Ship power supply management method based on particle swarm optimization

Abstract

The invention relates to a ship power supply management method based on a particle swarm algorithm. The method comprises the steps of constructing a sample set R, outputting a load prediction curve function based on current data, executing genetic algorithm global coarse search, searching an elite solution set P, executing particle swarm fine search, and judging whether continuous G generations in the particle swarm fine search update an optimal solution gbest or not. According to the ship power supply management method based on the particle swarm algorithm, through cooperation of genetic algorithm global coarse search and particle swarm fine search, high-dimensional space coverage and fine optimization precision are considered, compared with a single particle swarm algorithm search mode, the efficiency is higher, multi-target weight dynamic adjustment is matched, and the power supply efficiency is improved. According to the method, the single deviation that efficiency neglects cost preferentially or cost neglects stability preferentially is avoided, and a catastrophe mechanism and a closed-loop iteration mode are adopted to adapt to ship dynamic working conditions, so that the problems that high-dimensional space search efficiency is low, local extremum is easy to sink, and multi-target optimization deviates to a single target are solved.

Description

Technical Field

[0001] This invention relates to the field of marine power supply management technology, specifically a marine power supply management method based on particle swarm optimization algorithm. Background Technology

[0002] Ship power supply management is a core component in ensuring the safe, stable, and reliable operation of a ship's electrical system. It directly relates to the ship's navigation safety, normal equipment operation, and operational efficiency. Its management scope covers multiple dimensions, including power configuration, load regulation, fault response, and emergency support. It requires refined control throughout the entire process, taking into account the ship's operating conditions. Ship power supply management consists of the power supply section, power distribution network, monitoring and control system, protection devices, and electrical loads. Through the power supply providing energy, the power distribution network transmitting and distributing energy, the monitoring system coordinating and controlling energy, and the protection devices ensuring safety, it ultimately provides stable and reliable power to various electrical loads.

[0003] Chinese Patent Publication No. CN114180023A discloses a multi-energy ship control and management method and device based on a load prediction algorithm. By using the least squares support vector machine algorithm to establish a load prediction model, and then using the particle swarm optimization algorithm to optimize the parameters and determine the optimal load prediction model, the method controls the switching status of the ship's electric propulsion, ship energy equipment and primary, secondary and tertiary loads based on the input power, ship load prediction value and switching status data. This enables timely monitoring of changes in ship load and allows for the addition or shutdown of diesel generators. In practical applications, electric ships often have multiple power sources, such as diesel generators, lithium batteries, supercapacitors, and photovoltaics. The output ratio of each power source constitutes a high-dimensional vector, resulting in a large number of local extrema in the objective function. This causes the search space volume to grow exponentially with each increase of 1 in the dimension n of the solution vector. Furthermore, in high-dimensional space, the number of local extrema of the objective function increases exponentially with the dimension, making the algorithm prone to getting trapped in local extrema with low energy barriers. This leads to a sharp drop in search efficiency in high-dimensional space, making it impossible to cover all possible power allocation schemes. Moreover, when simultaneously optimizing both power supply reliability and operating costs, the algorithm tends to favor a single objective. Therefore, a ship power supply management method based on particle swarm optimization is proposed to solve the above problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a ship power supply management method based on particle swarm optimization (PSO). This method offers advantages such as efficient search and improved decision-making. It solves the problem that in electric ships, multiple power sources such as diesel generators, lithium batteries, supercapacitors, and photovoltaics exist, each with its own output ratio forming a high-dimensional vector. This results in numerous local extrema in the objective function, causing the search space volume to grow exponentially with each increase in the solution vector dimension n. Furthermore, in high-dimensional space, the number of local extrema of the objective function increases exponentially with the dimension, making the algorithm prone to getting trapped in low-energy-barrier local optima. This leads to a sharp drop in search efficiency in high-dimensional space, making it unable to cover all possible power allocation schemes. Additionally, when simultaneously optimizing both power supply reliability and operating costs, it tends to favor a single objective.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a ship power supply management method based on particle swarm optimization algorithm, comprising the following steps: S1. Collect payload data A, navigation data B, and environmental data C, and construct a sample set R based on payload data A, navigation data B, and environmental data C; S2. A load forecasting model is established using the least squares support vector machine algorithm, and the load forecasting model is incrementally trained using the sample set R. Then, a load forecasting curve function based on the current data is output. S3. Using the load prediction curve function output by S2 as input, and adjusting the fitness function weights based on the navigation data B, execute a global coarse search using a genetic algorithm to initialize the particle population. After constraint correction, the elite solution set P is obtained; S4. Using the elite solution set P output by S3 as the initial region, perform a fine-grained particle swarm search; S5. Determine whether the optimal solution gbest has been updated in consecutive generations of fine-tuning particle swarm optimization (FSO). If yes, output the updated optimal solution gbest; otherwise, trigger the catastrophe mechanism to form a new population. ’ As a particle population in S3, and then back to S3; S6 applies the optimal solution gbest output by S5 to the actual energy system, and at the same time collects real-time data after the strategy is executed as input data for the next round of S1.

[0006] Furthermore, the load data A in S1 = Navigation data B= Environmental data C= ; in ΔU represents real-time load power, ΔU represents voltage fluctuation, and Δf represents frequency fluctuation. For load mutation value, For the voltage at the sampling time, S represents the sampling time frequency, S represents the ship's operating conditions, V represents the ship's speed, and M represents the operating mode switching signal. For steady-state sampling number, For ambient temperature, This represents the sea state disturbance coefficient.

[0007] Furthermore, the global coarse search of the S3 genetic algorithm uses a real-number encoding method, and the solution vector is... ,in The output power ratio of the i-th type of power supply satisfies =1, the initial solution space is expanded to [−θ, 1+θ]: θ= ⋅Var( ), Var( is the scaling factor) ) represents the variance of the load mutation values, which is subsequently corrected to the feasible region [0,1] through constraint projection. The fitness function is: Fit(X) = ⋅ - ⋅ - ⋅ - ⋅ ; Overall Energy System Efficiency = , For the k-th class power supply efficiency, it is affected by ambient temperature. Influence; Energy consumption cost per unit time = , The unit energy consumption cost of the k-th type of power supply is affected by operating condition S and speed V. Battery life loss factor = , The loss coefficient is... This refers to the percentage of battery power. Static fluctuation variance = ( )⋅Var(Δ )+ ( )⋅Var(Δ ), For ship type, Δ Δ steady state stage Voltage / frequency fluctuations at each sampling time, + =1.

[0008] Furthermore, the weight coefficients in the global coarse search of the genetic algorithm satisfy... + + + =1, and the switching is monitored in real time by the working condition S in the navigation data B, and updated in a linear transition manner; The power balance constraint is forcibly satisfied using the projection method. The power limit constraint uses the mirror reflection method to correct out-of-bounds particles; The S4 process, which uses the elite solution set P as the initial region, includes performing feature analysis on the elite solution set P, determining the boundary and center of the initial search region for fine-grained particle swarm search, and superimposing random perturbations that follow a Gaussian distribution on each dimension of each solution vector X in the elite solution set P to generate new particles.

[0009] Furthermore, the velocity constraint of the new particle is: ∈[− ( ), ( )], =1 represents the theoretical maximum value of the power proportion of the i-th type of power supply, which is affected by ambient temperature. Dynamic correction; The inertia weight is: =0.9⋅exp{−k⋅t / [ ]}; t is the current iteration number. The upper limit for the number of iterations is dynamically set based on the ship's operating condition S and speed V: =30⋅[1+ ⋅min(V / ,1)]; in, The maximum speed of the ship. As an adjustment factor, 0 < <1, when V approaches hour, Increase the number of search iterations.

[0010] Furthermore, the fitness function of the disaster recovery mechanism in S5 is: F= (S, )⋅ + (S, )⋅ + (S, )⋅ ; in, + + =1, the weight is based on the sea state disturbance coefficients of navigation data B and environmental data C. Dynamically adjusted and updated using a linear transition method; ~ These are normalized quantitative indicators for energy efficiency, operational stability, and response speed, respectively. Energy efficiency =1-( - ) / ( - ); Let the energy consumption per unit time of each power module be E. k Total energy consumption: E 总 = , Let k be the energy consumption per unit time of the k-th type of power source. , It is a historical extreme value, and f1≥f 1min f 1min The minimum energy efficiency for the elite solution set P; Operational stability =1-(σ− ) / ( - ); Total variance σ = ε )⋅Var(ΔU)+ζ( )⋅Var(Δf); in, For device sensitivity, ε + ζ = 1. , These are the minimum permissible and safety critical variances, respectively. Response speed =1−( - ) / [ - ]; in, For load from Mutation to Adjustment time =t( )−t( ), , These represent the theoretical minimum and the maximum allowable response time in operating mode M, respectively.

[0011] Furthermore, the catastrophe operation in S5 that triggers the catastrophe mechanism includes sorting all individuals in the current population in descending order of their fitness function values, and taking the top [number of individuals]. Individuals are considered elite individuals, and the remaining individuals form an elite candidate set; Before judgment If there is high similarity among individuals, then the similar solution is replaced by calculating the Euclidean distance between individuals, and then the remaining individuals are randomly reset; otherwise, the remaining individuals are directly randomly reset. Among them, the Euclidean distance substitution similarity solution between individuals is calculated: Let X = ( , ,..., ) and Y=( , ,..., ); The Euclidean distance formula is: d(X,Y)= m is the number of power modules. and These represent the power allocation values ​​of the i-th power module in the two individuals, respectively. In addition, a bias sampling strategy is used during random reset: X i =λ⋅ +(1−λ)⋅rand[ , ]; Let λ be the i-th dimension of the optimal individual in the elite set, λ be the weighting factor, and rand be a random value within the feasible region determined by the sample set R. Finally, the elite group is mixed with the reset individuals to form a new population. ’ To ensure that the population size remains unchanged, the fitness values ​​of each individual are recalculated.

[0012] Furthermore, the core parameters of the particle swarm fine search are determined based on the weight coefficients of the global coarse search of the genetic algorithm. The inertial weight decay coefficient, iteration limit, and velocity constraint coefficient of the fine search are dynamically corrected by weight priority mapping to optimize the connection between the two types of search targets. Define priority indicators , , , These correspond to the optimization priorities of efficiency, cost, battery degradation, and fluctuation stability, respectively: = ; = ; = ; = ; Maximum number of iterations for fine search Dynamically adjusted based on the priority of the coarse search weights: like at this time, =30⋅[1+ ⋅min(V / ,1)]; like If the value is ≥0.4, then the base coefficient of 30 will be adjusted to 30⋅(1-0.2). ); at this time, =30⋅(1-0.2 )⋅[1+ ⋅min(V / ,1)]; Inertia weights for fine-grained search =0.9⋅exp{−k⋅t / [ In the context of [}], the attenuation coefficient k is dynamically adjusted based on the priority of the coarse search weights: like The corrected inertia weight formula is: =0.9⋅exp{(− ⋅t) / [ ]}; In the particle velocity constraint of the fine search, the coefficient 0.1 is dynamically adjusted according to the priority of the coarse search weights: like The corrected speed constraint is: ∈[− ( ), ( )].

[0013] Compared with existing technologies, this invention provides a ship power supply management method based on particle swarm optimization algorithm, which has the following beneficial effects: 1. This ship power supply management method based on particle swarm optimization (PSO) algorithm combines global coarse search using genetic algorithm with fine search using PSO algorithm, taking into account both high-dimensional space coverage and fine optimization accuracy. Compared with the single PSO algorithm search mode, it is more efficient. With dynamic adjustment of multi-objective weights, it avoids the single bias of "efficiency first, cost second" or "cost first, stability second". Furthermore, it adopts a disaster mechanism and closed-loop iteration to adapt to the dynamic operating conditions of ships, thereby solving the problems of low efficiency in high-dimensional space search, easy to get trapped in local extrema, and multi-objective optimization biased towards a single objective. 2. This ship power supply management method based on particle swarm optimization improves the coverage of the feasible region by expanding the initial solution space and correcting constraints, thus avoiding the neglect of potential solutions. At the same time, the fitness function quantifies multiple objectives, which can emphasize different objectives according to the working conditions. The linear transition of weights optimizes the solution adjustment time and avoids the oscillation of power allocation scheme caused by sudden changes in weights when switching working conditions. This solves the problem of incomplete multi-power search in high-dimensional solution space and the inability to balance efficiency, cost, battery loss and stability. 3. This ship power supply management method based on particle swarm optimization algorithm retains the previous... Individuals, as elite entities, avoid losing high-quality solutions, thereby improving the basic fitness of the new population, by defining priority indicators. , , , It directly maps the weights of the particle swarm optimization (PSO) fine search, ensuring that the PSO fine search target is consistent with the global coarse search of the genetic algorithm, dynamically corrects the core parameters of the PSO fine search, avoids "futile work in fine search", and optimizes the consistency of the search. Attached Figure Description

[0014] Figure 1 This is a flowchart of a ship power supply management method based on particle swarm optimization algorithm proposed in this invention. Figure 2 This is a flowchart of the global coarse search of the genetic algorithm in this invention; Figure 3 This is a flowchart of the fine-grained particle swarm search in this invention; Figure 4 This is a flowchart of the disaster triggering mechanism in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: Please refer to Figure 1-4 This embodiment of a ship power supply management method based on particle swarm optimization includes the following steps: S1. Collect payload data A, navigation data B, and environmental data C, and construct a sample set R based on payload data A, navigation data B, and environmental data C; S2. A load forecasting model is established using the least squares support vector machine algorithm, and the load forecasting model is incrementally trained using the sample set R. Then, a load forecasting curve function based on the current data is output. S3. Using the load prediction curve function output by S2 as input, and adjusting the fitness function weights based on the navigation data B, execute a global coarse search using a genetic algorithm to initialize the particle population. After constraint correction, the elite solution set P is obtained; S4. Using the elite solution set P output by S3 as the initial region, perform a fine-grained particle swarm search; S5. Determine whether the optimal solution gbest has been updated in consecutive generations of fine-tuning particle swarm optimization (FSO). If yes, output the updated optimal solution gbest; otherwise, trigger the catastrophe mechanism to form a new population. ’ As a particle population in S3, and then back to S3; S6 applies the optimal solution gbest output by S5 to the actual energy system, and at the same time collects real-time data after the strategy is executed as input data for the next round of S1.

[0017] Wherein, the load data A in S1 = Navigation data B= Environmental data C= ; It should be noted that, ΔU represents real-time load power, ΔU represents voltage fluctuation, and Δf represents frequency fluctuation. For load mutation value, For the voltage at the sampling time, S represents the sampling time frequency, S represents the ship's operating conditions, V represents the ship's speed, and M represents the operating mode switching signal. For steady-state sampling number, For ambient temperature, This represents the sea state disturbance coefficient.

[0018] Furthermore, the S3 genetic algorithm uses a real-number encoding method for global coarse search, with the solution vector being X=[X1,X2,⋯, ],in The output power ratio of the i-th type of power supply satisfies =1, the initial solution space is expanded to [−θ, 1+θ], enabling coverage of edge feasible solutions caused by load mutations: θ= ⋅Var( ), Var( is the scaling factor) ) represents the variance of the load mutation values, which is subsequently corrected to the feasible region [0,1] through constraint projection. The fitness function is: Fit(X) = ⋅ - ⋅ - ⋅ - ⋅ ; Overall Energy System Efficiency = , For the k-th class power supply efficiency, it is affected by ambient temperature. Influence; Energy consumption cost per unit time = , The unit energy consumption cost of the k-th type of power supply is affected by operating condition S and speed V. Battery life loss factor = , The loss coefficient is... This refers to the percentage of battery power. Static fluctuation variance = ( )⋅Var(Δ )+ ( )⋅Var(Δ ), For ship type, Δ Δ steady state stage Voltage / frequency fluctuations at each sampling time, + =1.

[0019] Furthermore, the weight coefficients in the global coarse search of the genetic algorithm satisfy... + + + =1, and the switching is monitored in real time by the working condition S in the navigation data B, and updated in a linear transition manner; In S4, the initial region is the elite solution set P. This includes performing feature analysis on the elite solution set P, determining the boundary and center of the initial search region for fine-grained particle swarm search, and superimposing random perturbations that follow a Gaussian distribution on each dimension of each solution vector X in the elite solution set P to generate new particles.

[0020] The velocity constraint for the new particle is: ∈[− ( ), ( )], =1 represents the theoretical maximum value of the power proportion of the i-th type of power supply, which is affected by ambient temperature. Dynamic correction; The inertia weight is: =0.9⋅exp{−k⋅t / [ ]}; t is the current iteration number. The upper limit for the number of iterations is dynamically set based on the ship's operating condition S and speed V: =30⋅[1+ ⋅min(V / ,1)]; in, The maximum speed of the ship. As an adjustment factor, 0 < <1, when V approaches hour, Increase the number of search iterations.

[0021] It should be noted that the core parameters of the particle swarm optimization fine search are determined based on the weight coefficients of the global coarse search of the genetic algorithm. The inertial weight decay coefficient, iteration limit, and velocity constraint coefficient of the fine search are dynamically corrected by weight priority mapping to optimize the connection between the two types of search targets. Define priority indicators , , , These correspond to the optimization priorities of efficiency, cost, battery degradation, and fluctuation stability, respectively: = ; = ; = ; = ; Maximum number of iterations for fine search Dynamically adjusted based on the priority of the coarse search weights: If a coarse search prioritizes optimizing energy efficiency... If the value is ≥0.4, then the adjustment factor will be applied. correction, = ⋅(1+0.5 ); at this time, =30⋅[1+ ⋅(1+0.5 )⋅min(V / ,1)]; If a coarse search prioritizes reducing battery wear... If the value is ≥0.4, then the adjustment factor will be applied. correction, = ⋅(1-0.3 ); at this time, =30⋅[1+ ⋅(1-0.3 )⋅min(V / ,1)]; If a coarse search is preferred to ensure the stability of fluctuations... If the value is ≥0.4, then the base coefficient of 30 will be adjusted to 30⋅(1-0.2). ); at this time, =30⋅(1-0.2 )⋅[1+ ⋅min(V / ,1)]; Inertia weights for fine-grained search =0.9⋅exp{−k⋅t / [ In the context of [}], the attenuation coefficient k is dynamically adjusted based on the priority of the coarse search weights: If efficiency is the priority ≥ , ≥ and ≥ Then k will be modified to =k⋅(1−0.4 ); If cost is the priority ≥ , ≥ and ≥ Then k will be modified to =k⋅(1+0.3 ); If stability is the priority ≥ , ≥ and ≥ Then k will be modified to =k⋅(1+0.5 ); The corrected inertia weight formula is: =0.9⋅exp{(− ⋅t) / [ ]}; In the particle velocity constraint of the fine search, the coefficient 0.1 is dynamically adjusted according to the priority of the coarse search weights: If battery loss has a high weighting, ≥0.3, and the coefficient 0.1 is corrected to 0.1⋅(1−0.4). ); If battery loss has a high weighting, ≥0.3, and the coefficient 0.1 is corrected to 0.1⋅(1−0.5). ); If there is no clear priority 0.3, 0.3 0.3 and 0.3, keeping the original coefficient of 0.1 unchanged; The corrected speed constraint is: ∈[− ( ), ( )].

[0022] It should be noted that the dynamic adaptation of particle swarm parameters is based on the weight priority of the genetic algorithm. When efficiency is prioritized, the number of iterations is extended and the decay of inertial weights is slowed down to enhance the global exploration capability. When stability is prioritized, the speed constraint is reduced to avoid scheme oscillation. This strategy improves the accuracy of fine-grained particle swarm search under different optimization objectives and adapts to the complex working conditions of ships.

[0023] In addition, the fitness function for the catastrophe mechanism in S5 is: F= (S, )⋅ + (S, )⋅ + (S, )⋅ ; in, + + =1, the weight is based on the sea state disturbance coefficients of navigation data B and environmental data C. Dynamically adjusted and updated using a linear transition method; ~ These are normalized quantitative indicators for energy efficiency, operational stability, and response speed, respectively. Energy efficiency =1-( - ) / ( - ); Let the energy consumption per unit time of each power module be E. k Total energy consumption: E 总 = , Let k be the energy consumption per unit time of the k-th type of power source. , It is a historical extreme value, and f1≥f 1min f 1min The minimum energy efficiency for the elite solution set P; Operational stability =1-(σ− ) / ( - ); Total variance σ = ε )⋅Var(ΔU)+ζ( )⋅Var(Δf); in, For device sensitivity, ε + ζ = 1. , These are the minimum permissible and safety critical variances, respectively. Response speed =1−( - ) / [ - ]; in, For load from Mutation to Adjustment time =t( )−t( ), , These represent the theoretical minimum and the maximum allowable response time in operating mode M, respectively.

[0024] It should be noted that the catastrophe operation that triggers the catastrophe mechanism in S5 includes sorting all individuals in the current population in descending order of their fitness function values, and taking the top [number of individuals]. Individuals are considered elite individuals, and the remaining individuals form an elite candidate set; Before judgment If there is high similarity among individuals, then the similar solution is replaced by calculating the Euclidean distance between individuals, and then the remaining individuals are randomly reset; otherwise, the remaining individuals are directly randomly reset. Among them, the Euclidean distance substitution similarity solution between individuals is calculated: Let X = ( , ,..., ) and Y=( , ,..., ); The Euclidean distance formula is: d(X,Y)= m is the number of power modules. and These represent the power allocation values ​​of the i-th power module in the two individuals, respectively. In addition, a bias sampling strategy is used during random reset: X i =λ⋅ +(1−λ)⋅rand[ , ]; Let λ be the i-th dimension of the optimal individual in the elite set, λ be the weighting factor, and rand be a random value within the feasible region determined by the sample set R. Finally, the elite group is mixed with the reset individuals to form a new population. ’ To ensure that the population size remains unchanged, the fitness values ​​of each individual are recalculated.

[0025] Using the above technical solution, when the particle swarm has no optimal solution updated for G consecutive generations, a catastrophe mechanism is triggered. Similar solutions are identified and replaced by Euclidean distance, thereby improving the diversity of the population. Furthermore, bias sampling is used to reset individuals, ensuring that the new population is distributed around the elite solution. This improves the convergence speed compared to random reset and effectively solves the problem of "the algorithm being trapped in local extreme values ​​at low energy barriers".

[0026] Example 2: Please refer to Figure 1-4 This embodiment presents a ship power supply management method based on particle swarm optimization, assuming the typical ship operating condition set is S∈{ , , , },in, For navigation status, For berthing status, For emergency situations, For each working state, the preset baseline weight is as follows: When the ship is in working condition Transition to At that time, the weight varies with the transition time. Smooth transition, total transition time T, current transition time t∈[0,T], t=0 corresponds to t=T corresponds to ; Weight update formula: = +(t / T)⋅( - ); = +(t / T)⋅( - ); = +(t / T)⋅( - ); = +(t / T)⋅( - ); Subsequently, the corrected population will be... Sort by fitness value in descending order and select the top [values]. The optimal solutions form an elite solution set P; P={ , ,⋯, |Fit[ ]≥Fit[ ]≥⋯≥Fit[ ]}.

[0027] By adopting the above technical solution, the weights are updated smoothly in a linear manner when the operating conditions are switched, avoiding voltage or frequency oscillations caused by sudden changes in the power distribution scheme, and making the voltage fluctuation during the transition phase stable.

[0028] Example 3: Please refer to Figure 1-4 Based on Example 2, a ship power supply management method based on particle swarm optimization algorithm is proposed. In S4, the elite solution set P is used as the initial region, including feature analysis of the elite solution set P to determine the boundary and center of the initial search region. Calculate the power percentage of each power source in the elite solution set P. minimum value and maximum value To obtain the boundary of each dimension [ , ], calculate the mean vector μ of the elite solution set P. , ,..., ], = The mean vector μ is used as the center of the initial search region; It should be noted that by anchoring the boundary and center of the initial search region near the globally high-quality solutions selected by GA, the search range is focused on the "verified high-fitness region", avoiding ineffective iteration of the particle swarm in low-quality regions, reducing the computational cost of fine search and improving the convergence speed. The search range for each dimension is defined as [μ-0.1,μ+0.1], where 0.1 is the fine-tuning radius set based on the power ratio range [0,1]. For each solution vector X in the elite solution set P, a random perturbation following a Gaussian distribution is superimposed on each dimension to generate a new particle in the initial region [ -0.1, Within +0.1], random samples are taken from a uniform distribution to generate supplementary particles, which further fill the initial region; It should be noted that the search scope is defined as [ -0.1, +0.1] ensures both the "precision" of the search and avoids the "local extremum trap caused by an excessively narrow range".

[0029] In addition, the power balance constraint is forcibly satisfied through the projection method. The power limit constraint uses the mirror reflection method to correct out-of-bounds particles; like >1 =2- ;like <0 ′=- After correction, the power balance constraint is executed again; Let Pi be the rated maximum power of the i-th power module. max The actual output power must satisfy Pi ≤ Pi max When it appears > At that time, the power output is fixed at Pi. max Then, the remaining power demand is corrected using a scaling method.

[0030] Using the above technical solution, the projection method is used to forcibly satisfy the condition. This ensures that the total power distribution is without deviation, avoiding power excess or insufficient power. The mirror reflection method corrects out-of-bounds particles, which can better retain the search direction information of the particles, so that the corrected solution still has optimization potential. It also increases the feasible solution ratio after constraint correction. Furthermore, for the power supply rated power limit, a fixed overload power and proportional scaling of the remaining demand are adopted to avoid practical application failure caused by the theoretical optimization solution exceeding the hardware capability.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A ship power supply management method based on particle swarm optimization algorithm, characterized in that, Includes the following steps: S1. Collect payload data A, navigation data B, and environmental data C, and construct a sample set R based on payload data A, navigation data B, and environmental data C; S2. A load forecasting model is established using the least squares support vector machine algorithm, and the load forecasting model is incrementally trained using the sample set R. Then, a load forecasting curve function based on the current data is output. S3. Using the load prediction curve function output by S2 as input, and adjusting the fitness function weights based on the navigation data B, execute a global coarse search using a genetic algorithm to initialize the particle population. After constraint correction, the elite solution set P is obtained; S4. Using the elite solution set P output by S3 as the initial region, perform a fine-grained particle swarm search; S5. Determine whether the optimal solution gbest has been updated in consecutive generations of fine-tuning particle swarm optimization (FSO). If yes, output the updated optimal solution gbest; otherwise, trigger the catastrophe mechanism to form a new population. ’ As a particle population in S3, and then back to S3; S6 applies the optimal solution gbest output by S5 to the actual energy system, and at the same time collects real-time data after the strategy is executed as input data for the next round of S1.

2. The ship power supply management method based on particle swarm optimization algorithm according to claim 1, characterized in that: The load data A in S1 = Navigation data B= Environmental data C= ; in ΔU represents real-time load power, ΔU represents voltage fluctuation, and Δf represents frequency fluctuation. For load mutation value, For the voltage at the sampling time, S represents the sampling time frequency, S represents the ship's operating conditions, V represents the ship's speed, and M represents the operating mode switching signal. For steady-state sampling number, For ambient temperature, This represents the sea state disturbance coefficient.

3. The ship power supply management method based on particle swarm optimization algorithm according to claim 2, characterized in that: The S3 genetic algorithm uses a global coarse search with real-number encoding to find the solution vector. ,in The output power ratio of the i-th type of power supply satisfies =1, the initial solution space is expanded to [−θ, 1+θ], where θ is a dynamically adjusted parameter for the initial solution space of the global coarse search of the extended genetic algorithm: θ= ⋅Var( ), Var( is the scaling factor) ) represents the variance of the load mutation values, which is subsequently corrected to the feasible region [0,1] through constraint projection, and the fitness function is: Fit(X)= ⋅ − ⋅ − ⋅ − ⋅ ; Overall efficiency of energy systems = , For the k-th class power supply efficiency, it is affected by ambient temperature. Influence; Energy consumption cost per unit time = , The unit energy consumption cost of the k-th type of power supply is affected by operating condition S and speed V. Battery life loss factor = , The loss coefficient is... This refers to the percentage of battery power. Static fluctuation variance = ( )⋅Var(Δ )+ ( )⋅Var(Δ ), For ship type, Δ Δ steady state stage Voltage / frequency fluctuations at each sampling time, + =1.

4. The ship power supply management method based on particle swarm optimization algorithm according to claim 3, characterized in that: The weight coefficients in the global coarse search of the genetic algorithm satisfy the following... + + + =1, and the switching is monitored in real time by the working condition S in the navigation data B, and updated in a linear transition manner; The power balance constraint is forcibly satisfied using the projection method. The power limit constraint uses the mirror reflection method to correct out-of-bounds particles; The S4 process, which uses the elite solution set P as the initial region, includes performing feature analysis on the elite solution set P, determining the boundary and center of the initial search region for fine-grained particle swarm search, and superimposing random perturbations that follow a Gaussian distribution on each dimension of each solution vector X in the elite solution set P to generate new particles.

5. The ship power supply management method based on particle swarm optimization algorithm according to claim 4, characterized in that: The velocity constraint of the new particle is: ∈[− ( ), ( )], =1 represents the theoretical maximum value of the power proportion of the i-th type of power supply, which is affected by ambient temperature. Dynamic correction; The inertia weight is: =0.9⋅exp{−k⋅t / [ ]}; t is the current iteration number. The upper limit for the number of iterations is dynamically set based on the ship's operating condition S and speed V: =30⋅[1+ ⋅min(V / ,1)]; in, The maximum speed of the ship. As an adjustment factor, 0 < <1, when V approaches hour, Increase the number of search iterations.

6. The ship power supply management method based on particle swarm optimization algorithm according to claim 2, characterized in that: The fitness function of the catastrophe mechanism in S5 is: F= (S, )⋅ + (S, )⋅ + (S, )⋅ ; in, + + =1, weighted based on the sea state disturbance coefficients of navigation data B and environmental data C. Dynamically adjusted and updated using a linear transition method; ~ These are normalized quantitative indicators for energy efficiency, operational stability, and response speed, respectively. Energy efficiency =1-( - ) / ( - ); Let the energy consumption per unit time of each power module be E. k Total energy consumption: E 总 = , Let k be the energy consumption per unit time of the k-th type of power source. , It is a historical extreme value, and f1≥f 1min f 1min The minimum energy efficiency for the elite solution set P; Operational stability =1-(σ− ) / ( - ); Total variance σ = ε )⋅Var(ΔU)+ζ( )⋅Var(Δf); in, For device sensitivity, ε + ζ = 1. , These are the minimum permissible and safety critical variances, respectively. Response speed =1−( - ) / [ - ]; in, For load from Mutation to Adjustment time =t( )−t( ), , These represent the theoretical minimum and the maximum allowable response time in operating mode M, respectively.

7. The ship power supply management method based on particle swarm optimization algorithm according to claim 6, characterized in that: The catastrophe operation in S5 that triggers the catastrophe mechanism includes sorting all individuals in the current population in descending order of their fitness function values ​​and taking the top [number of individuals]. Individuals are considered elite individuals, and the remaining individuals form an elite candidate set; Before judgment If there is high similarity among individuals, then the similar solution is replaced by calculating the Euclidean distance between individuals, and then the remaining individuals are randomly reset; otherwise, the remaining individuals are directly randomly reset. Among them, the Euclidean distance substitution similarity solution between individuals is calculated: Let X = ( , ,..., ) and Y=( , ,..., ); The Euclidean distance formula is: d(X,Y)= m is the number of power modules. and These represent the power allocation values ​​of the i-th power module in the two individuals, respectively. In addition, a bias sampling strategy is used during random reset: X i =λ⋅ +(1−λ)⋅rand[ , ]; Let λ be the i-th dimension of the optimal individual in the elite set, λ be the weighting factor, and rand be a random value within the feasible region determined by the sample set R. Finally, the elite group is mixed with the reset individuals to form a new population. ’ To ensure that the population size remains unchanged, the fitness values ​​of each individual are recalculated.

8. A ship power supply management method based on particle swarm optimization algorithm according to claims 3-4, characterized in that: The core parameters of the particle swarm optimization fine search are determined based on the weight coefficients of the global coarse search of the genetic algorithm. The inertial weight decay coefficient, iteration limit, and velocity constraint coefficient of the fine search are dynamically corrected by weight priority mapping to optimize the connection between the two types of search targets. Define priority indicators , , , These correspond to the optimization priorities of efficiency, cost, battery degradation, and fluctuation stability, respectively: = ； = ； = ； = ； Maximum number of iterations for fine search Dynamically adjusted based on the priority of the coarse search weights: like at this time, =30⋅[1+ ⋅min(V / ,1)]; like If the value is ≥0.4, then the base coefficient of 30 will be adjusted to 30⋅(1-0.2). ); at this time, =30⋅(1-0.2 )⋅[1+ ⋅min(V / ,1)]; Inertia weights for fine-grained search =0.9⋅exp{−k⋅t / [ In the context of [}], the decay coefficient k is dynamically adjusted based on the priority of the coarse search weights: like The corrected inertia weight formula is: =0.9⋅exp{(− ⋅t) / [ ]}; In the particle velocity constraint of the fine search, the coefficient 0.1 is dynamically adjusted according to the priority of the coarse search weights: like The corrected speed constraint is: ∈[− ( ), ( )].

Images