Nonlinear dynamic reservoir network time series prediction system and implementation method

Abstract

The embodiment of the specification provides a kind of nonlinear dynamic reservoir network timing prediction system and implementation method, wherein the system includes: microcontroller, built-in pulse width modulation PWM module, for receiving input signal, the input signal is preprocessed, obtain the input signal after preprocessing, the input signal after processing is encoded by the pulse width modulation PWM module, obtain PWM encoding signal;Multiple linear resistance-capacitance R-C network, wherein each R-C network has different resistance value and capacitance value;For realizing reservoir layer with distributed nonlinear characteristics and diversified time dynamic characteristics, each R-C network in the storage pool is connected in parallel, the PWM encoding signal is refitted, and the final output signal is obtained;Linear regression network is used to obtain the final output signal, and the classification and prediction task of time series data is executed.

Description

Technical Field

[0001] This document relates to the field of computational modeling technology, and in particular to a nonlinear dynamic reservoir network time series prediction system and its implementation method. Background Technology

[0002] At the forefront of current research in artificial intelligence and machine learning, deep neural networks (DNNs) have been widely applied in various disciplines such as computer vision, speech recognition, and image processing. However, with the diversification of computing demands, especially in resource-constrained edge computing environments, the need for lightweight model design and efficient inference capabilities is becoming increasingly urgent. Against this backdrop, reservoir computing (RC), as a parameter-economical and simplified neural network architecture, is gradually becoming a promising alternative to traditional DNN architectures. RC models, based on optimizations of recurrent neural networks (RNNs), effectively handle temporal data, significantly reduce training complexity, and exhibit excellent hardware compatibility.

[0003] Physical Reservoir Computing (PRC), as an innovative neuromorphic computing paradigm, fully leverages the dynamic physical characteristics of electronic devices to achieve highly parallelized signal processing and optimized energy efficiency. Based on nonlinear dynamical systems theory, PRC performs complex nonlinear transformations and information encoding operations on input signals through a recursively interconnected physical network. Its significant features include high computational efficiency, structural simplicity, and support for fully analog computing architectures. In the field of brain-inspired computing, research on PRC systems is becoming a key issue, with its application prospects in high-tech fields such as artificial intelligence, edge computing, and nonlinear dynamical systems attracting considerable attention. Compared to traditional digital computing methods, PRC systems theoretically offer significant improvements in computational performance and reductions in energy consumption, providing new scientific basis and research paths for the future development of computing technology.

[0004] Analog circuits, such as rotating neuron architectures, and dynamic systems consisting of a single nonlinear node subject to delayed feedback, can replace the typical structure with multiple connected nodes. However, analog circuits are susceptible to factors such as temperature, power supply fluctuations, and component aging, which can lead to system instability and reduced reliability. Their accuracy is typically lower than that of digital circuits, which may limit the computational accuracy of PRC (Programmable Logic Controller).

[0005] Using an analog memristor array as the output layer, its nonlinearity and memory properties are leveraged to simulate the behavior of neurons and synapses in the human brain, thereby reducing overall power consumption and hardware overhead and achieving a highly integrated memristor storage system. However, such RC systems typically have a non-adjustable timescale and a relatively fixed nonlinear transition function, which limits their information processing capabilities. Furthermore, memristor devices are expensive and currently remain in the laboratory stage, unable to be commercialized.

[0006] Optical circuits utilize precise control over the generation, modulation, transmission, and detection of light waves, employing optical elements to construct complex optical paths to achieve dynamic interference and diffraction of light waves. Such high-precision optical instruments require precise optical elements, highly pure materials, and specialized processing techniques. Furthermore, the complex optical paths involved in their construction make them prone to errors, resulting in high research and manufacturing costs. Optical signals may also encounter losses such as scattering, diffraction, and absorption during transmission, affecting computational accuracy.

[0007] Small-scale quantum computers are being developed, such as nanoparticle-permeable networks (PNNs) composed of metallic nanoparticles deposited on atomically smooth insulating substrates. Nanoparticles are novel self-assembled nanoscale systems with scale-free and small-world properties. Neuromorphic devices based on memristor nanowire networks (NWNs) utilize synaptic-like changes in conductance at nanowire-nanowire intersections to exhibit resistive memory switching in response to electrical input. However, most of these approaches remain in the theoretical and simulation stages and have not yet been used to perform actual computational tasks; practical applications are still a long way off.

[0008] In summary, existing PRC systems typically rely on nonlinear semiconductor devices or emerging nanodevices, which are susceptible to challenges such as temperature instability and device variability, including stochastic cycle-to-cycle variations and limited robustness. Furthermore, task mismatch and limited device characteristics reduce the performance and efficiency of PRC systems across various computational tasks. While stochastic device-to-device (D2D) variations can enhance reservoir state diversity, they hinder reproducibility and complicate large-scale inference applications. Summary of the Invention

[0009] The purpose of this invention is to provide a nonlinear dynamic reservoir network time series prediction system and implementation method, aiming to solve the above-mentioned problems in the prior art.

[0010] This invention provides a nonlinear dynamic reservoir network time series prediction system, comprising: a microcontroller, multiple linear resistor-capacitor (RC) networks connected to the microcontroller, and a linear regression network.

[0011] The microcontroller has a built-in pulse width modulation (PWM) module for receiving input signals, preprocessing the input signals to obtain preprocessed input signals, and then performing PWM encoding on the processed input signals through the PWM module to obtain PWM encoded signals.

[0012] Multiple linear resistor-capacitor (RC) networks, each with different resistance and capacitance values, are used to implement a reservoir layer with distributed nonlinear characteristics and diverse time-dynamic properties. Each RC network in the reservoir is connected in parallel to refit the PWM encoded signal to obtain the final output signal.

[0013] A linear regression network is used to obtain the final output signal and perform classification and prediction tasks for time series data.

[0014] This invention provides a method for implementing time series prediction of nonlinear dynamic reservoir networks, used in the aforementioned time series prediction system for nonlinear dynamic reservoir networks. The method specifically includes:

[0015] The microcontroller receives the input signal, preprocesses the input signal to obtain the preprocessed input signal, and then performs PWM encoding on the processed input signal through the pulse width modulation (PWM) module to obtain the PWM encoded signal.

[0016] A reservoir layer with distributed nonlinear characteristics and diverse time dynamic properties is realized by using multiple linear resistor-capacitor RC networks. The PWM encoded signal is then refitted to obtain the final output signal.

[0017] The final output signal is obtained through a linear regression network, and the task of classifying and predicting time series data is performed.

[0018] This invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-described nonlinear dynamic reservoir network timing prediction method.

[0019] This invention also provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, implements the steps of the above-described nonlinear dynamic reservoir network timing prediction method.

[0020] By employing embodiments of the present invention, a PRC system with advanced performance in multiple computational tasks is achieved through optimizing the parameter configuration of the linear resistor-capacitor (RC) network, introducing nonlinear characteristics, simplifying the training process, and conducting performance evaluation and system robustness testing. These embodiments not only reduce the number of parameter nodes and simplify the training process, but also construct a grouped reservoir with highly nonlinear characteristics by precisely adjusting the RC network parameters, thereby demonstrating superior performance and environmental adaptability in multiple fields such as arrhythmia classification and time series prediction. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of a nonlinear dynamic reservoir network time series prediction system according to an embodiment of the present invention;

[0023] Figure 2 A physical schematic diagram of the nonlinear dynamic reservoir network time-series prediction system according to an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram illustrating the comprehensive analysis of the nonlinear and time-dynamic characteristics of the RC combined reservoir system according to an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram illustrating the frequency adaptability and prediction results of the grouped RC circuit reservoir according to an embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram of the inter-cell transformation and temperature sensitivity response of the grouped RC circuit reservoir system according to an embodiment of the present invention;

[0027] Figure 6 This is a flowchart of the nonlinear dynamic reservoir network time-series prediction implementation method according to an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0030] System Implementation Examples

[0031] According to embodiments of the present invention, a nonlinear dynamic reservoir network time series prediction system is provided. Figure 1 This is a schematic diagram of a nonlinear dynamic reservoir network time-series prediction system according to an embodiment of the present invention, as shown below. Figure 1 As shown, the nonlinear dynamic reservoir network time series prediction system according to an embodiment of the present invention specifically includes: a microcontroller 10, multiple linear resistor-capacitor RC networks 12 connected to the microcontroller, and a linear regression network 14.

[0032] The microcontroller 10 has a built-in pulse width modulation (PWM) module 100 for receiving input signals, preprocessing the input signals to obtain preprocessed input signals, and then performing PWM encoding on the processed input signals through the PWM module 100 to obtain PWM encoded signals. Specifically, the microcontroller 10 is used to: perform mask processing on the input signals by multiplying the input signals by a mask matrix composed of 1s and -1s to generate N virtual nodes; and perform PWM encoding on the processed input signals within the input amplitude range of 0-T through the PWM module, while maintaining a constant pulse voltage, to obtain the PWM encoded signal, and the output voltage V at the last moment T. out(T) As shown in Formula 1:

[0033]

[0034] Where t is the encoded input width, τ is determined by the hardware, and τ is the time unit, which is calculated as: τ=R*C, where R is the resistance and C is the capacitance.

[0035] Multiple linear resistor-capacitor RC networks 12, each with different resistance and capacitance values, are used to realize a reservoir layer with distributed nonlinear characteristics and diverse time dynamic properties. Each RC network in the reservoir is connected in parallel to refit the PWM encoded signal to obtain the final output signal. The multiple linear resistor-capacitor RC networks 12 consist of 8 parallel RC networks, each with a fixed resistance of 40kΩ and capacitance values ​​of 10nF, 25nF, 40nF, 55nF, 70nF, 85nF, 100nF, and 115nF, respectively. The 8 parallel RC networks are specifically used to output the final output signal of 8*N reservoir states, where N is the amount of data after mask processing during preprocessing.

[0036] Linear regression network 14 is used to obtain the final output signal and perform time series data classification and prediction tasks. Specifically, the linear regression network is used for:

[0037] Obtain the final output signal and perform the classification and prediction task of the time series data according to Formula 2:

[0038] W out =Y target X T (XX T ) + Formula 2;

[0039] Among them, W out Y represents the weight values ​​trained using linear regression. target X represents the target value of the corresponding prediction task. T This represents the transpose of the matrix obtained after passing through the RC network, where X represents the matrix obtained after passing through the RC network, T represents the transpose of the matrix, and + represents the pseudo-inverse of the matrix.

[0040] Specifically, in this embodiment of the invention, the RC model (Reservoir Computing) typically follows a three-layer architecture: an input layer for receiving and preprocessing data, a reservoir layer where the input signal triggers nonlinear dynamics, and an output layer that recombines the signal to produce the final output. Since the input layer is directly connected to the reservoir layer, only the weights need to be trained, thus significantly reducing the number of training weight parameters and computation. In general physical reservoir computing devices, nonlinearity and memory effects are typically required. In this product, a resistor-capacitor series connection is used, based on the capacitor charging formula:

[0041] V out(T) =V in (1-e -t / τ )

[0042] Voltage and time exhibit an exponential function relationship; the voltage at one time point affects the voltage at the next time point, thus satisfying the characteristic of memory. Furthermore, since τ = RC, the voltage-time curve differs when the resistance value R or the capacitance C changes, resulting in different time characteristics for each circuit.

[0043] like Figure 2 As shown, the system in this embodiment of the invention mainly consists of a microcontroller system board (STM32 microcontroller [MCU] development board, blue part) serving as the input layer and eight parallel RC circuits mounted on a custom test board. For the eight parallel RC circuits, the resistance is fixed at 40kΩ, and the capacitance values ​​are 10nF, 25nF, 40nF, 55nF, 70nF, 85nF, 100nF, and 115nF, respectively. The test board also provides eight output pins for sampling and measuring the capacitor voltage for subsequent linear regression calculations. The eight resistor-capacitor series circuits have different parameters, thereby increasing the diversity of time dynamic responses.

[0044] Since RC circuits are often considered linear systems, this embodiment of the invention introduces pulse width modulation (PWM) coding at the input layer to address this limitation, introducing necessary nonlinearity into the RC circuit. The input signals originate from different tasks, such as arrhythmia detection, where the input signals are heartbeat data from different patients at different time points. During input signal preprocessing, a masking process is used, multiplying the input signal by a mask matrix composed of 1s and -1s to generate N virtual nodes. After preprocessing, the input signal is converted into a PWM signal by the PWM module inside the microcontroller. This signal is characterized by mapping the magnitude of the input voltage to the input time t; therefore, the larger the voltage, the longer the input voltage time. For example, if the input signal amplitude is encoded as 0–3.3V, after PWM, it will be mapped to 0–T. Specifically, the input amplitude is encoded within the range of 0–T while maintaining a constant pulse voltage of 3.3V (MCU power supply voltage). In this case, the output voltage at the last moment T can be expressed as:

[0045]

[0046] Where t is the encoded input width. In this way, each input signal will generate 8 (i.e., 8 RC circuits) × N (N is the size of the data after masking) reservoir states, which are then fed into a linear regression network for calculation, thereby performing time series classification or prediction tasks.

[0047] The subsequent linear regression is a simple calculation used to determine the weights in the prediction equation, as shown in the following formula: W out =Ytarget X T (XX T ) + This allows us to obtain the weight values.

[0048] To address the aforementioned technical problems, this invention relates to an efficient computational model, specifically a Physical Reservoir Computing (PRC) system. This system aims to optimize a miniaturized, highly efficient, and fast inference computational architecture, and is adapted for resource-constrained environments. The PRC system of this invention significantly reduces the number of parameter nodes and simplifies the training process. The detailed technical implementation scheme is as follows:

[0049] Reservoir Network Structure: The PRC system in this embodiment of the invention consists of eight unique linear resistor-capacitor (RC) networks, each with different resistance and capacitance parameter values, realizing multi-performance combinations of device-to-device (D2D).

[0050] Introduction of nonlinear characteristics: The input signal is encoded using the pulse width modulation (PWM) module built into the microcontroller, thereby introducing nonlinear characteristics and enhancing the system's computing power.

[0051] Training process optimization: By inputting the PWM encoded signal into the linear regression network to perform the classification and prediction tasks of time series data, this invention uses a simplified regression operation to replace the traditional backpropagation algorithm, which greatly shortens the training cycle and reduces resource consumption.

[0052] Parameter tuning and characteristic configuration: By precisely adjusting the resistance and capacitance parameters in the RC network, the distributed nonlinearity and diverse time dynamic characteristics of the reservoir were realized, and a grouped reservoir with highly nonlinear characteristics was constructed.

[0053] Performance Evaluation: In multiple computational tasks, including but not limited to arrhythmia classification, NARMA2 time series prediction, Mackey-Glass time series analysis, and Henon mapping prediction, this PRC system demonstrated state-of-the-art performance while minimizing reservoir size.

[0054] System robustness testing:

[0055] To verify the robustness of the present invention, the following test procedure was performed:

[0056] Device fabrication: Two RC circuit devices with the same parameters but different manufacturing processes were fabricated and labeled as device A and device B, respectively.

[0057] Weight training and transfer: Linear regression training is performed on device A under room temperature (RT) conditions to obtain reservoir weights.

[0058] Performance verification: The weights obtained from training device A are applied to device B, and performance tests are conducted at RT and a high temperature of 85℃.

[0059] Test results show that device B can achieve accurate prediction performance under different environmental conditions, which confirms the high robustness and environmental adaptability of this PRC system.

[0060] The technical solutions of the embodiments of the present invention will be further described below with reference to the illustrative figures.

[0061] 1. Heart Arrhythmia Detection Task

[0062] Heart diseases were categorized into four classes (A, L, V, and N) according to the standards provided by the American Association for Medical Devices (AAMI), and training and testing were conducted using the MIT-BIH arrhythmia database established by MIT. The MIT-BIH dataset contains electrocardiogram (ECG) recordings from 48 different subjects. First, 750 heartbeat data segments were selected from each of the four classes and randomly concatenated to form a long ECG waveform sequence of 3000 heartbeat segments. Then, the data was resampled at a frequency of 80Hz (input data interval = 12.5ms) and normalized. The preprocessing process is as follows: Figure 3 As shown in a and b in the figure.

[0063] For this classification task, this invention employs a D2D (device-to-device) method (combining eight RC circuits in parallel to form a reservoir layer), using virtual nodes with a node count of 5, and setting the mask data interval to T = 2.5 ms. The reservoir states of 40 nodes are then applied to the output layer (a 41×4 network), and classification data is generated at each time step. When the output of the corresponding neuron in the readout layer reaches its maximum value within a set time window, we consider the classification correct. Figure 3 As shown in c, the recognition results of the target signals and output signals of the four types of tags are as follows. Figure 3 The confusion matrix is ​​shown in the figure, where d is the output result of eight identical RC circuits connected in parallel, with a time constant of 2.2 ms and a nonlinear coefficient of 1.8 for each. Each RC sub-reservoir has a different mask sequence. Based on this, the present invention further proposes to combine eight RC circuits with different parameters in parallel, such as... Figure 3The output result can be seen in the figure of e. When using eight RC circuits with different parameter values, the output signal can fit the target signal more effectively and produce a peak at the corresponding heartbeat. Ultimately, the embodiment of the present invention achieved a 94% accuracy rate in classifying these four types of arrhythmias. In contrast, when the input signal is directly input into the linear regression network, the recognition accuracy is only 25%, and when using eight identical sub-reservoirs, an accuracy rate of less than 67% is ultimately achieved. The recognition accuracy comparison results are as follows: Figure 3 As shown in f.

[0064] These results highlight the feasibility and effectiveness of the grouped reservoir proposed in the embodiments of the present invention, which has extensive nonlinear and temporal characteristics and can process complex time-series signals with high accuracy.

[0065] 2. The versatility and applicability of grouped RC circuits

[0066] Real-world applications, such as predicting motion trajectories, typically require the ability to handle a wide range of velocity conditions, which in turn demands high adaptability from the PRC system. To simulate such scenarios, an improved Henon map prediction task was implemented, where the signal frequency varied from 10 Hz to 50 Hz, with each frequency segment consisting of 100 time steps. This variation was designed to test the reservoir's ability to manage different temporal dynamics. Reservoirs with fixed temporal dynamics are generally unsuitable for such tasks. For example, in experiments, a reservoir consisting of eight parallel, identical sub-reservoirs, each with a t of 400 μs and different mask sequences, predicted a 50 Hz signal quite well. However, as... Figure 4 As shown in a and d, this configuration performs poorly with 10Hz signals. This illustrates the limitations of using reservoirs with a narrow time constant range, which are effective at high frequencies but ineffective at slower dynamics. In contrast, another configuration using parallel reservoirs and setting a longer t (4.6ms) accurately predicts 10Hz signals, as shown in a diagram. Figure 4 As shown in b and e in the diagram. However, it cannot effectively handle 50Hz signals, further highlighting the limitations of a single device in processing dynamic data.

[0067] The reservoir optimized in this embodiment of the invention provides a wide range of time constants, enabling it to handle multiple time scales simultaneously. For example... Figure 4 As shown in c and f, this configuration allows the system to accurately predict 10Hz and 50Hz signals within the same frame without requiring any feedback control. Furthermore, this broad adaptability can be further enhanced by configuring the t-distribution to span a wider range, thus providing an effective solution for complex multi-frequency tasks.

[0068] 3. Robust performance of the PRC system

[0069] To verify the robustness of the embodiments of the present invention, two RC reservoir devices, labeled Circuit A and Circuit B, were fabricated, and they have the same parameters. Figure 5 As shown in 'a', to simulate maximum C2C ​​(copy-to-copy) variability, circuit A was assembled using machine soldering, while circuit B was assembled using hand soldering techniques. Therefore, despite setting the same parameters, the difference in manufacturing processes resulted in differences in system parameters. Here, circuit A was used as a reference, and linear regression was trained at room temperature (RT) to read out the layer weights. The trained weights were then applied to circuit B, and circuit B was tested at RT and 85°C, respectively.

[0070] Figure 5 The figure below (bd) shows the prediction performance of circuit B under RT for three time series tasks (NARMA2, Mackey-Glass, and Henon map). It can be seen that the predicted output (blue curve) closely matches the actual situation (grey curve), indicating that when the weights trained by circuit A are applied to circuit B, circuit B can effectively capture complex temporal dynamics. Figure 5 The figure in eg shows the performance of circuit B performing the same task at 85°C. The predicted output (red curve) still matches the actual situation very well, indicating that circuit B maintains stable and accurate performance even at high temperatures. Figure 5 The value of 'h' in the figure compares the relative error of the embodiments of the present invention with that of memristors and ReLU networks in three prediction tasks under C2C (copy-to-copy), 80°C, and a combination of the above two conditions. It can be seen that the embodiments of the present invention achieve the lowest relative error in all three tasks and various operating environments. This indicates that the method of the embodiments of the present invention ensures the stable performance of the present invention under different tasks and temperature variations, verifying the robustness of the present invention.

[0071] In summary, the Physical Reservoir (PRC) system employed in this embodiment of the invention introduces the necessary nonlinear characteristics of the reservoir network by utilizing Pulse Width Modulation (PWM), enabling the reservoir to better capture complex temporal dynamic information, such as chaotic sequences. By modulating the pulse width, a richer dynamic response is generated, thereby enhancing classification and prediction performance. Secondly, by directly using analog PWM, the need for a digital-to-analog converter (ADC) can be eliminated, avoiding problems such as high power consumption, high latency, and impact on system performance, achieving low power consumption, low cost, and greater commercial viability. Furthermore, this embodiment of the invention can flexibly adapt to multiple time scales, employing a wider range of nonlinearities and time constants, thereby improving the performance and versatility of the RC system. By covering a broader range, the reservoir can effectively handle various tasks without requiring task-specific modulation. This embodiment of the invention uses a digital readout layer to create a hybrid system, where Kirchhoff's laws can be directly applied since the output is already in voltage form, without the need for resistors to convert current to voltage. Furthermore, the robustness of the device to parameter and temperature changes makes the fully analog loop of the PRC system in this embodiment of the invention an ideal candidate for large-scale, low-power, high-speed edge computing applications in real-world scenarios.

[0072] Method Implementation Examples

[0073] According to embodiments of the present invention, a method for implementing time series prediction of nonlinear dynamic reservoir networks is provided, which is used in the aforementioned time series prediction system for nonlinear dynamic reservoir networks. Figure 6 This is a flowchart of the nonlinear dynamic reservoir network time-series prediction implementation method according to an embodiment of the present invention, as follows: Figure 6 As shown, the specific features of the embodiments of the present invention include:

[0074] Step S601 involves receiving an input signal via a microcontroller, preprocessing the input signal to obtain a preprocessed input signal, and then using the pulse width modulation (PWM) module to perform PWM encoding on the processed input signal to obtain a PWM encoded signal; specifically including:

[0075] The input signal is processed by masking, which involves multiplying the input signal by a mask matrix consisting of 1s and -1s to generate N virtual nodes.

[0076] The pulse width modulation (PWM) module performs PWM encoding on the processed input signal within the input amplitude range of 0-T, while maintaining a constant pulse voltage, to obtain the PWM encoded signal. The output voltage V at the last moment T is... out(T) As shown in Formula 1:

[0077]

[0078] Where t is the encoded input width, τ is determined by the hardware, and τ is the time unit, which is calculated as: τ=R*C, where R is the resistance and C is the capacitance.

[0079] Step S602 involves implementing a reservoir layer with distributed nonlinear characteristics and diverse time dynamic properties through multiple linear resistor-capacitor RC networks, and refitting the PWM encoded signal to obtain the final output signal; specifically, this includes outputting the final output signal of 8*N reservoir states, where N is the amount of data after mask processing during preprocessing;

[0080] Step S603 involves obtaining the final output signal through a linear regression network and performing the classification and prediction task for the time series data. Specifically, this includes:

[0081] Obtain the final output signal and perform the classification and prediction task of the time series data according to Formula 2:

[0082] W out =Y target X T (XX T ) + Formula 2;

[0083] Among them, W out Y represents the weight values ​​trained using linear regression. target X represents the target value of the corresponding prediction task. T The expression represents the transpose of the matrix obtained after passing through the RC network. X represents the matrix obtained after passing through the RC network, T represents the transpose of the matrix, and + represents the pseudo-inverse of the matrix.

[0084] The embodiments of the present invention are method embodiments corresponding to the system embodiments described above. The specific processing of each step can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0085] Device Example 1

[0086] This invention provides an electronic device, such as... Figure 7 As shown, it includes: a memory 70, a processor 72, and a computer program stored in the memory 70 and executable on the processor 72, wherein the computer program, when executed by the processor 72, performs the steps as described in the method embodiment.

[0087] Device Example 2

[0088] This invention provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor 72, performs the steps described in the method embodiment.

[0089] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A nonlinear dynamic reservoir network timing prediction system, characterized by, The method comprises the following steps: a microcontroller, a plurality of linear resistance-capacitance (R-C) networks connected to the microcontroller, and a linear regression network: the microcontroller is internally provided with a pulse width modulation (PWM) module, is configured to receive an input signal, pre-process the input signal to obtain a pre-processed input signal, and perform PWM coding on the pre-processed input signal through the PWM module to obtain a PWM coded signal; the plurality of linear R-C networks, each of which has different resistance and capacitance values, are configured to implement a reservoir layer with distributed nonlinear characteristics and diversified time dynamic characteristics, each R-C network in the reservoir layer is connected in parallel, and the PWM coded signal is refitted to obtain a final output signal; the linear regression network is configured to obtain the final output signal and perform a classification and prediction task on time series data.

2. The nonlinear dynamic reservoir network timing prediction system of claim 1, wherein, The microcontroller is specifically configured to perform mask processing on the input signal, multiply the input signal by a mask matrix composed of 1 and -1, and generate N virtual nodes.

3. The nonlinear dynamic reservoir network timing prediction system of claim 1, wherein, The microcontroller is specifically used for: through the pulse width modulation (PWM) module, the input signal is processed, and the PWM encoding is carried out in the range of input amplitude 0-T, while the constant pulse voltage is kept, the PWM encoding signal is obtained, and the output voltage of the last moment As shown in formula 1: Formula 1 ; Wherein, t is the coded input width, Indicated in time units, τ = R * C, R is the size of the resistance, C is the size of the capacitor.

4. The nonlinear dynamic reservoir network timing prediction system of claim 1, wherein, The plurality of linear R-C networks are eight parallel R-C networks, the resistance of each R-C network is fixed at 40 kΩ, and the capacitance values are 10 nF, 25 nF, 40 nF, 55 nF, 70 nF, 85 nF, 100 nF, and 115 nF, respectively. The eight parallel R-C networks are specifically configured to output 8*N final output signals of reservoir states, where N is the data size after mask processing during pre-processing.

5. The nonlinear dynamic reservoir network timing prediction system of claim 1, wherein, The linear regression network is specifically configured to: obtain the final output signal, and perform a classification and prediction task on time series data according to Formula 2: Equation 2; wherein, represents a weight value trained by linear regression, represents a target value of a corresponding prediction task, represents a transposed matrix obtained by transposing a matrix output after the RC network, X represents a matrix obtained after the RC network, T represents a transposing operation on a matrix, and + represents a pseudo-inverse operation on a matrix.

6. A method for implementing non-linear dynamic reservoir network timing prediction, characterized in that, The method for the nonlinear dynamic reservoir network time series prediction system according to any one of claims 1 to 5 specifically comprises: receiving an input signal through a microcontroller, pre-processing the input signal to obtain a pre-processed input signal, and performing PWM coding on the pre-processed input signal through a pulse width modulation (PWM) module to obtain a PWM coded signal; implementing a reservoir layer with distributed nonlinear characteristics and diversified time dynamic characteristics through a plurality of linear resistance-capacitance (R-C) networks, and refitting the PWM coded signal to obtain a final output signal; obtaining the final output signal through a linear regression network and performing a classification and prediction task on time series data.

7. The method of claim 6, wherein, Receiving an input signal through a microcontroller, pre-processing the input signal to obtain a pre-processed input signal, and performing PWM coding on the pre-processed input signal through a pulse width modulation (PWM) module to obtain a PWM coded signal specifically comprises: performing mask processing on the input signal, multiplying the input signal by a mask matrix composed of 1 and -1, and generating N virtual nodes; The PWM module encodes the processed input signal in the range of input amplitude 0-T, while maintaining constant pulse voltage, to obtain a PWM encoded signal, and the output voltage at the last moment As shown in formula 1: Formula 1: t is the coded input width, denoted in units of time, τ = R * C, R being the size of the resistance and C being the size of the capacitance.

8. The method according to claim 6, wherein: refitting the PWM coded signal to obtain a final output signal specifically comprises: outputting 8*N final output signals of reservoir states, where N is the data size after mask processing during pre-processing; The final output signal is obtained through a linear regression network, and the classification and prediction of time series data specifically includes: The final output signal is obtained, and the classification and prediction of time series data is performed according to formula 2: Formula 2: wherein, represents a weight value trained by linear regression, represents a target value of a corresponding prediction task, represents a transposed matrix obtained by transposing a matrix output after the RC network, X represents a matrix obtained after the RC network, T represents a transposition process on a matrix, and + represents a pseudo-inverse of a matrix.

9. An electronic device, comprising: including: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the nonlinear dynamic reservoir network time series prediction implementation method according to any one of claims 6 to 8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores an information transmission implementation program, and the program, when executed by the processor, implements the steps of the nonlinear dynamic reservoir network time series prediction implementation method according to any one of claims 6 to 8.

Images