A Dynamic Power Allocation Method for MPSV Hybrid Systems Based on Operating Condition Prediction

Abstract

This invention relates to a dynamic power allocation method for MPSV hybrid systems based on operating condition prediction. Data is collected using lidar, hydrological sensors, and load sensors. The collected data is fused using time synchronization technology to form a standardized dataset. A historical operating dataset is used to train an LSTM operating condition prediction model. The model is optimized using regularization and cross-validation. The fused real-time standardized dataset is then input into the optimized LSTM model to predict the operating type 5 minutes in advance. An initial weight ratio for power allocation between the diesel engine and battery is set based on the predicted operating type. A dynamic threshold is used to determine whether load fluctuations are small, stable fluctuations or large, abrupt changes. Conventional PID correction is used for small, stable fluctuations, while adaptive PID gain adjustment is used for large, abrupt changes, adjusting the power output of the diesel engine and battery in real time. This invention solves the problem of balancing energy efficiency and power performance in MPSV hybrid systems, effectively reducing operating costs.

Description

Technical Field

[0001] This invention relates to the field of marine power control technology, and in particular to a dynamic power allocation method for MPSV hybrid systems based on operating condition prediction. Background Technology

[0002] As a core operating vessel in the marine engineering field, the MPSV multi-functional platform supply vessel needs to complete a variety of operational tasks such as material replenishment, light engineering, and emergency rescue. Different operating conditions have significantly different power requirements and response speed requirements for the power system, and the dynamic changes in wind speed, wave height, and ocean currents in the marine environment will further affect the load stability of the propulsion system.

[0003] Existing hybrid systems for offshore vessels mostly employ a real-time load passive adjustment mode for energy management. This mode adjusts the power distribution between the diesel engine and battery based solely on currently collected load data, failing to consider the load characteristics differences of MPSVs under various operating conditions or the impact of changes in marine environmental parameters on power demand. This results in several technical defects: First, diesel consumption remains high, especially under DPS-2 positioning conditions, where frequent passive power adjustments cause the diesel engine to operate in uneconomical conditions for extended periods, leading to significant fuel waste. Second, power response is sluggish; when load changes abruptly, the system cannot quickly match power demands, affecting operational efficiency and safety. Third, positioning stability is poor; the lag in power distribution easily causes uneven propulsion output, resulting in DPS-2 positioning errors exceeding allowable limits and failing to meet the requirements for high-precision positioning operations.

[0004] While some existing technologies have attempted to introduce power distribution schemes based on operational condition prediction, significant technical shortcomings remain: data acquisition using a single sensor results in insufficient data accuracy and completeness; the training data for operational condition prediction models lacks environmental labels, leading to weak model generalization ability and low prediction accuracy; and weight correction employs only conventional PID algorithms, failing to differentiate between load fluctuation types, resulting in insufficient correction accuracy. Consequently, these technologies cannot solve the challenge of balancing energy efficiency and power performance in MPSV hybrid systems operating in complex marine environments, making them unsuitable for the actual operational needs of MPSVs. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic power allocation method for MPSV hybrid systems based on operating condition prediction, which can solve the problem of balancing energy efficiency and power performance in hybrid systems and effectively reduce operating costs.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction, comprising the following steps: Step S1: Collect distance and orientation data of obstacles in the surrounding environment using lidar; collect wind speed, wave height, and ocean current speed parameters using hydrological sensors and perform filtering and noise reduction processing; collect real-time power and power change rate data of the propulsion system using load sensors; and fuse the collected data using time synchronization technology to form a standardized dataset. Step S2: Construct a historical operation dataset consisting of 1200 sets of historical operation data with environmental labels and load change characteristics. Train an LSTM operation condition prediction model based on the historical operation dataset. Optimize the model using regularization and cross-validation. Input the fused real-time standardized dataset into the optimized LSTM model to predict the operation type 5 minutes in advance. Step S3: Set the initial weighting ratio of power distribution between the diesel engine and the battery according to the predicted operation type; Step S4: Compare the difference and rate of change between the actual load of the propulsion system and the load predicted by the model in real time. Use dynamic thresholds to determine whether the load fluctuation is a small, stable fluctuation or a large, sudden fluctuation. Then, use the PID algorithm to dynamically correct the power allocation weights. Use conventional PID correction for small, stable fluctuations and adaptive PID gain adjustment for large, sudden fluctuations. The weight correction accuracy is controlled within ±1%. Adjust the power output of the diesel engine and battery in real time.

[0007] Furthermore, the conventional PID correction for small, stable fluctuations in step S4 further includes the following steps: I. Deviation Calculation: Real-time acquisition of the actual load P of the propulsion system a The deviation from the load Pp predicted by the LSTM model is e(t) = P. a -P p Simultaneously, the deviation change rate ec(t) = [e(t) - e(t-1)] / Δt is recorded, where Δt is the data acquisition time step; II. PID Three-Stage Calculation: Substituting Preset Fixed Parameters: Proportional Coefficient Kp o Integral coefficient Ki o Differential coefficient Kd, calculate the correction ΔW=Kp o ×e(t)+Ki o × +Kd o ×ec(t), where τ is any past time point from 0 to the current time t. To integrate and accumulate all historical load deviations from time 0 to the current time t, Kd o The basic fixed differential gain coefficient; III. Weighting Adjustment: The initial power allocation weighting ratio is adjusted to W. ne w=W o+ΔW, the corrected weights must satisfy the condition that the sum of the power ratios of the diesel engine and the battery is 1, where... After predicting the operating conditions for the LSTM, the initial power allocation weights for the diesel engine and battery are set. This is the weight fine-tuning correction amount calculated using conventional PID correction; This is the newly obtained power allocation weight after regular PID correction; IV. Accuracy Verification and Output: Verify whether the absolute value of the weight correction is ≤1%. If it meets the requirement, adjust the diesel-electric power output according to the corrected weight. If it exceeds the requirement, take ±1% as the maximum correction amount and then perform power output adjustment.

[0008] Furthermore, the step of adaptively adjusting the PID gain to correct large fluctuations further includes the following steps: a. Sudden Change Identification and Deviation Quantification: After determining that the load has experienced a large sudden change, calculate the real-time load deviation e(t) = P a -P p At the same time, the mutation level is classified according to the absolute value of the load change rate ec(t); b. Adaptive matching of gain parameters: The preset gain parameter mapping table is retrieved according to the mutation level, and the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd are adjusted in real time. The parameter adjustment range is 1.5-3 times that of the fixed parameters of conventional PID. c. Calculation of Variable Parameter PID Correction: Substitute the matched Kp, Ki, and Kd into the formula, where Ki is the integral gain coefficient and Kd is the derivative gain coefficient, to calculate the weight correction ΔW' = Kp × e(t) + Ki × +Kd×ec(t), and at the same time, an anti-integral saturation circuit is added. When the integral term exceeds the preset threshold, the integral calculation is cut off to avoid correction overshoot; d. Weighting and Limiting: Adjust the initial weighting ratio to W. ne w'=W o +ΔW', where ΔW' is the weight correction amount calculated by the variable parameter PID in PID gain adaptive adjustment mode, W ne w' represents the final dynamic power allocation weight after adaptive PID correction; the absolute value of the weight correction is forcibly constrained to be ≤1% to ensure correction accuracy, while ensuring that the sum of the power ratios of the diesel engine and the battery is 1. e. Rapid power output and parameter stabilization: The diesel-electric power output is immediately adjusted according to the corrected weight to achieve a power response within 0.2 seconds; when the load fluctuation tends to stabilize, the PID gain parameter is automatically switched back to the normal fixed parameter to restore the normal PID correction mode. f. Closed-loop iterative optimization: Complete steps a-e above with a period of 0.005s to shorten the iteration cycle and improve the speed of sudden change response until the load recovers to a stable state.

[0009] Furthermore, the hydrological sensor has a data acquisition range of wind speed 0-25m / s, wave height 0-5m, and ocean current speed 0-3m / s, and the load sensor has a data acquisition range of propulsion system real-time power 0-7500kW. Furthermore, the filtering and denoising process in step S1 further includes: removing outliers and smoothing the marine environmental parameters collected by the hydrological sensor.

[0010] Furthermore, the input to the LSTM condition prediction model in step S2 is the fused environmental parameters, obstacle data, real-time power of the propulsion system, and power change rate data.

[0011] Furthermore, the initial weighting ratios are as follows: 6:4 for material supply conditions, 7:3 for light engineering conditions, and 3:7 for emergency rescue conditions.

[0012] Furthermore, in step S4, determining the load fluctuation type using a dynamic threshold is further defined as follows: when the load change rate is ≤ a preset threshold, it is determined to be a small, stable fluctuation; when the load change rate is > a preset threshold, it is determined to be a large, sudden fluctuation.

[0013] Furthermore, the allocation method is applicable to the hybrid system of the 88-meter MPSV multi-functional platform supply vessel and is adapted to the DPS-2 positioning conditions.

[0014] The beneficial effects of this invention are: 1. Significantly reduced fuel consumption: This invention predicts operating conditions in advance and sets initial weights for specific operating conditions. Combined with high-precision dynamic weight correction, it ensures that the diesel engine always operates in an economical condition. Under the DPS-2 positioning condition, the fuel saving rate is increased by more than 20%, effectively reducing the operating cost of MPSV. 2. Significantly improved power response speed: The PID-dynamic threshold dual algorithm collaborative correction can achieve differentiated weight adjustment according to the type of load fluctuation, with a power response delay of ≤0.2 seconds. It can quickly match the power demand of sudden load changes and completely solve the problem of power response lag in existing technologies. 3. Significantly improved positioning stability: Precise power distribution makes the propulsion system output more stable, and the positioning error of DPS-2 is strictly controlled within ±0.3 meters, meeting the requirements of MPSV high-precision positioning operations; 4. High prediction accuracy and correction precision: The optimized LSTM model has a prediction accuracy of ≥98% and a weight correction precision within ±1%, which greatly improves the accuracy of power allocation. 5. High adaptability and practicality: This method covers all operational conditions of MPSV material supply, light engineering, and emergency rescue. It can adapt to complex marine environments with wind speeds of 0-25m / s and wave heights of 0-5m. Moreover, it does not require large-scale modification of the existing hybrid system of 88-meter MPSV and can be directly applied, making it highly valuable for promotion. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method flow of the present invention; Detailed Implementation

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Please see Figure 1 The present invention provides an embodiment: a method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction, characterized by the following steps: Step S1: The system collects distance and orientation data of surrounding environmental obstacles using lidar; wind speed, wave height, and ocean current speed parameters using hydrological sensors, and performs filtering and noise reduction processing; and real-time power and power change rate data of the propulsion system using load sensors. The collected data are fused using time synchronization technology to form a standardized dataset. Specifically, lidar collects precise distance and orientation data of environmental obstacles around the port and operating area, providing environmental characteristic basis for operational condition prediction. Hydrological sensors collect marine environmental parameters such as wind speed, wave height, and ocean current speed, and perform outlier removal and smoothing filtering to eliminate parameter errors caused by environmental interference. Load sensors collect real-time power and power change rate data of the propulsion system, reflecting the real-time load status of the propulsion system. The data collected by lidar, hydrological sensors, and load sensors are fused using time synchronization technology to ensure consistent timestamps across all dimensions, forming a standardized dataset containing environmental parameters, obstacle data, and load data. This improves the accuracy, completeness, and consistency of data collection, providing a high-quality data foundation for subsequent operational condition prediction.

[0018] Step S2: Organize 1200 sets of historical data from actual MPSV operations to construct a labeled historical operation dataset. This dataset consists of 1200 sets of historical operation data with environmental labels and load change characteristics. An LSTM operation condition prediction model is trained based on this dataset. Regularization and cross-validation are used to optimize the model. The fused real-time standardized dataset is input into the optimized LSTM model to predict the operation type 5 minutes in advance. The trained LSTM operation condition prediction model achieves a prediction accuracy of ≥98%, covering operation types such as material supply, light engineering, and emergency rescue. The 1200 sets of historical data from actual MPSV operations are organized, and environmental labels (wind speed, wave height, ocean current speed) and load change characteristic labels are added to each set of historical data to construct a labeled historical operation dataset. An LSTM Long Short-Term Memory network operation condition prediction model is trained based on this dataset. To address the overfitting problem of the LSTM model, regularization and cross-validation are used to optimize the model, improving its generalization ability and prediction accuracy. The fused real-time standardized dataset is input into the optimized LSTM working condition prediction model. The model extracts the correlation between environmental features, load features and working conditions, and accurately predicts the current and subsequent work types 5 minutes in advance. The work types cover material supply, light engineering and emergency rescue, and the model's prediction accuracy is ≥98%, providing accurate working condition basis for setting power allocation weights.

[0019] Step S3: Set the initial weight ratio of power distribution between diesel engine and battery according to the predicted operation type; the initial weight distribution is set according to the priority of power demand and the optimal energy efficiency target of each working condition. Emergency rescue working conditions give priority to ensuring power output, while material supply and light engineering working conditions take into account both energy efficiency and power.

[0020] Step S4: Real-time comparison of the difference and rate of change between the actual load of the propulsion system and the model-predicted load. Using dynamic thresholds, determine whether the load fluctuation is a small, stable fluctuation or a large, abrupt change. Then, use a PID algorithm to dynamically correct the power allocation weights. For small, stable fluctuations, conventional PID correction is used; for large, abrupt changes, adaptive PID gain adjustment is used. The weight correction accuracy is controlled within ±1%, and the power output of the diesel engine and battery is adjusted in real time. Differentiated PID correction strategies are used for different load fluctuation types: for small, stable fluctuations, conventional PID algorithms are used to fine-tune the power allocation weights to ensure power allocation stability; for large, abrupt changes, an adaptive PID gain adjustment algorithm is used to improve the PID algorithm's response speed and correction magnitude, quickly matching the sudden changes in load demand. The accuracy of the entire weight correction process is controlled within ±1%. The power output of the diesel engine and battery is adjusted in real time through the dynamically corrected weights to ensure that the power allocation accurately matches the actual load demand of the propulsion system, avoiding power over- or under-power.

[0021] The MPSV hybrid system in this invention is a complete ship propulsion system, encompassing core units such as diesel engines, batteries, power distribution control modules, propulsion systems, and auxiliary power systems. The propulsion system is the execution end of the MPSV hybrid system, responsible for converting the power output of the hybrid system into the ship's propulsion power. It is also the core carrier of the hybrid system load. The propulsion system load data collected in this invention is the core basis for reflecting the actual power demand of the MPSV hybrid system.

[0022] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, step S4, which involves using conventional PID correction for small, stable fluctuations, further includes the following steps: I. Deviation Calculation: Real-time acquisition of the actual load P of the propulsion system a The load P predicted by the LSTM model p Calculate the deviation e(t) = P a -P p Simultaneously, the rate of change of deviation ec(t) = [e(t) - e(t-1)] / Δt is recorded, where Δt is the data acquisition time step; Δt can be 0.01s. II. PID Three-Stage Calculation: Substituting Preset Fixed Parameters: Proportional Coefficient Kp o Integral coefficient Ki o Differential coefficient Kd, calculate the correction ΔW=Kp o ×e(t)+Ki o × +Kdo×ec(t), where τ is any past time point from 0 to the current time t. To integrate and accumulate all historical load deviations from time 0 to the current time t, Kd o The basic fixed differential gain coefficient; III. Weighting Adjustment: The initial power allocation weighting ratio is adjusted to W. na w=W o +ΔW, the corrected weights must satisfy the condition that the sum of the power ratios of the diesel engine and the battery is 1, where... After predicting the operating conditions for the LSTM, the initial power allocation weights for the diesel engine and battery are set. The first part represents the weight fine-tuning correction amount calculated by the conventional PID correction; the second part represents the newly updated power allocation weights obtained after the conventional PID correction. IV. Accuracy Verification and Output: Verify whether the absolute value of the weight correction is ≤1%. If it meets the requirement, adjust the diesel-electric power output according to the corrected weight. If it exceeds the requirement, take ±1% as the maximum correction amount, and then perform power output adjustment. This correction is a fixed parameter proportional-integral-derivative closed-loop correction, which achieves fine-tuning of power allocation weights to ensure power output stability.

[0023] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the step of adaptively adjusting the PID gain to correct large-scale sudden fluctuations further includes the following steps: a. Sudden Change Identification and Deviation Quantification: After determining that the load has experienced a large sudden change, calculate the real-time load deviation e(t) = P a -P p The mutation level is determined based on the absolute value of the load change rate ec(t); the larger the ec(t), the higher the mutation level, with a total of 3 levels. b. Adaptive matching of gain parameters: The preset gain parameter mapping table is retrieved according to the mutation level, and the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd are adjusted in real time. The parameter adjustment range is 1.5-3 times that of the fixed parameters of conventional PID. Among them, the higher the mutation level, the larger Kp and Kd are, and Ki is appropriately reduced to avoid integral saturation. c. Calculation of Variable Parameter PID Correction: Substitute the matched Kp, Ki, and Kd into the formula, where Ki is the integral gain coefficient and Kd is the derivative gain coefficient, to calculate the weight correction ΔW' = Kp × e(t) + Ki × +Kd×ec(t), and at the same time, an anti-integral saturation circuit is added. When the integral term exceeds the preset threshold, the integral calculation is cut off to avoid correction overshoot; d. Weighting and Limiting: Adjust the initial weighting ratio to W. ne w'=W o +ΔW', where ΔW' is the weight correction amount calculated by the variable parameter PID in PID gain adaptive adjustment mode, W ne w' represents the final dynamic power allocation weight after adaptive PID correction; the absolute value of the weight correction is forcibly constrained to be ≤1% to ensure correction accuracy, while ensuring that the sum of the power ratios of the diesel engine and the battery is 1. e. Rapid power output and parameter stabilization: The diesel-electric power output is immediately adjusted according to the corrected weight to achieve a power response within 0.2 seconds; when the load fluctuation tends to be stable, the PID gain parameter is automatically switched back to the normal fixed parameter to restore the normal PID correction mode; where the load fluctuation tends to be stable, i.e. ec(t) ≤ preset threshold.

[0024] f. Closed-loop iterative optimization: Steps a-e above are completed with a period of 0.005s, shortening the iteration cycle and improving the response speed to sudden changes until the load returns to stability. This correction is a variable parameter proportional-integral-derivative closed-loop correction. The core is to improve the response speed and correction magnitude by adjusting the PID gain parameter in real time, so as to quickly match the power demand of sudden load changes.

[0025] The preset thresholds for both correction methods are calibration values ​​based on 1200 sets of historical operation data, and the preset threshold for load change rate is 2kW / s; the conventional PID fixed parameter (Kp) o =0.008, Ki o =0.002、Kd o =0.004) is the optimal value obtained through multiple simulations and actual ship tests. The parameter mapping table for adaptive adjustment of PID gain is calibrated based on the mutation level and the actual ship dynamic response requirements to ensure that the correction effect matches the safety of the power system.

[0026] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the hydrological sensor has a data acquisition range of wind speed 0-25m / s, wave height 0-5m, and ocean current speed 0-3m / s, and the load sensor has a data acquisition range of propulsion system real-time power 0-7500kW. Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the filtering and denoising process in step S1 further comprises: outlier removal and smoothing filtering of the marine environmental parameters collected by the hydrological sensor. This is used to ensure the accuracy of parameter acquisition.

[0027] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the input to the LSTM condition prediction model in step S2 is the fused environmental parameters, obstacle data, real-time power of the propulsion system, and power change rate data.

[0028] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the initial weighting ratio is further defined as follows: 6:4 for material supply operation, 7:3 for light engineering operation, and 3:7 for emergency rescue operation. Combining the power demand priority and optimal energy efficiency target for each operation, the initial weighting ratio of diesel engine and battery power allocation is set. Specifically, for the material supply operation: the load during operation is relatively stable, balancing energy efficiency and power output, the initial weighting ratio of diesel to electric power is set at 6:4; for the light engineering operation: the power output requirement is slightly higher, so the proportion of diesel engine power is appropriately increased, and the initial weighting ratio of diesel to electric power is set at 7:3; for the emergency rescue operation: the operation needs to ensure a rapid and sufficient power response, prioritizing the rapid charging and discharging advantages of the battery, and the initial weighting ratio of diesel to electric power is set at 3:7.

[0029] Please continue reading. Figure 1 As shown, in one embodiment of the present invention, the determination of the load fluctuation type by dynamic threshold in step S4 is further as follows: when the load change rate is ≤ preset threshold, it is determined to be a small and stable fluctuation; when the load change rate is > preset threshold, it is determined to be a large and sudden fluctuation.

[0030] Please continue reading. Figure 1As shown, in one embodiment of the present invention, the allocation method is applicable to the hybrid system of an 88-meter MPSV multi-functional platform supply vessel and is adapted to the DPS-2 positioning condition. Under the DPS-2 positioning condition, fuel saving rate is increased by more than 20%, power response delay is ≤0.2 seconds, and positioning error is controlled within ±0.3 meters. Specific Implementation

[0031] A method for dynamic power allocation of a hybrid system for 88-meter MPSV material supply based on operational condition prediction. 1. Multi-sensor data fusion acquisition The LiDAR system collected precise data on the distance and orientation of obstacles around the port, with no obvious obstacles interfering. The hydrological sensor collected marine environmental parameters, measuring wind speed of 12 m / s, wave height of 2 m, and ocean current speed of 1.5 m / s. The parameters were then processed for noise reduction by outlier removal and smoothing filtering to eliminate environmental interference. The load sensor collected the real-time power of the propulsion system at 4200 kW, and also collected the power change rate of 0.5 kW / s.

[0032] The data collected by lidar, hydrological sensors, and load sensors are fused using time synchronization technology to ensure that the timestamps of each data are consistent, forming a standardized dataset that includes port environment, marine parameters, and load status.

[0033] 2. High-generalization LSTM operating condition prediction The real-time standardized dataset fused in step 1 was input into the LSTM condition prediction model optimized by regularization and cross-validation. The model extracted environmental features (wind speed 12m / s, wave height 2m, ocean current speed 1.5m / s) and load features (real-time power 4200kW, power change rate 0.5kW / s). Through the correlation analysis between features and operating conditions, the model predicted the current operation type as material replenishment 5 minutes in advance. The prediction accuracy was 99.2%.

[0034] 3. Initial weight allocation for operating conditions Based on the material supply conditions predicted by the model, and in combination with the goal of "balancing energy efficiency and power output" for this condition, the initial weighting ratio of power distribution between the diesel engine and the battery is set to 6:4; accordingly, the initial power output is calculated as follows: diesel engine output 4200kW × 60% = 2520kW, battery output 4200kW × 40% = 1680kW.

[0035] 4. Collaborative Correction of PID and Dynamic Threshold Algorithms During the material supply operation, the actual load of the propulsion system fluctuated to 4500kW, and the real-time load change rate was 1.2kW / s. Based on the preset dynamic threshold, the fluctuation was judged to be a small and stable fluctuation, and the power allocation weight was dynamically corrected using a conventional PID algorithm.

[0036] The corrected power distribution weight ratio between the diesel engine and the battery is 6.1:3.9, with a correction accuracy of 0.8%, controlled within ±1%. Based on this, the power output is adjusted as follows: diesel engine output 4500kW × 61% = 2745kW, and battery output 4500kW × 39% = 1755kW. This achieves precise matching between power distribution and actual load, ensuring the smooth operation of material supply and maintaining the diesel engine in economical operating conditions.

[0037] In this embodiment, the method of the present invention achieves precise dynamic power allocation in the 88-meter MPSV material supply operation. During the operation, the propulsion system output is stable, without problems such as power lag or power excess. Fuel consumption is reduced by 22% compared with the existing passive adjustment mode, achieving the expected technical effect.

[0038] The scope of protection of this invention is not limited to the above embodiments. For light engineering and emergency rescue conditions, the implementation logic of this method is the same as that for material supply conditions. Only the initial weight ratio is adjusted according to the condition type, and the corresponding PID correction strategy is adopted according to the load fluctuation type.

[0039] For example, in emergency rescue situations, the LSTM model predicts and sets the initial weight ratio to 3:7 to prioritize the rapid power output of the battery. If the load changes drastically during operation (e.g., the load change rate is 5kW / s), the weights are quickly corrected through a PID gain adaptive adjustment algorithm to achieve an instantaneous power boost and meet the power requirements of emergency rescue.

[0040] For DPS-2 positioning conditions, this method can continuously predict the operating conditions and dynamically adjust the weights to ensure that the propulsion system always maintains a stable low-power output, effectively offsetting the impact of the marine environment (wind speed, wave height, ocean currents) on ship positioning, controlling the positioning error within ±0.3 meters, and achieving a fuel saving rate of more than 20%, thus solving the problem of balancing positioning stability and energy efficiency in existing technologies.

[0041] This invention operates on the following principle: It integrates data from multiple sensors, including lidar, hydrological sensors, and load sensors, and generates a standardized dataset through time synchronization and filtering. Based on 1200 sets of historical data with environmental labels and load variation characteristics, it trains and optimizes an LSTM operational condition prediction model, predicting the operation type 5 minutes in advance with an accuracy ≥98%. Initial diesel-electric power weights are set according to a 6:4 ratio for material replenishment, 7:3 for light engineering, and 3:7 for emergency rescue. The weights are collaboratively corrected using a PID-dynamic threshold dual algorithm, achieving a correction accuracy within ±1%, adapting to actual load requirements. This invention improves fuel economy by over 20% under DPS-2 positioning conditions, with a power response delay ≤0.2 seconds and a positioning error within ±0.3 meters. It effectively solves the challenge of balancing energy efficiency and power performance in MPSV hybrid systems under complex marine environments, and is suitable for all operating conditions of 88-meter MPSVs, demonstrating strong practicality and adaptability.

[0042] The above description is only a preferred embodiment of the present invention and should not be construed as a limitation of this application. All equivalent changes and modifications made in accordance with the scope of the patent application of the present invention should be covered by the present invention.

Claims

1. A method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction, characterized in that: Includes the following steps: Step S1: Collect distance and orientation data of obstacles in the surrounding environment using lidar; collect wind speed, wave height, and ocean current speed parameters using hydrological sensors and perform filtering and noise reduction processing; collect real-time power and power change rate data of the propulsion system using load sensors; and fuse the collected data using time synchronization technology to form a standardized dataset. Step S2: Construct a historical operation dataset consisting of 1200 sets of historical operation data with environmental labels and load change characteristics. Train an LSTM operation condition prediction model based on the historical operation dataset. Optimize the model using regularization and cross-validation. Input the fused real-time standardized dataset into the optimized LSTM model to predict the operation type 5 minutes in advance. Step S3: Set the initial weighting ratio of power distribution between the diesel engine and the battery according to the predicted operation type; Step S4: Compare the difference and rate of change between the actual load of the propulsion system and the load predicted by the model in real time. Use dynamic thresholds to determine whether the load fluctuation is a small, stable fluctuation or a large, sudden fluctuation. Then, use the PID algorithm to dynamically correct the power allocation weights. Use conventional PID correction for small, stable fluctuations and adaptive PID gain adjustment for large, sudden fluctuations. The weight correction accuracy is controlled within ±1%. Adjust the power output of the diesel engine and battery in real time.

2. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: Step S4, which involves applying conventional PID correction to small, stable fluctuations, further includes the following steps: I. Deviation Calculation: Real-time acquisition of the actual load P of the propulsion system a The load P predicted by the LSTM model p Calculate the deviation e(t) = P a -P p Simultaneously, the deviation change rate ec(t) = [e(t) - e(t-1)] / Δt is recorded, where Δt is the data acquisition time step; II. PID Three-Stage Calculation: Substituting Preset Fixed Parameters: Proportional Coefficient Kp o Integral coefficient Ki o The differential coefficient Kd is used to calculate the correction ΔW = Kpo × e(t) + Ki. o × +Kd o ×ec(t), where τ is any past time point from 0 to the current time t. To integrate and accumulate all historical load deviations from time 0 to the current time t, Kd o The basic fixed differential gain coefficient; III. Weighting Adjustment: The initial power allocation weighting ratio is adjusted to W. ne w=W o +ΔW, the corrected weights must satisfy the condition that the sum of the power ratios of the diesel engine and the battery is 1, where... After predicting the operating conditions for the LSTM, the initial power allocation weights for the diesel engine and battery are set. This is the weight fine-tuning correction amount calculated using conventional PID correction; This is the newly obtained power allocation weight after regular PID correction; IV. Accuracy Verification and Output: Verify whether the absolute value of the weight correction is ≤1%. If it meets the requirement, adjust the diesel-electric power output according to the corrected weight. If it exceeds the requirement, take ±1% as the maximum correction amount and then perform power output adjustment.

3. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 2, characterized in that: The step of adaptively adjusting the PID gain to correct large-scale sudden fluctuations further includes the following steps: a. Sudden Change Identification and Deviation Quantification: After determining that the load has experienced a large sudden change, calculate the real-time load deviation e(t) = P a -P p At the same time, the mutation level is classified according to the absolute value of the load change rate ec(t); b. Adaptive matching of gain parameters: The preset gain parameter mapping table is retrieved according to the mutation level, and the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd are adjusted in real time. The parameter adjustment range is 1.5-3 times that of the fixed parameters of conventional PID. c. Calculation of Variable Parameter PID Correction: Substitute the matched Kp, Ki, and Kd into the formula, where Ki is the integral gain coefficient and Kd is the derivative gain coefficient, to calculate the weight correction ΔW' = Kp × e(t) + Ki × +Kd×ec(t), and at the same time, an anti-integral saturation circuit is added. When the integral term exceeds the preset threshold, the integral calculation is cut off to avoid correction overshoot; d. Weighting and Limiting: Adjust the initial weighting ratio to W. ne w'=W o +ΔW', where ΔW' is the weight correction amount calculated by the variable parameter PID in PID gain adaptive adjustment mode, W ne w' represents the final dynamic power allocation weight after adaptive PID correction; the absolute value of the weight correction is forcibly constrained to be ≤1% to ensure correction accuracy, while ensuring that the sum of the power ratios of the diesel engine and the battery is 1. e. Rapid power output and parameter stabilization: The diesel-electric power output is immediately adjusted according to the corrected weight to achieve a power response within 0.2 seconds; when the load fluctuation tends to stabilize, the PID gain parameter is automatically switched back to the normal fixed parameter to restore the normal PID correction mode. f. Closed-loop iterative optimization: Complete steps a-e above with a period of 0.005s to shorten the iteration cycle and improve the speed of response to sudden changes until the load recovers to a stable state.

4. The method for dynamic power allocation of an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: The hydrological sensor has a data acquisition range of wind speed 0-25m / s, wave height 0-5m, and ocean current speed 0-3m / s, while the load sensor has a data acquisition range of propulsion system real-time power 0-7500kW.

5. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: The filtering and denoising process in step S1 further involves: removing outliers and smoothing the marine environmental parameters collected by the hydrological sensor.

6. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: The inputs to the LSTM condition prediction model in step S2 are the fused environmental parameters, obstacle data, and real-time power and power change rate data of the propulsion system.

7. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: The initial weighting ratios are further defined as follows: 6:4 for material supply conditions, 7:3 for light engineering conditions, and 3:7 for emergency rescue conditions.

8. The method for dynamic power allocation of an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: In step S4, determining the load fluctuation type using a dynamic threshold is further defined as follows: when the load change rate is less than or equal to a preset threshold, it is determined to be a small, stable fluctuation; when the load change rate is greater than the preset threshold, it is determined to be a large, sudden fluctuation.

9. The method for dynamic power allocation in an MPSV hybrid system based on operating condition prediction according to claim 1, characterized in that: The allocation method is applicable to the hybrid system of the 88-meter MPSV multi-functional platform supply vessel and is adapted to the DPS-2 positioning conditions.

Images