Anomaly detection device, reservoir device, anomaly detection method, and program

Abstract

To accurately detect an abnormality of an observation target apparatus with a simple configuration even when an observation signal includes noise.SOLUTION: An abnormality detection apparatus comprises an input unit, a reservoir unit, an output unit, and a determination unit. The input unit outputs a time-series input signal corresponding to an observation signal. The reservoir unit outputs a plurality of output spike signals on the basis of the input signal. The output unit outputs a plurality of prediction signals on the basis of the plurality of output spike signals. The determination unit outputs a determination signal indicating a determination result as to whether the observation target apparatus is normal or abnormal on the basis of the plurality of prediction signals. The reservoir unit includes a plurality of input firing circuits. Each of the plurality of input firing circuits increases a membrane potential according to the input signal, decreases the membrane potential with time in accordance with a preset time constant, and outputs an intermediate spike signal by firing when the membrane potential exceeds a preset threshold potential. Each of the plurality of input firing circuits is set with a different time constant.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] Embodiments of the present invention relate to an anomaly detection apparatus, a reservoir apparatus, an anomaly detection method and a program. [Background Art]

[0002] Artificial Intelligence (AI) is used in various automated processing and labor-saving processing. A neural network is known as a typical AI algorithm. A Deep Neural Network (DNN) with multiple layers of neural networks is used for deep learning algorithms. In addition, a Recurrent Neural Network (RNN), which is recursively connected from neurons closer to the output to neurons closer to the input, is used for time-series data processing. Moreover, LSTM (Long-Short-Term Memory) is known as a neural network that enhances the expressive capability of short-term memory and long-term memory. LSTM has even higher applicability in time-series data processing.

[0003] RNN and LSTM used for time-series data processing are calculated using a CPU (Central Processing Unit), which is a general-purpose arithmetic unit. However, since RNN and LSTM require a larger amount of computation than ordinary neural networks, they are often calculated using a General Purpose Graphical Processing Unit (GPU). Furthermore, in order to achieve high performance for deep learning, RNN and LSTM, it was necessary to perform fine parameter tuning during learning. Therefore, RNN and LSTM further increase the amount of computation during learning. As a result, RNN and LSTM have long learning times and high power consumption during learning.

[0004] On the other hand, reservoir computing is known as an algorithm for processing time-series data that requires less computation during learning. Reservoir computing comprises an input unit, a reservoir unit, and an output unit. Reservoir computing does not require training the reservoir unit. However, in order to output the desired signal, reservoir computing must first determine the optimal reservoir unit based on the input, and then ensure high accuracy in the weights between the reservoir unit and the output unit.

[0005] The reservoir section is implemented in hardware using various media such as electronic circuits. Furthermore, the signals output from the reservoir section are generally analog signals. Reservoir computing enables high-speed processing by implementing the calculation of output weights in the output section using an analog circuit-based multiply-accumulate unit.

[0006] Such reservoir computing can detect anomalies in infrastructure equipment that operates continuously. For example, reservoir computing can detect when infrastructure equipment has operated in a way that deviates from normal conditions intermittently and instantaneously, even during periods when the equipment is functioning normally.

[0007] However, conventional reservoir computing has struggled to detect irregular, instantaneous anomalies when the observed signals from infrastructure equipment contain noise. Therefore, when applying reservoir computing to such anomaly detection systems, it was necessary to decompose the signal into multiple frequency components using the Fast Fourier Transform (FFT) and remove noise from each of these components. Consequently, such anomaly detection systems required digital processing circuits to perform the FFT, resulting in large and expensive circuits that negated the advantage of reservoir computing, which can be implemented with analog circuits. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] Melika Payvand, Melika Payvand1, Filippo Moro1, Kumiko Nomura, Thomas Dalgaty, Elisa Vianello, Yoshifumi Nishi, Giacomo Indiveri, “Self-organization of an inhomogeneous memristive hardware for sequence learning”, October 2, 2022, Nature Communications volume 13, Article number:5793 (2022) [Overview of the Initiative] [Problems that the invention aims to solve]

[0009] The problem that this invention aims to solve is to accurately detect abnormalities in the target device with a simple configuration, even when the observed signal acquired from the target device contains noise, without decomposing the observed signal into multiple frequency components using a digital processing circuit. [Means for solving the problem]

[0010] An anomaly detection device according to the embodiment detects an anomaly in a device under observation. The anomaly detection device comprises an input unit, a reservoir unit, an output unit, and a determination unit. The input unit acquires an observation signal obtained by observing the device under observation and outputs a time-series input signal corresponding to the acquired observation signal. The reservoir unit outputs a plurality of output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal. The output unit outputs a plurality of prediction signals, each obtained by multiplying each of the plurality of output spike signals by a preset output weight and summing them up. The determination unit determines whether the device under observation is normal or abnormal based on the plurality of prediction signals and outputs a determination signal representing the determination result. The reservoir unit includes a plurality of input firing circuits. Each of the plurality of input firing circuits increases the membrane potential in accordance with the input signal and decreases the membrane potential over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and resets the membrane potential for a certain period of time in accordance with the output of the intermediate spike signal. Each of the aforementioned plurality of input firing circuits is set to have a different time constant. [Brief explanation of the drawing]

[0011] [Figure 1] A diagram showing the anomaly detection system according to the first embodiment together with the device under observation. [Figure 2] A diagram illustrating the configuration of the anomaly detection device according to the first embodiment. [Figure 3] A diagram showing the configuration of the reservoir section according to the first embodiment. [Figure 4] Configuration diagram of the input ignition circuit. [Figure 5] A diagram showing an example of input current, relative threshold, and firing time. [Figure 6] This diagram shows the relationship between the frequency of the intermediate spike signal and the frequency of the input signal. [Figure 7] A diagram showing the normal waveform, abnormal waveform, relative threshold, and firing interval of the input current. [Figure 8]A diagram for explaining the content of anomaly detection processing. [Figure 9] A configuration diagram of a modified example of the reservoir unit according to the first embodiment. [Figure 10] A configuration diagram of the anomaly detection device according to the second embodiment. [Figure 11] A configuration diagram of a partial reservoir unit. [Figure 12] A configuration diagram of a modified example of the partial reservoir unit according to the second embodiment. [Figure 13] A hardware configuration diagram of an information processing apparatus. DESCRIPTION OF EMBODIMENTS

[0012] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0013] (First Embodiment) FIG. 1 is a diagram showing the configuration of an anomaly detection system 10 according to the first embodiment, together with an observation target device 100.

[0014] When an anomaly occurs in the observation target device 100, the anomaly detection system 10 notifies, for example, an administrator or another device that the anomaly has occurred.

[0015] The observation target device 100 operates continuously. For example, the observation target device 100 is a power line switching device or a power supply device in power conversion equipment or power supply equipment. The observation target device 100 is not limited to such devices, and may be any device as long as it operates continuously.

[0016] The anomaly detection system 10 includes an anomaly detection device 20 and a notification device 22.

[0017] The anomaly detection device 20 detects anomalies in the observation target device 100. For example, the anomaly detection device 20 detects anomalies that occur instantaneously in the observation target device 100. Even during periods of normal operation, the observation target device 100 may occasionally and instantaneously exhibit behavior or actions that differ from its normal state. Such behavior or actions may be a precursor to a failure or significant performance degradation in the observation target device 100. The anomaly detection device 20 according to this embodiment can detect precursors to failure or significant performance degradation in the observation target device 100 by continuously detecting such instantaneous anomalies in the observation target device 100 in real time.

[0018] The anomaly detection device 20 acquires an observation signal, i.e., a time-series observation signal, which is obtained by observing the device 100 under observation, and whose value changes in the time direction. In this embodiment, the time-series observation signal is an analog signal. The time-series observation signal may also be a digital signal generated by converting the analog signal from analog to digital at a predetermined sampling interval.

[0019] If the device under observation 100 is a device that outputs AC power, the observation signal may be a signal representing the amplitude of the AC power output by the device under observation 100. Alternatively, the observation signal may be a signal detecting the ground potential or power supply potential, etc., at the device under observation 100. The observation signal may also contain white noise, etc. Based on the acquired observation signal, the anomaly detection device 20 outputs a determination signal indicating whether an anomaly has occurred in the device under observation 100 or whether the device under observation 100 is functioning normally.

[0020] The notification device 22 acquires a judgment signal from the anomaly detection device 20. The notification device 22 notifies the administrator of the judgment result indicated by the judgment signal, for example, by displaying it on a display device or by outputting it as audio from an audio output device. The notification device 22 may also transmit the judgment result indicated by the judgment signal to other devices, such as a server, via a network.

[0021] Figure 2 shows the configuration of the anomaly detection device 20 according to the first embodiment.

[0022] The anomaly detection device 20 includes an input unit 30, a reservoir unit 32, an output unit 34, a determination unit 36, and a learning unit 38.

[0023] The input unit 30 acquires the observation signal. The input unit 30 then outputs a time-series input signal corresponding to the observation signal. The input unit 30 may output the observation signal as is, or it may output the difference between the observation signal and the reference signal as the input signal. For example, the reference signal may be a triangular wave, a square wave, or a sine wave with a predetermined period. The input unit 30 supplies the input signal to the reservoir unit 32.

[0024] The reservoir unit 32 acquires the input signal and outputs multiple output spike signals, each with a waveform that is reproducible with respect to the waveform of the input signal. Each of the multiple output spike signals is a binary signal representing firing or non-firing.

[0025] Furthermore, the reservoir unit 32 outputs multiple intermediate spike signals, each with a waveform that is reproducible with respect to the waveform of the input signal. Each of the multiple intermediate spike signals is a binary signal representing firing or non-firing.

[0026] The reservoir unit 32 is a recurrent neural network that includes multiple neurons and multiple synapses, which are recursively connected internally. In this embodiment, the reservoir unit 32 is a recurrent neural network realized by analog circuits, and more specifically, a spiking neural network.

[0027] Each of the multiple output spike signals represents whether or not one of the multiple neurons has fired.

[0028] Furthermore, each of the multiple intermediate spike signals indicates whether or not a preceding neuron among multiple neurons has fired. A preceding neuron is a neuron that fires in response to an input signal, or a neuron that fires in response to one or more synaptic signals output from one or more synapses that receive an input signal. In other words, a preceding neuron is a neuron that fires only in response to the influence of an input signal, without receiving signal feedback from other neurons.

[0029] In this embodiment, the reservoir unit 32 outputs M output spike signals from the 1st to the Mth (where M is an integer of 2 or more). In addition, in this embodiment, the reservoir unit 32 outputs N intermediate spike signals from the 1st to the Nth (where N is an integer of 2 or more).

[0030] The reservoir unit 32 may be implemented using a digital computer. Alternatively, the reservoir unit 32 may be a physical reservoir composed of responsive electronic components or materials that output a signal corresponding to the input signal.

[0031] Such a reservoir unit 32 has reproducibility in that, when it receives an input signal with the same waveform, it operates in the same way and outputs multiple intermediate spike signals and multiple output spike signals with the same waveform. In other words, the reservoir unit 32 operates deterministically. Therefore, each of the multiple intermediate spike signals and multiple output spike signals output from the reservoir unit 32 is nonlinear with respect to the input signal, but it is a signal corresponding to the input signal. The reservoir unit 32 temporarily holds the information represented by the input signal internally. The internal coupling strength and connection relationships of the reservoir unit 32 are set randomly. Therefore, the reservoir unit 32 can output multiple output spike signals with randomness that mimics the mechanisms of nature. Such a reservoir unit 32 can perform recursive and dynamic signal processing, like the human brain.

[0032] The output unit 34 acquires multiple output spike signals output from the reservoir unit 32. Based on the acquired multiple output spike signals, the output unit 34 outputs multiple prediction signals. Each of the multiple prediction signals is obtained by multiplying each of the multiple output spike signals by a predetermined output weight and summing them up. In other words, the output unit 34 is a fully connected layer in a neural network that acquires multiple output spike signals and outputs multiple prediction signals.

[0033] The output unit 34 outputs multiple intermediate spike signals and the same number of prediction signals. In this embodiment, the output unit 34 outputs N prediction signals, from the 1st to the Nth.

[0034] The determination unit 36 ​​receives multiple prediction signals output from the output unit 34 and multiple intermediate spike signals output from the reservoir unit 32. Based on the multiple prediction signals and the multiple intermediate spike signals, the determination unit 36 ​​determines whether the observation target device 100 is normal or abnormal.

[0035] Here, the output weights of the output unit 34 are adjusted so that, when an observation signal obtained by observing the normally operating observation target device 100 is provided to the input unit 30, each of the multiple prediction signals matches the corresponding intermediate spike signal among the multiple intermediate spike signals generated in response to the input of the observation signal for the timing to be predicted. In this case, the intermediate spike signal to be compared is the signal generated at the timing when the observation signal for one sample after or later of the prediction signal is input. For example, the output weights of the output unit 34 are adjusted so that the first prediction signal among the N prediction signals matches the first intermediate spike signal among the N intermediate spike signals. Also, the output weights of the output unit 34 are adjusted so that the second prediction signal among the N prediction signals matches the second intermediate spike signal among the N intermediate spike signals. Furthermore, the output weights of the output unit 34 are adjusted so that the Nth prediction signal among the N prediction signals matches the Nth intermediate spike signal among the N intermediate spike signals. In other words, the output unit 34 is adjusted so that the nth prediction signal out of the N prediction signals (where n is any integer from 1 to N) matches the nth intermediate spike signal out of the N intermediate spike signals.

[0036] Therefore, the determination unit 36 ​​determines that the system is normal if each of the multiple prediction signals matches the corresponding intermediate spike signal among the multiple intermediate spike signals, and abnormal if they do not match. For example, the determination unit 36 ​​determines that the system is normal if the first prediction signal matches the first intermediate spike signal, the second prediction signal matches the second intermediate spike signal, and the Nth prediction signal matches the Nth intermediate spike signal, and abnormal if any of them do not match. In other words, the determination unit 36 ​​determines that the system is normal if all of the multiple prediction signals match the corresponding intermediate spike signals, and abnormal if any one of them does not match.

[0037] The determination unit 36 ​​then outputs a determination signal representing the determination result.

[0038] Prior to performing the abnormality detection operation, the learning unit 38 sets parameters for the input unit 30, the reservoir unit 32, the output unit 34, and the determination unit 36.

[0039] For example, the learning unit 38 sets synaptic weights for each of the multiple synapses included in the reservoir unit 32. The learning unit 38 may also set the internal connection relationships between the multiple synapses included in the reservoir unit 32 and the multiple neurons. For example, the learning unit 38 may receive random numbers when the anomaly detection device 20 is shipped from the factory or prior to its implementation, and set synaptic weights for each of the multiple synapses or set the internal connection relationships in the reservoir unit 32 according to the received random numbers.

[0040] For example, during a learning period that precedes the anomaly detection operation, the learning unit 38 provides the input unit 30 with observation signals obtained by observing the normally operating target device 100. During the learning period, the learning unit 38 adjusts the output weights in the output unit 34 so that each of the multiple prediction signals matches the corresponding intermediate spike signal among the multiple intermediate spike signals at the time the prediction signal is input. For example, the learning unit 38 adjusts the output weights so that the first prediction signal matches the first intermediate spike signal, the second prediction signal matches the second intermediate spike signal, and the nth prediction signal matches the nth intermediate spike signal. In other words, the learning unit 38 adjusts the output weights so that the nth prediction signal matches the nth intermediate spike signal.

[0041] Figure 3 shows the configuration of the reservoir section 32 according to the first embodiment.

[0042] The reservoir unit 32 includes a plurality of input firing circuits 42, a plurality of synaptic circuits 44, a plurality of adder circuits 46, and a plurality of output firing circuits 48. In this embodiment, the reservoir unit 32 includes N input firing circuits 42, (M × (N + M)) synaptic circuits 44, M adder circuits 46, and M output firing circuits 48.

[0043] Each of the multiple input firing circuits 42 receives an input signal and outputs an intermediate spike signal in response to the input signal. Each of the multiple input firing circuits 42 functions as a front-stage neuron in a recurrent neural network.

[0044] Each of the multiple input firing circuits 42 is a firing circuit that follows the leakage integral firing model. A firing circuit that follows the leakage integral firing model maintains a film potential. For example, a firing circuit that follows the leakage integral firing model maintains charge using a capacitor and maintains a voltage as the film voltage corresponding to the amount of charge stored in the capacitor.

[0045] Furthermore, a firing circuit following the leakage integral firing model increases the membrane potential in response to the input signal and decreases the membrane potential over time according to a predetermined time constant. For example, a firing circuit following the leakage integral firing model charges the capacitor by supplying an input current corresponding to the input signal, and discharges the charge stored in the capacitor through a resistor connected in parallel with the capacitor.

[0046] A firing circuit following the leakage integral firing model fires and outputs an intermediate spike signal when the membrane potential exceeds a preset threshold potential. Then, in response to the output of the intermediate spike signal, the firing circuit following the leakage integral firing model resets the membrane potential it has maintained. For example, after firing, the firing circuit following the leakage integral firing model discharges the charge accumulated in the capacitor for a predetermined time, reducing the membrane potential to below the threshold potential.

[0047] As a result, each of the multiple input firing circuits 42 can function as a circuit that models the firing function of a neuron in a spiking neural network.

[0048] Here, each of the multiple input firing circuits 42 is set to have a different time constant. For example, each of the multiple input firing circuits 42 has a different value for the resistor connected in parallel with the capacitor. As a result, each of the multiple input firing circuits 42 decreases the membrane potential at a different rate. Therefore, each of the multiple input firing circuits 42 has a different level of input signal that outputs an intermediate spike signal, i.e., a different level of input signal that fires.

[0049] In this embodiment, the first input firing circuit 42-1 among the N input firing circuits 42 outputs a first intermediate spike signal. The second input firing circuit 42-2 among the N input firing circuits 42 outputs a second intermediate spike signal. The nth input firing circuit 42-N among the N input firing circuits 42 outputs the nth intermediate spike signal.

[0050] Multiple synaptic circuits 44 are grouped into multiple groups. Each of these groups contains one or more synaptic circuits 44. Each of the multiple synaptic circuits 44 belongs to one of these groups.

[0051] In this embodiment, the multiple synaptic circuits 44 are grouped into M groups. In this embodiment, each of the M groups contains (N+M) synaptic circuits 44. For example, the (N+M) synaptic circuits 44 arranged in the row direction (horizontal direction) shown in Figure 3 belong to the same group.

[0052] Each of the multiple synaptic circuits 44 is assigned a synaptic weight. The synaptic weight may be binary, multi-level (three or more levels), or an analog value.

[0053] Each of the multiple synaptic circuits 44 receives an intermediate spike signal or an output spike signal from one of the multiple input firing circuits 42 and the multiple output firing circuits 48. For example, the intermediate spike signals output from each of the multiple input firing circuits 42 are supplied to at least one of the multiple synaptic circuits 44. Also, for example, the output spike signals output from each of the multiple output firing circuits 48 are output to at least one of the multiple synaptic circuits 44.

[0054] In this embodiment, N of the (N+M) synaptic circuits 44 included in each of the multiple groups correspond one-to-one with N input firing circuits 42. In this case, each of the N synaptic circuits 44 acquires an intermediate spike signal from the corresponding input firing circuit 42 among the N input firing circuits 42.

[0055] Furthermore, of the (N+M) synaptic circuits 44 contained in each of the multiple groups, M synaptic circuits 44 correspond one-to-one with M output firing circuits 48. In this case, each of the M synaptic circuits 44 obtains an output spike signal from the corresponding output firing circuit 48 among the M output firing circuits 48.

[0056] Each of the multiple synaptic circuits 44 outputs a synaptic signal in which the received signal is influenced by a set synaptic weight. In this embodiment, each of the multiple synaptic circuits 44 may receive a binary intermediate spike signal or a binary output spike signal and output a synaptic signal that is delayed by a time corresponding to the set synaptic weight. Alternatively, each of the multiple synaptic circuits 44 may output a synaptic signal obtained by multiplying the level of the received signal by the synaptic weight. Furthermore, each of the multiple synaptic circuits 44 may receive a pulsed signal and output a synaptic signal in which the pulse width of the received signal is changed according to the synaptic weight.

[0057] Each of the multiple adder circuits 46 receives a synaptic signal output from a synaptic circuit 44 belonging to a corresponding group of multiple synaptic circuits 44. In this embodiment, each of the multiple adder circuits 46 receives a synaptic signal output from each of the (N+M) synaptic circuits 44 belonging to a corresponding group. Then, each of the multiple adder circuits 46 outputs an added signal obtained by adding the received synaptic signals.

[0058] In this embodiment, the first adder circuit 46-1 among the M adder circuits 46 outputs a first adder signal. The second adder circuit 46-2 among the M adder circuits 46 outputs a second adder signal. The Mth adder circuit 46-M among the M adder circuits 46 outputs the Mth adder signal.

[0059] Multiple output firing circuits 48 correspond one-to-one with multiple adding circuits 46. Each of the multiple output firing circuits 48 receives an addition signal output from the corresponding adding circuit 46 among the multiple adding circuits 46.

[0060] Each of the multiple output firing circuits 48 fires and outputs an output spike signal when the value obtained by integrating the received summing signal exceeds a threshold. Each of the multiple output firing circuits 48 functions as a neuron other than the pre-stage neuron in a recurrent neural network.

[0061] Each of the multiple output firing circuits 48 is a firing circuit that follows a leakage integral firing model. Each of the multiple output firing circuits 48 may have the same circuit configuration as the input firing circuit 42. That is, each of the multiple output firing circuits 48 increases the membrane potential in response to the summing signal and decreases the membrane potential over time according to a preset time constant. Then, each of the multiple output firing circuits 48 fires and outputs an output spike signal when the membrane potential exceeds a preset threshold potential, and resets the membrane potential for a certain period of time in response to the output of the output spike signal.

[0062] However, each of the multiple output ignition circuits 48 is set to the same time constant as the others. Therefore, each of the multiple output ignition circuits 48 has the same value of the resistor connected in parallel with the capacitor. As a result, each of the multiple output ignition circuits 48 decreases the membrane potential at the same rate. Therefore, each of the multiple output ignition circuits 48 has the same level of the summing signal that outputs an output spike signal, i.e., the level of the summing signal that ignites.

[0063] In this embodiment, the first output ignition circuit 48-1 of the M output ignition circuits 48 outputs a first output spike signal when a signal is input at a specified timing. The second output ignition circuit 48-2 of the M output ignition circuits 48 outputs a second output spike signal when a signal is input at a specified timing. The Mth output ignition circuit 48-M of the M output ignition circuits 48 outputs the Mth output spike signal when a signal is input at a specified timing.

[0064] Figure 4 shows the configuration of the input ignition circuit 42.

[0065] Each of the multiple input firing circuits 42 includes a membrane potential holding circuit 52, a comparator 54, and a reset circuit 56.

[0066] The membrane potential holding circuit 52 has one terminal connected to a reference potential and the other terminal connected to the first terminal 60. The membrane potential holding circuit 52 includes a membrane capacitor 62 and a membrane resistor 64. Each of the membrane capacitor 62 and the membrane resistor 64 has one terminal connected to a reference potential and the other terminal connected to the first terminal 60. That is, the membrane capacitor 62 and the membrane resistor 64 are connected in parallel between the first terminal 60 and the reference potential.

[0067] The membrane potential holding circuit 52 receives an input current from the first terminal 60 in accordance with the input signal. This membrane potential holding circuit 52 generates a membrane potential from the first terminal 60 corresponding to the charge stored in the membrane capacitor 62.

[0068] Furthermore, the membrane potential holding circuit 52 increases the membrane potential by accumulating charge in the membrane capacitor 62 in accordance with the input current. In addition, the membrane potential holding circuit 52 discharges the charge accumulated in the membrane capacitor 62 through the membrane resistor 64, thereby decreasing the membrane potential over time.

[0069] The comparator 54 compares the membrane potential with a preset threshold potential. The comparator 54 outputs an intermediate spike signal when the membrane potential becomes greater than the threshold potential. This allows the comparator 54 to fire and output an intermediate spike signal when the membrane potential becomes greater than the threshold potential.

[0070] The reset circuit 56 resets the charge stored in the film capacitor 62 in response to the output of an intermediate spike signal. For example, after an intermediate spike signal is output and a certain amount of time has elapsed, the reset circuit 56 discharges a predetermined amount of charge stored in the film capacitor 62 and maintains that state for a certain period of time. This allows the reset circuit 56 to reduce the film potential to below the threshold potential after an intermediate spike signal is output, thereby enabling the film capacitor 62 to store charge again in accordance with the input current.

[0071] Each of these multiple input firing circuits 42 can function as a firing circuit according to the leakage integral firing model. Note that each of the multiple input firing circuits 42 may have a configuration other than that shown in Figure 4. For example, each of the multiple input firing circuits 42 may have a configuration in which the sign is reversed compared to the configuration shown in Figure 4.

[0072] Here, each of the multiple input firing circuits 42 has a different product of the capacitance of the film capacitor 62 and the resistance of the film resistor 64. For example, each of the multiple input firing circuits 42 may have the same capacitance of the film capacitor 62 but different resistance values ​​of the film resistor 64. This allows each of the multiple input firing circuits 42 to function as a firing circuit with different time constants in the leakage integral firing model.

[0073] Figure 5 shows an example of the relative thresholds and firing periods for multiple input firing circuits 42 with different input current waveforms and time constants.

[0074] In firing circuits following the leakage integral firing model, if the time constants are different, even if the threshold potential set in the comparator 54 is the same, the level of the input current at which an intermediate spike signal is output, i.e., the relative threshold, will change. For example, in firing circuits following the leakage integral firing model, the smaller the time constant, the higher the level of input current at which firing occurs. Therefore, because each of the multiple input firing circuits 42 has a different time constant, even when the same waveform input current is applied, the firing period at which an intermediate spike signal is output will differ from one another.

[0075] Figure 6 shows the simulation results of the relationship between the frequency of the intermediate spike signal and the frequency of the input signal for each of the multiple input firing circuits 42 with different time constants.

[0076] As shown in Figure 6, the output rate of the intermediate spike signal, i.e., the firing rate, of the input firing circuit 42 increases as the frequency of the input signal increases. However, the firing rate of the input firing circuit 42 increases as the time constant decreases, and decreases as the time constant increases. For example, in the input firing circuit 42 shown in Figure 4, if the capacitance of the film capacitor 62 is kept the same, the firing rate increases as the resistance value of the film resistor 64 decreases, and the firing rate decreases as the resistance value of the film resistor 64 increases.

[0077] Therefore, the multiple input firing circuits 42 output multiple intermediate spike signals with different frequency components, similar to how the input signal is frequency-divided by, for example, an FFT.

[0078] Figure 7 shows an example of a normal input current waveform, an abnormal input current waveform, and the relative threshold and firing period of multiple input firing circuits 42 with different time constants.

[0079] If, for example, an instantaneous anomaly occurs in the observation target device 100, the input current may contain an anomaly component that is only present in specific frequency components. In such cases, the multiple intermediate spike signals will have some waveforms that differ from normal, similar to when the input signal is frequency-divided by FFT, but all others will have waveforms that are identical to normal.

[0080] For example, suppose that multiple input firing circuits 42 receive an abnormal waveform that contains an abnormal component that instantaneously exceeds the relative threshold of the first time constant during a period of the normal waveform as shown in Figure 7. In this case, the intermediate spike signal output from the input firing circuit 42 with the first time constant set will differ between normal and abnormal waveforms. However, the intermediate spike signals output from each of the three input firing circuits 42 with the second to fourth time constants set will be the same waveform between normal and abnormal waveforms.

[0081] Figure 8 is a diagram illustrating the contents of the abnormality detection process in the abnormality detection device 20.

[0082] The anomaly detection device 20 generates multiple prediction signals that correspond one-to-one with multiple intermediate spike signals. The anomaly detection device 20 then learns that each of the multiple prediction signals is identical to the normal waveform of the corresponding intermediate spike signal among the multiple intermediate spike signals.

[0083] Therefore, the anomaly detection device 20 can determine whether or not an abnormal component is present for each frequency component by comparing each of the multiple predicted signals with the corresponding intermediate spike signal among the multiple intermediate spike signals at the predicted timing. If the anomaly detection device 20 generates a predicted signal whose waveform differs from the corresponding intermediate spike signal at the predicted timing, it outputs a determination signal indicating that it is an anomaly.

[0084] Thus, the anomaly detection device 20 according to this embodiment determines whether or not an abnormal component is included in the input signal for each frequency component, similar to when the input signal is frequency-divided by FFT. As a result, the anomaly detection device 20 according to this embodiment can accurately detect anomalies in the observation target device 100 with a simple configuration, even if the observation signal acquired from the observation target device 100 contains noise, without decomposing the observation signal into multiple frequency components using a digital processing circuit or the like.

[0085] (Modification of the first embodiment) Figure 9 shows a modified configuration of the reservoir section 32 according to the first embodiment.

[0086] In the first embodiment, the reservoir section 32 may have the configuration shown in Figure 9.

[0087] The modified reservoir unit 32 further includes a plurality of input synaptic circuits 72 and a plurality of input summing circuits 74. In the example shown in Figure 9, the reservoir unit 32 further includes N × N input synaptic circuits 72 and N input summing circuits 74.

[0088] Multiple input synaptic circuits 72 are grouped into multiple input groups. Each of the multiple input groups contains one or more input synaptic circuits 72. Each of the multiple input synaptic circuits 72 belongs to one of the input groups.

[0089] In this example, the multiple input synaptic circuits 72 are grouped into N input groups. Each of the N input groups contains N input synaptic circuits 72. For example, the N input synaptic circuits 72 arranged in the row (horizontal direction) as shown in Figure 9 belong to the same input group.

[0090] Each of the multiple input synaptic circuits 72 is assigned an input synaptic weight. The input synaptic weight may be binary, multi-level (three or more levels), or an analog value.

[0091] Each of the multiple input synaptic circuits 72 acquires an input signal. Each of the multiple input synaptic circuits 72 outputs an input synaptic signal to which the input signal has been influenced by a set input synaptic weight. For example, each of the multiple input synaptic circuits 72 outputs an input synaptic signal to which the input signal has been delayed by a time corresponding to the set input synaptic weight.

[0092] Each of the multiple input summing circuits 74 receives input synaptic signals output from one or more input synaptic circuits 72 that belong to the corresponding input group among the multiple input synaptic circuits 72. Then, each of the multiple input summing circuits 74 outputs an input sum signal obtained by adding the one or more input synaptic signals it has received.

[0093] In this example, the first input adder 74-1 of the N input adder circuits 74 outputs the first input adder signal. The second input adder 74-2 of the N input adder circuits 74 outputs the second input adder signal. The nth input adder 74-N of the N input adder circuits 74 outputs the nth input adder signal.

[0094] The multiple input firing circuits 42 in the modified configuration correspond one-to-one with the multiple input adding circuits 74. Each of the multiple input firing circuits 42 receives an input adding signal output from the corresponding input adding circuit 74 among the multiple input adding circuits 74, instead of an input signal.

[0095] The reservoir unit 32 according to this modified example can have reproducibility, meaning that when it receives an input signal with the same waveform, it will perform the same operation and output multiple intermediate spike signals and multiple output spike signals with the same waveform.

[0096] Furthermore, in the modified reservoir unit 32, multiple input firing circuits 42 can be arranged in parallel with multiple output firing circuits 48. As a result, the modified reservoir unit 32 can be easily designed by, for example, changing the time constant of some of the reservoir devices implemented using existing semiconductor circuits.

[0097] (Second Embodiment) Next, a second embodiment will be described. The anomaly detection system 10 according to the second embodiment has substantially the same functions and configuration as the anomaly detection system 10 according to the first embodiment. Therefore, elements having substantially the same functions and configurations are given the same reference numerals, and detailed explanations will be omitted except for differences.

[0098] Figure 10 shows the configuration of the anomaly detection device 20 according to the second embodiment.

[0099] The reservoir section 32 according to the second embodiment includes a plurality of partial reservoir sections 82.

[0100] Each of the multiple partial reservoir units 82 receives an input signal from the input unit 30. Each of the multiple partial reservoir units 82 outputs one intermediate spike signal and multiple output spike signals. In this embodiment, the reservoir unit 32 includes N partial reservoir units 82-1 to 82-N, numbered from the first to the Nth. Each of the N partial reservoir units 82-1 to 82-N outputs M output spike signals. The first partial reservoir unit 82-1 outputs a first intermediate spike signal. The second partial reservoir unit 82-2 outputs a second intermediate spike signal. The Nth partial reservoir unit 82-N outputs an Nth intermediate spike signal.

[0101] Each of the multiple partial reservoir units 82 is a recurrent neural network that includes multiple neurons and multiple synapses and is recursively connected internally. In this embodiment, the reservoir unit 32 is a recurrent neural network realized by analog circuits, and more specifically, a spiking neural network.

[0102] Each of the multiple partial reservoir units 82 may be implemented by a digital computer. Alternatively, each of the multiple partial reservoir units 82 may be a physical reservoir composed of electronic components or materials that have responsiveness to output a signal corresponding to an input signal.

[0103] Each of these multiple partial reservoir units 82 has reproducibility in that, when it receives an input signal of the same waveform, it operates identically and outputs an intermediate spike signal and multiple output spike signals of the same waveform. In other words, each of the multiple partial reservoir units 82 operates deterministically. Therefore, the intermediate spike signal and multiple output spike signals output from each of the multiple partial reservoir units 82 are nonlinear with respect to the input signal, but they are signals that correspond to the input signal. Each of the multiple partial reservoir units 82 temporarily holds information represented by the input signal internally. The internal coupling strength and connection relationships of each of the multiple partial reservoir units 82 are set randomly. Therefore, each of the multiple partial reservoir units 82 can output multiple output spike signals with randomness that mimics the mechanisms of nature. Each of these multiple partial reservoir units 82 can perform recursive and dynamic signal processing, like the human brain.

[0104] The output unit 34 according to the second embodiment includes a plurality of partial output units 84.

[0105] Multiple partial output units 84 correspond one-to-one with multiple partial reservoir units 82. Each of the multiple partial output units 84 acquires multiple output spike signals from the corresponding partial reservoir unit 82 among the multiple partial reservoir units 82. Each of the multiple partial output units 84 outputs one prediction signal based on the acquired multiple output spike signals.

[0106] In this embodiment, the output unit 34 includes N partial output units 84-1 to 84-N, numbered from the first to the Nth. The first partial output unit 84-1 outputs a first prediction signal. The second partial output unit 84-2 outputs a second prediction signal. The Nth partial output unit 84-N outputs the Nth prediction signal.

[0107] Each of the multiple prediction signals is obtained by multiplying each of the multiple output spike signals by a predetermined output weight and summing them up. In other words, each of the multiple sub-output units 84 is a fully connected layer in a neural network that acquires multiple output spike signals and outputs a single prediction signal.

[0108] Here, each of the multiple partial output units 84 is adjusted so that when an observation signal obtained by observing the normally operating observation target device 100 is given to the input unit 30, the output weight of the predicted signal it outputs matches the corresponding intermediate spike signal among the multiple intermediate spike signals at the predicted time. In other words, the multiple partial output units 84 are adjusted so that the nth predicted signal among the N predicted signals matches the nth intermediate spike signal among the N intermediate spike signals at the predicted time.

[0109] The determination unit 36 ​​acquires multiple intermediate spike signals from multiple partial reservoir units 82. The determination unit 36 ​​also acquires multiple prediction signals from multiple partial output units 84. The determination unit 36 ​​determines that the system is normal if all of the prediction signals match the corresponding intermediate spike signals, and abnormal if any one of them does not match.

[0110] Figure 11 shows the configuration of the partial reservoir section 82.

[0111] The partial reservoir unit 82 includes one input firing circuit 42, a plurality of synaptic circuits 44, a plurality of adder circuits 46, and a plurality of output firing circuits 48. In this embodiment, the partial reservoir unit 82 includes an input firing circuit 42, (M × (1 + M)) synaptic circuits 44, M adder circuits 46, and M output firing circuits 48.

[0112] The input ignition circuit 42 receives an input signal and outputs an intermediate spike signal according to the input signal.

[0113] The input firing circuit 42 has the same configuration as one of the multiple input firing circuits 42 included in the reservoir unit 32 according to the first embodiment. That is, the input firing circuits 42 included in each of the multiple partial reservoir units 82 have different time constants set compared to the input firing circuits 42 included in the other partial reservoir units 82.

[0114] In this embodiment, the multiple synaptic circuits 44 are grouped into M groups. In this embodiment, each of the M groups contains (1+M) synaptic circuits 44. For example, the (1+M) synaptic circuits 44 arranged in the row direction (horizontal direction) shown in Figure 11 belong to the same group.

[0115] Each of the multiple synaptic circuits 44 receives an intermediate spike signal or an output spike signal from either the input firing circuit 42 or one of the multiple output firing circuits 48. For example, the intermediate spike signal output from the input firing circuit 42 is supplied to at least one of the multiple synaptic circuits 44. Also, for example, the output spike signals output from each of the multiple output firing circuits 48 are output to at least one of the multiple synaptic circuits 44.

[0116] In this embodiment, one of the (1+M) synaptic circuits 44 included in each of the multiple groups acquires an intermediate spike signal from the input firing circuit 42. The M synaptic circuits 44 included in each of the multiple groups correspond one-to-one with the M output firing circuits 48. In this case, each of the M synaptic circuits 44 acquires an output spike signal from the corresponding output firing circuit 48 among the M output firing circuits 48.

[0117] In this embodiment, each of the multiple adder circuits 46 receives (1+M) synaptic signals output from (1+M) synaptic circuits 44 included in the corresponding group. Then, each of the multiple adder circuits 46 outputs an added signal obtained by adding the received (1+M) synaptic signals.

[0118] The anomaly detection device 20 according to this second embodiment, similar to the first embodiment, determines whether or not an abnormal component is included in the input signal for each frequency component, just as when the input signal is frequency-divided by FFT. As a result, the anomaly detection device 20 according to the second embodiment can accurately detect anomalies in the observation target device 100 with a simple configuration, even if the observation signal acquired from the observation target device 100 contains noise, without decomposing the observation signal into multiple frequency components using a digital processing circuit or the like.

[0119] (Modified version of the second embodiment) Figure 12 shows a modified configuration of the partial reservoir section 82 according to the second embodiment.

[0120] In the second embodiment, the partial reservoir section 82 may have the configuration shown in Figure 12.

[0121] The modified partial reservoir unit 82 further includes one or more input synaptic circuits 72 and an input summing circuit 74. In the example in Figure 12, the partial reservoir unit 82 further includes N input synaptic circuits 72 and an input summing circuit 74.

[0122] The modified input summing circuit 74 receives one or more input synaptic signals output from one or more input synaptic circuits 72. The input summing circuit 74 then outputs an input summed signal obtained by adding the one or more input synaptic signals it has received.

[0123] The modified input firing circuit 42 receives an input sum signal instead of an input signal and outputs an intermediate spike signal in accordance with the input sum signal.

[0124] The partial reservoir unit 82 in this modified form can have reproducibility, meaning that when it receives an input signal with the same waveform, it operates in the same way and outputs multiple intermediate spike signals and multiple output spike signals with the same waveform.

[0125] Furthermore, in the modified partial reservoir section 82, the input ignition circuit 42 can be arranged in parallel with a plurality of output ignition circuits 48. As a result, the modified partial reservoir section 82 can be easily designed by, for example, changing the time constant of a part of a reservoir device implemented with existing semiconductor circuits.

[0126] (Hardware configuration of information processing equipment) Figure 13 shows an example of the hardware configuration of an information processing device.

[0127] The anomaly detection device 20 may be implemented by a computer (information processing device) with a hardware configuration such as that shown in Figure 13, instead of an analog circuit. In this case, the anomaly detection device 20 comprises a CPU (Central Processing Unit) 301, RAM (Random Access Memory) 302, ROM (Read Only Memory) 303, an operation input device 304, a display device 305, a storage device 306, and a communication device 307. These components are connected by a bus.

[0128] The CPU 301 is a processor that performs arithmetic and control processing according to a program. The CPU 301 uses a predetermined area of ​​the RAM 302 as a working area and performs various processes in cooperation with programs stored in the ROM 303 and storage device 306, etc.

[0129] RAM302 is a type of memory such as SDRAM (Synchronous Dynamic Random Access Memory). RAM302 functions as a workspace for the CPU301. ROM303 is a memory that stores programs and various information in a non-rewritable format.

[0130] The operation input device 304 is an input device such as a mouse and a keyboard. The operation input device 304 receives information input from the user as an instruction signal and outputs the instruction signal to the CPU 301.

[0131] The display device 305 is a display device such as an LCD (Liquid Crystal Display). The display device 305 displays various information based on display signals from the CPU 301.

[0132] The storage device 306 is a device that writes and reads data to and from a storage medium made of semiconductors such as flash memory, or a storage medium that can record magnetically or optically. The storage device 306 writes and reads data to and from the storage medium in response to control from the CPU 301. The communication device 307 communicates with external devices via a network in response to control from the CPU 301.

[0133] A program executed on a computer has a modular structure that includes an input module, a reservoir module, an output module, a decision module, and a learning module.

[0134] This program is loaded and executed on RAM 302 by the CPU 301 (processor), thereby causing the computer to function as an input unit 30, a reservoir unit 32, an output unit 34, a determination unit 36, and a learning unit 38. Note that some or all of the input unit 30, reservoir unit 32, output unit 34, determination unit 36, and learning unit 38 may be implemented as hardware circuits.

[0135] Furthermore, programs executed on a computer are provided as files in a format that can be installed on a computer or in an executable format, recorded on computer-readable recording media such as CD-ROMs, flexible disks, CD-Rs, and DVDs (Digital Versatile Disks).

[0136] Furthermore, this program may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Alternatively, this program may be configured to be provided or distributed via a network such as the Internet. Furthermore, the program executed by the anomaly detection device 20 may be pre-installed in a ROM 303 or the like and provided accordingly.

[0137] Although embodiments of the present invention have been described above, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These novel embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents.

[0138] (Note) Furthermore, the above embodiments can be summarized in the following technical proposal.

[0139] [Technical proposal 1] An anomaly detection device for detecting abnormalities in the equipment being observed, An input unit that acquires an observation signal obtained by observing the aforementioned observation target device and outputs a time-series input signal corresponding to the acquired observation signal, A reservoir unit that outputs multiple output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. A determination unit that determines whether the observed device is normal or abnormal based on the plurality of prediction signals and outputs a determination signal representing the determination result, Equipped with, The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Anomaly detection device.

[0140] [Technical proposal 2] The determination unit determines that the observed device is abnormal if the first prediction signal among the plurality of prediction signals does not match the first intermediate spike signal, which is the intermediate spike signal output from the first input firing circuit among the plurality of input firing circuits at the prediction timing. An anomaly detection device as described in Technical Proposal 1.

[0141] [Technical proposal 3] The system further includes a learning unit that, during the learning period, provides the observation signal obtained by observing the normally operating target device to the input unit, and modifies the output weight set in the output unit so that the first prediction signal matches the first intermediate spike signal in the input at the time of prediction. An anomaly detection device as described in Technical Proposal 2.

[0142] [Technical proposal 4] The output weights are adjusted such that when the observation signal obtained by observing the normally operating target device is supplied to the input unit, the first prediction signal matches the first intermediate spike signal at the time of prediction. An anomaly detection device as described in Technical Proposal 2 or 3.

[0143] [Technical proposal 5] The reservoir unit is a recurrent neural network. An anomaly detection device described in any one of the technical proposals 1 to 4.

[0144] [Technical proposal 6] The reservoir unit is a spiking neural network. An anomaly detection device described in any one of the technical proposals 1 to 5.

[0145] [Technical proposal 7] Each of the above-mentioned input firing circuits is a firing circuit that follows a leakage integral firing model, and the time constants are different from each other. An anomaly detection device as described in Technical Proposal 6.

[0146] [Technical proposal 8] Each of the above-mentioned plurality of input firing circuits is A membrane capacitor and a membrane resistor are connected in parallel, and an input current corresponding to the input signal flows through the membrane capacitor, generating a membrane potential corresponding to the charge stored in the membrane capacitor. A comparator that outputs an intermediate spike signal when the membrane potential becomes greater than or less than the threshold potential, A reset circuit that resets the charge stored in the film capacitor in response to the output of the aforementioned intermediate spike signal, Includes, Each of the above-mentioned input firing circuits has a different product value obtained by multiplying the capacitance of the film capacitor by the resistance of the film resistor. An anomaly detection device as described in Technical Proposal 7.

[0147] [Technical proposal 9] The reservoir section is, Multiple synaptic circuits grouped into multiple groups, Multiple adding circuits provided one-to-one for each of the aforementioned groups, Multiple output ignition circuits are provided one-to-one with the aforementioned multiple addition circuits, each outputting an output spike signal. It further includes, Each of the plurality of synaptic circuits acquires the intermediate spike signal or the output spike signal from any one of the plurality of input firing circuits and the plurality of output firing circuits, and outputs a synaptic signal multiplied by a preset synaptic weight. Each of the plurality of summing circuits outputs an added signal obtained by acquiring and adding the synaptic signals from each of the one or more synaptic circuits included in the corresponding group of the plurality of groups. Each of the multiple output firing circuits receives an addition signal from the corresponding addition circuit among the multiple addition circuits, and fires to output the output spike signal when the value obtained by integrating the addition signal exceeds a threshold. An anomaly detection device described in any one of the technical proposals 1 through 8.

[0148] [Technical proposal 10] Each of the above-mentioned multiple output firing circuits is, The membrane potential is increased in accordance with the summing signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, the device fires and outputs the output spike signal, and in response to the output of the output spike signal, the membrane potential is reset for a certain period of time. Each of the aforementioned multiple output firing circuits is set to have the same time constant. An anomaly detection device as described in Technical Proposal 9.

[0149] [Technical proposal 11] The reservoir section is, Multiple input synaptic circuits grouped into multiple input groups, Multiple input summing circuits are provided one-to-one for each of the aforementioned multiple input groups, It further includes, Each of the aforementioned plurality of input synaptic circuits acquires the input signal and outputs an input synaptic signal multiplied by a preset input synaptic weight. Each of the plurality of input summing circuits outputs an input summing signal obtained by acquiring and summing the input synaptic signals from each of the one or more input synaptic circuits included in the corresponding input group among the plurality of input groups. Each of the plurality of input firing circuits receives the input sum signal from the corresponding input sum circuit among the plurality of input sum circuits instead of the input signal. An anomaly detection device as described in Technical Proposal 9 or 10.

[0150] [Technical proposal 12] The reservoir unit includes a plurality of partial reservoir units, each of which outputs the plurality of output spike signals. The output unit includes a plurality of partial output units that correspond one-to-one with the plurality of partial reservoir units, Each of the plurality of partial output units acquires the plurality of output spike signals output from the corresponding partial reservoir unit among the plurality of partial reservoir units, and outputs a prediction signal obtained by multiplying each of the acquired plurality of output spike signals by a preset output weight and summing them up. Each of the plurality of partial reservoir sections includes one of the plurality of input firing circuits. An anomaly detection device described in any one of the technical proposals 1 through 8.

[0151] [Technical proposal 13] Each of the aforementioned plurality of partial reservoir sections is One or more input synaptic circuits, Input summing circuit, It further includes, Each of the one or more input synaptic circuits acquires the input signal and outputs an input synaptic signal multiplied by a preset input synaptic weight. The input summing circuit outputs an input summed signal obtained by acquiring and adding the input synaptic signals from each of the one or more input synaptic circuits. An anomaly detection device as described in Technical Proposal 12.

[0152] [Technical proposal 14] A reservoir unit that outputs multiple output spike signals, each with a waveform that is reproducible with respect to the waveform of the time-series input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. Equipped with, The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Reservoir device.

[0153] [Technical proposal 15] An anomaly detection method for detecting anomalies in an observation target device using an anomaly detection device, The input unit of the anomaly detection device acquires an observation signal obtained by observing the device under observation, and outputs a time-series input signal corresponding to the acquired observation signal. The reservoir unit of the anomaly detection device outputs a plurality of output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal. The input section of the anomaly detection device outputs a plurality of prediction signals, each of which is obtained by multiplying each of the plurality of output spike signals by a preset output weight and summing them up. The determination unit of the abnormality detection device determines whether the observed device is normal or abnormal based on the plurality of prediction signals, and outputs a determination signal representing the determination result. The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Anomaly detection device.

[0154] [Technical proposal 16] A program for causing an information processing device to function as an anomaly detection device that detects abnormalities in the device being observed, The aforementioned information processing device An input unit that acquires an observation signal obtained by observing the aforementioned observation target device and outputs a time-series input signal corresponding to the acquired observation signal, A reservoir unit that outputs multiple output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. A determination unit that determines whether the observed device is normal or abnormal based on the plurality of prediction signals and outputs a determination signal representing the determination result, and make it work The reservoir section includes a plurality of input firing sections, Each of the aforementioned plurality of input firing units is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing units is set to have a different time constant. program. [Explanation of Symbols]

[0155] 10 Anomaly detection system 100 Observation Target Devices 20 Anomaly detection device 22 Notification device 30 Input section 32 Reservoir section 34 Output section 36 Judgment section 38 Learning Department 42 Input ignition circuit 44 Synaptic Circuits 46 Adding Circuit 48 Output ignition circuit 52 Membrane potential holding circuit 54 Comparator 56 Reset circuit 60 1st terminal 62 Membrane Capacitors 64 Film Resistors 72 Input Synaptic Circuits 74 Input Adder Circuit 82 Partial reservoir section 84 Partial output section

Claims

1. An anomaly detection device for detecting abnormalities in the equipment being observed, An input unit that acquires an observation signal obtained by observing the aforementioned observation target device and outputs a time-series input signal corresponding to the acquired observation signal, A reservoir unit that outputs multiple output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. A determination unit that determines whether the observed device is normal or abnormal based on the plurality of prediction signals and outputs a determination signal representing the determination result, Equipped with, The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Anomaly detection device.

2. The determination unit determines that the observation target device is abnormal if the first prediction signal among the plurality of prediction signals does not match the first intermediate spike signal, which is the intermediate spike signal output from the first input firing circuit among the plurality of input firing circuits at the time of input of the prediction timing. An anomaly detection device according to claim 1.

3. The system further includes a learning unit that, during the learning period, provides the observation signal obtained by observing the normally operating target device to the input unit, and modifies the output weight set in the output unit so that the first prediction signal matches the first intermediate spike signal at the time of input for prediction timing. An anomaly detection device according to claim 2.

4. The output weights are adjusted such that when the observation signal obtained by observing the normally operating observation target device is supplied to the input unit, the first prediction signal matches the first intermediate spike signal at the time of input for prediction timing. An anomaly detection device according to claim 2.

5. The reservoir unit is a recurrent neural network. An anomaly detection device according to claim 1.

6. The reservoir unit is a spiking neural network. An anomaly detection device according to claim 1.

7. Each of the above-mentioned input firing circuits is a firing circuit that follows a leakage integral firing model, and the time constants are different from each other. An anomaly detection device according to claim 6.

8. Each of the above-mentioned plurality of input firing circuits is A membrane capacitor and a membrane resistor are connected in parallel, and an input current corresponding to the input signal flows through the membrane capacitor, generating a membrane potential corresponding to the charge stored in the membrane capacitor. A comparator that outputs an intermediate spike signal when the membrane potential becomes greater than or less than the threshold potential, A reset circuit that resets the charge stored in the film capacitor in response to the output of the aforementioned intermediate spike signal, Includes, Each of the above-mentioned input firing circuits has a different product value obtained by multiplying the capacitance of the film capacitor by the resistance of the film resistor. An anomaly detection device according to claim 7.

9. The reservoir section is, Multiple synaptic circuits grouped into multiple groups, Multiple adding circuits provided one-to-one for each of the aforementioned groups, Multiple output ignition circuits are provided one-to-one with the aforementioned multiple addition circuits, each outputting an output spike signal. It further includes, Each of the plurality of synaptic circuits acquires the intermediate spike signal or the output spike signal from any one of the plurality of input firing circuits and the plurality of output firing circuits, and outputs a synaptic signal multiplied by a preset synaptic weight. Each of the plurality of summing circuits outputs an added signal obtained by acquiring and adding the synaptic signals from each of the one or more synaptic circuits included in the corresponding group of the plurality of groups. Each of the multiple output firing circuits receives an addition signal from the corresponding addition circuit among the multiple addition circuits, and fires to output the output spike signal when the value obtained by integrating the addition signal exceeds a threshold. An anomaly detection device according to claim 1.

10. Each of the above-mentioned multiple output firing circuits is, The membrane potential is increased in accordance with the summing signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, the device fires and outputs the output spike signal, and in response to the output of the output spike signal, the membrane potential is reset for a certain period of time. Each of the aforementioned multiple output firing circuits is set to have the same time constant. An anomaly detection device according to claim 9.

11. The reservoir section is, Multiple input synaptic circuits grouped into multiple input groups, Multiple input summing circuits are provided one-to-one for each of the aforementioned multiple input groups, It further includes, Each of the aforementioned plurality of input synaptic circuits acquires the input signal and outputs an input synaptic signal multiplied by a preset input synaptic weight. Each of the plurality of input summing circuits outputs an input summing signal obtained by acquiring and summing the input synaptic signals from each of the one or more input synaptic circuits included in the corresponding input group among the plurality of input groups. Each of the plurality of input firing circuits receives the input sum signal from the corresponding input sum circuit among the plurality of input sum circuits instead of the input signal. An anomaly detection device according to claim 9.

12. The reservoir unit includes a plurality of partial reservoir units, each of which outputs the plurality of output spike signals. The output unit includes a plurality of partial output units that correspond one-to-one with the plurality of partial reservoir units, Each of the plurality of partial output units acquires the plurality of output spike signals output from the corresponding partial reservoir unit among the plurality of partial reservoir units, and outputs a prediction signal obtained by multiplying each of the acquired plurality of output spike signals by a preset output weight and summing them up. Each of the plurality of partial reservoir sections includes one of the plurality of input firing circuits. An anomaly detection device according to claim 1.

13. Each of the aforementioned plurality of partial reservoir sections is One or more input synaptic circuits, Input summing circuit, It further includes, Each of the one or more input synaptic circuits acquires the input signal and outputs an input synaptic signal multiplied by a preset input synaptic weight. The input summing circuit outputs an input summed signal obtained by acquiring and adding the input synaptic signals from each of the one or more input synaptic circuits. An anomaly detection device according to claim 12.

14. A reservoir unit that outputs multiple output spike signals, each with a waveform that is reproducible with respect to the waveform of the time-series input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. Equipped with, The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Reservoir device.

15. An anomaly detection method for detecting anomalies in an observation target device using an anomaly detection device, The input unit of the anomaly detection device acquires an observation signal obtained by observing the device under observation, and outputs a time-series input signal corresponding to the acquired observation signal. The reservoir unit of the anomaly detection device outputs a plurality of output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal. The output unit of the anomaly detection device outputs a plurality of prediction signals, each of which is obtained by multiplying each of the plurality of output spike signals by a preset output weight and summing them up. The determination unit of the abnormality detection device determines whether the observed device is normal or abnormal based on the plurality of prediction signals, and outputs a determination signal representing the determination result. The reservoir section includes a plurality of input firing circuits, Each of the above-mentioned plurality of input firing circuits is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing circuits is set to have a different time constant. Anomaly detection method.

16. A program for causing an information processing device to function as an anomaly detection device that detects abnormalities in the device being observed, The aforementioned information processing device An input unit that acquires an observation signal obtained by observing the aforementioned observation target device and outputs a time-series input signal corresponding to the acquired observation signal, A reservoir unit that outputs multiple output spike signals, each having a waveform that is reproducible with respect to the waveform of the input signal, Each of the output units outputs multiple prediction signals obtained by multiplying each of the multiple output spike signals by a preset output weight and summing them up. A determination unit that determines whether the observed device is normal or abnormal based on the plurality of prediction signals and outputs a determination signal representing the determination result, and make it work The reservoir section includes a plurality of input firing sections, Each of the aforementioned plurality of input firing units is The membrane potential is increased in response to the input signal, and the membrane potential is decreased over time according to a preset time constant. When the membrane potential exceeds a preset threshold potential, it fires and outputs an intermediate spike signal, and in response to the output of the intermediate spike signal, the membrane potential is reset for a certain period of time. Each of the above-mentioned input firing units is set to have a different time constant. program.

Images