Electric drive control method and device

Abstract

The application discloses an electric drive control method and device, relates to the technical field of vehicle drive optimization control, and comprises the following steps: high-bandwidth real-time sampling is performed on the bus voltage and current of an electric drive system, and a transient energy fluctuation component represented by a phase and an amplitude vector is extracted; the fluctuation component is input into a multi-layer frequency band electric drive memory matrix structure, node state updating with historical fluctuation memory is completed, and a multi-frequency band fluctuation memory vector is obtained; time sequence prediction is performed on the memory vector, and an energy deviation estimation value of the next control period is generated; a dynamic correction instruction is generated based on the estimation value, an inverter modulation signal is adjusted to realize active compensation, through the memory matrix structure with self-learning capability, advanced prediction and correction of high-frequency fluctuation are realized, control delay is greatly reduced, and the stability and energy efficiency of the electric drive system are improved.

Description

Technical Field

[0001] This invention relates to the technical field of vehicle drive optimization control, and in particular to an electric drive control method and apparatus. Background Technology

[0002] The motor controller is a core component of the electric drive system in new energy vehicles, requiring high energy efficiency and high dynamic response to drive the motor across a wide speed range (0–20,000 rpm). Current systems must cope with high-frequency voltage ripples (10–100 kHz), nonlinear load abrupt changes (such as rapid acceleration / deceleration), and strong coupling ripple interference caused by rapid changes in drive frequency (such as 10 kHz → 15 kHz) under complex operating conditions, placing stringent requirements on real-time performance, robustness, and adaptability.

[0003] Currently, electric drive systems for new energy vehicles generally rely on techniques such as LC filtering, adaptive PID, or frequency domain decomposition to suppress high-frequency energy fluctuations (1–50kHz). However, LC filtering has a fixed bandwidth and is difficult to respond to dynamic shifts in the fluctuation spectrum; adaptive PID is based only on instantaneous error adjustment and has inherent phase lag; and frequency domain decomposition is computationally complex and cannot meet the microsecond-level real-time requirements.

[0004] Research has revealed three structural limitations in existing technologies: First, they lack memory: the control system does not store historical fluctuation patterns, leading to repeated trial and error and non-convergence in corrections under repetitive operating conditions; second, they exhibit weak coupling: treating voltage ripple and current jitter as isolated signals ignores their multi-physics coupling nature in electromagnetic energy flow, easily causing secondary disturbances; and third, they suffer from open-loop uncontrollability: lacking a mechanism for quantitatively evaluating the energy level of the correction effect, they cannot determine whether high-frequency energy is absorbed, transferred, or dissipated, and reliability assurance lacks a physical basis. Summary of the Invention

[0005] The purpose of this invention is to provide an electric drive control method and apparatus to alleviate the technical problems of delayed response and low control accuracy and reliability caused by high frequency energy fluctuations (HFEF) in electric drive control.

[0006] In a first aspect, the present invention provides an electric drive control method, comprising: The bus voltage and current of the electric drive system are sampled in real time, and the transient energy fluctuation components that deviate from the steady-state mean are extracted in the form of phase and amplitude vectors. The transient energy fluctuation components are input into a multi-band electric drive memory matrix structure, and the nodes carrying historical fluctuation memories are updated in a multi-band manner to obtain a fluctuation memory vector that integrates multi-band fluctuation features. The fluctuation memory vector is used for time-series prediction to generate an energy deviation estimate for the next control cycle. Based on the energy deviation estimate, a dynamic correction command is generated, which is applied to the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

[0007] In an optional implementation, the step of inputting the transient energy fluctuation component into a multi-band frequency-modulated electric drive memory matrix structure, and performing multi-band frequency-modulated state updates on the nodes carrying historical fluctuation memories to obtain a fluctuation memory vector that fuses multi-band fluctuation characteristics includes: According to the preset frequency band division rules corresponding to the electric drive memory matrix structure, the input transient energy fluctuation component is divided into at least two non-overlapping frequency intervals and mapped to the multi-layer energy node clusters corresponding to the frequency intervals to obtain the fluctuation signal replicas that are directionally allocated to the corresponding frequency band node clusters. Based on each copy of the fluctuation signal, the historical fluctuation memory state stored by each node in the node cluster, and the coupling state signal from the adjacent frequency band cluster, the state of each node in the node cluster is updated to form a local memory representation of each node cluster in each frequency band. The local memory representations of each node cluster are weighted and combined according to a preset weight, and the topological structure information of the inter-layer coupling relationship is injected to generate a multi-band high-dimensional fluctuation memory vector.

[0008] In an optional implementation, the step of updating the state of each node in the node cluster based on each copy of the fluctuation signal, the historical fluctuation memory state stored by each node in the node cluster, and the coupling state signal from adjacent frequency band clusters, to constitute a local memory representation of each node cluster in each frequency band, includes: Using the fluctuation signal replica as an external stimulus, the response strength of each node cluster corresponding to the fluctuation signal replica is adjusted; Using the historical fluctuation memory state stored in each node of the node cluster as an inertial reference, the old memory is decayed according to the forgetting coefficient; Using the coupling state signals of adjacent frequency band clusters of each node cluster as correlation constraints, the collaborative evolution relationship of cross-frequency band fluctuation characteristics is strengthened; The response intensity, the old memory, and the co-evolutionary relationship are used as the updated set of node states to constitute the local memory representation of the node cluster.

[0009] In an optional implementation, the step of performing time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle includes: For each frequency band in the wave memory vector, linear slope fitting and periodic detection are performed respectively. Using the slow frequency band as a benchmark, the periodic fluctuation sequence of the mid-frequency band and the fast frequency band after phase synchronization with the slow frequency band is determined, and a composite deviation trajectory of fusion trend evolution characteristics and phase synchronization periodic characteristics is generated. Based on the trajectory value corresponding to the time node at the start of the next control cycle of the composite deviation trajectory, the energy deviation estimate for the next control cycle is obtained.

[0010] In an optional implementation, the step of determining the periodic fluctuation sequence of the mid-frequency band and the fast frequency band after phase synchronization with the slow frequency band, using the slow frequency band as a reference, and generating a composite deviation trajectory that integrates trend evolution characteristics and phase synchronization periodic characteristics, includes: Using the main period of the slow frequency band as the reference time axis, the main period lengths detected in the mid-frequency band and the fast frequency band are uniformly mapped onto the reference time axis. Based on the initial phase offset of the slow frequency band, the phase difference between the mid-frequency band and the fast frequency band is corrected, and the periodic fluctuation sequence of each frequency band after phase synchronization is output. Based on the contribution weight of each frequency band in the system energy imbalance, the trend sequence corresponding to the linear slope of each frequency band and the periodic fluctuation sequence of each frequency band after phase synchronization are weighted and superimposed to generate a composite deviation trajectory that integrates trend evolution characteristics and phase synchronization periodic characteristics.

[0011] In an optional implementation, the step of generating a dynamic correction command based on the energy deviation estimate and applying it to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system includes: Based on the energy imbalance tendency and severity corresponding to the energy deviation estimate, the correction direction and correction step size of the inverter's modulation signal are calibrated. Based on the current operating conditions of the electric drive system, hardware safety boundary constraints are applied to the correction step size to generate the final dynamic correction instruction; The final dynamic correction command is applied synchronously to the modulation signal of the inverter to dynamically adjust the duty cycle and switching timing, thereby correcting the bus voltage ripple and phase current fluctuation.

[0012] In an optional implementation, the method further includes: The energy balance index of the electric drive system is calculated in real time based on the ratio of the absolute value of the difference between the output energy and the input energy of the electric drive system to the input energy. When the energy balance index is lower than the preset stability threshold, the electric drive system is determined to have entered an energy stable state, and the fluctuation parameter characteristics of the current period are stored in the historical fluctuation pattern library of the electric drive memory matrix structure. When the energy balance index continuously exceeds the preset abnormal threshold, the reward function is determined based on the fluctuation characteristics of the current period, and the node core parameters of the electric drive memory matrix structure are adaptively updated through a reinforcement learning strategy. If the energy balance index does not drop below the preset stability threshold within a preset number of control cycles, the memory weights of each frequency band node cluster in the electric drive memory matrix structure are redistributed.

[0013] In a second aspect, the present invention provides an electric drive control device, comprising: The extraction module samples the bus voltage and current of the electric drive system in real time and extracts the transient energy fluctuation components that deviate from the steady-state mean, characterized in the form of phase and amplitude vectors. The update module inputs the transient energy fluctuation components into a multi-band frequency-based electric drive memory matrix structure, performs multi-band frequency-based state updates on the nodes carrying historical fluctuation memories, and obtains a fluctuation memory vector that integrates multi-band fluctuation characteristics. The prediction module performs time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle. The correction module generates a dynamic correction command based on the energy deviation estimate, which is applied to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

[0014] Thirdly, the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the method as described in any of the foregoing embodiments.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed, implements the method described in any of the foregoing embodiments.

[0016] This invention provides an electric drive control method and apparatus. First, it uses high-bandwidth real-time sampling, at least 10 times the inverter's PWM carrier frequency, to extract transient energy fluctuation vectors with phase and amplitude information. This accurately captures high-frequency energy disturbances across the entire frequency band of the electric drive system, such as 1-50kHz, solving the problems of insufficient sampling bandwidth and inability to fully capture high-frequency fluctuation characteristics in traditional solutions. Second, through a multi-layered frequency band electric drive memory matrix structure, it performs frequency-band-based state updates with historical memory for the fluctuation components, fusing historical fluctuation patterns with current fluctuation characteristics to generate a fluctuation memory vector. This fundamentally solves the core defects of traditional electric drive control, such as lack of memory and inability to utilize historical data to optimize control performance. Third, by predicting the timing of the fluctuation memory vector, it obtains the energy deviation trend for the next control cycle in advance, replacing the post-event error feedback compensation of traditional solutions and completely solving the problem of lag in traditional filtering and compensation algorithms. Finally, based on the predicted energy deviation estimate, it generates a dynamic correction command that directly acts on the inverter modulation signal, achieving active fine-tuning compensation of the power transmission path. This maintains the energy balance of the electric drive system even under high-frequency switching conditions above 20kHz. The core technical effects of this overall technical solution are: improving the accuracy of high-frequency energy fluctuation correction in the electric drive system, reducing control response delay, and significantly improving the long-term operational stability of the system under complex operating conditions through the self-learning characteristics of the memory matrix. The energy balance index can be maintained below 0.01 for a long time, effectively reducing EMI interference and heat loss of the inverter, and extending the service life of power devices.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 A flowchart of an electric drive control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the functional modules of an electric drive control device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware architecture of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] With the rapid development of new energy vehicles and intelligent electric drive systems, the motor controller, as a core component, directly impacts the vehicle's energy efficiency and driving experience. Modern electric drive systems need to achieve efficient energy conversion across a wide speed range (e.g., 0-20,000 rpm) while coping with high-frequency voltage pulsations (typically 10-100kHz) and nonlinear load fluctuations (e.g., rapid acceleration / deceleration scenarios) under complex operating conditions. Current mainstream solutions employ silicon-based IGBTs or silicon carbide power modules to construct a three-phase inverter topology, using space vector modulation (SVPWM) to control the motor. However, this approach faces significant challenges in high-frequency energy fluctuation (HFEF) scenarios.

[0023] In existing technologies, there are schemes that use LC filter networks with fixed parameters to suppress high-frequency ripple, but their non-adjustable cutoff frequency leads to poor adaptability to multi-spectral fluctuations. Additionally, while proposed dynamic PID parameter tuning mechanisms can improve transient response, the control loop delay (typically 50-100 μs) still cannot meet the requirements of high-frequency fluctuations. Furthermore, although some frequency domain decomposition methods proposed in academic journals can theoretically handle wideband disturbances, they require floating-point DSPs to implement Fast Fourier Transform (FFT), resulting in increased computational delay (>200 μs), making it difficult to meet the real-time requirements of automotive systems.

[0024] In actual operating conditions, when the motor drive frequency changes rapidly (e.g., from 10kHz to 15kHz), the voltage ripple at the inverter output will couple to the DC bus through parasitic parameters, forming a feedback peak with an amplitude of 15%-20% of the nominal voltage. Traditional digital control systems, limited by ADC sampling rates (typically below 1MHz) and control cycles (50-100kHz), cannot accurately capture nanosecond-level transient processes, leading to false triggering of overcurrent protection thresholds (false alarm rates can reach 5%-8%). Furthermore, existing energy distribution models generally employ instantaneous error compensation strategies, requiring recalculation of control quantities each time under repetitive operating conditions (e.g., urban start-stop cycles). This wastes computational resources (increasing CPU utilization by 30%-40%) and prevents control accuracy from improving over time.

[0025] Based on this, the electric drive control method and device provided in this embodiment of the invention achieve active compensation for high-frequency energy fluctuations in the electric drive system through frequency band processing with historical memory and timing advance prediction of the multi-layer frequency band electric drive memory matrix, which greatly improves the fluctuation correction accuracy and control response speed. At the same time, through self-learning optimization, the energy balance and operational stability of the electric drive system under all operating conditions are improved.

[0026] To facilitate understanding of this embodiment, a detailed description of an electric drive control method disclosed in this embodiment of the invention will be provided first.

[0027] Figure 1 A flowchart of an electric drive control method provided in an embodiment of the present invention; Reference Figure 1 The electric drive control method includes the following steps: Step S102: Real-time sampling of the bus voltage and current of the electric drive system, and extraction of transient energy fluctuation components that deviate from the steady-state mean, characterized in the form of phase and amplitude vectors.

[0028] The core technical approach in this step is to synchronously sample the bus voltage and current using a high sampling frequency, for example, no less than 10 times the inverter's PWM carrier frequency. By calculating the difference between the real-time sampled values ​​and the system's steady-state operating average, a fluctuation vector with both phase and amplitude information is extracted, fully preserving the frequency, amplitude, and phase characteristics of the fluctuation. The technical benefits of this step are: firstly, the high sampling rate ensures distortion-free capture of high-frequency fluctuations across the entire frequency band, avoiding the loss of high-frequency characteristics caused by traditional low sampling rates; and secondly, the vector representation can fully reflect the dynamic characteristics of the fluctuation, providing accurate basic data for subsequent frequency band processing and prediction.

[0029] Among them, the transient energy fluctuation component refers to the energy change corresponding to the instantaneous voltage and current fluctuations that deviate from the steady-state operating average of the system during the operation of the electric drive system, caused by factors such as rapid changes in motor drive frequency, load fluctuations, and inverter switching actions. It is characterized in the form of a vector of phase and amplitude, and can fully reflect the frequency, amplitude and phase characteristics of the fluctuation. The frequency range usually covers 150kHz.

[0030] The extracted transient fluctuation component is:

[0031] In the formula, The real-time transient energy value of the DC bus side of the electric drive system at time t is calculated from the sampled values ​​of the instantaneous voltage and instantaneous current of the bus. To ensure the average electrical energy within a sliding time window that matches the PWM carrier cycle of the motor controller, the number of sampling points within the sliding window shall not be less than 10. The transient energy fluctuation component at time t is used as the input vector of the electric drive memory matrix.

[0032] Step S104: Input the transient energy fluctuation component into the multi-band electric drive memory matrix structure, perform multi-band state update on the nodes carrying historical fluctuation memory, and obtain the fluctuation memory vector that integrates multi-band fluctuation characteristics.

[0033] The core technology of this step is to use a multi-layered frequency band matrix topology as the core computing carrier. The input fluctuation components are split into frequency bands and input to the corresponding layered energy node clusters. The node state is updated by combining the historical fluctuation patterns stored in the nodes and the coupling signals of adjacent layers. Finally, the features of each frequency band are fused to generate a high-dimensional memory vector. The technical effects of this step are: firstly, it replaces the traditional memoryless control model, enabling the learning and reuse of historical fluctuation patterns, thus solving the defect that the correction accuracy of traditional schemes cannot improve with running time; and secondly, frequency band processing can achieve precise adaptation to fluctuations of different frequencies, avoiding the indiscriminate suppression of signals across the entire frequency band by traditional single-filter schemes.

[0034] Among them, the multi-level frequency band electric drive memory matrix structure, abbreviated as DMM, is the core computing and storage architecture of this invention. It consists of a multi-level energy node cluster layered by frequency band. Each node can store the historical energy fluctuation pattern of the electric drive system. There is an interconnection and coupling relationship between the layers. It can perform frequency band feature extraction and memory state update on the input energy fluctuation components. The fluctuation memory vector refers to the high-dimensional feature vector output by the electric drive memory matrix. It integrates the historical fluctuation pattern of each frequency band of the electric drive system, the current fluctuation characteristics and the cross-frequency band coupling relationship. It is the core input for subsequent time series prediction.

[0035] In some embodiments, step S104 may be implemented by the following steps: Step 2.1: According to the preset frequency band division rules corresponding to the electric drive memory matrix structure, the input transient energy fluctuation component is divided into at least two non-overlapping frequency intervals and mapped to the multi-layer energy node clusters corresponding to the frequency intervals to obtain the fluctuation signal replicas that are directionally allocated to the corresponding frequency band node clusters.

[0036] Here, based on the frequency characteristics of the electric drive system fluctuations, a pre-defined non-overlapping frequency band division rule is used to perform time-frequency decomposition on the input transient fluctuation components, splitting them into multiple frequency band signals corresponding to various frequency ranges. Each frequency band signal is then mapped to a corresponding layered energy node cluster in the matrix structure, generating a replica of the fluctuation signal for that frequency band. This layered decoupling of full-band fluctuations avoids mutual interference between fluctuations of different frequencies, overcoming the limitation of traditional single-filter schemes in adapting to multi-spectral fluctuations. The directional mapping mechanism ensures that the fluctuation characteristics of each frequency band are processed by a dedicated node cluster, improving the accuracy of feature extraction.

[0037] In addition, a wave signal replica can refer to a frequency band wave signal obtained by decomposing the transient energy wave component into a single frequency range after time-frequency decomposition. It fully preserves the phase and amplitude characteristics of the wave within that frequency range and is a dedicated processing object for the corresponding frequency band node cluster.

[0038] Step 2.2: Based on each fluctuation signal copy, the historical fluctuation memory state stored by each node in the node cluster, and the coupling state signal from the adjacent frequency band cluster, update the state of each node in the node cluster to form a local memory representation of each node cluster in each frequency band.

[0039] Using a copy of the fluctuation signal in the current frequency band as input, and combining the node's own stored historical fluctuation memory state with the coupling signals of adjacent frequency band clusters, the node state is iteratively updated. The current fluctuation characteristics are fused with historical memory and cross-frequency band correlation information to generate a local memory representation for the corresponding frequency band. Here, the independent memorization and updating of fluctuation characteristics for each frequency band is achieved while preserving cross-frequency band coupling correlations, fully restoring the multi-frequency coupling characteristics of the electric drive system's fluctuations. Furthermore, using the historical memory state as the update benchmark, continuous learning and reuse of historical fluctuation patterns are realized.

[0040] Among them, historical fluctuation memory state refers to the energy fluctuation pattern data stored in the energy nodes of the electric drive memory matrix during the historical operation of the corresponding frequency band, including fluctuation amplitude, period, phase characteristics, etc. under different operating conditions, which is the core benchmark for node state update; local memory representation refers to the feature set generated after a single frequency band node cluster completes state update, which integrates the current fluctuation characteristics, historical memory, and cross-frequency band coupling information of the frequency band, and is a complete representation of the fluctuation characteristics of the frequency band.

[0041] The electrically driven memory matrix described in this invention is a distributed temporal memory structure composed of multiple neural energy nodes. Each layer of nodes corresponds to a fixed energy fluctuation frequency band, and the layers adopt a forward coupling architecture. The matrix adopts a frequency band layering and intra-layer parallel architecture, meaning that each layer corresponds to a fixed energy fluctuation frequency band, realizing accurate and independent memorization of multi-spectral fluctuations, and completely solving the shortcomings of existing technologies that cannot adapt to multi-frequency band fluctuations.

[0042] For example, the electric drive memory matrix can be a 4-layer architecture, with each layer corresponding to the following frequency bands: Layer 1 1-10kHz, Layer 2 10-20kHz, Layer 3 20-30kHz, and Layer 4 30-50kHz. It can be expanded to 8 layers to cover the 1-100kHz ultra-high frequency band as needed. Each layer contains 64 independent neural energy nodes, and a single matrix has a total of 256 nodes. Each node corresponds to a typical fluctuation mode in its frequency band (such as ripple amplitude, fluctuation period, and phase characteristics under different speeds and torques).

[0043] Within a layer, nodes within the same layer are independent of each other, processing different fluctuation characteristics within the same frequency band in parallel without cross-interference, ensuring multi-mode parallel memory. Between layers, a forward-coupled architecture is adopted; the state updates of higher-level nodes (high-frequency bands) are affected by the state updates of lower-level nodes (low-frequency bands), as shown in the corresponding formula. The interlayer correlation terms enable coupled memory of fluctuations in different frequency bands, overcoming the shortcomings of existing technologies that can only handle single-layer signal coupling and cannot cope with multi-band modulation fluctuations.

[0044] The state of each neural energy node can be updated using the following iterative formula:

[0045] In the formula, Let be the state value of the j-th energy node in the i-th layer at time t+1, α be the forgetting coefficient (0 < α < 1), β be the gain coefficient, γ be the interconnection coupling weight, and ΔE(t) be the transient energy fluctuation component at time t. is the inter-layer activation function, and the input is the state value of the corresponding node in the (i-1)th layer at time t.

[0046] This formula is the core equation for the temporal state iteration of a single neural energy node and is the underlying rule for the operation of the entire electric drive memory matrix (DMM). It is a lightweight iterative model with long short-term memory specifically designed for the temporal characteristics of high-frequency energy fluctuations in electric drives.

[0047] As the system continues to operate, the node states are iteratively updated using the above formula. For recurring stable fluctuation patterns, the node memory strength is continuously improved; for occasional invalid fluctuations, a forgetting mechanism is used to gradually filter them out. The final matrix automatically clusters to form an energy fluctuation operating condition mapping map bound to vehicle operating conditions. That is, different operating conditions and different fluctuation characteristics correspond to different node activation combinations, enabling the system to achieve energy self-sensing. No additional operating condition identification is required; the current fluctuation type can be determined solely by the node activation state, and the optimal correction strategy can be output.

[0048] As an exemplary embodiment, the scheme for determining the local memory representation of each node cluster in each frequency band in step 2.2 can be further refined as follows: Step 3.1: Using the fluctuation signal replica as an external stimulus, adjust the response strength of each node cluster corresponding to the fluctuation signal replica.

[0049] Understandably, by using a replica of the fluctuation signal in the current frequency band as the external input stimulus, and through a preset gain coefficient, the response sensitivity of the corresponding node cluster to the current fluctuation signal is adjusted so that the node output matches the amplitude of the current fluctuation. Here, the accurate capture of the current fluctuation characteristics allows for dynamic adjustment of the node response intensity based on the fluctuation amplitude, avoiding the loss of weak fluctuation characteristics or saturation of strong fluctuation signals; and it fully corresponds to the gain stimulus term in the node update formula, ensuring that the current fluctuation characteristics are completely integrated into the node state.

[0050] Among them, response strength refers to the sensitivity of the energy node cluster to the input fluctuation signal copy. It is controlled by a preset gain coefficient β and reflects the node's ability to capture the current fluctuation characteristics. The larger the gain coefficient, the stronger the node's response to the current fluctuation.

[0051] Step 3.2: Using the historical fluctuation memory state stored in each node of the cluster as an inertial benchmark, decay the old memory according to the forgetting coefficient.

[0052] Here, the historical fluctuation memory state stored by the node itself is used as the inertial benchmark for updates. A forgetting coefficient between 0 and 1 is used to decay the historical memory stored by the node, balancing the weight of historical memory and current fluctuation characteristics. This selective retention of historical fluctuation patterns filters outdated and invalid historical data through the forgetting coefficient, preserving valid historical fluctuation characteristics. Furthermore, it perfectly corresponds to the forgetting decay term in the node update formula, addressing the shortcomings of traditional control schemes that lack historical memory and cannot reuse historical data. The forgetting coefficient is an adjustable parameter with a value range of 0 < α < 1, used to control the degree of decay of historical memory by the node. The closer the coefficient is to 1, the more historical memory is retained; the closer the coefficient is to 0, the faster the historical memory decays, and the more sensitive the node's response to current fluctuations.

[0053] Step 3.3: Using the coupling state signals of adjacent frequency band clusters of each node cluster as correlation constraints, strengthen the cooperative evolution relationship of cross-frequency band fluctuation characteristics.

[0054] Here, the coupled state signals output by adjacent frequency band node clusters are used as correlation constraints. Through preset interconnection coupling weights, the collaborative correlation between fluctuation characteristics of different frequency bands is strengthened, restoring the cross-frequency band coupling characteristics of electric drive system fluctuations. This collaborative processing of cross-frequency band fluctuation characteristics avoids feature fragmentation caused by frequency band processing, fully restoring the true characteristics of electric drive system fluctuations; and it completely corresponds to the interconnection coupling term in the node update formula, improving the ability of node states to represent the overall system fluctuations.

[0055] Among them, the interconnection coupling weight refers to the adjustable parameter γ used to control the signal correlation strength between adjacent frequency band clusters. The larger the weight, the stronger the influence of the fluctuation characteristics of adjacent frequency bands on the current node state update. It can be dynamically adjusted according to the coupling characteristics of different frequency bands.

[0056] Step 3.4: Use the response intensity, old memory, and co-evolutionary relationship as the updated set of node states to form the local memory representation of the node cluster.

[0057] Here, the adjusted node response strength, the attenuated updated historical memory, and the cross-frequency band co-evolution relationship are combined to generate an updated set of node states. This set forms the basis for the local memory representation of the corresponding frequency band node cluster. This not only fully reproduces all elements of the node update formula, ensuring a closed-loop logic for node state updates, but also fully integrates the three core features of current, historical, and cross-frequency band characteristics, providing accurate foundational data for subsequent full-frequency band feature fusion.

[0058] Wherein, the response intensity corresponds to the formula in The transient fluctuation update term characterizes the node's response to current real-time energy fluctuations and its short-term learning intensity; it is jointly determined by the gain coefficient β and the transient fluctuation component ΔE(t). The old memory corresponds to the formula in... Long-term memory retention terms represent the historical memory reserves of nodes and the fluctuating characteristics and experiences accumulated through long-term learning; the co-evolutionary relationship corresponds to the formula in... Interlayer coupling correlation terms characterize the cooperative coupling and co-evolution relationship between different layers and different nodes, that is, the frequency band correlation and operating condition linkage characteristics between multi-layer energy nodes; the essence of old memory can be understood as the storage of the node's historical state at time t mentioned in the aforementioned embodiment. .

[0059] In practical applications, step 2.2 can also be implemented through the following example steps: First, for the replica of the fluctuation signal in the corresponding frequency band, it is used as the external input stimulus for the node cluster. Through a preset gain coefficient β, the response intensity of the node cluster to the current fluctuation signal is dynamically adjusted to ensure that the amplitude and phase characteristics of the current fluctuation are fully captured. This corresponds to the node update formula... Gain excitation term. Secondly, using the historical fluctuation memory states stored in the MRAM of each node in the node cluster as an inertial benchmark, a pre-configured forgetting coefficient α is retrieved to attenuate the historical old memories stored within the nodes, filtering out outdated and invalid historical data, while retaining valid historical fluctuation patterns that match the current operating conditions, corresponding to the node update formula in... Forgotten attenuation term. Then, the coupling state signal output by the node cluster of adjacent frequency bands is obtained and used as the correlation constraint for the current node state update. Through the preset interconnection coupling weight γ, the co-evolution relationship of the fluctuation characteristics of the current frequency band and adjacent frequency bands is strengthened, restoring the cross-frequency coupling characteristics of the electric drive system fluctuations, corresponding to the node update formula in... Interconnection coupling terms. Furthermore, the adjusted node response strength, the attenuated updated historical memory, and the cross-frequency band cooperative evolution relationship are merged to generate an updated node state set. Based on this, a local memory representation of the node cluster in this frequency band is constructed, fully reflecting the fluctuation characteristics and historical memory features of this frequency band. In this embodiment, node state updates can be implemented using 256 parallel computing cores of an FPGA, with each core corresponding to one energy node. The single-cycle computation latency is <0.2μs, supporting real-time processing of 1100kHz high-frequency fluctuations. Simultaneously, the number of matrix layers can be expanded using FPGA logic resources, up to a maximum of 8 layers and 512 nodes, adapting to higher frequency fluctuation correction requirements.

[0060] Step 2.3: The local memory representations of each node cluster are weighted and combined according to preset weights, and the topological structure information of inter-layer coupling relationship is injected to generate a multi-band high-dimensional fluctuation memory vector.

[0061] Based on the contribution of fluctuations in each frequency band to the system's energy imbalance, corresponding weights are preset, and the local memory representations of each frequency band are weighted and fused. Simultaneously, inter-layer coupling topology information of each node cluster is supplemented, ultimately generating a high-dimensional fluctuation memory vector that fuses features from all frequency bands. Firstly, weight allocation ensures focused attention on core influencing frequency bands, improving the accuracy of subsequent predictions. Furthermore, the injection of inter-layer coupling information ensures that the final output memory vector fully preserves the cross-frequency band correlation characteristics of fluctuations, avoiding feature fragmentation caused by frequency band-specific processing.

[0062] In practical applications, S104 can be implemented through the following steps, for example: First, based on the frequency distribution characteristics of high-frequency fluctuations in the electric drive system, the frequency band division rules of the pre-configured electric drive memory matrix structure are used to divide the 150kHz fluctuation range into four non-overlapping frequency intervals: the first layer at 110kHz, the second layer at 1020kHz, the third layer at 2030kHz, and the fourth layer at 3050kHz. Each frequency interval corresponds to an independent energy node cluster in the matrix structure. Each node consists of an FPGA programmable logic unit and an MRAM non-volatile memory unit. The MRAM read / write speed is ≤10ns, allowing for rapid retrieval of historical fluctuation data. Then, wavelet packet time-frequency decomposition is performed on the input transient energy fluctuation components to convert the time-domain vector signal into a joint time-domain and frequency-domain feature. According to the preset frequency band division rules, the signal is split into four sub-band signals corresponding to the four frequency intervals. Each sub-band signal is then mapped to the corresponding layer of the energy node cluster, generating a dedicated fluctuation signal copy for each frequency band. The data retrieval delay is ≤0.5μs. Secondly, for each frequency band's fluctuation signal replica, using it as input stimulus, the historical fluctuation memory state stored in the MRAM of each node within the corresponding node cluster is retrieved. Simultaneously, the coupling state signals output by adjacent frequency band clusters are acquired. Based on this, the state of each node is iteratively updated, fully fusing the current fluctuation characteristics, historical memory data, and cross-frequency band coupling information to generate a local memory representation of the corresponding frequency band node cluster. Furthermore, according to the contribution of each frequency band's fluctuations to the system's energy imbalance, corresponding weights are assigned to the local memory representation of each frequency band. In this embodiment, the 2030kHz frequency band, which has the greatest impact, is assigned the highest weight. All local memory representations are weighted and combined, and the topological structure information of the inter-layer coupling relationships of each node cluster is supplemented. Finally, a high-dimensional fluctuation memory vector covering the entire frequency band and carrying complete fluctuation characteristics is generated. This embodiment also supports the adaptation of alternative solutions, allowing the use of one-dimensional convolutional memory networks (1DCNN) or sparse matrix structures to replace traditional fully connected matrices, achieving higher prediction efficiency or lower power consumption and computational load.

[0063] Step S106: Perform time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle.

[0064] Based on a memory vector that integrates historical and current fluctuation characteristics, the trend and periodic characteristics of fluctuations in each frequency band are fitted to predict the energy fluctuation trend of the next control cycle and quantify the estimated value of energy deviation from steady state. Here, the advance prediction of energy fluctuations replaces the traditional ex-post feedback compensation, fundamentally solving the control lag problem of traditional algorithms; and the prediction based on multi-band fusion characteristics has higher accuracy than single signal prediction, and can accurately predict the tendency and severity of energy imbalance.

[0065] Among them, the energy deviation estimate It refers to the estimated value of energy fluctuations that the electric drive system will experience in the next control cycle, obtained through time-series prediction. It can reflect the tendency and severity of energy imbalance in the next cycle.

[0066] In some embodiments, step S106 may be implemented by the following steps: Step 4.1: Perform linear slope fitting and periodicity detection on the wave intensity components corresponding to each frequency band in the wave memory vector.

[0067] The wave intensity components corresponding to each frequency band are extracted from the wave memory vector. Linear slope fitting is performed on each component to extract the trend evolution characteristics of the wave. Simultaneously, autocorrelation detection is used to identify the periodic characteristics of the wave, obtaining the main period and phase information. This decoupled extraction of the trend and periodic characteristics of the wave in each frequency band provides a foundation for subsequent accurate prediction. Furthermore, the linear fitting and periodic detection have low computational cost, requiring no complex frequency domain operations, thus meeting the real-time requirements of automotive embedded systems.

[0068] Among them, linear slope fitting refers to fitting a straight line to the time series of the wave intensity component using the least squares method to extract the upward / downward trend of the wave. The positive and negative signs and absolute values ​​of the slope reflect the trend direction and rate of change of the wave, respectively. Periodicity detection refers to identifying the periodicity of the wave intensity component using an autocorrelation algorithm to extract periodic characteristic parameters such as the main period length and initial phase offset of the wave.

[0069] Step 4.2: Using the slow frequency band as a benchmark, determine the periodic fluctuation sequence of the mid-frequency band and the fast frequency band after phase synchronization with the slow frequency band, and generate a composite deviation trajectory of the fusion trend evolution characteristics and the phase synchronization periodic characteristics.

[0070] The core technical approach in this step is to use the main period of the slow-frequency band fluctuation with the slowest rate of change as the reference time axis, and to perform phase correction and time axis mapping on the periodic fluctuation sequences of the mid-frequency and fast-frequency bands to achieve phase synchronization of fluctuations across the entire frequency band. Then, the trend sequences of each frequency band are weighted and superimposed with the synchronized periodic sequences to generate a composite deviation trajectory. Firstly, achieving phase synchronization across the entire frequency band using the slow-frequency band as the reference solves the prediction error problem caused by phase misalignment of fluctuations in different frequency bands. Furthermore, the composite deviation trajectory, which integrates trend and periodic characteristics, fully restores the evolutionary law of energy fluctuations, significantly improving prediction accuracy.

[0071] Among them, the slow frequency band refers to the lowest fluctuation frequency band, usually corresponding to a fluctuation of 110kHz. It has a slow rate of change and a long period, reflecting the overall energy change trend of the electric drive system and serving as the benchmark for phase synchronization across the entire frequency band. The composite deviation trajectory refers to a time-series curve that integrates the trend evolution characteristics of fluctuations in each frequency band with the periodic characteristics after phase synchronization. It can fully reflect the past change patterns and future evolution trends of the energy deviation of the electric drive system.

[0072] Step 4.3: Based on the trajectory values ​​corresponding to the time nodes at the start of the next control cycle of the composite deviation trajectory, obtain the energy deviation estimate for the next control cycle.

[0073] Here, the time node corresponding to the start time of the next control cycle is located on the composite deviation trajectory. The trajectory value corresponding to this node is read and quantified into an estimated energy deviation value for the next control cycle. This precise quantification of the energy deviation for the next cycle allows for an advance prediction of the direction and severity of energy imbalance. Furthermore, the predicted value is directly read from the composite deviation trajectory, resulting in extremely low computational latency, which can meet the real-time requirements of high-frequency control cycles.

[0074] For example, step S106 may include the following example: First, the fluctuation intensity component corresponding to each frequency band is extracted from the fluctuation memory vector output by the electric drive memory matrix. For each component, a linear slope fitting is performed using the least squares method to extract the trend evolution characteristics of the fluctuation in that frequency band, clarifying the rising / falling direction and rate of change of the fluctuation. At the same time, the periodicity of the fluctuation intensity component is detected by the autocorrelation algorithm to identify periodic characteristic parameters such as the main period length and initial phase offset of the fluctuation. Second, using the main period of the slow-frequency band fluctuation with the slowest rate of change (110kHz) as the reference time axis, the main period lengths detected in the mid-frequency and fast-frequency bands are uniformly mapped onto this reference time axis. Simultaneously, based on the initial phase offset of the slow-frequency band, the phase difference between the mid-frequency and fast-frequency bands is corrected to complete the phase synchronization of the full-band periodic fluctuation sequence. Then, based on the contribution weight of each frequency band to the system energy imbalance, the linear trend sequence of each frequency band is weighted and superimposed with the phase-synchronized periodic fluctuation sequence to generate a composite deviation trajectory that integrates trend and periodic characteristics. Furthermore, on the composite deviation trajectory, the time node corresponding to the start time of the next control cycle is located, and the trajectory value corresponding to that node is read. After quantization, the energy deviation estimate for the next control cycle is obtained, clarifying the tendency and severity of energy imbalance in the next cycle. In this embodiment, the prediction module can adopt the Bayesian temporal network in the alternative scheme to improve the prediction ability of unknown fluctuation patterns. At the same time, it can achieve higher accuracy temporal prediction through the improved LSTM long short-term memory neural network on the preset platform. The model input layer is a 64-dimensional feature vector, the hidden layer has 3 layers, and the inference time is ≤1ms, which can adapt to the real-time control requirements of vehicles.

[0075] Based on the foregoing embodiments, step 4.2 can be further refined into the following steps: Step 5.1: Using the main period of the slow frequency band as the reference time axis, map the main period lengths detected in the mid-frequency band and the fast frequency band to the reference time axis. Based on the initial phase offset of the slow frequency band, correct the phase difference between the mid-frequency band and the fast frequency band, and output the periodic fluctuation sequence of each frequency band after phase synchronization.

[0076] Here, the main period of the slow frequency band is used as a unified time reference to perform a unified time-scale mapping on the periodic sequences of the mid-frequency and fast frequency bands. Simultaneously, the initial phase of the slow frequency band is used as a reference to correct the phase shifts in the mid-frequency and fast frequency bands, eliminating phase differences between different frequency bands and achieving phase synchronization of the periodic fluctuation sequences across the entire frequency band. This not only completely solves the prediction error problem caused by period and phase misalignment in fluctuations across different frequency bands, ensuring a unified time reference for fluctuations across the entire frequency band, but also, by using the slow frequency band as a reference, it aligns with the characteristics of energy fluctuations in the electric drive system. The slow frequency band reflects the overall energy trend of the system, and using it as a reference significantly improves the stability of predictions.

[0077] Phase synchronization can be understood as unifying the periodic fluctuation sequences of different frequency bands to the same time and phase reference, eliminating the phase difference between different frequency bands, ensuring that the time nodes of fluctuations in each frequency band are aligned, and avoiding feature fusion errors caused by phase misalignment.

[0078] Step 5.2: Based on the contribution weight of each frequency band in the system energy imbalance, the trend sequence corresponding to the linear slope of each frequency band and the periodic fluctuation sequence of each frequency band after phase synchronization are weighted and superimposed to generate a composite deviation trajectory that integrates the trend evolution characteristics and the phase synchronization periodic characteristics.

[0079] This invention, based on the impact of fluctuations in each frequency band on the overall energy imbalance of the system, pre-configures corresponding contribution weights and weights the trend sequences and synchronized periodic sequences of each frequency band, ultimately generating a composite deviation trajectory that integrates the trend and periodic characteristics of the entire frequency band. This not only achieves focused attention on core influencing frequency bands through the allocation of contribution weights, further improving the accuracy of the predicted trajectory, but also features low computational complexity, strong real-time performance, and suitability for the requirements of vehicle-mounted high-frequency control scenarios.

[0080] Among them, the contribution weight can refer to the parameter used to characterize the degree of influence of fluctuations in different frequency bands on the overall energy imbalance of the system. The greater the influence of a frequency band, the higher its weight and the greater its proportion in the composite deviation trajectory. It can be dynamically optimized through historical data statistics or reinforcement learning.

[0081] In practical applications, step 4.2 can be implemented using the following example: First, the main period detected by the 110kHz slow-frequency band fluctuation is determined and used as the reference time axis for the full-band periodic sequence. The lengths of the main periods detected by the mid-frequency and fast-frequency bands are uniformly mapped to this reference time axis through time scale transformation to ensure the time scale uniformity of the full-band fluctuations. Then, based on the initial phase offset of the slow-frequency band periodic fluctuation, the phase difference between the mid-frequency and fast-frequency band periodic sequences and the reference time axis is calculated. The phase correction of the mid-frequency and fast-frequency band periodic sequences is performed to eliminate the phase misalignment between different frequency bands, and finally, a periodic fluctuation sequence of each frequency band with completely synchronized phase is output. Secondly, based on the degree of influence of fluctuations in each frequency band on the system's energy imbalance, a corresponding contribution weight is assigned to each frequency band. For example, in this embodiment, the core frequency band of motor drive (20-30kHz) is assigned the highest weight of 0.4, the slow frequency band (1-10kHz) is assigned a weight of 0.3, the mid-frequency band (10-20kHz) is assigned a weight of 0.2, and the fast frequency band (30-50kHz) is assigned a weight of 0.1. The trend sequences obtained by fitting the linear slope of each frequency band, as well as the periodic fluctuation sequences after phase synchronization, are weighted and superimposed to fully integrate the trend characteristics and periodic characteristics of the entire frequency band, ultimately generating a composite deviation trajectory that reflects the evolution law of system energy deviation. Bench tests have verified that the composite deviation trajectory generated in this step has a prediction accuracy of over 98% for the energy fluctuation of the next cycle, which is far higher than that of traditional single-signal prediction schemes.

[0082] Step S108: Generate a dynamic correction command based on the energy deviation estimate, and apply it to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

[0083] Specifically, based on the predicted direction and amplitude of the energy deviation, correction parameters for the corresponding inverter PWM modulation signal are generated, dynamically adjusting the switching timing and duty cycle of the inverter bridge arms to actively compensate for fluctuations in bus voltage and phase current. This active correction of high-frequency energy fluctuations avoids problems such as oscillation propagation and false triggering of overcurrent protection caused by the accumulation of fluctuating energy; and it directly affects the inverter modulation stage, resulting in a short control link, fast response speed, and adaptability to high-frequency switching conditions above 20kHz.

[0084] In some embodiments, step S108 may be implemented by the following steps: Step 6.1: Based on the energy imbalance tendency and severity corresponding to the energy deviation estimate, calibrate the correction direction and correction step size of the inverter's modulation signal.

[0085] The core technical approach in this step is to determine the tendency of energy imbalance based on the sign of the estimated energy deviation value. A positive deviation corresponds to energy overload, and a negative deviation corresponds to insufficient energy supply, thus calibrating the correction direction. The severity of the imbalance is determined based on the absolute value of the estimated energy deviation value, thereby calibrating the correction step size of the PWM modulation signal. Here, the correction parameters are precisely matched with the predicted deviation, the correction direction corresponds perfectly to the imbalance tendency, and the correction step size is positively correlated with the severity of the imbalance, avoiding under-correction or over-correction. Furthermore, the correction parameters are directly calibrated based on the predicted value, replacing the traditional feedback correction based on real-time error, completely solving the control lag problem.

[0086] Among them, the correction direction refers to the adjustment direction of the inverter's PWM modulation signal, which corresponds completely to the energy imbalance tendency. When the power supply is overloaded, the duty cycle is adjusted downward, and when the power supply is insufficient, the duty cycle is adjusted upward. The correction step size refers to the single adjustment range of the PWM modulation signal duty cycle, which is positively correlated with the severity of the energy imbalance. The more severe the imbalance, the larger the correction step size.

[0087] Step 6.2: Based on the current operating conditions of the electric drive system, apply hardware safety boundary constraints to the correction step size and generate the final dynamic correction instruction.

[0088] Here, this embodiment of the invention combines the current operating parameters of the electric drive system, such as motor speed, torque, bus rated voltage, and power device safety thresholds, to set upper and lower safety boundaries for the correction step size. This prevents the correction step size from exceeding the hardware safety range and generates a final dynamic correction command that meets automotive-grade safety requirements. This not only ensures that the correction command always remains within the safe operating range of the power devices, avoiding hardware failures such as overmodulation, overcurrent, and overvoltage, and meeting automotive-grade safety requirements, but also dynamically adjusts the safety boundaries based on the current operating conditions, balancing correction effectiveness with hardware safety.

[0089] Among them, hardware safety boundary constraints can refer to the upper and lower limit thresholds set for the correction step size based on the rated parameters of the inverter power devices and the safe operating range of the electric drive system under the current operating conditions, to ensure that the corrected modulation signal will not cause the hardware to exceed the safe operating range.

[0090] Step 6.3: Apply the final dynamic correction command synchronously to the inverter's modulation signal, dynamically adjust the duty cycle and switching timing, and correct the bus voltage ripple and phase current fluctuation.

[0091] The core technology of this step involves synchronously inputting the final dynamic correction command into the inverter's PWM modulator. This dynamically adjusts the duty cycle and switching sequence of the PWM signals for each bridge arm, actively fine-tuning the inverter's output voltage and current to compensate for predicted energy deviations and correct bus voltage ripple and phase current fluctuations. This proactive compensation for high-frequency energy fluctuations suppresses their amplification and propagation at the source. Furthermore, it acts synchronously on each bridge arm of the inverter, resulting in a short control link, fast response, and adaptability to high-frequency switching conditions above 20kHz. This step also supports the adaptation of alternative solutions. In addition to PWM modulation correction, energy buffering can be achieved through DC bus capacitor charge redistribution, achieving the same fluctuation correction effect.

[0092] In practical applications, step S108 can be implemented through the following examples: First, based on the energy deviation estimate obtained from time-series prediction, the tendency and severity of system energy imbalance are determined. A positive energy deviation estimate indicates an overload tendency, while a negative estimate indicates an underload tendency. This is used to calibrate the correction direction of the inverter's PWM modulation signal. Simultaneously, based on the absolute value of the energy deviation estimate, the severity of the imbalance is quantified, thus calibrating the correction step size of the PWM signal duty cycle. The more severe the imbalance, the larger the correction step size. Second, the current operating parameters of the electric drive system are retrieved via the DSP, including motor speed, torque, bus rated voltage, and the safety threshold of IGBT power devices. Based on this, upper and lower hardware safety boundary constraints are applied to the calibrated correction step size. In this embodiment, the maximum adjustment range of the correction step size is set to no more than ±20% of the original duty cycle to avoid exceeding the hardware safety operating range. Correction parameters that do not meet automotive-grade safety requirements are filtered out, ultimately generating compliant dynamic correction instructions. In addition, the generated final dynamic correction command is synchronously input into the inverter's PWM modulator to dynamically adjust the duty cycle and switching sequence of the PWM signals of each bridge arm of the inverter, actively fine-tuning the voltage and current distribution in the power transmission path, compensating for the predicted energy deviation, and realizing active correction of bus voltage ripple and phase current fluctuation.

[0093] Based on the aforementioned embodiments, in a preferred embodiment of practical application, firstly, after the system is powered on and started, the initial configuration of the sampling module is completed. High-precision voltage and current sensors are used to synchronously sample the voltage of the DC bus of the electric drive system and the output current of the three-phase permanent magnet synchronous motor in real time. The sampling frequency is configured to be more than 10 times the inverter PWM carrier frequency. In this embodiment, the PWM switching frequency can be configured to 20kHz, and the corresponding sampling frequency is set to 200kHz. The analog signal is converted into a digital signal through a high-speed ADC. The built-in RC low-pass filter circuit with a cutoff frequency of 100kHz suppresses sampling noise and ensures that the 150kHz full-band high-frequency fluctuation signal is captured without distortion. Then, the real-time voltage and current data obtained by sampling are preprocessed. After filtering out the sampling noise, the difference between the current sampling value and the system steady-state operating mean is calculated, and the transient energy fluctuation component represented in the form of phase and amplitude vectors is extracted, thus completely preserving the dynamic characteristics of the fluctuation. Secondly, the extracted transient energy fluctuation components are input into a pre-configured multi-layer frequency band electric drive memory matrix structure. In this embodiment, the matrix adopts a 4-layer architecture (not limited to 4 layers), with each layer containing 64 energy nodes. Each layer corresponds to an independent frequency range. The matrix splits the fluctuation components according to a preset frequency band division rule and maps them to the corresponding layer's energy node cluster. Combining the historical fluctuation patterns stored in the nodes and the coupling signals of adjacent frequency band clusters, the state update of each node is completed. Finally, the processing results of each layer are weighted and fused to obtain a fluctuation memory vector carrying historical and current fluctuation characteristics. Then, the fluctuation memory vector is subjected to time-series trend fitting and periodic characteristic analysis to predict the energy fluctuation trend of the next control cycle, quantify and generate the energy deviation estimate for the next cycle, and clarify the direction and severity of energy imbalance. Furthermore, based on the generated energy deviation estimate, the correction direction and step size of the inverter PWM modulation signal are calculated. After generating a dynamic correction command, it is synchronously applied to the control units of each bridge arm of the inverter IGBT module to dynamically adjust the switching on-time and duty cycle, and actively fine-tune the voltage and current distribution in the power transmission path to achieve real-time compensation for high-frequency energy fluctuations. At the same time, the excess high-frequency energy recovered during the correction process is converted into low-voltage power through an integrated bidirectional DC / DC converter and replenished to the vehicle's 12V low-voltage battery, with an energy recovery efficiency of over 85%.

[0094] In some embodiments, the electric drive control method may further include: Step 7.1: Calculate the energy balance index of the electric drive system in real time based on the ratio of the absolute value of the difference between the output energy and the input energy of the electric drive system to the input energy.

[0095] First, the input electrical energy and output mechanical energy of the electric drive system are collected in real time. The absolute value of the difference between the input and output energy is calculated, and then the ratio is calculated with the input energy to obtain the real-time energy balance index, quantifying the energy balance state of the system. This real-time, quantitative assessment of the energy balance state of the electric drive system provides a unified quantitative benchmark for subsequent state determination and self-learning optimization. Furthermore, the calculation method is simple and efficient, capable of real-time calculation in each control cycle, adapting to the real-time requirements of embedded systems.

[0096] Among them, the Energy Balance Index (EBI) is a core indicator for quantifying the energy balance state of an electric drive system, and its calculation formula is as follows: ,in Input energy into the system, For the system to output energy, the closer the EBI is to 0, the better the system's energy balance and the smaller the fluctuations.

[0097] Step 7.2: When the energy balance index is lower than the preset stability threshold, the electric drive system is determined to have entered an energy stable state, and the fluctuation parameter characteristics of the current cycle are stored in the historical fluctuation pattern library of the electric drive memory matrix structure.

[0098] Here, when the real-time calculated energy balance index falls below a preset stability threshold, the system is deemed to have achieved the required correction effect and enters a stable energy state. The fluctuation parameter characteristics, operating parameters, and correction execution parameters for the current cycle are then stored in the historical fluctuation pattern library of the electric drive memory matrix, completing the accumulation of effective samples. This continuous accumulation of effective fluctuation patterns not only provides high-quality training samples for the self-learning optimization of the memory matrix but also ensures that samples are stored only when the system is stable, guaranteeing that the data in the historical fluctuation pattern library consists entirely of high-quality, effectively corrected samples, thus avoiding interference from invalid data.

[0099] The preset stability threshold refers to the EBI critical value used to determine whether the system has entered an energy stable state. It is usually set to 0.02. When EBI < 0.02, the system is determined to have entered a stable state and the fluctuation correction effect meets the standard.

[0100] Step 7.3: When the energy balance index continuously exceeds the preset abnormal threshold, the reward function is determined based on the fluctuation characteristics of the current cycle, and the node core parameters of the electric drive memory matrix structure are adaptively updated through a reinforcement learning strategy.

[0101] Here, when the energy balance index exceeds a preset abnormal threshold for multiple consecutive control cycles, the system's current correction effect is deemed substandard, triggering a self-learning optimization process. The reward function is determined based on the fluctuation characteristics of the current cycle, with the inverse ratio of the sum of squared fluctuation energy as the reward objective. Through reinforcement learning, the system adaptively updates core node parameters of the memory matrix, such as the forgetting coefficient, gain coefficient, and interconnection coupling weights. This adaptive optimization of the memory matrix's core parameters not only ensures the system's correction effect continuously improves over time, overcoming the limitations of traditional fixed-parameter schemes in adapting to complex operating conditions, but also, by using the energy balance index as the trigger condition and the fluctuation suppression effect as the reward objective, ensures that the self-learning process always revolves around the core objective of minimizing energy fluctuations.

[0102] The preset anomaly threshold refers to the EBI critical value used to trigger the system's self-learning optimization process, typically set to 0.3. When the EBI continuously exceeds this threshold, the current correction effect is deemed unsatisfactory, triggering parameter self-learning optimization. The reward function refers to the optimization objective function of the reinforcement learning strategy, defined in this invention as the inverse ratio of the sum of squares of fluctuation energy. The smaller the fluctuation, the higher the reward value, guiding the algorithm to optimize in the direction of minimum energy fluctuation. This step also supports the adaptation of alternative solutions; the reinforcement learning module can be replaced with fuzzy control logic or sliding mode control algorithm to achieve the same parameter adaptive optimization effect.

[0103] For example, the core reward function of the self-learning control module of the Drive Memory Matrix (DMM) is the only objective function that guides the self-optimization and self-convergence of the core parameters of the memory matrix (forgetting coefficient α, gain coefficient β, interconnection coupling weight γ). The parameters of the electric drive memory matrix are optimized through self-learning using a reinforcement learning reward function. The formula for calculating the reward function is as follows:

[0104] In the formula, R is the reinforcement learning reward value, which ranges from 0 to 1. <R≤1; Let be the transient energy fluctuation component of the electric drive system at time t; The sum of squares of energy fluctuation components within a sliding time window that matches the PWM carrier cycle of the motor controller is used. The reward function aims to maximize R and drives the iterative optimization of the core parameters of the electric drive memory matrix to achieve continuous convergence of high-frequency energy fluctuations.

[0105] Step 7.4: If the energy balance index does not drop below the preset stable threshold within a preset number of control cycles, the memory weights of each frequency band node cluster in the electric drive memory matrix structure are redistributed.

[0106] Here, if the energy balance index fails to drop below the stable threshold after multiple consecutive control cycles following parameter self-learning optimization, the current weight allocation is deemed unsuitable for the current operating conditions, triggering a matrix reconstruction mechanism. This mechanism globally redistributes the memory weights of the memory matrix's frequency band node clusters. If the reconstructed weights still fail to meet the requirements, the system switches to a fixed-parameter safety net compensation mode. This progressive fallback protection under extreme conditions ensures that the system maintains a basic fluctuation correction effect under any operating condition, preventing system loss of control. Furthermore, the global weight redistribution can thoroughly reconstruct the matrix's feature extraction logic, adapting to specific fluctuation patterns under extreme conditions.

[0107] Among them, the matrix reconstruction mechanism refers to a fallback optimization mechanism that globally redistributes the memory weights of the node clusters of each frequency band in the electric drive memory matrix when conventional parameter optimization fails to achieve a stable state. It can thoroughly adjust the feature extraction and memory logic of the matrix to adapt to the fluctuation characteristics under extreme operating conditions.

[0108] For example, the following steps may be included: First, within each control cycle, the electrical energy on the input side and the mechanical energy on the output side of the electric drive system are collected in real time. Based on the ratio of the absolute value of the difference between the input and output energy to the input energy, the energy balance index of the electric drive system is calculated in real time, quantifying the current energy balance state of the system. Second, when the real-time calculated energy balance index is lower than the preset stability threshold of 0.02, the electric drive system is determined to have entered an energy stable state, and the current fluctuation correction effect is deemed satisfactory. The fluctuation parameter characteristics, operating parameters, and correction execution parameters of the current cycle are stored in the historical fluctuation pattern library of the electric drive memory matrix structure, completing the continuous accumulation of effective fluctuation samples. Then, when the energy balance index exceeds the preset abnormal threshold of 0.3 for five consecutive control cycles, the current correction effect is deemed unsatisfactory, triggering a self-learning optimization process. The reward function is determined based on the fluctuation characteristics of the current cycle, and the inverse ratio of the sum of squares of the fluctuation energy is used as the optimization objective. Through reinforcement learning strategy, the forgetting coefficient, gain coefficient, interconnection coupling weight, and other core node parameters of the electric drive memory matrix structure are adaptively updated to optimize the fluctuation feature extraction and memory capability of the matrix. In this embodiment, multi-objective optimization can be achieved through an improved LSTM neural network, constructing an LSTM multi-objective function J=λ1. EBI+λ2 ΔT_loss dynamically adjusts the weight coefficients based on operating conditions to achieve synergistic optimization of energy balance and thermal management. After 10,000 kilometers of real-vehicle road testing, the model's accuracy in identifying high-frequency fluctuations increased from 85% to 98%, the average EBI decreased from 0.25 to 0.12, and energy loss decreased by 52%. Furthermore, if, after self-learning optimization, the energy balance index still does not fall below the preset stability threshold within 20 consecutive control cycles, a matrix reconstruction mechanism is triggered. This mechanism globally redistributes the memory weights of each frequency band node cluster in the electric drive memory matrix structure. If stability is still not achieved after reconstruction, the system immediately switches to a fixed-parameter PID baseline compensation mode to ensure system stability and safety. This solution can be extended to high-dynamic power systems such as electric aircraft, electric-driven robotic arms, and energy storage converters. It can also be embedded with SiC or GaN power modules to achieve ultra-high frequency power correction control above 100kHz.

[0109] In some embodiments, such as Figure 2 As shown, an embodiment of the present invention provides an electric drive control device, comprising: The extraction module samples the bus voltage and current of the electric drive system in real time and extracts the transient energy fluctuation components that deviate from the steady-state mean, characterized in the form of phase and amplitude vectors. The update module inputs the transient energy fluctuation components into a multi-band frequency-based electric drive memory matrix structure, performs multi-band frequency-based state updates on the nodes carrying historical fluctuation memories, and obtains a fluctuation memory vector that integrates multi-band fluctuation characteristics. The prediction module performs time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle. The correction module generates a dynamic correction command based on the energy deviation estimate, which is applied to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

[0110] The present invention provides an embodiment for implementing an electronic device. In this embodiment, the electronic device may be, but is not limited to, a personal computer (PC), a laptop computer, a monitoring device, a server, or other computer device with analysis and processing capabilities.

[0111] As an exemplary embodiment, see [link to example]. Figure 3 The electronic device 110 includes a communication interface 111, a processor 112, a memory 113, and a bus 114. The processor 112, the communication interface 111, and the memory 113 are connected via the bus 114. The memory 113 is used to store a computer program that supports the processor 112 in executing the above-described method. The processor 112 is configured to execute the program stored in the memory 113.

[0112] The machine-readable storage medium mentioned in this article can be any electronic, magnetic, optical, or other physical storage device that can contain or store information such as executable instructions, data, etc. For example, machine-readable storage media can be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drives (such as hard disk drives), any type of storage disk (such as optical discs, DVDs, etc.), or similar storage media, or combinations thereof.

[0113] Non-volatile media can be non-volatile memory, flash memory, storage drives (such as hard disk drives), any type of storage disk (such as optical discs, DVDs, etc.), or similar non-volatile storage media, or combinations thereof.

[0114] It is understood that the specific operation methods of each functional module in this embodiment can be referred to the detailed description of the corresponding steps in the above method embodiment, and will not be repeated here.

[0115] The computer-readable storage medium provided in the embodiments of the present invention stores a computer program. When the computer program code is executed, it can implement the method of any of the above embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0117] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0118] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0119] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.

Claims

1. An electric drive control method, characterized in that, include: The bus voltage and current of the electric drive system are sampled in real time, and the transient energy fluctuation components that deviate from the steady-state mean are extracted in the form of phase and amplitude vectors. The transient energy fluctuation components are input into a multi-band electric drive memory matrix structure, and the nodes carrying historical fluctuation memories are updated in a multi-band manner to obtain a fluctuation memory vector that integrates multi-band fluctuation features. The fluctuation memory vector is used for time-series prediction to generate an energy deviation estimate for the next control cycle. Based on the energy deviation estimate, a dynamic correction command is generated, which is applied to the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

2. The method according to claim 1, characterized in that, The steps of inputting the transient energy fluctuation components into a multi-band frequency-modulated electric drive memory matrix structure, updating the state of nodes carrying historical fluctuation memories in a multi-band frequency-modulated manner, and obtaining a fluctuation memory vector that fuses multi-band fluctuation characteristics include: According to the preset frequency band division rules corresponding to the electric drive memory matrix structure, the input transient energy fluctuation component is divided into at least two non-overlapping frequency intervals and mapped to the multi-layer energy node clusters corresponding to the frequency intervals to obtain the fluctuation signal replicas that are directionally allocated to the corresponding frequency band node clusters. Based on each copy of the fluctuation signal, the historical fluctuation memory state stored by each node in the node cluster, and the coupling state signal from the adjacent frequency band cluster, the state of each node in the node cluster is updated to form a local memory representation of each node cluster in each frequency band. The local memory representations of each node cluster are weighted and combined according to a preset weight, and the topological structure information of the inter-layer coupling relationship is injected to generate a multi-band high-dimensional fluctuation memory vector.

3. The method according to claim 2, characterized in that, The step of updating the state of each node in the node cluster based on each copy of the fluctuation signal, the historical fluctuation memory state stored by each node in the node cluster, and the coupling state signal from adjacent frequency band clusters, to constitute a local memory representation of each node cluster in each frequency band, includes: Using the fluctuation signal replica as an external stimulus, the response strength of each node cluster corresponding to the fluctuation signal replica is adjusted; Using the historical fluctuation memory state stored in each node of the node cluster as an inertial reference, the old memory is decayed according to the forgetting coefficient; Using the coupling state signals of adjacent frequency band clusters of each node cluster as correlation constraints, the collaborative evolution relationship of cross-frequency band fluctuation characteristics is strengthened; The response intensity, the old memory, and the co-evolutionary relationship are used as the updated set of node states to constitute the local memory representation of the node cluster.

4. The method according to claim 1, characterized in that, The step of performing time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle includes: For each frequency band in the wave memory vector, linear slope fitting and periodic detection are performed respectively. Using the slow frequency band as a benchmark, the periodic fluctuation sequence of the mid-frequency band and the fast frequency band after phase synchronization with the slow frequency band is determined, and a composite deviation trajectory of fusion trend evolution characteristics and phase synchronization periodic characteristics is generated. Based on the trajectory value corresponding to the time node at the start of the next control cycle of the composite deviation trajectory, the energy deviation estimate for the next control cycle is obtained.

5. The method according to claim 4, characterized in that, The steps of determining the periodic fluctuation sequences of the mid-frequency band and the fast-frequency band after phase synchronization with the slow-frequency band, and generating a composite deviation trajectory that integrates trend evolution characteristics and phase synchronization periodic characteristics, based on the slow-frequency band, include: Using the main period of the slow frequency band as the reference time axis, the main period lengths detected in the mid-frequency band and the fast frequency band are uniformly mapped onto the reference time axis. Based on the initial phase offset of the slow frequency band, the phase difference between the mid-frequency band and the fast frequency band is corrected, and the periodic fluctuation sequence of each frequency band after phase synchronization is output. Based on the contribution weight of each frequency band in the system energy imbalance, the trend sequence corresponding to the linear slope of each frequency band and the periodic fluctuation sequence of each frequency band after phase synchronization are weighted and superimposed to generate a composite deviation trajectory that integrates trend evolution characteristics and phase synchronization periodic characteristics.

6. The method according to claim 1, characterized in that, The step of generating a dynamic correction command based on the energy deviation estimate and applying it to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system includes: Based on the energy imbalance tendency and severity corresponding to the energy deviation estimate, the correction direction and correction step size of the inverter's modulation signal are calibrated. Based on the current operating conditions of the electric drive system, hardware safety boundary constraints are applied to the correction step size to generate the final dynamic correction instruction; The final dynamic correction command is applied synchronously to the modulation signal of the inverter to dynamically adjust the duty cycle and switching timing, thereby correcting the bus voltage ripple and phase current fluctuation.

7. The method according to claim 1, characterized in that, The method further includes: The energy balance index of the electric drive system is calculated in real time based on the ratio of the absolute value of the difference between the output energy and the input energy of the electric drive system to the input energy. When the energy balance index is lower than the preset stability threshold, the electric drive system is determined to have entered an energy stable state, and the fluctuation parameter characteristics of the current period are stored in the historical fluctuation pattern library of the electric drive memory matrix structure. When the energy balance index continuously exceeds the preset abnormal threshold, the reward function is determined based on the fluctuation characteristics of the current period, and the node core parameters of the electric drive memory matrix structure are adaptively updated through a reinforcement learning strategy. If the energy balance index does not drop below the preset stability threshold within a preset number of control cycles, the memory weights of each frequency band node cluster in the electric drive memory matrix structure are redistributed.

8. An electric drive control device, characterized in that, include: The extraction module samples the bus voltage and current of the electric drive system in real time and extracts the transient energy fluctuation components that deviate from the steady-state mean, characterized in the form of phase and amplitude vectors. The update module inputs the transient energy fluctuation components into a multi-band frequency-based electric drive memory matrix structure, performs multi-band frequency-based state updates on the nodes carrying historical fluctuation memories, and obtains a fluctuation memory vector that integrates multi-band fluctuation characteristics. The prediction module performs time-series prediction on the fluctuation memory vector to generate an energy deviation estimate for the next control cycle. The correction module generates a dynamic correction command based on the energy deviation estimate, which is applied to the modulation signal of the inverter to adjust the voltage or current distribution of the power transmission path in the electric drive system.

9. An electronic device, characterized in that, It includes a memory, a processor, and a program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed, implements the method described in any one of claims 1-7.

Images