An AI adaptive-based multi-source power cooperative control method and system

Abstract

The application provides an AI adaptive multi-source power cooperative control method and system, which comprises collecting multi-dimensional operation parameters of a power system and preprocessing the multi-dimensional operation parameters; using an artificial intelligence adaptive model combining a time series prediction algorithm and a reinforcement learning algorithm, predicting the operation condition in the future period of time according to the preprocessed multi-dimensional operation parameters, and dynamically generating a control instruction with at least one of the overall efficiency of the power system, battery loss and emission as the optimization target; according to the control instruction, cooperatively controlling the power matching and energy distribution among the range extender, battery and motor; in the cooperative control process, the combustion state feedback of the range extender is obtained in real time, and if it is judged that there is an abnormal combustion phenomenon, the control instruction is corrected in real time through the artificial intelligence adaptive model, realizing millisecond-level accurate cooperation of the range extender-battery-motor, and improving the system operation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of new energy power control technology, specifically relating to an AI-adaptive multi-source power collaborative control method and system. Background Technology

[0002] With the rapid development of new energy vehicles and distributed power generation technologies, range-extended electric power systems have become an important technological route in the new energy power field due to their combination of the smoothness of pure electric drive and long driving range. Among them, methanol range extenders have been increasingly widely used in commercial vehicles, construction machinery, agricultural machinery, and stationary power generation equipment due to the advantages of methanol fuel's wide availability, low carbon and cleanliness, convenient storage and transportation, and ability to effectively solve the range anxiety of pure electric equipment. This has become an important support for promoting the diversified development of new energy power. However, methanol fuel itself has inherent characteristics such as high latent heat of vaporization, poor low-temperature fluidity, low calorific value, and combustion characteristics that differ significantly from gasoline. This has caused existing methanol range extenders to generally face many prominent problems in actual industrial operation, which seriously restricts their technological upgrading and large-scale promotion, and makes it difficult to meet the current stringent requirements for energy consumption, emissions, and reliability.

[0003] Currently, traditional methanol range extender systems mostly adopt fixed threshold control strategies, simply controlling the start and stop of the range extender based on the battery's SOC (state of charge). This fails to dynamically and accurately adjust based on load changes, ambient temperature, altitude, and methanol combustion status, resulting in severely insufficient coordinated control precision among the range extender, battery, and motor. When the vehicle or equipment is under transient conditions such as rapid acceleration, heavy load, or climbing, the range extender's power generation response is lagging, and it can only rely on the battery's high-current discharge to supplement power, leading to severe battery overheating and rapid degradation of cycle life. Under light load and idling conditions, the range extender generates excess power, and the excess electrical energy cannot be efficiently utilized, resulting in energy waste and ultimately leading to low overall system efficiency. Meanwhile, due to the slow flame propagation speed and high requirements for mixture formation of methanol, methanol range extenders are prone to combustion lag, local oxygen deficiency, and incomplete combustion under complex operating conditions such as sudden load changes, low temperature, and high-altitude hypoxia. This not only further reduces power generation efficiency but also produces excessive formaldehyde, CO, and other unconventional emissions. Existing control strategies cannot identify the in-cylinder combustion state in real time, nor can they form a closed-loop correction through dynamic adjustments of fuel injection quantity, ignition timing, and speed. As a result, the reliability and emission levels of the range extender cannot meet the stringent environmental regulations.

[0004] Furthermore, methanol range extenders are used in various fields, including passenger cars, commercial vehicles, construction machinery, and stationary power generation. The load characteristics of these different scenarios vary significantly, and extreme environments such as high altitudes, frigid zones, and high temperatures can also significantly impact power performance. Traditional control algorithms lack self-learning and adaptive capabilities, making it difficult for a single set of control parameters to adapt to diverse application scenarios and extreme operating conditions. This leads to problems such as starting difficulties, insufficient power, and increased risk of thermal runaway in extreme environments, resulting in extremely poor adaptability. Simultaneously, existing technologies often treat the range extender, battery, and motor as independent units for separate control, lacking a globally optimal energy allocation mechanism. This fails to achieve efficient coordination between power generation, energy storage, and energy consumption, preventing the maximum conversion of methanol fuel energy into effective electrical energy. Moreover, the battery is frequently subjected to deep charge and discharge states, further exacerbating energy consumption and component aging, contradicting the original intention of methanol fuel's low-carbon, high-efficiency, and economical application.

[0005] While existing technologies have attempted localized improvements to address the aforementioned issues by modifying engine structure, optimizing individual control parameters, and adding preheating devices, none have fundamentally resolved the core technological bottlenecks of "multi-component collaboration, operating condition prediction, combustion closed-loop control, and adaptive learning." Most improvement solutions remain passive response controls, unable to anticipate load changes for proactive regulation, lacking AI-level adaptive adjustment, and failing to adapt to complex operating conditions and different types of methanol fuels. Furthermore, they lack a real-time closed-loop correction mechanism for combustion status, failing to balance system operating efficiency and emission levels, and thus struggling to overcome the limitations of current methanol range extender technology. Therefore, the industry urgently needs an intelligent control method and system capable of multi-source power collaboration between the methanol range extender, battery, and motor; possessing AI adaptive prediction and self-learning capabilities; real-time combustion state correction; and adaptability to multiple scenarios and operating conditions. This would effectively address the numerous shortcomings of existing technologies, improve the operating efficiency, reliability, and emission levels of methanol range extender systems, and promote the further development and large-scale industrial application of methanol range extender technology. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an AI-adaptive multi-source power cooperative control method and system, which solves the technical problems in the prior art.

[0007] On the one hand, the invention provides the following technical solution: a multi-source power cooperative control method based on AI adaptive control, the method comprising: Collect multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters; An artificial intelligence adaptive model that integrates time-series prediction and reinforcement learning algorithms is used to predict the operating conditions for a future period of time based on the preprocessed multi-dimensional operating parameters, and dynamically generate control commands with at least one of the overall efficiency of the power system, battery consumption and emissions as optimization targets. According to the control instructions, the power matching and energy distribution between the range extender, battery and motor are coordinated and controlled. During the coordinated control process, the combustion status feedback of the range extender is obtained in real time. If an abnormal combustion phenomenon is detected, the control command is corrected in real time through the artificial intelligence adaptive model.

[0008] Compared with existing technologies, the beneficial effects of this application are as follows: This invention breaks through the limitations of traditional fixed threshold control and passive response control. By integrating time-series prediction algorithms and reinforcement learning algorithms into an artificial intelligence adaptive model, it achieves accurate prediction of operating conditions and dynamic optimization of control commands. The response time is ≤50ms, and the accuracy of operating condition prediction is ≥95%. It effectively solves the problems of insufficient collaborative control precision and lag in transient operating condition response in existing technologies, achieving millisecond-level precise coordination among the range extender, battery, and motor, significantly improving the overall system operating efficiency by 15% to 20% compared to existing technologies. A real-time closed-loop correction mechanism for methanol combustion status is established. By real-time monitoring of combustion parameters such as cylinder pressure and exhaust temperature, it promptly identifies abnormal phenomena such as incomplete combustion and knocking. The control parameters are corrected in real time through an artificial intelligence adaptive model, effectively improving the completeness and stability of methanol combustion. Methanol consumption is reduced by 10% to 15%, formaldehyde emissions are reduced by more than 40%, and CO emissions are reduced by more than 30%, fully complying with the stringent emission regulations of China VI b and above. This invention avoids the risk of component damage caused by knocking. It possesses strong autonomous learning and adaptive capabilities, capable of autonomously optimizing control parameters based on long-term operating data without manual adjustment. It is adaptable to various application scenarios, including passenger cars, commercial vehicles, stationary power generation, and agricultural machinery, automatically switching between corresponding control modes. It can also automatically identify traditional methanol and green methanol, adjusting control parameters accordingly to adapt to complex fuel conditions such as high proportions of methanol and impurity methanol. Furthermore, it is adaptable to extreme environments such as -40℃ cold, ≥35℃ high, and ≥3000m high altitude, significantly improving the system's scenario adaptability and versatility, overcoming the shortcomings of poor adaptability in existing technologies. This invention constructs a globally optimal energy allocation mechanism, achieving efficient coordination between power generation, energy storage, and power consumption, avoiding energy waste under light load and idling conditions. It also reduces instantaneous high-current discharge and deep charge / discharge of the battery, reducing battery wear by 30%–35%, effectively extending battery life and the overall system maintenance cycle by more than 30%, and reducing the system's total lifespan cost. The system structure of this invention is rationally designed, with each module working together efficiently. It uses sensors made of methanol-resistant materials to adapt to complex fuel conditions. At the same time, it is equipped with redundant protection modules to ensure that the system can still operate stably when the AI ​​control module fails or the component parameters are abnormal, reducing the failure rate by more than 25% and improving the reliability and stability of the system. Moreover, the system adopts a modular software design, which can be adapted to different models of range extenders, batteries and motors without the need for overall redevelopment. It has low industrialization costs, is easy to scale up and promote application, and aligns with the development trend of the "dual carbon" policy, thus having extremely high industrialization transformation value.

[0009] Furthermore, the multi-dimensional operating parameters include: Range extender related parameters: combustion efficiency, fuel injection quantity, ignition timing, engine speed, exhaust temperature, and cylinder pressure; Battery-related parameters: remaining capacity, charge / discharge rate, individual cell voltage, battery temperature; Motor drive related parameters: motor load power, motor speed, motor torque; and Environmental parameters: ambient temperature, altitude, atmospheric pressure.

[0010] Furthermore, the time series prediction algorithm employs a long short-term memory network, and the prediction of operating conditions specifically includes: Based on the time-series data of motor load power and speed over a period of time, the cell state is iteratively updated using the LSTM cell state update formula to filter sensor noise and retain the effective time-series trend of motor load and speed. The hidden state is extracted from the updated cell state using the LSTM hidden state output formula, and used as the feature vector for condition prediction. The hidden state feature vector is mapped to the probability distribution of each preset working condition type through a fully connected layer and the working condition prediction output formula, and the predicted working condition in the next 3 to 5 seconds is determined according to the maximum probability; wherein, the preset working condition types include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idle speed / light load.

[0011] Furthermore, the LSTM cell state update formula includes:

[0012] in, for LSTM cell state at any given time Output the value for the forget gate. for Cellular state at all times The input gate output value, This refers to the state of candidate cells for the input gate. For Hadamah accumulation; The LSTM hidden state output formula includes:

[0013] in, for LSTM hidden state at time step The output value is the output gate value. The hyperbolic tangent activation function is used. The state of the LSTM cells at time t; The operating condition prediction output formula includes:

[0014] in, For the future The operating condition prediction results at any given time. For normalization function, This is the output layer weight matrix. For output layer bias terms, Let t be the hidden state of the LSTM.

[0015] Furthermore, the working logic of the aforementioned artificial intelligence adaptive model is as follows: The timing prediction algorithm predicts the operating conditions for the next 3 to 5 seconds based on the trend of motor load power and speed changes over the past 5 seconds. The operating conditions include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idling / light load. The reinforcement learning algorithm dynamically adjusts the control commands based on the predicted operating condition type and real-time collected multi-dimensional operating parameters. The control commands include the following adjustment strategies: When the predicted operating conditions are rapid acceleration or heavy load, the range extender's power generation is increased to 110% to 120% of the required power, the fuel injection quantity is increased by 40% to 60% compared to the idling condition, and the ignition timing is advanced by 2 to 3 degrees of crankshaft angle. When the operating condition is predicted to be idling or light load, the range extender's power generation is reduced to the minimum stable power, the fuel injection is reduced by 30% to 50%, the ignition timing is delayed by 1 to 2 degrees of crankshaft rotation, and the excess power is used for battery equalization charging. When the predicted operating condition is constant speed, the error between the range extender's power generation and the motor's drive power should not exceed 5%, and the remaining battery charge should be maintained in the range of 40% to 60%. When the anticipated operating condition is deceleration or braking, the range extender is reduced to the idle speed power generation level, and braking energy is recovered to charge the battery.

[0016] Furthermore, the step of real-time correction of the control command includes: Real-time monitoring of cylinder pressure and exhaust temperature is performed, and corrections are made according to the following rules: If the exhaust temperature is below 600℃ and the peak cylinder pressure is below 8 Bar, it is determined that the combustion is incomplete. At this time, the following corrections are performed by the artificial intelligence adaptive model: adjusting the fuel injection quantity, advancing the ignition timing, and increasing the range extender speed. If the cylinder pressure mutation rate is detected to be greater than or equal to 1 Bar / μs, it is determined to be knocking. At this time, the following corrections are performed through the artificial intelligence adaptive model: reducing the amount of fuel injection, delaying the ignition timing, and reducing the power generation of the range extender. If an abnormal peak value of 9 Bar is detected in the cylinder pressure before spark plug ignition, and the peak value is greater than or equal to 9 Bar, and the instantaneous rise rate of exhaust temperature is greater than or equal to 50℃ / s, it is determined to be pre-ignition. At this time, the following correction is performed by the artificial intelligence adaptive model: stop the current injection cycle, delay the ignition timing and reduce the range extender speed, and resume injection and ignition after the cylinder pressure and exhaust temperature return to normal.

[0017] Furthermore, the method also includes: The AI ​​adaptive model learns and optimizes the control parameter thresholds based on long-term accumulated multi-dimensional operating parameters, and updates the optimized thresholds into the AI ​​adaptive model for subsequent operation condition prediction and control command generation. The steps of the self-learning optimization include: Once the cumulative operating time of the power system reaches a preset duration, the artificial intelligence adaptive model performs at least one of the following autonomous learning optimization operations based on the multi-dimensional operating parameters from historical operation: Autonomously optimizes the fuel injection quantity adjustment range, ignition timing offset, and power generation matching threshold; Automatically switches to high-efficiency energy-saving mode, high-reliability mode, or voltage regulation priority mode based on long-term operating scenarios; It automatically identifies conventional methanol and green methanol, and automatically adjusts the fuel injection quantity and ignition timing based on the difference in calorific value between the conventional methanol and the green methanol.

[0018] Secondly, this invention provides the following technical solution: an AI-adaptive multi-source power cooperative control system, the system comprising: A multi-parameter acquisition module is used to acquire multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters. The AI ​​control module is used to use an artificial intelligence adaptive model that integrates time-series prediction algorithm and reinforcement learning algorithm to predict the operating conditions in the future based on the preprocessed multi-dimensional operating parameters, and dynamically generate control commands with at least one of the overall efficiency of the power system, battery consumption and emissions as optimization targets. The execution module is used to coordinate the power matching and energy distribution between the range extender, battery and motor according to the control instructions. The redundancy protection module includes a fault monitoring unit and a backup control unit. The fault monitoring unit is used to monitor the operating status of the AI ​​control module. When the AI ​​control module is detected to be faulty or malfunctioning, it automatically takes over the control function. In the takeover state, it obtains the combustion status feedback of the range extender in real time. If it is determined that there is an abnormal combustion phenomenon, it corrects the control command in real time through the backup control unit.

[0019] Thirdly, the invention provides the following technical solution: a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the AI-based adaptive multi-source power cooperative control method as described above.

[0020] Fourthly, the invention provides the following technical solution: a storage medium storing a computer program, which, when executed by a processor, implements the AI-adaptive multi-source power cooperative control method described above. Attached Figure Description

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

[0022] Figure 1 A flowchart of the AI-adaptive multi-source power cooperative control method provided in the first embodiment of the present invention; Figure 2 The logic block diagram of the AI-adaptive multi-source power cooperative control method provided in the first embodiment of the present invention illustrates the closed-loop relationship between data flow and control flow; Figure 3 This is a structural block diagram of the AI-adaptive multi-source power cooperative control system provided in the second embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of a computer provided in the third embodiment of the present invention.

[0023] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0024] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0025] In the description of the embodiments of the present invention, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of the present invention and 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 present invention.

[0026] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] Example 1 In the first embodiment of the present invention, please refer to Figure 1 and Figure 2 As shown, an AI-adaptive multi-source power cooperative control method includes the following steps S01 to S05: S01, Collect multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters; Specifically, the multi-dimensional operating parameters include: Range extender related parameters: combustion efficiency, fuel injection quantity, ignition timing, engine speed, exhaust temperature, and cylinder pressure; Battery-related parameters: remaining capacity, charge / discharge rate, individual cell voltage, battery temperature; Motor drive related parameters: motor load power, motor speed, motor torque; and Environmental parameters: ambient temperature, altitude, atmospheric pressure.

[0028] In this embodiment, firstly, a multi-dimensional parameter acquisition system is established. Taking a methanol-fueled range-extended commercial vehicle as an example, the following four dimensions of operating parameters are collected in real time through a multi-parameter acquisition module: Range extender related parameters: combustion efficiency, fuel injection quantity, ignition timing, range extender speed, exhaust temperature, and cylinder pressure. The fuel injection quantity is indirectly calculated using an intake manifold absolute pressure sensor (MAP) combined with the range extender speed and a preset air-fuel ratio. Cylinder pressure is collected by a cylinder pressure sensor installed inside the range extender cylinder. Exhaust temperature is collected by an exhaust temperature sensor installed in the exhaust pipe. The sampling frequency is 100Hz, and the acquisition accuracies are: combustion efficiency ±1%, fuel injection quantity ±0.1mL / min, ignition timing ±1°CA, range extender speed ±10r / min, exhaust temperature ±5℃, and cylinder pressure ±0.1Bar.

[0029] Battery-related parameters: State of Charge (SOC), charge / discharge rate, individual cell voltage, and battery temperature. Data are collected via a Battery Management System (BMS) at a sampling frequency of 50Hz, with the following accuracies: SOC ±1%, charge / discharge rate ±0.1C, individual cell voltage ±0.01V, and battery temperature ±0.5℃.

[0030] Motor drive related parameters: motor load power, motor speed, and motor torque. These are acquired through a motor controller (MCU) at a sampling frequency of 100Hz, with accuracies of ±0.5kW for power, ±10r / min for speed, and ±5N·m for torque.

[0031] Environmental parameters: ambient temperature, altitude, and atmospheric pressure. Data were collected using environmental sensors at a sampling frequency of 10Hz, with accuracies of ±0.5℃ for temperature, ±10m for altitude, and ±1kPa for atmospheric pressure.

[0032] The acquired raw data is transmitted to the data preprocessing unit. This unit first uses a Kalman filter algorithm with a 5ms filter window to remove sensor noise (such as occasional fluctuations in motor load power and instantaneous changes in exhaust temperature). Then, denoising is performed, deleting abnormal data exceeding reasonable ranges (such as cylinder pressure > 15 Bar, SOC > 100% or < 0%) and replacing them with the average of the previous three frames. Finally, normalization is performed, normalizing all parameters to the [0,1] interval using the normalization formula: ,in The original data, The maximum and minimum reasonable values ​​for this parameter are determined to eliminate the influence of dimensions. Preprocessing delay ≤ 10ms.

[0033] S02, using an artificial intelligence adaptive model that integrates time-series prediction algorithm and reinforcement learning algorithm, predicts the operating conditions for a future period of time based on the preprocessed multi-dimensional operating parameters, and dynamically generates control commands with at least one of the overall efficiency of the power system, battery loss and emissions as optimization targets. Specifically, the time series prediction algorithm employs a long short-term memory network, and the prediction of operating conditions specifically includes: Based on the time-series data of motor load power and speed over a period of time, the cell state is iteratively updated using the LSTM cell state update formula to filter sensor noise and retain the effective time-series trend of motor load and speed. The hidden state is extracted from the updated cell state using the LSTM hidden state output formula, and used as the feature vector for condition prediction. The hidden state feature vector is mapped to the probability distribution of each preset working condition type through a fully connected layer and the working condition prediction output formula, and the predicted working condition in the next 3 to 5 seconds is determined according to the maximum probability; wherein, the preset working condition types include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idle speed / light load.

[0034] More specifically, the LSTM cell state update formula includes:

[0035] in, for LSTM cell state at any given time Output the value for the forget gate. for Cellular state at all times The input gate output value, This refers to the state of candidate cells for the input gate. For Hadamah accumulation; The LSTM hidden state output formula includes:

[0036] in, for LSTM hidden state at time step The output value is the output gate value. The hyperbolic tangent activation function is used. The state of the LSTM cells at time t; The operating condition prediction output formula includes:

[0037] in, For the future The operating condition prediction results at any given time. For normalization function, This is the output layer weight matrix. For output layer bias terms, Let t be the hidden state of the LSTM.

[0038] Specifically, the working logic of the artificial intelligence adaptive model is as follows: The timing prediction algorithm predicts the operating conditions for the next 3 to 5 seconds based on the trend of motor load power and speed changes over the past 5 seconds. The operating conditions include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idling / light load. The reinforcement learning algorithm dynamically adjusts the control commands based on the predicted operating condition type and real-time collected multi-dimensional operating parameters. The control commands include the following adjustment strategies: When the predicted operating conditions are rapid acceleration or heavy load, the range extender's power generation is increased to 110% to 120% of the required power, the fuel injection quantity is increased by 40% to 60% compared to the idling condition, and the ignition timing is advanced by 2 to 3 degrees of crankshaft angle. When the operating condition is predicted to be idling or light load, the range extender's power generation is reduced to the minimum stable power, the fuel injection is reduced by 30% to 50%, the ignition timing is delayed by 1 to 2 degrees of crankshaft rotation, and the excess power is used for battery equalization charging. When the predicted operating condition is constant speed, the error between the range extender's power generation and the motor's drive power should not exceed 5%, and the remaining battery charge should be maintained in the range of 40% to 60%. When the anticipated operating condition is deceleration or braking, the range extender is reduced to the idle speed power generation level, and braking energy is recovered to charge the battery.

[0039] In this embodiment, the preprocessed multi-dimensional operating parameters are input to the AI ​​control module. This module has a built-in adaptive AI model that integrates time series prediction algorithms and reinforcement learning algorithms. After receiving the preprocessed parameter data, the AI ​​control module uses the adaptive AI model that integrates time series prediction algorithms and reinforcement learning algorithms to complete the prediction of operating conditions and output control commands. The specific implementation is as follows: First, extract the motor load power within the past 5 seconds. and motor speed Time series data form the input feature vector First, load the weight matrix and bias terms that have been pre-trained using 1.2 million sets of working condition samples. The specific parameters are as follows: The weight matrix includes the forget gate weight matrix. (Dimension 128×1001), Input gate weight matrix (Dimension 128×1001), Candidate Cell State Weight Matrix (Dimension 128×1001), Output gate weight matrix (Dimension 128×1001), Output layer weight matrix (Dimension 4×128); Bias terms include forget gate bias terms. (Dimension 1×128), Input gate bias term (Dimension 1×128), Candidate cell state bias term (Dimension 1×128), Output gate bias term (Dimension 1×128), Output layer bias term (Dimension 1×4); The weight matrix and bias terms mentioned above are stored in the chip memory of the AI ​​control module. The loading process takes ≤1ms. After loading is complete, the initial cell states of the LSTM network are initialized. and initial hidden state (All vectors are set to all zeros, with a dimension of 1×128).

[0040] Next, the cell state is updated by executing the cell state update formula. ,in (Forget Gate Output) via formula Calculations show that It is a sigmoid activation function with an output value range of 0 to 1. 0 means that the corresponding historical data is completely forgotten, and 1 means that the corresponding historical data is completely retained. Its core function is to filter sensor noise and retain the continuous trend of motor load and speed. (Input gate input) via formula Calculations show that It is a sigmoid activation function with an output value range of 0 to 1. Its core function is to determine which new time series data are worth retaining. (Input gate candidate cell state) via formula Calculations show that It is a hyperbolic tangent activation function with an output value range of -1 to 1. Its core function is to quantify and preserve the trend strength of the data. Hadamard product (element-wise multiplication) ensures cell state. It can integrate "filtered historical memory" and "new effective trends" to achieve iterative updates of core memory, and finally output the cell state at time t. (Dimension 1×128) is used to store the effective temporal patterns within the first 5 seconds.

[0041] After the LSTM time-series prediction algorithm completes the "cell state update", it executes the hidden state output formula. ,in (Output gate output) via formula Calculations show that It is a sigmoid activation function with an output value ranging from 0 to 1. Its core function is to determine the cell state after the update. In the process, effective features related to working condition prediction are selected; This is used to normalize the cell state values ​​to -1 to 1, avoiding computational distortion caused by excessively large values; finally, the hidden state at time t is obtained through the Hadamard product. (Dimension 1×128) This hidden state is the core feature extracted by the LSTM network and is directly used for subsequent working condition prediction.

[0042] After the LSTM timing prediction algorithm completes the "hidden state output", it executes the operating condition prediction output formula. ,in That is, predicting the operating conditions in the next 3 to 5 seconds. Hide state (1×128 dimensions) is transformed into a 4-dimensional vector (corresponding to 4 types of working conditions); The normalization function normalizes each element of the 4-dimensional vector to 0-1, and the sum of all elements is 1, resulting in probability vectors for four types of operating conditions (probabilities for emergency acceleration / heavy load, rapid deceleration / braking, constant speed, and idling / light load operating conditions, respectively). For example, if the output probability vector is [0.96, 0.01, 0.02, 0.01], then the operating condition in the next 3-5 seconds is determined to be rapid acceleration. After obtaining the predicted operating condition type through the LSTM time series prediction algorithm, the result is transmitted to the reinforcement learning algorithm.

[0043] In this embodiment, the reinforcement learning algorithm adopts the Q-learning algorithm framework, with "highest overall system efficiency, lowest battery loss, and lowest emissions" as the global optimal objective. It runs in conjunction with the LSTM time-series prediction algorithm, and all calculations are completed on the same MCU chip, ensuring that the instruction output delay is ≤30ms and seamlessly connecting with the LSTM prediction results and the actions of the execution module. First, the global optimal objective of reinforcement learning is defined, and the objective parameters are quantified to ensure that the control direction is clear. The specific objectives are as follows: ① Overall system efficiency ≥45% (system efficiency = effective power output of motor / methanol energy consumed by methanol range extender); ② Minimize battery loss, with the SOC change ΔSOC ≤5% in a single control cycle, avoiding deep charging and discharging and high-current charging and discharging of the battery; ③ Lowest emissions, with formaldehyde emissions reduced by more than 40% compared to traditional control methods and CO emissions reduced by more than 35%, complying with the China VI b emission regulations. The weights of the three objectives are 40%, 30%, and 30%, respectively. The objective function is obtained by weighted summation, and the core of the algorithm optimization is to maximize the value of this objective function.

[0044] The reinforcement learning algorithm receives the operating condition type predicted by the LSTM time-series prediction algorithm, and simultaneously receives real-time operating parameters (preprocessed in-cylinder pressure, exhaust temperature, intake manifold absolute pressure, SOC, motor load power, range extender speed, etc.) transmitted by the multi-parameter acquisition module. Combined with preset control parameter thresholds, it dynamically optimizes the control commands. The optimization process consists of three steps: "operating condition matching → parameter calculation → command verification." The specific control logic is as follows: 1. Rapid acceleration / heavy load condition (predicted condition code = 1): Under this condition, the motor load increases instantaneously, requiring avoidance of a sudden large current discharge of the battery, while ensuring that the range extender's power generation keeps up. First, based on the real-time motor load power... Calculate the target power generation of the range extender = ×115% (15% redundancy reserved to avoid insufficient power), and The maximum power output of the range extender (30kW in this embodiment) shall not exceed the maximum power generation of the range extender; secondly, the target fuel injection quantity shall be calculated based on the range extender speed and the absolute pressure of the intake manifold. =Idle fuel injection quantity × 150% (50% increase compared to idle condition to ensure complete methanol combustion and meet power requirements); Next, adjust the ignition timing. =Base ignition timing (10°CA) + 2.5°CA (pre-ignition to improve combustion efficiency); Finally, the battery charging and discharging strategy is "stop discharging and maintain the current SOC" to avoid instantaneous high current output from the battery and reduce battery wear.

[0045] 2. Rapid deceleration / braking condition (predicted condition code = 2): Under this condition, the motor load drops sharply. The core objective is to recover braking energy and reduce the load on the range extender. The target power output of the range extender is [not specified]. =5kW (minimum stable power generation to avoid range extender shutdown); target fuel injection quantity = Idle fuel injection quantity × 50% (further reduce fuel injection quantity and reduce emissions); Ignition timing maintains the baseline ignition timing; Battery charging and discharging strategy is "braking energy recovery", charging current is controlled at 10~15A (not exceeding the maximum charging current of the battery), recovering the electrical energy generated by motor braking to the battery, improving energy utilization, and reducing the external charging demand of the battery.

[0046] 3. Constant speed operation (predicted operation code = 3): Under this operation, the motor load is stable. The core is to achieve precise matching between the range extender's power generation and the motor load power, and the target power generation of the range extender. = ±5% (error ≤5%, ensuring power matching and avoiding energy waste); target fuel injection quantity The system dynamically adjusts the air-fuel ratio based on real-time intake manifold absolute pressure and range extender speed to ensure it remains at 14.7:1 (the optimal air-fuel ratio for complete methanol combustion). The ignition timing is maintained at the baseline ignition timing (10°CA). The battery charging and discharging strategy is to "maintain stable SOC," keeping the SOC between 40% and 60%. If the SOC is below 40%, the range extender's power generation is increased by 5% to 10% to supplement charging. If the SOC is above 60%, the range extender's power generation is reduced by 5% to 10% to decrease charging.

[0047] 4. Idle / Light Load Condition (Predicted Condition Code = 4): Under this condition, the motor load is relatively small, and the range extender's power output needs to be reduced to the lowest stable value. Excess energy is used for battery equalization charging. Specific optimization logic: Target power output of the range extender. =5kW (minimum stable power generation to ensure the range extender operates stably without shutting down); target fuel injection quantity =Idle fuel injection quantity × 60% (40% reduction compared to idle condition, avoiding methanol waste and reducing emissions); Ignition sequence =Base ignition timing (10°CA) - 1.5°CA (delayed ignition, adapted to low load combustion requirements); the battery charging and discharging strategy is "equalized charging", with the charging current controlled at 5~8A, maintaining the SOC at around 50% (the optimal SOC range of the battery), and avoiding overcharging.

[0048] All of the above-mentioned control parameters (such as fuel injection quantity adjustment ratio, ignition timing offset, and power generation redundancy) are trained based on 1.2 million sets of operating condition samples. After the system has been running for a long time (cumulative ≥100h), it will be further iterated and adjusted through subsequent self-learning optimization functions to ensure that it is adapted to different scenarios and different fuels.

[0049] S03, according to the control command, coordinate the power matching and energy distribution between the range extender, battery and motor; In this embodiment, after optimization, the control commands are converted into digital signals recognizable by the execution module. Specifically, these include: range extender power generation control commands (accuracy ±0.1kW), methanol fuel injection quantity control commands (accuracy ±0.2mL / min), ignition timing control commands (accuracy ±0.1°CA), and battery charge / discharge rate control commands (accuracy ±0.5A). The commands are transmitted to the execution modules (methanol range extender ECU, battery charge / discharge controller, and motor MCU) via the CAN bus. The transmission time is ≤5ms, ensuring that the execution modules receive the commands and act in a timely manner, forming a closed loop with LSTM prediction and subsequent feedback correction.

[0050] The execution module receives control commands from the AI ​​control module, and each controller (methanol range extender ECU, battery charge / discharge controller, and motor MCU) operates synchronously and collaboratively, strictly following the command parameters to achieve seamless coordination of "methanol range extender power generation - battery energy storage / power supply - motor drive," ensuring stable system operation and optimal efficiency. The actions of each controller are as follows: Methanol range extender controller: Based on the power generation command issued by the AI ​​module and its own real-time speed, it dynamically adjusts the throttle opening to achieve precise matching between the range extender speed and power generation. At the same time, based on the target fuel injection quantity and ignition timing commands, it controls the fuel injector to inject fuel and the spark plug to ignite, ensuring complete combustion of methanol; Battery charge / discharge controller: Based on the charge / discharge rate command issued by the AI ​​module, it adjusts the battery charge / discharge current (charging current ≤ 10A, discharging current ≤ 20A) to avoid deep charging and discharging and maintain stable SOC; Motor controller: Based on the motor load power command issued by the AI ​​module, it adjusts the motor torque and speed to ensure precise matching between the motor speed and the range extender's power generation, achieving coordination of "power generation - energy storage - power consumption."

[0051] S04, During the coordinated control process, the combustion status feedback of the range extender is obtained in real time. If it is determined that there is an abnormal combustion phenomenon, the control command is corrected in real time through the artificial intelligence adaptive model.

[0052] Specifically, the step of real-time correction of the control command includes: Real-time monitoring of cylinder pressure and exhaust temperature is performed, and corrections are made according to the following rules: If the exhaust temperature is below 600℃ and the peak cylinder pressure is below 8 Bar, it is determined that the combustion is incomplete. At this time, the following corrections are performed by the artificial intelligence adaptive model: adjusting the fuel injection quantity, advancing the ignition timing, and increasing the range extender speed. If the cylinder pressure mutation rate is detected to be greater than or equal to 1 Bar / μs, it is determined to be knocking. At this time, the following corrections are performed through the artificial intelligence adaptive model: reducing the amount of fuel injection, delaying the ignition timing, and reducing the power generation of the range extender. If an abnormal peak value of 9 Bar is detected in the cylinder pressure before spark plug ignition, and the peak value is greater than or equal to 9 Bar, and the instantaneous rise rate of exhaust temperature is greater than or equal to 50℃ / s, it is determined to be pre-ignition. At this time, the following correction is performed by the artificial intelligence adaptive model: stop the current injection cycle, delay the ignition timing and reduce the range extender speed, and resume injection and ignition after the cylinder pressure and exhaust temperature return to normal.

[0053] In this embodiment, while performing coordinated control, the multi-parameter acquisition module collects feedback parameters in real time (such as: cylinder pressure, exhaust temperature, intake manifold absolute pressure, SOC change, power matching error), and sends them back to the AI ​​control module. Combined with the calculated fuel injection quantity, the methanol combustion state and the degree of coordination and matching of the three parameters are determined. Combustion status is determined as follows: If the exhaust temperature is detected to be <600℃ and the peak cylinder pressure is detected to be <8Bar, it is determined to be incomplete combustion. The AI ​​adaptive model immediately corrects the control command—increasing the fuel injection quantity by 10% and advancing the ignition timing by 0.5°CA; If a sudden change in cylinder pressure is detected to be ≥1Bar / μs, it is determined to be knock. The AI ​​adaptive model immediately corrects the control command—reducing the fuel injection quantity by 15%, delaying the ignition timing by 0.5°CA, and reducing the range extender's power generation by 10%; If an abnormal peak value (≥9Bar) is detected in the cylinder pressure before spark plug ignition and the exhaust temperature rises instantaneously by ≥50℃ / s, it is determined to be pre-ignition. The AI ​​adaptive model immediately corrects the control command—stopping the current fuel injection cycle (pausing fuel injection for one cylinder working cycle), delaying the ignition timing by 1.0°CA, and reducing the range extender speed by 200r / min. After the cylinder pressure and exhaust temperature return to the normal range (peak cylinder pressure <8Bar, exhaust temperature <700℃), normal fuel injection and ignition are resumed to avoid damage to the range extender cylinder block caused by pre-ignition. After the control command is corrected based on the real-time methanol combustion status, it is directly fed back to the execution module for real-time adjustment, forming a short closed loop of "monitoring-judgment-correction-execution-re-monitoring" to ensure the system operates stably in real time.

[0054] S05, the artificial intelligence adaptive model learns and optimizes the control parameter thresholds based on the long-term accumulated multi-dimensional operating parameters, and updates the optimized thresholds into the artificial intelligence adaptive model for subsequent operation condition prediction and control command generation; Specifically, the steps of the self-learning optimization include: Once the cumulative operating time of the power system reaches a preset duration, the artificial intelligence adaptive model performs at least one of the following autonomous learning optimization operations based on the multi-dimensional operating parameters from historical operation: Autonomously optimizes the fuel injection quantity adjustment range, ignition timing offset, and power generation matching threshold; Automatically switches to high-efficiency energy-saving mode, high-reliability mode, or voltage regulation priority mode based on long-term operating scenarios; It automatically identifies conventional methanol and green methanol, and automatically adjusts the fuel injection quantity and ignition timing based on the difference in calorific value between the conventional methanol and the green methanol.

[0055] In this embodiment, if the monitored combustion state is normal, the system continues to run. When the cumulative running time of the system reaches a preset duration (≥100 hours in this embodiment), the artificial intelligence adaptive model autonomously learns and optimizes the control parameter thresholds based on historical data from long-term operation. The optimized parameters are synchronized to the entire control system through the "loop control" stage. When the next round of control process begins, the optimized parameters are directly used for execution, continuously iterating to achieve scenario adaptation and fuel adaptation. No manual debugging is required, realizing a long closed loop of autonomous learning and optimization.

[0056] In summary, an AI-adaptive multi-source power cooperative control method has the following effects: By integrating time-series prediction algorithms and reinforcement learning algorithms into an AI adaptive model, this invention achieves accurate prediction of operating conditions and dynamic optimization of control commands. Specifically, based on real-time collected multi-dimensional operating parameters, this invention can predict operating conditions over a future period and dynamically generate control commands with at least one of the following as optimization objectives: overall system efficiency, battery wear, and emissions. This effectively solves the problems of insufficient collaborative control accuracy and delayed response to transient conditions in existing technologies. Actual measurements show that the system response time is ≤50ms, the operating condition prediction accuracy is ≥95%, and millisecond-level precise coordination among the range extender, battery, and motor is achieved. The overall system operating efficiency is improved by 15%–20% compared to existing technologies.

[0057] Differentiated control strategies for various operating conditions, including rapid acceleration, heavy load, idling, light load, constant speed, and deceleration, ensure precise matching of the range extender's power generation, fuel injection quantity, ignition timing, and motor requirements. This avoids energy waste under light load and idling conditions while reducing instantaneous high-current discharge and deep charge / discharge of the battery. By real-time monitoring of combustion parameters such as cylinder pressure and exhaust temperature, abnormal phenomena such as incomplete combustion, knocking, and pre-ignition are promptly identified. An AI adaptive model then corrects the control parameters in real time, effectively improving the completeness and stability of methanol combustion. Based on these technical solutions, this invention achieves a 10%–15% reduction in methanol consumption, a more than 40% reduction in formaldehyde emissions, and a more than 30% reduction in CO emissions, fully complying with the stringent emission regulations of China VI b and above, while mitigating the risk of component damage caused by knocking.

[0058] By employing an AI adaptive model that autonomously learns and optimizes control parameter thresholds based on long-term accumulated operational data, the system further limits the autonomous optimization of fuel injection quantity adjustment, ignition timing offset, and power generation matching thresholds. It automatically switches to high-efficiency energy-saving mode, high-reliability mode, or voltage stabilization priority mode based on long-term operating scenarios. Simultaneously, it can automatically identify traditional methanol and green methanol and adjust the fuel injection quantity and ignition timing accordingly. Based on this solution, this invention requires no manual debugging and can be adapted to various application scenarios such as passenger cars, commercial vehicles, stationary power generation, and agricultural machinery. It is also adaptable to complex fuel conditions such as high proportions of methanol and impurity methanol, and can withstand extreme environments such as -40℃ cold, ≥35℃ high, and ≥3000m high altitudes. This significantly improves the system's scenario adaptability and versatility, overcoming the shortcomings of poor operating condition adaptability in existing technologies.

[0059] By employing techniques such as maintaining a range extender power generation and motor drive power error of ≤5% under constant speed conditions while keeping the battery SOC between 40% and 60%, recovering braking energy to charge the battery during deceleration and braking, and using excess energy for battery equalization charging under idling and light load conditions, highly efficient coordination of "power generation-storage-consumption" is achieved. Based on the above solutions, battery losses are reduced by 30% to 35%, effectively extending battery life and the overall system maintenance cycle, extending the maintenance cycle by more than 30%, and reducing the total lifespan cost of the system.

[0060] Example 2 like Figure 3 As shown, in a second embodiment of the present invention, an AI-adaptive multi-source power cooperative control system is provided, the system comprising: A multi-parameter acquisition module is used to acquire multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters. The AI ​​control module is used to use an artificial intelligence adaptive model that integrates time-series prediction algorithm and reinforcement learning algorithm to predict the operating conditions in the future based on the preprocessed multi-dimensional operating parameters, and dynamically generate control commands with at least one of the overall efficiency of the power system, battery consumption and emissions as optimization targets. The execution module is used to coordinate the power matching and energy distribution between the range extender, battery and motor according to the control instructions. The redundancy protection module includes a fault monitoring unit and a backup control unit. The fault monitoring unit is used to monitor the operating status of the AI ​​control module. When the AI ​​control module is detected to be faulty or malfunctioning, it automatically takes over the control function. In the takeover state, it obtains the combustion status feedback of the range extender in real time. If it is determined that there is an abnormal combustion phenomenon, it corrects the control command in real time through the backup control unit.

[0061] The collaborative control system in this embodiment is applied to methanol-powered range-extended commercial vehicles. To ensure efficient and coordinated operation of each module, the core hardware configuration is as follows: Multi-parameter acquisition module: Employing sensors made of methanol-resistant materials to avoid the impact of methanol fuel on acquisition accuracy, this module includes: a cylinder pressure sensor for acquiring cylinder pressure in the range extender, with an accuracy of ±0.1 Bar and a sampling frequency of 100 Hz, installed inside the range extender cylinder; an exhaust temperature sensor for acquiring exhaust temperature in the range extender, with an accuracy of ±5℃ and a sampling frequency of 100 Hz, installed in the exhaust pipe; an intake manifold absolute pressure sensor for acquiring the intake air volume into the cylinder, indirectly calculating the fuel injection quantity based on the theoretical air-fuel ratio and engine speed, with an accuracy of ±0.2 mL / min and a sampling frequency of 100 Hz, installed in the intake manifold; a voltage sensor for acquiring the voltage of individual battery cells or the entire battery pack, with an accuracy of ±0.01V and a sampling frequency of 50 Hz, installed in the battery module; and a current sensor for monitoring charging and discharging current, with an accuracy of... ±0.01A, sampling frequency 50Hz, installed in the battery module; Battery temperature sensor, used to collect real-time battery temperature, sampling accuracy ±0.5℃, sampling frequency 50Hz, installed inside the battery pack; Motor load power sensor, used to collect the current, voltage and phase relationship at the motor input terminal, and then calculate the real-time load power, sampling accuracy ±0.5kW, sampling frequency 100Hz, installed in the motor power supply line; Motor speed sensor, used to collect motor speed, sampling accuracy ±10r / min, sampling frequency 100Hz, installed on the motor output shaft; Ambient temperature sensor, used to collect the temperature of the environment in which the vehicle is located, sampling accuracy ±0.3℃, sampling frequency 10Hz, installed in the front compartment of the vehicle; Ambient altitude sensor, used to collect the atmospheric pressure of the environment in which the vehicle is located, and then calculate the corresponding altitude, sampling accuracy ±10m, sampling frequency 10Hz, installed in the front compartment of the vehicle.

[0062] AI control module: It adopts an MCU chip with a main frequency of ≥1GHz, and has built-in artificial intelligence adaptive model that integrates timing prediction algorithm and reinforcement learning algorithm. The chip memory is ≥16GB, which is used to store the working condition training samples (a total of 1.2 million sets, covering 4 types of working conditions: rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idling / light load), algorithm weight matrix and running data, ensuring that the LSTM algorithm prediction delay is ≤10ms and the overall control response time is ≤50ms.

[0063] The execution module includes a methanol range extender controller, a battery charge / discharge controller, and a motor controller. These controllers receive control commands from the AI ​​control module and adjust the power generation, fuel injection quantity, ignition timing, and charge / discharge rate to achieve coordinated operation of the three components. Redundancy protection module: It adopts a dual backup design, including a fault monitoring unit (real-time monitoring of the operating status of the AI ​​control module and the execution module) and a backup control unit (built-in traditional fixed threshold control program). When the AI ​​control module fails, it automatically switches to the backup mode with a switching time of ≤100ms to avoid system shutdown. At the same time, temperature and pressure protection thresholds are set, and power is immediately reduced or the system is shut down when an abnormality is triggered.

[0064] The AI-adaptive multi-source power cooperative control system provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0065] It should be noted that the above embodiments are only illustrated using methanol-powered range-extended commercial vehicles as an example, but the present invention is not limited thereto. Other embodiments: The range extender can be a methanol range extender or a range extender for other fuels (such as gasoline, ethanol, and hydrogen), only requiring adjustment of the combustion parameters accordingly.

[0066] The time series prediction algorithm is not limited to LSTM, but can also use network structures with time series prediction capabilities such as GRU and Transformer.

[0067] The reinforcement learning algorithm is not limited to Q-learning, but can also use deep reinforcement learning algorithms such as DQN and PPO.

[0068] The specific thresholds of the multi-dimensional parameters (such as temperature 600℃, pressure 8 Bar, mutation rate 1 Bar / μs, etc.) can be adjusted according to the actual application scenario, as long as they do not deviate from the core concept of the present invention.

[0069] The system can be applied to various power systems such as passenger cars, commercial vehicles, construction machinery, agricultural machinery, and stationary power generation equipment.

[0070] Example 3 like Figure 4 As shown, in the third embodiment of the present invention, the present invention provides the following technical solution: a computer, including a memory 202, a processor 201, and a computer program stored in the memory 202 and executable on the processor 201, wherein the processor 201 executes the computer program to implement the AI-adaptive multi-source power cooperative control method as described above.

[0071] Specifically, the processor 201 may include a central processing unit, a specific integrated circuit, or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0072] Memory 202 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 202 may include a hard disk drive, floppy disk drive, solid-state drive, flash memory, optical disk drive, magneto-optical disk drive, magnetic tape drive, or Universal Serial Bus drive, or a combination of two or more of these. Where appropriate, memory 202 may include removable or non-removable media. Where appropriate, memory 202 may be internal or external to a data processing device. In a particular embodiment, memory 202 is non-volatile memory. In a particular embodiment, memory 202 includes read-only memory and random access memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM, an erasable PROM, an electrically erasable PROM, an electrically rewritable ROM, or flash memory, or a combination of two or more of these. Where appropriate, the RAM may be static random access memory (SRAM) or dynamic random access memory (DRAM), wherein DRAM may be fast page-mode DRAM, extended data output DRAM, synchronous DRAM, etc.

[0073] The memory 202 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 201.

[0074] The processor 201 reads and executes the computer program instructions stored in the memory 202 to implement the above-mentioned AI-adaptive multi-source power cooperative control method.

[0075] In some embodiments, the computer may further include a communication interface 203 and a bus 200. For example, Figure 4 As shown, the processor 201, memory 202, and communication interface 203 are connected through bus 200 and complete communication with each other.

[0076] The communication interface 203 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication interface 203 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0077] Bus 200 includes hardware, software, or both, that couples computer components together. Bus 200 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, local bus. For example, and not limitingly, bus 200 may include a graphics acceleration interface or other graphics bus, an enhanced industry standard architecture bus, a front-side bus, HyperTransport interconnect, an industry standard architecture bus, a wireless bandwidth interconnect, a low pin count bus, a memory bus, a WeChat architecture bus, a peripheral component interconnect bus, a PCI Express bus, a Serial Advanced Technology Attached Bus, a Video Electronics Standards Association local bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 200 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0078] Example 4 In the fourth embodiment of the present invention, in conjunction with the above-described AI-adaptive multi-source power cooperative control method, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described AI-adaptive multi-source power cooperative control method.

[0079] Those skilled in the art will understand that the data in the flowchart, or the logic and / or steps otherwise described herein, such as a sequence of instructions or a program list that can be considered as executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device. For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0080] More specific examples of readable media include: electrical connections with one or more wires, portable computer disk drives, random access memory, read-only memory, erasable and editable read-only memory, fiber optic devices, and portable optical disc read-only memory. Furthermore, computer-readable media can even be printed on paper or other suitable media, as the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0081] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0082] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multi-source power cooperative control method based on AI adaptive control, characterized in that, The method includes: Collect multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters; An artificial intelligence adaptive model that integrates time-series prediction and reinforcement learning algorithms is used to predict the operating conditions for a future period of time based on the preprocessed multi-dimensional operating parameters, and dynamically generate control commands with at least one of the overall efficiency of the power system, battery consumption and emissions as optimization targets. According to the control instructions, the power matching and energy distribution between the range extender, battery and motor are coordinated and controlled. During the coordinated control process, the combustion status feedback of the range extender is obtained in real time. If an abnormal combustion phenomenon is detected, the control command is corrected in real time through the artificial intelligence adaptive model. The step of real-time correction of the control command includes: Real-time monitoring of cylinder pressure and exhaust temperature is performed, and corrections are made according to the following rules: If the exhaust temperature is below 600℃ and the peak cylinder pressure is below 8 Bar, it is determined that the combustion is incomplete. At this time, the following corrections are performed by the artificial intelligence adaptive model: adjusting the fuel injection quantity, advancing the ignition timing, and increasing the range extender speed. If the cylinder pressure mutation rate is detected to be greater than or equal to 1 Bar / μs, it is determined to be knocking. At this time, the following corrections are performed through the artificial intelligence adaptive model: reducing the amount of fuel injection, delaying the ignition timing, and reducing the power generation of the range extender. If an abnormal peak value of 9 Bar is detected in the cylinder pressure before spark plug ignition and the peak value is greater than or equal to 9 Bar, and the instantaneous rise rate of exhaust temperature is greater than or equal to 50℃ / s, it is determined to be pre-ignition. At this time, the following correction is performed by the artificial intelligence adaptive model: stop the current injection cycle, delay the ignition timing and reduce the range extender speed, and resume injection and ignition after the cylinder pressure and exhaust temperature return to normal. The method further includes: The AI ​​adaptive model learns and optimizes the control parameter thresholds based on long-term accumulated multi-dimensional operating parameters, and updates the optimized thresholds into the AI ​​adaptive model for subsequent operation condition prediction and control command generation. The steps of the self-learning optimization include: Once the cumulative operating time of the power system reaches a preset duration, the artificial intelligence adaptive model performs at least one of the following autonomous learning optimization operations based on the multi-dimensional operating parameters from historical operation: Autonomously optimizes the fuel injection quantity adjustment range, ignition timing offset, and power generation matching threshold; Automatically switches to high-efficiency energy-saving mode, high-reliability mode, or voltage regulation priority mode based on long-term operating scenarios; Automatically identifies conventional methanol and green methanol, and automatically adjusts the fuel injection quantity and ignition timing based on the difference in calorific value between the conventional methanol and the green methanol; The time series prediction algorithm employs a long short-term memory network, and the prediction of operating conditions specifically includes: Based on the time-series data of motor load power and speed over a period of time, the cell state is iteratively updated using the LSTM cell state update formula to filter sensor noise and retain the effective time-series trend of motor load and speed. The hidden state is extracted from the updated cell state using the LSTM hidden state output formula, and used as the feature vector for condition prediction. The hidden state feature vector is mapped to the probability distribution of each preset working condition type through a fully connected layer and the working condition prediction output formula, and the predicted working condition in the next 3 to 5 seconds is determined according to the maximum probability; wherein, the preset working condition types include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idle speed / light load.

2. The AI-adaptive multi-source power cooperative control method according to claim 1, characterized in that, The multi-dimensional operating parameters include: Range extender related parameters: combustion efficiency, fuel injection quantity, ignition timing, engine speed, exhaust temperature, and cylinder pressure; Battery-related parameters: remaining capacity, charge / discharge rate, individual cell voltage, battery temperature; Motor drive related parameters: motor load power, motor speed, motor torque; and Environmental parameters: ambient temperature, altitude, atmospheric pressure.

3. The AI-adaptive multi-source power cooperative control method according to claim 1, characterized in that, The LSTM cell state update formula includes: in, for LSTM cell state at any given time Output the value for the forget gate. for Cellular state at all times The input gate output value, This refers to the state of candidate cells for the input gate. For Hadamah accumulation; The LSTM hidden state output formula includes: in, for LSTM hidden state at time step The output value is the output gate value. The hyperbolic tangent activation function is used. The state of the LSTM cells at time t; The operating condition prediction output formula includes: in, For the future The operating condition prediction results at any given time. For normalization function, This is the output layer weight matrix. For output layer bias terms, Let t be the hidden state of the LSTM.

4. The AI-adaptive multi-source power cooperative control method according to claim 1, characterized in that, The working logic of the artificial intelligence adaptive model is as follows: The timing prediction algorithm predicts the operating conditions for the next 3 to 5 seconds based on the trend of motor load power and speed changes over the past 5 seconds. The operating conditions include rapid acceleration / heavy load, rapid deceleration / braking, constant speed, and idling / light load. The reinforcement learning algorithm dynamically adjusts the control commands based on the predicted operating condition type and real-time collected multi-dimensional operating parameters. The control commands include the following adjustment strategies: When the predicted operating conditions are rapid acceleration or heavy load, the range extender's power generation is increased to 110% to 120% of the required power, the fuel injection quantity is increased by 40% to 60% compared to the idling condition, and the ignition timing is advanced by 2 to 3 degrees of crankshaft angle. When the operating condition is predicted to be idling or light load, the range extender's power generation is reduced to the minimum stable power, the fuel injection is reduced by 30% to 50%, the ignition timing is delayed by 1 to 2 degrees of crankshaft rotation, and the excess power is used for battery equalization charging. When the predicted operating condition is constant speed, the error between the range extender's power generation and the motor's drive power should not exceed 5%, and the remaining battery charge should be maintained in the range of 40% to 60%. When the anticipated operating condition is deceleration or braking, the range extender is reduced to the idle speed power generation level, and braking energy is recovered to charge the battery.

5. A multi-source power cooperative control system based on AI adaptive technology, characterized in that, The system for executing the AI-adaptive multi-source dynamic cooperative control method as described in any one of claims 1 to 4, the system comprising: A multi-parameter acquisition module is used to acquire multi-dimensional operating parameters of the power system and preprocess the multi-dimensional operating parameters. The AI ​​control module is used to use an artificial intelligence adaptive model that integrates time-series prediction algorithm and reinforcement learning algorithm to predict the operating conditions in the future based on the preprocessed multi-dimensional operating parameters, and dynamically generate control commands with at least one of the overall efficiency of the power system, battery consumption and emissions as optimization targets. The execution module is used to coordinate the power matching and energy distribution between the range extender, battery and motor according to the control instructions. The redundancy protection module includes a fault monitoring unit and a backup control unit. The fault monitoring unit is used to monitor the operating status of the AI ​​control module. When the AI ​​control module is detected to be faulty or malfunctioning, it automatically takes over the control function. In the takeover state, it obtains the combustion status feedback of the range extender in real time. If it is determined that there is an abnormal combustion phenomenon, it corrects the control command in real time through the backup control unit.

6. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the AI-adaptive multi-source power cooperative control method as described in any one of claims 1 to 4.

7. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the AI-adaptive multi-source power cooperative control method as described in any one of claims 1 to 4.

Images