality of different locations (x) of the static electric induction device assembly (10) as a function of time (t) for a reference time range (ΔTref) when the static electric induction device assembly (10) is in a condition in which at least a portion of the static electric induction device (12) generates heat during at least a portion of the reference time range (ΔTref).
The method further comprises:
- using a time dependent partial differential equation representing a physical condition of the static electric induction device assembly (10) during the reference time range (ΔTref), wherein the physical quantity forms a source term of the partial differential equation;
- generating a temperature model for estimated temperature data, the estimated temperature data corresponding to an estimated temperature (Test(x, t)) in each one of the plurality of different locations (x) of the static electric induction device assembly (10) as a function of time (t), the temperature model comprising a first neural network (NN1) representing the estimated temperature data (Test(x, t)) as well as the measured temperature data (Ttrue(x, t)), and
- estimating the physical quantity by training a neural network system that uses at least the following entities: the time dependent partial differential equation, information from the temperature model and a second neural network (NN2) for the physical quantity.-
公开(公告)号：EP4402548A1
公开(公告)日：2024-07-24
申请号：EP22801807.3
申请日：2022-10-14
申请人： Hitachi Energy Ltd
IPC分类号： G05B23/02 , G06N3/08 , H02J13/00 , H02H7/26 , G05B19/042
CPC分类号： G05B23/0221 , G05B23/024 , G05B19/0426 , H02H1/0092 , H02H3/40 , H02J13/00002 , H02J13/00036 , H02J3/003 , H02J2203/2020200101 , G06N3/0985 , G06N3/0442
-
4.发明公开
公开(公告)号：EP4386628A1
公开(公告)日：2024-06-19
申请号：EP22217410.4
申请日：2022-12-30
申请人： Hitachi Energy Ltd
IPC分类号： G06N3/0464 , G06N3/0495 , G06N3/0442 , G06N3/09
CPC分类号： G06N3/0464 , G06N3/0495 , G06N3/0442 , G06N3/09
摘要： Conventional methods for reducing the size of a machine-learning model generally involve a costly and slow search for a new architecture, or pruning model parameters to produce a sparse machine-learning model. However, some environments, such as regression tasks to be embedded in hardware for performance in real time on time series data, require the pruned machine-learning model to be dense. Accordingly, automated progressive tuning is disclosed to jointly tune both the architecture and the parameters of a machine-learning model, initially trained for a first environment, for deployment in a second environment. In an embodiment, a constraint is imposed on the dimensions of the machine-learning model, during the tuning, to prune the machine-learning model into a smaller, dense machine-learning model that is suitable for embedding in hardware, as well as other potential environments.
-
-
-
搜索结果
4 条记录 (耗时 0.017 秒)
