Ship stall simulation estimation method based on multi-dimensional data fusion

Abstract

The invention discloses a ship stall simulation estimation method based on multi-dimensional data fusion, and belongs to the field of ship stall estimation. Aiming at the problems that stalling points of ships under complex sea conditions are difficult to accurately judge and identify and navigational speed loss is difficult to dynamically estimate, a physically guided deep fusion network is constructed, a ship dynamic model is organically combined with a multi-source environment and operation data, and a fuzzy progressive clustering and multi-state Bayesian correction mechanism is introduced, so that the ship speed loss is dynamically estimated. And intelligent clustering identification of stall points and real-time adaptive estimation of navigational speed loss are realized. The method has physical consistency constraint and uncertainty quantification capability, can be widely applied to the fields of ship dynamic performance monitoring, intelligent navigation decision, extreme working condition stall early warning and the like, and greatly improves the accuracy and reliability of ship stall risk prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of ship stall estimation, and more particularly relates to a ship stall simulation estimation method based on multi-dimensional data fusion. BACKGROUND

[0002] With the rapid development of global shipping industry, the power reliability and safety of ships under harsh sea conditions, complex ocean currents and variable weather conditions have become one of the key technical problems that need to be solved in the field of maritime engineering. Especially when large ocean-going ships sail in strong wind and heavy wave environment, the additional resistance induced by waves increases dramatically and the instability of ship motion easily leads to a sharp decrease in speed or even stall phenomenon, which not only affects the efficiency of shipping and increases fuel consumption, but also may cause serious damage to the ship and safety risks to the crew. Traditional ship stall analysis mainly relies on static design parameters, linear resistance estimation methods or test data under single working condition, which is difficult to fully reflect the influence of multi-source heterogeneous environment and complex nonlinear coupling factors of ship power system, resulting in serious lack of prediction accuracy and early identification ability of ship stall under extreme sea conditions. In recent years, with the gradual popularization of sensing and control network, intelligent data acquisition and shipping big data technology, a large amount of multi-modal, multi-source and multi-temporal and spatial scale data can be obtained in real time during the operation of the ship, which provides a data basis for dynamic modeling, accurate estimation and active warning of ship stall. However, the existing multi-dimensional data fusion and intelligent modeling technology has not yet combined the ship motion physical process, wave environment characteristics and historical stall event data organically, and generally has problems such as data redundancy, weak feature expression and damaged physical mechanism. Therefore, it is urgent to develop a ship stall simulation and estimation method that can systematically fuse ship parameters, power state, environmental disturbance and historical experience, has physical consistency constraint and can adapt to dynamic change scenarios, in order to better protect the safety and efficient operation of maritime shipping. SUMMARY

[0003] The present application aims to solve the technical problems of low information utilization efficiency, insufficient feature expression, inability to effectively fuse ship power physical process and complex environmental changes, and lack of physical consistency and self-adaptive ability in stall identification of the existing ship stall estimation method in the face of multi-source heterogeneous data, and proposes a new method that can dynamically fuse ship parameters, power performance, multi-dimensional environmental monitoring and historical stall event information, and realize accurate identification of ship stall point and adaptive estimation of speed loss under extreme wave conditions by combining physical constraints and intelligent clustering algorithm, thereby significantly improving the prediction accuracy and risk warning ability of ship stall under complex working conditions.

[0004] In order to achieve the above purpose, the application adopts the following technical scheme: the method comprises:

[0005] Multi-modal multi-source data collection and preprocessing, fusion of ship body parameters, real-time power performance data, external environment monitoring, historical stall event multi-source data;

[0006] Multi-dimensional data fusion based on physical constraints, a physical constraint deep neural network is constructed, embedding the ship motion physical equation and the wave resistance model, self-learning the weight of each dimension data on stall, realizing deep adaptive fusion of multi-dimensional features;

[0007] Establish an estimation model of ship stall in waves, for ships that have not done static water rapid performance model test, use the approximate estimation formula of transport ship stall in waves to calculate;

[0008] Stall point clustering identification and adaptive estimation, automatically identify the stall point corresponding to each typical speed in waves, and automatically adjust the class centroid to adapt to the new distribution of data, so as to obtain the optimal stall estimation interval in dynamic environment.

[0009] In one scheme, the preprocessing includes:

[0010] Using statistical correlation (Corr(a, b)) and causal discrimination; According to the event chain time sequence and the physical causal chain, a multi-level, multi-scale feature nested combination is constructed;

[0011] Specifically, the original multi-modal data is mapped to a nested label feature set through feature recombination, where each is mapped by a combination of physical quantities in time sequence and causal coupling.

[0012] In one scheme, the multi-dimensional data fusion based on physical constraints specifically includes:

[0013] For each time, the input feature vector is encoded into a latent representation through the network, and the current stall risk or power performance target output is predicted;

[0014] The network structure is specially inserted with physical constraints derived from the basic equation of ship motion and the wave resistance model; For the power balance of speed V, the constraint network output should satisfy the approximation of the mechanical equation;

[0015] In the design of the network loss function, the key is the integration of physical priori, the network continuously self-learns and optimizes the stall contribution of each input feature dimension through attention mechanism or adaptive weight distribution function, and the network automatically adjusts the action strength of different physical quantities according to the history and current state, realizing adaptive fusion and discrimination of multi-dimensional features from a deep level.

[0016] In one scheme, the establishment of the estimation model of ship stall in waves specifically includes:

[0017] It is proposed according to the regression analysis of model test results of multiple transport ships.

[0018]

[0019] where, ΔV is the speed loss in waves, V is the designed speed, L is the length between two columns, B is the width, T is the draft, C W is the factor of wave level;

[0020]

[0021] K is a correction coefficient, which is calculated by the following formula:

[0022] when L, L, 150, K = -0.05·H / 3 +0.9

[0023] when 150 < L < 200, K = 1.3

[0024] when L…200,

[0025] H 1 / 3 Hs is the significant wave height, and T1 is the characteristic period of waves.

[0026] In one scheme, the stall point clustering identification and adaptive estimation comprises:

[0027] For the power-speed curves under different working conditions obtained by S3, a "fuzzy-progressive stall clustering analysis" method (FGA) constrained by physical laws is used to automatically identify the stall point corresponding to each typical speed in waves, and the clustering centroid is automatically adjusted to adapt to the new distribution of data, so that the optimal stall estimation interval under dynamic environment is obtained;

[0028] In the preliminary clustering stage, a weighted fuzzy clustering algorithm is used to define the membership degree of each data point to the clustering centroid.

[0029] The clustering number and initial centroid setting are guided by the characteristics of the critical turning point of the power-speed curve; in order to fuse the physical constraints, a progressive constraint condition is constructed: only when the samples corresponding to a certain clustering center meet the dynamic imbalance criticality, it is considered as a stall center; the clustering centroid is dynamically adjusted in each iteration with progressive weights, so as to adapt to the changes of environment and data distribution.

[0030] After step-by-step clustering optimization, the optimal stall estimation interval under dynamic environment is identified.

[0031] In one scheme, the stall point clustering identification and adaptive estimation further comprises: the speed loss adaptive reconstruction uses the intersection point method of effective power curve-static water power as the basis, introduces a multi-state Bayesian dynamic correction model, captures the speed loss ΔV under the influence of waves, real-time corrects the uncertainty and data drift, and outputs the final stall estimation value and confidence interval.

[0032] The present application has the beneficial effects of:

[0033] The present application introduces a multi-modal multi-source data fusion and physical constraint intelligent analysis method, which not only significantly improves the automatic identification accuracy of the loss speed point of the ship power system in complex sea conditions, but also realizes the real-time adaptive estimation of the speed loss, providing strong data analysis and decision support for ship safety operation and accident warning. Compared with the existing method, the present application can fully excavate the deep coupling relationship between the ship's own parameters, power performance, environmental disturbance and historical loss speed events, and enhance the physical interpretation of feature expression and the reliability of model reasoning. Through the physical guided adaptive fusion network, the failure problem of pure data driven model in limited training samples or environmental mutation is effectively avoided, which guarantees the physical consistency and generalization of the loss speed identification. Intelligent clustering and Bayesian dynamic correction algorithm make the estimation result have uncertainty quantization ability and adaptive correction performance, greatly improving the timeliness and accuracy of loss speed risk warning in extreme and novel sea conditions. Therefore, the present application not only promotes the development of ship power performance monitoring, loss speed mechanism identification and fault prediction and other frontier technologies, but also provides solid technical support for maritime transportation safety and green energy efficiency management. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The present application has the beneficial effects of:

[0035] Figure 2 The effective power Pe~speed V curve diagram is shown in Figure 1.

[0036] Figure 3 The navigation decision path result is shown in Figure 2.

[0037] Figure 4 The ship state change result is shown in Figure 3.

[0038] Figure 5 The navigation risk result is shown in Figure 4. DETAILED DESCRIPTION

[0039] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the related drawings. The drawings show typical embodiments of the present application. However, the present application can be realized in many different forms and is not limited to the embodiments described in the present application. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0040] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0041] like Figure 1 As shown, a method for simulating and estimating ship stall based on multidimensional data fusion specifically includes the following steps:

[0042] S1. Multimodal and multi-source data acquisition and preprocessing

[0043] It integrates multi-source data such as ship body parameters (main dimensions, propulsion system, etc.), real-time dynamic performance data (speed, main engine output, rudder angle), external environmental monitoring (wave parameters, wind conditions, flow field), and historical stall events.

[0044] In the multimodal, multi-source data acquisition and preprocessing stage, the first step is to construct a comprehensive data acquisition system. This system integrates ship body parameters (such as overall length L, beam B, depth D, displacement T, and propulsion system parameters P). prop (etc.), to acquire various data on ship dynamic performance in real time (such as speed V(t), main engine output power P, etc.). eng (t), rudder angle δ(t), and simultaneously integrate external environmental monitoring information, including wave parameters (such as significant wave height Hs, wave period T). w Wave direction θ w Wind conditions (wind speed U) wind ), wind direction θ wind ), flow field (flow velocity U) current ), flow direction θ current In addition, historical stall event data, accident-related information, and trigger times are also recorded and denoted as the stall event marker set E. stall All the above data sources are aligned at an accurate timestamp t to form a multidimensional heterogeneous original dataset D. raw (t).

[0045] The "relationship-oriented feature recombination" method is adopted to nest and combine the original data of different modalities in a unified data space according to time sequence and causality, and automatically generate interactive label feature sets, providing input variables with more physical meaning and interpretation for subsequent modeling.

[0046] First, the feature pairs are screened by statistical correlation (Corr(a, b)) and causal discrimination (e.g. Granger causality test or mutual information based discrimination MI(a, b)). Then, the multi-level and multi-scale feature nested combinations are constructed according to the event chain time series [t1, t2,..., t N ] and physical causal chain (propulsion force - speed - load - environmental interaction). Specifically, let the original multi-modal data X(t) = [X1(t), X2(t),..., X M (t)], where each feature X i belongs to a different physical quantity space. Through feature reorganization mapping F: X(t)→ Z(t), the nested label feature set Z(t) = [Z1(t),..., Z K (t)] is generated, where each Z k is mapped by a combination of physical quantities in a time series and causally coupled manner: for example

[0047]

[0048] Here f k represents the feature generation function of correlation and causal coupling, and τ j is the optimal time delay between different features. Through machine learning method knowledge embedding, this process automatically generates feature labels that can express complex interactions and discriminant significance, achieving high physical interpretability and discriminant effectiveness of input variables, and providing a solid multi-dimensional and multi-scale data basis for subsequent modeling.

[0049] S2, multi-dimensional data fusion based on physical constraints

[0050] A physically constrained deep neural network is constructed, embedding the ship motion physical equation and the wave action resistance model. The dynamic balance of multi-source feature input is realized, and the network self-learns the weight of each dimension data on the stall effect, realizing deep adaptive fusion of multi-dimensional features. Prior knowledge based on kinetic energy conservation and ship-wave interaction is introduced into the network loss function to ensure the consistency of the reasoning physics.

[0051] In the multi-dimensional data fusion representation stage based on physical constraints, the nested label feature set Z(t) = [Z1(t),..., Z K (t)] generated in step one is first input, and a physics-guided adaptive fusion network (PGAFN) is designed and built. The structure of this network combines deep neural networks (such as multi-layer perceptron) and physical models such as ship motion-wave interaction, which strictly embeds the physical process into the information flow of the network. Specifically, for each time t, the input feature vector Z(t) is processed by the network N phyEncoding as potential representation H(t), and predicting current stall risk or power performance, etc. target outputs

[0052] H(t) = f(W - Z(t) + b),

[0053] where f(·) is the activation function, W, b are network parameters, g(·) is the output mapping. Physical constraints derived from ship motion fundamental equations and wave drag models are specially inserted in the network structure. For the power balance at speed V, the network output should satisfy the following mechanical equation approximation:

[0054]

[0055] where M is the ship mass, T prop is the total thrust of the propulsion system, which can be converted from the main engine output and propeller efficiency, (R total = R calm + R wave ) is the total resistance, including the calm water resistance R calm and the wave added resistance R wave calculated based on wave spectrum function (such as Longuet-Higgins spectrum). In the design of the network loss function, the integration of physical prior is extremely critical. Let the target output be y * (t), then the total loss is:

[0056] L total = L pred + l_1L phy1 + l_2L phy2

[0057] where is the prediction error; L phy1 characterizes the kinetic energy conservation constraint, for example ensures the consistency of mechanical balance; L phy2 then embodies the reasonable range of ship-wave interaction (such as the additional resistance does not exceed the physical threshold), l_1, l_2 are weight factors, balancing data learning and physical consistency requirements. In addition, the PGAFN network continuously self-learns and optimizes the stall contribution of each input feature dimension through attention mechanism or adaptive weight distribution function a k (t), that is, in the model structure

[0058]

[0059] where The network can automatically adjust the strength of different physical quantities according to the history and current state, and dynamically realize the adaptive fusion and discrimination of multi-dimensional features from the deep level. In this way, the network comprehensively reflects the nonlinear and time-varying interaction relationship of multi-source information, and gives more reliable and physically interpretable stall estimation results under the framework of ship dynamics rationality.

[0060] S3, Establishing a model for estimating the stall of a ship in waves

[0061] According to the increase in resistance of the ship in waves, multiplied by the corresponding speed, the effective power increase of the ship in waves is obtained. Adding the effective power value of the corresponding speed in still water, the total effective power value Pew under each wave condition is obtained. As for the stall of the ship in waves, the current common processing method is to draw the effective power Pew in waves and the effective power Pes in still water on the same "effective power Pe ~ speed V curve" graph. On this curve graph, a horizontal line is drawn according to the effective power value at the design speed in still water, and the effective power Pe under each wave condition is intersected. The difference ΔV between the speed value corresponding to each intersection point and the speed value in still water is the stall in waves. However, this processing method is based on the assumption that the self-propulsion elements of the ship in waves are equivalent to those in still water. Thus, the speed difference ΔV should originally be calculated on the "received power Pd ~ speed V curve", which is simplified to directly calculating the speed difference ΔV on the "effective power Pe ~ speed V curve". As shown in the following figure, it is the "effective power Pe ~ speed V curve" of a certain ship under each wave condition, from which the stall in waves can be calculated. Figure 2

[0062] For ships that have not undergone still water rapid performance model test, the approximate estimation formula for calculating the stall of a transport ship in waves can be directly used to calculate it. This formula is proposed based on the regression analysis of the model test results of a plurality of transport ships.

[0063]

[0064] Where ΔV is the stall in waves, V is the design speed, L is the length between two columns, B is the width, T is the draft, C W represents the factor of wave level;

[0065]

[0066] K is a correction coefficient, which is calculated by the following formula:

[0067] When L, 150, K = -0.05·H / 3 +0.9

[0068] When 150 < L < 200, K = 1.3

[0069] When L…200, ​

[0070] H 1 / 3 Significant wave height, T1 Wave characteristic period.

[0071] S4, Stall point clustering identification and adaptive estimation

[0072] For the different working condition power-speed curves obtained in S3, a "fuzzy-gradual aggregation" method (FGA) constrained by physical laws is used to automatically identify the stall point corresponding to each typical speed in waves. The cluster center is automatically adjusted to adapt to the new data distribution, so as to obtain the optimal stall estimation interval in dynamic environment.

[0073] In the stall point clustering identification and adaptive estimation stage, first, the multi-condition "effective power-speed" curve data obtained in S3 needs to be intelligently clustered. For this purpose, a "fuzzy-gradual aggregation" (FGA) method combined with physical constraints is proposed to describe the nonlinear changes and stall critical behavior of ship power performance in complex environments such as waves. In the implementation process, each group of working condition samples is represented as x i = [P eff,i , V i ], where P eff,i is the effective propulsion power, and V is the corresponding speed. In the preliminary clustering stage, a weighted fuzzy clustering algorithm (such as Fuzzy C-means, FCM) is used to define the membership μ j of each data point to the cluster center c ij :

[0074]

[0075] where C is the number of cluster centers, and m is the fuzzy index. The clustering number and initial center setting are guided by physical knowledge (such as power-speed critical turning point characteristics). To integrate physical constraints, a progressive constraint condition is constructed: that is, only when the sample corresponding to a certain cluster center satisfies the power imbalance critical (such as P eff,min <P eff,stall , and sharp decline), it is considered as a stall center. The cluster center c is dynamically adjusted by the following weight progressive formula in each iteration to adapt to the changes in environment and data distribution:

[0076]

[0077] where γ weight is used to integrate the theoretical physical expected stall point to better cope with the influence of actual different working conditions and waves. After step-by-step clustering optimization, the optimal stall estimation interval in dynamic environment can be identified.

[0078] Speed loss adaptive reconstruction: Based on the effective power curve-still water power intersection method, the "multi-state Bayesian dynamic correction model" is introduced, which not only captures the speed loss ΔV under the influence of waves, but also real-time corrects the uncertainty and data drift, and outputs the final speed loss estimate and confidence interval.

[0079] In the speed loss adaptive reconstruction part, based on the "effective power curve-still water power intersection method", let represent the critical speed of losing speed in waves, is the corresponding speed loss, where V calm is the expected speed in still water. The specific intersection process is:

[0080]

[0081] Through numerical iteration, we get Considering the environmental changes, data drift and uncertainty problems in real working conditions, the "multi-state Bayesian dynamic correction model" is further introduced. The target distribution p(ΔV|data,θ) is constructed, where θ is the Bayesian model hyperparameter and the observation is recursively updated over time. Bayesian filtering / particle filtering algorithm is used to dynamically update the speed loss ΔV and its confidence interval:

[0082] p(ΔV t |D 1:t )∝p(ΔV t |ΔV t-1 )·p(data t |ΔV t )

[0083] Finally, the real-time estimate and confidence interval [L,U] of ΔV are output, which improves the robustness of speed loss identification and provides risk warning ability of quantitative uncertainty. This process ensures the physical interpretation of clustering analysis and the reliable feedback of adaptive estimation results to real complex environment.

[0084] Simulation experiment

[0085] The effectiveness of the ship navigation auxiliary decision-making model based on deep reinforcement learning is verified by example. In all experiments, the ship uses the same dynamic parameters, and the verification platform is pycharm, which has powerful computing power and drawing functions to complete the work in this paper. The framework of ship navigation auxiliary decision-making consists of two parts: approaching the target and avoiding the meteorological risk area. When there is no meteorological risk area in the environment or the meteorological risk area is not in the safe area, the ship will adjust the direction to the turning point; When the meteorological risk area appears in the designed route range, the ship uses the Q learning strategy to make auxiliary decision and redesign the route. First, set the parameters: learning rate α = 0.5, decay factor γ = 0.8, parameter w = 0.02, ship speed vr = 18 kn.

[0086] No weather risk and ship speed loss

[0087] To evaluate the ship's path planning and decision-making ability in this environment, a two-dimensional grid ocean scene is constructed, which divides the sea area into a two-dimensional grid simulation experiment. The starting point and the end point are preset in the sea chart, and the autonomous agent needs to start from the starting point, avoid the navigation risk area, and finally reach the target point in the case of unknown or partially known. The obstacle risk area is distributed in the map in a static form, which is represented as an increase in passing cost or a decrease in reward. The sea chart also contains several dynamic obstacle points to increase the complexity of decision-making. In this environment, the ship takes the current position and its neighborhood grid information as the state input, and uses a discrete action space to make movement decisions. The reward function design fully considers the dual influence of obstacle risk on safety and efficiency: moving in ordinary grid points gives a slight negative reward, entering different types of obstacle risk areas gives corresponding degree of punishment, and reaching the end point gives a high positive reward.

[0088] DRL algorithm is used in the training process, and experience replay and target network are introduced to improve the training stability and convergence speed. As shown in Figures 3-5 Through multiple rounds of training and testing, the ship's avoidance ability and path optimization performance under different risk area distributions are evaluated. The experiment focuses on whether the ship can autonomously learn to identify risk areas and actively avoid them, as well as achieve a relatively optimal path length and cumulative reward under the premise of ensuring safety. The final results are quantitatively displayed through trajectory visualization, risk exposure frequency analysis, and other methods, providing a benchmark and reference for subsequent path planning research in dynamic risk environments.

[0089] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0090] It should be understood that the above detailed description of the technical solutions of the present application by means of preferred embodiments is illustrative rather than limiting. Those skilled in the art can modify the technical solutions recorded in each embodiment on the basis of the description of the present application, or make equivalent substitutions for part of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A ship stall simulation estimation method based on multi-dimensional data fusion, characterized in that: The method comprises: Multimodal multi-source data acquisition and preprocessing, fusion of ship body parameters, real-time power performance data, external environment monitoring, and historical stall event multi-source data; Multi-dimensional data fusion based on physical constraints, construction of a physical constraint deep neural network, embedding of ship motion physical equations and wave action resistance models, self-learning of the weight of each dimension data on stall influence, and realization of deep adaptive fusion of multi-dimensional features; Establishing a stall estimation model of a ship in waves, and using an approximate estimation formula of a transport ship in waves to calculate the stall of a ship that has not undergone a static water rapid performance model test; Stall point clustering identification and adaptive estimation, automatic identification of the stall point corresponding to each typical speed in waves, automatic adjustment of the cluster centroid to adapt to the new distribution of data, and thus obtaining the optimal stall estimation interval in a dynamic environment.

2. The ship stall simulation estimation method based on multi-dimensional data fusion according to claim 1, characterized in that: The preprocessing comprises: Using statistical correlation (Corr(a, b)) and causal discrimination; constructing multi-level and multi-scale feature nested combinations according to the event chain time sequence and physical causal chain; Specifically, the original multimodal data is mapped to generate a nested label feature set through feature reorganization, wherein each is mapped by a combination of physical quantities in a time sequence and a causal coupling manner.

3. The ship stall simulation estimation method based on multi-dimensional data fusion according to claim 1, characterized in that: The multi-dimensional data fusion based on physical constraints specifically comprises: For each time, the input feature vector is coded into a latent representation through the network, and the current stall risk or power performance target output is predicted; The network structure is specially inserted with physical constraints derived from the basic equations of ship motion and wave resistance models; for the power balance of the speed V, the constraint network output should satisfy the approximation of the mechanical equation; In the loss function design of the network, the key is the integration of physical priors. The network continuously self-learns and optimizes the stall contribution of each input feature dimension through attention mechanisms or adaptive weight distribution functions. The network automatically adjusts the action strength of different physical quantities according to the historical and current state, realizing the adaptive fusion and discrimination of multi-dimensional features from a deep level.

4. The ship stall simulation estimation method based on multi-dimensional data fusion according to claim 1, characterized in that: The establishment of the stall estimation model of a ship in waves specifically comprises: Proposed according to the regression analysis of model test results of multiple transport ships; where ΔV is the loss of speed in waves, V is the designed speed, L is the length between perpendiculars, B is the beam, T is the draft, and C is the block coefficient W factor indicating wave height The K correction coefficient is calculated by the following formula: When L, 150, K = -0.05-H / 3 +0.9 When 150 < L < 200, K = 1.3 When H 1 / 3 Significant wave height, T1 Wave characteristic period.

5. The ship stall simulation estimation method based on multi-dimensional data fusion according to claim 1, characterized in that: The stall point clustering identification and adaptive estimation comprises: For the power-speed curves under different conditions obtained by S3, a "fuzzy-progressive stall clustering analysis" method (FGA) constrained by physical laws is used to automatically identify the stall point corresponding to each typical speed in waves, and the cluster centroid is automatically adjusted to adapt to the new distribution of data, thereby obtaining the optimal stall estimation interval in a dynamic environment; In the preliminary clustering stage, a weighted fuzzy clustering algorithm is used to define the membership degree of each data point to the cluster centroid; The clustering number and initial centroid setting are guided by the power-speed critical turning point characteristics; to integrate physical constraints, progressive constraint conditions are constructed: only when the samples corresponding to a cluster center satisfy the power imbalance threshold, it is considered as a stall center; the cluster centroid is dynamically adjusted in each iteration to adapt to the changes in the environment and data distribution; After step-by-step clustering optimization, the optimal stall estimation interval in a dynamic environment is identified.

6. The ship stall simulation estimation method based on multi-dimensional data fusion according to claim 1, characterized in that: The stall point clustering identification and adaptive estimation further include: the speed loss adaptive reconstruction uses the effective power curve-still water power intersection point method as a basis, introduces a multi-state Bayesian dynamic correction model, captures the speed loss AV under the influence of waves, corrects the uncertainty and data drift in real time, and outputs the final stall estimation value and confidence interval.

Images