Hybrid power output control method and device based on emission optimization

Abstract

The invention discloses a hybrid power output control method and device based on emission optimization, and belongs to the field of ship power system control, and the method comprises the steps: predicting the propulsion power demand of a next time step according to the propulsion power data of a historical time step; determining a first combustion state of the main engine at the current time step according to the first tail gas component data, and obtaining a multi-modal feature vector in combination with a first load state of the engine; obtaining a first main engine output power decision of an optimal first prediction Q value based on a deep neural network model according to the multi-modal feature vector; wherein the deep neural network is obtained by using a deep reinforcement learning algorithm and training through a reward function with tail gas emission minimization as a target; and controlling the hybrid power output system to execute power distribution operation according to the propulsive power demand and the first main engine output power decision. According to the invention, the problem of excessive exhaust emission caused by unreasonable power distribution of the hybrid power system can be solved.

Description

Technical Field

[0001] The present application relates to the field of ship power system control, and in particular to a hybrid power output control method and device based on emission optimization. Background Art

[0002] Since traditional power systems that rely on internal combustion engines have high emissions and high energy consumption, hybrid propulsion systems have advantages such as lower energy consumption. Therefore, hybrid propulsion systems have gradually become the first choice for ship propulsion systems.

[0003] However, existing hybrid power output control methods typically prioritize power demand or energy efficiency, ignoring the real-time nature of exhaust emissions. This can lead to significant emissions fluctuations or excessive emissions due to improper power allocation under certain operating conditions. Furthermore, the main engine's combustion state is affected by a variety of factors, including load variations and fluctuating sea conditions. Traditional control strategies struggle to balance power response with emissions control.

[0004] Therefore, how to ensure power response while avoiding excessive exhaust emissions due to unreasonable power distribution of the hybrid system has become a technical problem that needs to be solved. Summary of the Invention

[0005] The present application provides a hybrid power output control method and device based on emission optimization, which can solve the problem of excessive tail gas emissions caused by unreasonable power distribution control of the hybrid power system in the prior art.

[0006] In one embodiment of the present application, a hybrid power output control method based on emission optimization is provided, comprising: Collect propulsion power data of several historical time steps of the ship, and use the pre-trained long short-term memory network model to obtain the propulsion power requirement for the next time step based on all the propulsion power data; collecting first exhaust gas component data of exhaust gas emitted by the ship at the current time step, and determining a first combustion state of the main engine at the current time step through a pre-trained support vector machine based on the first exhaust gas component data; Collecting the first load state of the main engine at the current time step, and weightedly fusing the first combustion state and the first load state through an attention mechanism module to obtain a multimodal feature vector; Obtaining a pre-trained deep neural network model, and obtaining, based on the deep neural network model and the multimodal feature vector, a first predicted Q value for each main engine output power decision within a preset discrete action space; wherein the deep neural network is obtained by constructing a reward function with the goal of minimizing exhaust emissions and training using a deep reinforcement learning algorithm based on the reward function; obtaining a first main engine output power decision of an optimal first predicted Q value, and controlling the hybrid output system to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision.

[0007] Compared with the prior art, the above-mentioned embodiment has the following beneficial effects: since the exhaust composition of the ship is related to the combustion state and the load state of the main engine, the combustion state of the current main engine is determined based on the support vector machine, the real running state of the current main engine is determined first, and accurate data input is provided for the subsequent deep neural network model. At the same time, since the power distribution decision at the current time step will affect the load state and the combustion state of the main engine at the next time step, and the load state and the combustion state of the main engine at the next time step determine the exhaust composition, the exhaust emission can be optimized by optimizing the power distribution decision. The deep reinforcement learning algorithm is adopted in the present application, the deep neural network model is trained by a reward function aiming at minimizing the exhaust emission, the predicted Q value of each main engine output power decision is evaluated by the deep neural network model, so that the first main engine output power decision obtained finally reduces the exhaust emission to the greatest extent. Finally, by predicting the propulsion power demand of the ship at the next time step in advance, the propulsion power demand and the first main engine output power decision are combined, the output power decision of the energy storage can be determined, so that the power distribution operation finally executed can meet the power response demand of the ship at the next time step.

[0008] Further, the determination of the first combustion state of the main engine at the current time step based on the first exhaust composition data through the pre-trained support vector machine comprises: selecting a corresponding first support vector machine according to the first load state, wherein the first support vector machine is obtained by training according to second exhaust composition data of the exhaust gas of the ship when the main engine is in different combustion states under the first load state; determining the first combustion state of the main engine at the current time step under the first load state based on the first support vector machine and the first exhaust composition data.

[0009] Compared with the prior art, the above-mentioned embodiment has the following beneficial effects: since different load states and combustion states are two independent factors affecting the composition of the final exhaust emission, a corresponding support vector machine model is selected based on the first load state of the main engine which can be actually observed first, so as to remove the interference of the load state in the combustion state judgment process; further, when the load state of the main engine is fixed, the composition of the exhaust gas under different combustion states has obvious difference, so the combustion state of the current main engine can be accurately reflected by the support vector machine, and a reliable input basis is provided for the subsequent power distribution optimization.

[0010] Further, the first load state of the current time step main engine is collected, and the first combustion state and the first load state are weighted and fused through an attention mechanism module to obtain a multi-modal feature vector, including: The first combustion state is embedded and coded through a preset embedding layer to obtain a first vector; The first load state is linearly transformed and activated through a preset feedforward neural network to obtain a second vector with the same dimension as the first vector; According to the correlation of the first combustion state and the first load state with the first exhaust component data, respectively, the first weight corresponding to the first vector and the second vector is determined; The first vector and the second vector are multiplied by the respective first weight and then added to obtain the multi-modal feature vector.

[0011] Compared with the prior art, the above embodiment has the following beneficial effects: Since the first combustion state is discrete data, embedding coding is required to make it no longer a single integer but a dense representation that can participate in calculation. At the same time, since the load state is real device running data that can be actually collected, it is continuous data. Therefore, through linear transformation and activation, the multi-dimensional continuous signal is compressed or expanded to the same dimension as the embedded vector, ensuring the feasibility of subsequent weighted fusion. In addition, since the load state and the combustion state are two key factors that affect the composition of the final exhaust emission, the first combustion state and the first load state are respectively extracted to obtain the first vector and the second vector, and the first vector and the second vector are weighted and fused through the attention mechanism to obtain the multi-modal data, which more truly reflects the running state of the current main engine, thereby improving the input quality and learning ability of the deep neural network.

[0012] Further, the power distribution operation of the hybrid power output system is controlled according to the propulsion power demand and the first main engine output power decision, including: The output power of the main engine is controlled according to the first main engine output power decision, and the output power of the energy storage is controlled according to the difference between the propulsion power demand and the first main engine output power decision; After the ship performs the power distribution operation, third exhaust component data of the exhaust gas emitted by the ship at the next time step are collected, and the experience pool is updated according to the difference between the first exhaust component data and the third exhaust component data; According to the updated experience pool, the parameters of the deep neural network model are fine-tuned through a supervised learning method.

[0013] Compared with the existing technology, the above embodiment has the following beneficial effects: after knowing the output power decision of the first main engine, the propulsion power gap of the current propulsion power demand can be determined based on the output power decision of the first main engine. The gap is the output power that needs to be allocated to the energy storage, thereby ensuring that the power output of the main engine and the energy storage can meet the power response requirements of the ship in the next time step; further, when the hybrid power system executes the corresponding output power decision, the operating state of the main engine will change, and the exhaust emission composition data will also change accordingly. The third exhaust composition data obtained at this time is real sample data. The experience pool is updated according to the real sample data, and the neural network parameters are continuously fine-tuned in combination with supervised learning, thereby enhancing the adaptability of the deep neural network model to real environmental changes, and at the same time improving the robustness and generalization ability of the deep neural network model under actual working conditions.

[0014] Furthermore, updating the experience pool according to the difference between the first exhaust gas component data and the third exhaust gas component data includes: Determining second weights corresponding to various exhaust gas components using an entropy weight method based on the first load state and the first combustion state to obtain a weight parameter combination; wherein a weight parameter combination corresponds to a combination of a load state and a combustion state; For each exhaust component, after calculating a first difference between its value in the first exhaust component data and its value in the third exhaust component data, accumulating a first product of the first difference corresponding to each exhaust component and a second weight to obtain a weighted sum; The weighted sum is used as the actual Q value, the first load state and the first combustion state are used as input data, a training sample is constructed according to the actual Q value, the input data and the first main engine output power decision, and the training sample is added to the experience pool.

[0015] Compared with the prior art, the above embodiment has the following beneficial effects: since the load state and the combustion state are two factors that simultaneously affect the composition of the final exhaust emissions, when a certain load state and combustion state are fixed, there is a fixed corresponding exhaust component in the exhaust composition data that accounts for a larger proportion. Therefore, the entropy weight method is used to assign component weights based on the exhaust samples under the load and combustion states to obtain a second weight. The second weight reflects the difference in the impact of each emission under different load states and combustion states on the total emissions, and the actual Q value is determined based on the second weight. The actual Q value is used as a supervisory signal in the subsequent deep neural network fine-tuning process, thereby improving the deep neural network's ability to adapt to real environmental changes.

[0016] Furthermore, determining second weights corresponding to various exhaust gas components by an entropy weight method according to the first load state and the first combustion state includes: Acquire a plurality of fourth exhaust gas component data when the main engine is in the first load state and the first combustion state, and for each exhaust gas component, add the corresponding value of the fourth exhaust gas component data to obtain a first cumulative sum corresponding to each exhaust gas component; Calculating a ratio of a quantity value of each exhaust gas component in each item of the fourth exhaust gas component data to the corresponding first cumulative sum, and obtaining a second product of each of the ratios and a logarithm of the ratio; After negating each of the second products, for each exhaust gas component, accumulating all corresponding second products to obtain a second cumulative sum corresponding to each exhaust gas component; For each exhaust gas component, a second difference between the corresponding second accumulated sum and a preset threshold is calculated, and the normalized second difference is used as a second weight corresponding to the exhaust gas component.

[0017] Compared with the prior art, the above embodiment has the following beneficial effects: for each exhaust gas component, the ratio of its value in the fourth exhaust gas component data and the corresponding first cumulative sum is calculated to determine its probability of occurrence, and the ratio and the corresponding logarithm of the ratio are further multiplied and negative and then added to obtain a second cumulative sum, thereby evaluating the discreteness and uncertainty of the distribution of each exhaust gas component in the sample. The larger the second cumulative sum, the smaller the fluctuation of the corresponding exhaust gas component under the same load state and combustion state, which means that the correlation between the exhaust gas component and the load state and the combustion state is smaller, and vice versa, the greater the correlation between the exhaust gas component and the load state and the combustion state is. Therefore, in order to ensure the rationality of the weight, the second difference between the second cumulative sum and the preset threshold is taken, so that the second difference is positively correlated with the above correlation, thereby improving the accuracy of the subsequent actual Q value estimation.

[0018] Furthermore, fine-tuning the parameters of the deep neural network model using a supervised learning method based on the updated experience pool includes: For each training sample in the experience pool, obtaining a second predicted Q value for each main engine output power decision based on its corresponding input data and the deep neural network model, and determining a second main engine output power decision corresponding to an optimal second predicted Q value based on each second predicted Q value; Filtering out a first training sample from all the training samples, in which the output power decision of the second main engine is consistent with the output power decision of the first main engine; For each of the first training samples, calculating a third difference between the corresponding actual Q value and the optimal second predicted Q value; The mean of the sum of squares of the third differences is used as a loss value, and the parameters of the deep neural network model are adjusted through back propagation according to the loss value.

[0019] Compared with the prior art, the above-mentioned embodiment has the following beneficial effects: since the deep neural network model is in a state of continuous iteration, the first main engine output power decision recorded in the experience pool may be the result output by the deep neural network model before being adjusted for many times, the second main engine output power decision output by the latest deep neural network model may have changed, at this time, the actual Q value corresponding to the second main engine output power decision is also different from that recorded in the experience pool, so it is necessary to screen the samples in the experience pool to avoid adjustment errors; the deep neural network model is continuously learned and updated online through the real sample data actually collected, and the self-adaptive ability of the deep neural network to real environment changes is improved.

[0020] Further, the obtaining of the propulsion power demand at the next time step according to all the propulsion power data by the pre-trained long short-term memory network model comprises: collecting first sea state data at the plurality of historical time steps and second sea state data at the next time step, and splicing the first sea state data and the propulsion power data according to time steps to obtain a two-dimensional time sequence feature tensor, and splicing the second sea state data to the tail of the two-dimensional time sequence feature tensor; extracting propulsion power trend change features from the two-dimensional time sequence feature tensor by a one-dimensional convolutional neural network; inputting the propulsion power trend change features into the long short-term memory network model to obtain the propulsion power demand at the next time step.

[0021] Compared with the prior art, the above-mentioned embodiment has the following beneficial effects: the time sequence tensor is constructed by the sea state data and the propulsion power data, the propulsion trend of the ship affected by external factors is more comprehensively reflected, the trend features are further extracted by one-dimensional convolution, and the long short-term memory network model is used to predict time sequence information, the accuracy and dynamic response capability of the propulsion power prediction are enhanced, thereby providing prospective input for subsequent power scheduling, responding in advance to the difference in power demand caused by different sea state changes, and ensuring that the finally executed power distribution operation can meet the power response demand of the ship at the next time step.

[0022] Another embodiment of the present application also provides a hybrid power output control device based on emission optimization, comprising: a propulsion power demand prediction module, a combustion state classification module, a feature fusion module, a Q value prediction module, and a power distribution module; The propulsion power demand prediction module is configured to collect propulsion power data of a ship at a plurality of historical time steps, and obtain a propulsion power demand at a next time step according to all the propulsion power data by a pre-trained long short-term memory network model. The combustion state classification module is configured to collect first exhaust component data of exhaust gas emitted by the ship at a current time step, and determine a first combustion state of the main engine at the current time step based on the first exhaust component data by using a pre-trained support vector machine. The feature fusion module is configured to collect a first load state of the main engine at the current time step, and obtain a multi-modal feature vector by weightedly fusing the first combustion state and the first load state through an attention mechanism module. The Q value prediction module is configured to obtain a pre-trained deep neural network model, and obtain a first predicted Q value of each main engine output power decision in a preset discrete action space based on the multi-modal feature vector by using the deep neural network model; the deep neural network is obtained by constructing a reward function with the objective of minimizing exhaust emission, and training the reward function based on a deep reinforcement learning algorithm. The power distribution module is configured to obtain a first main engine output power decision corresponding to an optimal first predicted Q value, and control the hybrid power output system to perform a power distribution operation based on the propulsion power demand and the first main engine output power decision.

[0023] Further, the combustion state classification module includes a support vector machine selection unit and a support vector machine calling unit; the combustion state classification module is configured to determine a first combustion state of the main engine at the current time step based on the first exhaust component data by using a pre-trained support vector machine, and includes: The support vector machine selection unit is configured to select a corresponding first support vector machine based on the first load state; the first support vector machine is obtained by training second exhaust component data of exhaust gas emitted by the ship when the main engine is in different combustion states under the first load state; The support vector machine calling unit is configured to determine the first combustion state of the main engine under the first load state at the current time step based on the first exhaust component data by using the first support vector machine. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0025] Figure 1 A flowchart of a hybrid power output control method based on emission optimization provided in some embodiments of the present application is shown in the figure. Figure 2This is a schematic structural diagram of a hybrid power output control device based on emission optimization provided in some embodiments of the present application. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0028] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0029] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0030] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0031] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0032] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0033] Because traditional power systems relying on internal combustion engines have high emissions and high energy consumption, hybrid propulsion systems have gradually become the preferred choice for ship propulsion systems due to their advantages such as lower energy consumption. Existing hybrid output control methods typically focus on power demand or energy efficiency as the primary optimization target, ignoring the real-time changes in exhaust emissions. This can lead to severe emission fluctuations or excessive emissions due to irrational power allocation under certain operating conditions. In addition, the combustion state of the main engine is affected by various factors such as load changes and fluctuating sea conditions. Traditional control strategies struggle to ensure both power response and emission control.

[0034] Example 1 Please refer to Figure 1 To solve the problem in the prior art of ensuring power response while avoiding excessive exhaust emissions due to unreasonable power distribution of the hybrid system, the present invention provides a hybrid output control method based on emission optimization, including S101 to S105, specifically: S101: Propulsion power data of a ship at several historical time steps are collected, and the propulsion power requirement for the next time step is obtained based on all the propulsion power data through a pre-trained long short-term memory network model.

[0035] Furthermore, in some embodiments of the present application, the pre-trained long short-term memory network model obtains the propulsion power requirement for the next time step based on all the propulsion power data, including: collecting first sea state data of the plurality of historical time steps and second sea state data of the next time step, splicing the first sea state data with the propulsion power data according to the time step to obtain a two-dimensional time series feature tensor, and splicing the second sea state data to the tail of the two-dimensional time series feature tensor; extracting propulsion power trend change features from the two-dimensional time series feature tensor through a one-dimensional convolutional neural network; The propulsion power trend change characteristics are input into the long short-term memory network model to obtain the propulsion power demand for the next time step.

[0036] Preferably, in some embodiments of the present application, the propulsion power data of the several historical time steps include at least: the sum of the output power of the main engine and the energy storage at the current time and several historical time steps; the sea condition data include at least: the offshore wind speed vector, wave height and wave flow velocity vector at the corresponding time step. It should be noted that the first sea condition data is the actual collected sea condition data, and the second sea condition data is the predicted sea condition data obtained through the meteorological platform.

[0037] Preferably, in some embodiments of the present application, the first sea state data and the propulsion power data are spliced ​​according to the time step to obtain a two-dimensional time series feature tensor, and the second sea state data is spliced ​​to the tail of the two-dimensional time series feature tensor, including: assuming that the shape of the first sea state data and the second sea state data is (T, D1), where T is the time step, and D1 is the characteristic dimension of the sea state data, such as the offshore wind speed vector, wave height, and wave current vector; assuming that the shape of the propulsion power data is (T, D2), D2 is the characteristic dimension of the propulsion power data, such as the sum of the output power of the main engine and the energy storage; then the first sea state data and the propulsion power data are spliced ​​according to the time step to obtain a two-dimensional time series feature tensor with a shape of (T, D1+D2); further, the second sea state data is spliced ​​to the tail of the two-dimensional time series feature tensor to obtain a two-dimensional time series feature tensor with a shape of (T+1, D1+D2), where the propulsion power data corresponding to the second sea state data is replaced by a placeholder.

[0038] Preferably, in some embodiments of the present application, the propulsion power trend change characteristics are extracted from the two-dimensional time series feature tensor through a one-dimensional convolutional neural network, including: performing sliding extraction in the time dimension, setting the convolution kernel size to (k, D1+D2), where k is the width of the sliding window, which can be set according to actual conditions, and this application does not limit this value.

[0039] Preferably, in some embodiments of the present application, the propulsion power trend change characteristics are input into the long short-term memory network model to obtain the propulsion power demand for the next time step, including: wherein the long short-term memory network model is used to capture the long-term dependence and dynamic pattern of the propulsion power in the time series. When the propulsion power trend change characteristics are input into the long short-term memory network model, after forward propagation, the long short-term memory network model outputs the propulsion power data corresponding to the second sea condition data, that is, the propulsion power demand value corresponding to the next time step.

[0040] It can be seen from the above embodiments that the application constructs a time sequence tensor by sea state data and propulsion power data, more comprehensively reflects the propulsion trend of the ship affected by external factors, further extracts trend characteristics by one-dimensional convolution and predicts time sequence information by a short-term memory network model, enhances the accuracy and dynamic response capability of the propulsion power prediction, thereby providing prospective input for subsequent power scheduling, in advance responding to the difference in power demand caused by different sea state changes, and ensuring that the finally executed power distribution operation can meet the power response demand of the ship at the next time step.

[0041] S102: Collect first exhaust component data of exhaust gas emitted by the ship at the current time step, and determine a first combustion state of the main engine at the current time step by a pre-trained support vector machine according to the first exhaust component data.

[0042] Further, in some embodiments of the application, the determination of the first combustion state of the main engine at the current time step by the pre-trained support vector machine according to the first exhaust component data comprises: selecting a corresponding first support vector machine according to the first load state, wherein the first support vector machine is obtained by training according to second exhaust component data of exhaust gas emitted by the ship when the main engine is in different combustion states under the first load state; determining the first combustion state of the main engine under the first load state at the current time step according to the first exhaust component data based on the first support vector machine.

[0043] Preferably, in some embodiments of the application, the combustion state comprises: a complete combustion state, in which the fuel is fully oxidized, the concentration of carbon oxides and carbon hydride compounds in the exhaust emission data is low, the concentration of nitrogen oxides is moderate, and the smoke degree is low; an incomplete combustion state, in which the concentration of carbon oxides and carbon hydride compounds is increased, which may be accompanied by an increase in smoke degree, and the concentration of nitrogen oxides is low; a pre-ignition / early combustion state, in which the combustion is advanced, the peak pressure is advanced, which may cause knocking or knocking phenomenon, and the concentration of nitrogen oxides and smoke degree is increased simultaneously; a cold combustion state, in which the combustion is incomplete, the exhaust temperature is low, and the concentration of carbon hydride compounds in the exhaust is increased; and a delayed combustion state, in which the combustion is delayed, and the concentration of carbon hydride compounds in the exhaust is increased.

[0044] Preferably, in some embodiments of the present application, the load state is a current operating condition parameter set of the main engine, including but not limited to main engine output power, speed, torque, fuel flow, intake pressure, exhaust temperature and other sensor collected data. Further, when selecting the corresponding first support vector machine according to the first load state, first set the classification interval of different load states (such as defining the interval of each operating condition parameter when the main engine is respectively in low, medium and high load), according to the actual operating condition parameter in the first load state, combining the above classification interval, determine the classification of the first load state of the current main engine, and further determine the corresponding support vector machine. When training the support vector machine, ensure that the second exhaust component data used for current training is the exhaust component data generated when the main engine is in the corresponding first load state.

[0045] Preferably, in some embodiments of the present application, the exhaust component data includes but is not limited to: carbon oxide, hydrocarbon, nitrogen oxide and smoke content data.

[0046] As can be seen from the above embodiments, since different load states and combustion states are two independent factors that affect the composition of the final exhaust emission, the corresponding support vector machine model is first selected based on the first load state of the main engine that can be actually observed, so as to remove the interference of the load state in the combustion state judgment process; further, when the load state of the main engine is fixed, the composition of the exhaust under different combustion states has obvious difference, so that the support vector machine can accurately reflect the combustion state of the current main engine, and provide reliable input basis for subsequent power distribution optimization.

[0047] S103: Collect the first load state of the main engine at the current time step, and obtain a multi-modal feature vector by weighting and fusing the first combustion state and the first load state through an attention mechanism module.

[0048] Further, in some embodiments of the present application, the collection of the first load state of the main engine at the current time step and the obtaining of the multi-modal feature vector by weighting and fusing the first combustion state and the first load state through the attention mechanism module include: embedding and encoding the first combustion state through a preset embedding layer to obtain a first vector; linearly transforming and activating the first load state through a preset feedforward neural network to obtain a second vector with the same dimension as the first vector; determining the first weight corresponding to the first vector and the second vector respectively according to the correlation of the first combustion state and the first load state with the first exhaust component data; The first vector and the second vector are added after being multiplied by the corresponding first weight to obtain the multi-modal feature vector.

[0049] Preferably, in some embodiments of the present application, the first vector is obtained by embedding and encoding the first combustion state through a preset embedding layer, comprising: taking the first combustion state identified by the support vector machine at the current time step as a discrete label (for example, for the complete combustion state, the corresponding discrete label is defined as the number "2"); inputting the discrete label into the preset embedding layer to obtain a first vector of a preset dimension.

[0050] Preferably, in some embodiments of the present application, the second vector with the same dimension as the first vector is obtained by linear transformation and activation operation of the first load state through a preset feedforward neural network, comprising: inputting the collected first load state into a preset feedforward neural network, which includes a fully connected layer and an activation layer, and the activation layer adopts a ReLU activation function, outputting a second vector with the same dimension as the first vector.

[0051] Preferably, in some embodiments of the present application, the first weight corresponding to the first vector and the second vector is determined according to the correlation of the first combustion state and the first load state with the first tail gas component data, comprising: constructing a fully connected attention network, connecting the fully connected attention network with a subsequent deep neural network, training the fully connected attention network while training the deep neural network, so that the fully connected attention network can automatically learn how to allocate the first weight, and improve the accuracy of the data input into the subsequent deep neural network. Further, the first vector and the second vector are added after being multiplied by the corresponding first weight to obtain the multi-modal feature vector, which is realized by constructing a fully connected layer, the input of which includes the first weight, the first vector and the second vector output by the fully connected attention network, and the weighted calculation is realized.

[0052] As can be seen from the above embodiments, since the first combustion state is discrete data, embedding coding needs to be performed, so that it is no longer a single integer, but a dense representation that can participate in calculation. At the same time, since the load state is real device running data that can be actually collected, it is continuous data, so through linear change and activation, the multi-dimensional continuous signal is compressed or expanded to the same dimension as the embedding vector, ensuring the feasibility of subsequent weighted fusion. In addition, since the load state and the combustion state are two key factors that affect the composition of the final exhaust emission, the first combustion state and the first load state are respectively extracted for feature extraction to obtain the first vector and the second vector, and the first vector and the second vector are weighted and fused through the attention mechanism to obtain the multi-modal data, which more truly reflects the current running state of the main engine, thereby improving the input quality and learning ability of the deep neural network.

[0053] S104: Obtain a pre-trained deep neural network model, and obtain a first prediction Q value of each main engine output power decision in a preset discrete action space based on the multi-modal feature vector according to the deep neural network model; wherein the deep neural network is obtained by constructing a reward function with the target of minimizing exhaust emission, and training using a deep reinforcement learning algorithm based on the reward function.

[0054] Preferably, in some embodiments of the present application, the deep neural network is a multi-layer perceptron structure (Multi-Layer Perceptron, MLP), the input is the fused multi-modal feature vector at the current time step (obtained by fusing the combustion state and the load state through the attention mechanism), and the output is a group of Q value vectors representing the action estimated value corresponding to each main engine power output decision in the discrete action space.

[0055] Preferably, in some embodiments of the present application, the reward function is obtained by setting a corresponding weight for each exhaust component and performing weighted calculation, and combining a simulation model. The input data of the simulation model is the current load state, combustion state and executed main engine output power decision of the main engine, and the output is exhaust component data. The modeling process and method of the simulation model are not limited in the present application.

[0056] Preferably, in some embodiments of the present application, the step of calculating the reward value for an input sample by the reward function comprises: inputting the load state, the combustion state and the executed main engine output power decision in the input sample into the simulation model to obtain the exhaust composition data; determining the second weight corresponding to each exhaust composition in the exhaust composition data by the entropy weight method; determining the maximum and minimum values of each exhaust composition under different load states and combustion states, and performing normalization processing on each exhaust composition in the exhaust composition data using the maximum and minimum values; and finally calculating the weighted sum according to the second weight and the normalization processing result to obtain the reward value. For example, the load state F, the combustion state B and the main engine output power decision P of a sample are known, and the exhaust composition data [carbon oxides: X1, hydrocarbons: X2, nitrogen oxides: X3, smoke: X4] can be obtained by the simulation model. Next, the maximum and minimum values of each exhaust composition when the main engine is in the load state F and the combustion state B are obtained. Assuming that the maximum value of carbon oxides is max, the minimum value is min, and the actual content of carbon oxides is X1, the normalization value is determined by (max-X1) / (max-min). After obtaining the normalization value of each exhaust composition, the normalization value is multiplied by the corresponding second weight and then added to obtain the reward value.

[0057] S105: Obtain the first main engine output power decision corresponding to the optimal first predicted Q value, and control the hybrid power output system to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision.

[0058] Further, in some embodiments of the present application, the step of controlling the hybrid power output system to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision comprises: controlling the output power of the main engine according to the first main engine output power decision, and controlling the output power of the energy storage according to the difference between the propulsion power demand and the first main engine output power decision; After the ship performs the power distribution operation, third exhaust composition data of exhaust gas emitted by the ship at the next time step is collected, and the experience pool is updated according to the difference between the first exhaust composition data and the third exhaust composition data; According to the updated experience pool, the parameters of the deep neural network model are fine-tuned by a supervised learning method.

[0059] Preferably, in some embodiments of the present application, the step of controlling the output power of the main engine according to the first main engine output power decision and controlling the output power of the energy storage device according to the difference between the propulsion power demand and the first main engine output power decision comprises: assuming that the first main engine output power decision is P1 and the propulsion power demand is P2, the output power of the energy storage device to be configured is P2-P1.

[0060] Preferably, in some embodiments of the present application, the data structure of the sample data in the experience pool is [input data: [load state, combustion state], label data: [actual Q value, first main engine output power decision]]. For the setting of the initial experience pool, the present application does not make any limitation, which can be set to be empty or can be constructed and obtained by using the data in the training data set used when training the deep neural network.

[0061] As can be seen from the above embodiments, when the first main engine output power decision is known, the propulsion power gap of the current propulsion power demand can be determined based on the first main engine output power decision, which is the output power that needs to be allocated to the energy storage device, so as to ensure that the output power of the main engine and the energy storage device can meet the power response demand of the ship at the next time step; further, when the hybrid power system executes the corresponding output power decision, the operating state of the main engine will change, so that the exhaust emission composition data also changes, at this time the third exhaust composition data obtained is the real sample data, the experience pool is updated according to the real sample data, and the neural network parameters are continuously fine-tuned combined with supervised learning, so as to enhance the self-adaptive ability of the deep neural network model to the real environment change, and improve the robustness and generalization ability of the deep neural network model under actual working conditions.

[0062] Further, in some embodiments of the present application, the step of updating the experience pool according to the difference between the first exhaust composition data and the third exhaust composition data comprises: determining a second weight corresponding to each exhaust composition by entropy weight method to obtain a weight parameter combination according to the first load state and the first combustion state; wherein one weight parameter combination corresponds to one combination of load state and combustion state; for each exhaust composition, after calculating the first difference between the amount value in the first exhaust composition data and the amount value in the third exhaust composition data, accumulating the first product of the first difference and the second weight corresponding to each exhaust composition to obtain a weighted sum; the weighted sum is taken as the actual Q value, the first load state and the first combustion state are taken as the input data, a training sample is constructed according to the actual Q value, the input data and the first main engine output power decision, and the training sample is added to the experience pool.

[0063] Preferably, in some embodiments of the present application, the step of calculating the reward value by the reward function for an input sample can also be: inputting the load state, the combustion state and the executed main engine output power decision in the input sample into the simulation model to obtain the exhaust component data; determining the second weight corresponding to each exhaust component in the exhaust component data by the entropy weight method; for each exhaust component, calculating the first difference value between the amount value thereof in the first exhaust component data and the amount value thereof in the third exhaust component data, then accumulating the first product of the first difference value of each exhaust component and the second weight to obtain a weighted sum, and taking the weighted sum as the reward value.

[0064] As can be seen from the above embodiments, since the load state and the combustion state are two factors that simultaneously affect the composition of the final exhaust emission, there is a certain exhaust component that accounts for more in the exhaust component data when a certain load state and combustion state are fixed. Therefore, the entropy weight method is used to assign component weights based on the exhaust samples under the load and combustion state to obtain the second weight, which reflects the difference in the influence of each emission on the total emission under different load states and combustion states through the second weight, and determines the actual Q value according to the second weight, which is used as a supervision signal in the subsequent deep neural network fine-tuning process, thereby improving the self-adaptive ability of the deep neural network to real environment changes.

[0065] Further, in some embodiments of the present application, the step of determining the second weight corresponding to each exhaust component by the entropy weight method according to the first load state and the first combustion state comprises: obtaining a plurality of fourth exhaust component data when the main engine is in the first load state and the first combustion state, and for each exhaust component, adding the amount value corresponding to each of the fourth exhaust component data to obtain a first cumulative sum corresponding to each exhaust component; calculating the ratio of the amount value of each exhaust component in each of the fourth exhaust component data to the first cumulative sum to obtain a second product of each of the ratios and the logarithm of the ratio; taking the negative of each of the second products, and for each exhaust component, accumulating all of the second products corresponding thereto to obtain a second cumulative sum corresponding to each exhaust component; for each exhaust component, calculating a second difference value between the second cumulative sum corresponding thereto and a preset threshold, and taking the normalized second difference value as the second weight corresponding to the exhaust component.

[0066] Exemplarily, the determining, according to the first load state and the first combustion state, the second weight corresponding to each exhaust component by the entropy weight method comprises: summing up the corresponding value of each exhaust component in each fourth exhaust component data, assuming that the first cumulative sum of each exhaust component is [carbon oxides: Sum1_1, hydrocarbons: Sum1_2, nitrogen oxides: Sum1_3, smoke: Sum1_4], then dividing the corresponding value of each exhaust component in each fourth exhaust component data by the above-mentioned corresponding first cumulative sum, to obtain the ratio data corresponding to each fourth exhaust component data [carbon oxides: X1 / Sum1_1, hydrocarbons: X2 / Sum1_2, nitrogen oxides: X3 / Sum1_3, smoke: X4 / Sum1_4], then multiplying the ratio corresponding to each exhaust component data by the logarithm corresponding to the ratio to obtain the second product [carbon oxides: IE1, hydrocarbons: IE2, nitrogen oxides: IE3, smoke: IE4], where the logarithm can be in the form of logarithm with base 2, since the ratio is a value less than 1, the value will be a negative value after taking the logarithm, so negative processing is required, and then the second cumulative sum [carbon oxides: Sum2_1, hydrocarbons: Sum2_2, nitrogen oxides: Sum2_3, smoke: Sum2_4] is obtained by accumulating each exhaust component in the second product corresponding to each fourth exhaust component data, then a preset threshold value is set according to the number of fourth exhaust component data involved in the current calculation (for example, assuming that the logarithm of the ratio is taken in the form of logarithm with base 2, and the number of fourth exhaust component data is , then the preset threshold value is ), and the second weight of each exhaust component is obtained by normalizing after subtracting the second cumulative sum corresponding to each exhaust component from the preset threshold value (assuming that the preset threshold value is , then the second weight of the above-mentioned carbon oxides is: (Sum2_1-Sum2_1) / Sum2, where Sum2=Sum2_1+Sum2_2+Sum2_3+Sum2_4).

[0067] ​As can be seen from the above embodiments, for each tail gas component, the application determines its occurrence probability by calculating the ratio of its value in the fourth tail gas component data and the corresponding first cumulative sum, further multiplies the ratio and the logarithm corresponding to the ratio to obtain the second cumulative sum after taking the negative, thereby evaluating the discrete degree and uncertainty of the distribution of each tail gas component in the sample, the larger the second cumulative sum, the smaller the fluctuation of the corresponding tail gas component under the same load state and combustion state, that is, the smaller the correlation of the tail gas component with the load state and the combustion state, and vice versa, therefore, in order to ensure the rationality of the weight, the second difference between the second cumulative sum and the preset threshold is taken, so that the second difference is positively correlated with the above-mentioned correlation, and the accuracy of subsequent actual Q value estimation is improved.

[0068] Further, in some embodiments of the application, the parameters of the deep neural network model are fine-tuned according to the updated experience pool through a supervised learning method, including: For each training sample in the experience pool, a second predicted Q value of each main engine output power decision is obtained according to its corresponding input data and the deep neural network model, and a second main engine output power decision corresponding to the optimal second predicted Q value is determined according to each second predicted Q value; The first training sample in which the second main engine output power decision is consistent with the first main engine output power decision is filtered out from all the training samples; For each first training sample, a third difference between its corresponding actual Q value and optimal second predicted Q value is calculated; The square sum of the third difference is taken as a loss value, and the parameters of the deep neural network model are adjusted through back propagation according to the loss value.

[0069] Preferably, in some embodiments of the application, the first training sample in which the second main engine output power decision is consistent with the first main engine output power decision is filtered out from all the training samples, including: for each training sample [input data: [load state, combustion state], label data: [actual Q value, first main engine output power decision]] in the experience pool, first input the multi-modal feature vector corresponding to the input data into the deep neural network to obtain the second main engine output power decision of the optimal second predicted Q value output by the current deep neural network, if the second main engine output power decision is different from the first main engine output power decision, it means that this training sample is not suitable for fine-tuning of the deep neural network, all samples with the same second main engine output power decision and first main engine output power decision are filtered out for subsequent fine-tuning of the deep neural network, and adjustment error is avoided.

[0070] The purpose of using a deep neural network model in this application is to enable the deep neural network model to accurately predict the quality of each main engine output power decision in the decision space, and the predicted Q value output by the deep neural network is the data for judging the quality of the main engine output power decision. Therefore, the third difference between the actual Q value and the optimal second predicted Q value is used as the loss value. The parameters of the deep neural network model are adjusted through back propagation based on this loss value to improve the accuracy of the deep neural network model.

[0071] It can be seen from the above embodiments that since the deep neural network model is in a state of continuous iteration, the first main engine output power decision recorded in the training samples in the experience pool may be the result output by the deep neural network model before multiple adjustments. The second main engine output power decision output by the latest deep neural network model may have changed. At this time, the actual Q value corresponding to the second main engine output power decision is also different from that recorded in the experience pool. Therefore, it is necessary to screen the samples in the experience pool to avoid adjustment errors; the deep neural network model is continuously learned and optimized and updated online through the real sample data actually collected, so as to improve the adaptability of the deep neural network to real environmental changes.

[0072] In summary, it can be seen that compared with the prior art, the hybrid power output control method based on emission optimization provided by the embodiment of the present application has the following beneficial effects: Since the exhaust composition of the ship is related to the combustion state and load state of the main engine, the combustion state of the current main engine is determined based on the support vector machine. First, the actual operating state of the current main engine is determined to provide accurate data input for the subsequent deep neural network model. At the same time, since the power allocation decision of the current time step will affect the load state and combustion state of the main engine in the next time step, and the load state and combustion state of the main engine in the next time step determine the exhaust composition, the exhaust emissions can be optimized by optimizing the power allocation decision. This application adopts a deep reinforcement learning algorithm to train a deep neural network model with a reward function that minimizes exhaust emissions. The predicted Q value of each main engine output power decision is evaluated through the deep neural network model, so that the final output power decision of the first main engine can reduce exhaust emissions to the greatest extent. Finally, by predicting the propulsion power demand of the ship in the next time step in advance and combining the propulsion power demand with the output power decision of the first main engine, the output power decision of the energy storage can be determined, thereby ensuring that the power allocation operation finally executed can meet the power response demand of the ship in the next time step.

[0073] Example 2 like Figure 2As shown, on the basis of the above method embodiment, an embodiment of the present application provides a hybrid output control device based on emission optimization, comprising: a propulsion power demand prediction module 201, a combustion state classification module 202, a feature fusion module 203, a Q value prediction module 204, and a power distribution module 205.

[0074] Further, in some embodiments of the present application, the propulsion power demand prediction module 201 is configured to collect propulsion power data of a plurality of historical time steps of the ship, and obtain propulsion power demand of a next time step based on all the propulsion power data through a pre-trained long short-term memory network model; the combustion state classification module 202 is configured to collect first exhaust component data of exhaust gas of the ship at a current time step, and determine a first combustion state of the main engine at the current time step based on the first exhaust component data through a pre-trained support vector machine; the feature fusion module 203 is configured to collect a first load state of the main engine at the current time step, and obtain a multi-modal feature vector by weighted fusion of the first combustion state and the first load state through an attention mechanism module; the Q value prediction module 204 is configured to obtain a pre-trained deep neural network model, and obtain a first predicted Q value of a main engine output power decision in a preset discrete action space based on the multi-modal feature vector through the deep neural network model; wherein the deep neural network is obtained by constructing a reward function with the objective of minimizing exhaust emission, and is trained based on the reward function using a deep reinforcement learning algorithm; and the power distribution module 205 is configured to obtain a first main engine output power decision of an optimal first predicted Q value, and control the hybrid power output system to perform a power distribution operation based on the propulsion power demand and the first main engine output power decision.

[0075] Further, in some embodiments of the present application, the combustion state classification module 202 comprises a support vector machine selection unit and a support vector machine calling unit; the combustion state classification module 202 is configured to determine a first combustion state of the main engine at a current time step based on the first exhaust component data through a pre-trained support vector machine, comprising: the support vector machine selection unit is configured to select a corresponding first support vector machine based on the first load state; wherein the first support vector machine is trained based on second exhaust component data of exhaust gas of the ship when the main engine is in different combustion states under the first load state; and the support vector machine calling unit is configured to determine the first combustion state of the main engine at the current time step under the first load state based on the first support vector machine and the first exhaust component data.

[0076] Further, in some embodiments of the present application, the feature fusion module 203 comprises a first vector extraction unit, a second vector extraction unit, a first weight determination unit, and a feature fusion unit; the feature fusion module 203 is configured to collect a first load state of the main engine at a current time step, and obtain a multi-modal feature vector by weighting and fusing the first combustion state and the first load state through an attention mechanism module, comprising: the first vector extraction unit is configured to embed and encode the first combustion state through a preset embedding layer to obtain a first vector; the second vector extraction unit is configured to linearly transform and activate the first load state through a preset feedforward network to obtain a second vector with the same dimension as the first vector; the first weight determination unit is configured to determine a first weight corresponding to each of the first vector and the second vector according to the correlation of the first combustion state and the first load state with the first exhaust component data; and the feature fusion unit is configured to multiply and add the first vector and the second vector and their respective first weights to obtain the multi-modal feature vector.

[0077] Further, in some embodiments of the present application, the power distribution module 205 comprises an output power control unit, an experience pool updating unit, and a parameter fine-tuning unit; the power distribution module 205 is configured to control the hybrid power output system to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision, comprising: the output power control unit is configured to control the output power of the main engine according to the first main engine output power decision, and control the output power of the energy storage device according to the difference between the propulsion power demand and the first main engine output power decision; the experience pool updating unit is configured to collect third exhaust component data of exhaust gas of the ship at a next time step after the ship performs the power distribution operation, and update an experience pool according to the difference between the first exhaust component data and the third exhaust component data; and the parameter fine-tuning unit is configured to fine-tune the parameters of the deep neural network model through a supervised learning method according to the updated experience pool.

[0078] Furthermore, in some embodiments of the present application, the experience pool updating unit is used to update the experience pool according to the difference between the first exhaust gas component data and the third exhaust gas component data, including: determining the second weights corresponding to various exhaust gas components by the entropy weight method according to the first load state and the first combustion state, and obtaining a weight parameter combination; wherein, a weight parameter combination corresponds to a combination of a load state and a combustion state; for each exhaust gas component, calculating the first difference between its value in the first exhaust gas component data and its value in the third exhaust gas component data, and then accumulating the first product of the first difference corresponding to various exhaust gas components and the second weight to obtain a weighted sum; taking the weighted sum as the actual Q value, taking the first load state and the first combustion state as input data, constructing a training sample according to the actual Q value, the input data and the first main engine output power decision, and adding the training sample to the experience pool.

[0079] Furthermore, in some embodiments of the present application, the second weights corresponding to various exhaust gas components are determined by the entropy weight method based on the first load state and the first combustion state, including: obtaining a number of fourth exhaust gas component data when the main engine is in the first load state and the first combustion state, and for each exhaust gas component, adding the corresponding values ​​in each of the fourth exhaust gas component data to obtain the first cumulative sum corresponding to the various exhaust gas components; calculating the ratio of the value of each exhaust gas component in each of the fourth exhaust gas component data to the corresponding first cumulative sum, and obtaining the second product of each ratio and the logarithm of the ratio; after negating each of the second products, for each exhaust component, adding all the corresponding second products to obtain the second cumulative sum corresponding to each exhaust component; for each exhaust component, calculating the second difference between the corresponding second cumulative sum and the preset threshold, and using the normalized second difference as the second weight corresponding to the exhaust component.

[0080] Further, in some embodiments of the present application, the parameter fine-tuning unit is configured to fine-tune the parameters of the deep neural network model according to the updated experience pool by a supervised learning method, including: for each training sample in the experience pool, obtaining a second predicted Q value of each main engine output power decision according to its corresponding input data and the deep neural network model, and determining a second main engine output power decision corresponding to an optimal second predicted Q value according to each second predicted Q value; screening first training samples in which the second main engine output power decision is consistent with the first main engine output power decision from all the training samples; for each first training sample, calculating a third difference value between the actual Q value corresponding to the first training sample and the optimal second predicted Q value; taking the square sum of the third difference values as a loss value, and adjusting the parameters of the deep neural network model by back propagation according to the loss value.

[0081] Further, in some embodiments of the present application, the propulsion power demand prediction module 201 includes a two-dimensional time series feature tensor acquisition unit, a propulsion power trend change feature extraction unit, and a propulsion power demand prediction unit. The propulsion power demand prediction module 201 is configured to obtain the propulsion power demand at the next time step according to all the propulsion power data by using a pre-trained long short-term memory network model, including: the two-dimensional time series feature tensor acquisition unit is configured to collect the first sea state data at the plurality of historical time steps and the second sea state data at the next time step, and splice the first sea state data and the propulsion power data according to time steps to obtain a two-dimensional time series feature tensor, and splice the second sea state data to the tail of the two-dimensional time series feature tensor; the propulsion power trend change feature extraction unit is configured to extract propulsion power trend change features from the two-dimensional time series feature tensor by using a one-dimensional convolutional neural network; and the propulsion power demand prediction unit is configured to input the propulsion power trend change features into the long short-term memory network model to obtain the propulsion power demand at the next time step.

[0082] It can be understood that the above-mentioned device embodiment is corresponding to the method embodiment of the present application, and can realize the hybrid power output control method based on emission optimization provided by any one of the above-mentioned method embodiments of the present application.

[0083] As can be seen from the above, compared with the prior art, the hybrid power output control device based on emission optimization provided by the embodiments of the present application has the following beneficial effects: Since the exhaust composition of the ship is related to the combustion state and load state of the main engine, the combustion state of the current main engine is determined based on the support vector machine. First, the actual operating state of the current main engine is determined to provide accurate data input for the subsequent deep neural network model. At the same time, since the power allocation decision of the current time step will affect the load state and combustion state of the main engine in the next time step, and the load state and combustion state of the main engine in the next time step determine the exhaust composition, the exhaust emissions can be optimized by optimizing the power allocation decision. This application adopts a deep reinforcement learning algorithm to train a deep neural network model with a reward function that minimizes exhaust emissions. The predicted Q value of each main engine output power decision is evaluated through the deep neural network model, so that the final output power decision of the first main engine can reduce exhaust emissions to the greatest extent. Finally, by predicting the propulsion power demand of the ship in the next time step in advance and combining the propulsion power demand with the output power decision of the first main engine, the output power decision of the energy storage can be determined, thereby ensuring that the power allocation operation finally executed can meet the power response demand of the ship in the next time step.

[0084] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided herein, the connection relationship between the modules indicates that there is a communication connection between them, which may be implemented as one or more communication buses or signal lines. Those skilled in the art may understand and implement the present invention without inventive effort.

[0085] Based on the above-mentioned embodiment of the hybrid power output control method based on emission optimization, another embodiment of the present application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the hybrid power output control method based on emission optimization of any embodiment of the present application is implemented.

[0086] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more module elements may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0087] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0088] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the terminal device, and connects all parts of the terminal device through various interfaces and lines.

[0089] On the basis of the above-mentioned method embodiment, another embodiment of the present application provides a computer readable storage medium, including a stored computer program, wherein when the computer program runs, the device where the computer readable storage medium is located executes the discharge optimization-based hybrid output control method provided in any one of the above-mentioned method embodiments of the present application.

[0090] The modules / units integrated in the apparatus / terminal device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

Claims

1. A hybrid power output control method based on emission optimization, characterized in that: include: Collect propulsion power data of several historical time steps of the ship, and use the pre-trained long short-term memory network model to obtain the propulsion power requirement for the next time step based on all the propulsion power data; collecting first exhaust gas component data of exhaust gas emitted by the ship at the current time step, and determining a first combustion state of the main engine at the current time step through a pre-trained support vector machine based on the first exhaust gas component data; Collecting the first load state of the main engine at the current time step, and weightedly fusing the first combustion state and the first load state through an attention mechanism module to obtain a multimodal feature vector; Obtaining a pre-trained deep neural network model, and obtaining, based on the deep neural network model and the multimodal feature vector, a first predicted Q value for each main engine output power decision within a preset discrete action space; wherein the deep neural network is obtained by constructing a reward function with the goal of minimizing exhaust emissions and training using a deep reinforcement learning algorithm based on the reward function; A first main engine output power decision of an optimal first predicted Q value is obtained, and a hybrid power output system is controlled to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision.

2. The hybrid power output control method based on emission optimization according to claim 1, characterized in that: The determining, based on the first exhaust gas composition data, a first combustion state of the main engine at the current time step by a pre-trained support vector machine includes: Selecting a corresponding first support vector machine according to the first load state; wherein the first support vector machine is trained based on second exhaust gas component data of exhaust gas emitted by the ship when the main engine is in different combustion states under the first load state; Based on the first support vector machine and according to the first exhaust gas composition data, a first combustion state of the main engine when it is in the first load state at a current time step is determined.

3. The hybrid power output control method based on emission optimization according to claim 1, characterized in that: The collecting of the first load state of the main engine at the current time step, and weighted fusion of the first combustion state and the first load state through an attention mechanism module to obtain a multimodal feature vector, includes: Embedding the first combustion state through a preset embedding layer to obtain a first vector; Performing a linear transformation and activation operation on the first load state through a preset feedforward neural network to obtain a second vector with the same dimension as the first vector; determining first weights corresponding to the first vector and the second vector respectively according to correlations between the first combustion state and the first load state and the first exhaust gas composition data; The first vector and the second vector are multiplied by the corresponding first weights and then added to obtain the multimodal feature vector.

4. The hybrid power output control method based on emission optimization according to claim 1, characterized in that: The controlling the hybrid power output system to perform a power distribution operation according to the propulsion power demand and the first main engine output power decision includes: controlling the output power of the main engine according to the first main engine output power decision, and controlling the output power of energy storage according to the difference between the propulsion power demand and the first main engine output power decision; After the ship performs the power allocation operation, third exhaust gas component data of the exhaust gas emitted by the ship in the next time step is collected, and the experience pool is updated according to the difference between the first exhaust gas component data and the third exhaust gas component data; Based on the updated experience pool, the parameters of the deep neural network model are fine-tuned through a supervised learning method.

5. The hybrid power output control method based on emission optimization according to claim 4, characterized in that: The updating of the experience pool according to the difference between the first exhaust gas component data and the third exhaust gas component data includes: Determining second weights corresponding to various exhaust gas components using an entropy weight method based on the first load state and the first combustion state to obtain a weight parameter combination; wherein a weight parameter combination corresponds to a combination of a load state and a combustion state; For each exhaust component, calculating a first difference between its value in the first exhaust component data and its value in the third exhaust component data, and then accumulating a first product of the first difference corresponding to each exhaust component and a second weight to obtain a weighted sum; The weighted sum is used as the actual Q value, the first load state and the first combustion state are used as input data, a training sample is constructed according to the actual Q value, the input data and the first main engine output power decision, and the training sample is added to the experience pool.

6. The hybrid power output control method based on emission optimization according to claim 5, characterized in that: The determining, according to the first load state and the first combustion state, second weights corresponding to various exhaust gas components by an entropy weight method includes: Acquire a plurality of fourth exhaust gas component data when the main engine is in the first load state and the first combustion state, and for each exhaust gas component, add the corresponding value of the fourth exhaust gas component data to obtain a first cumulative sum corresponding to each exhaust gas component; Calculating a ratio of a value of each exhaust gas component in each item of the fourth exhaust gas component data to the corresponding first cumulative sum, and obtaining a second product of each of the ratios and a logarithm of the ratio; After negating each of the second products, for each exhaust gas component, accumulating all corresponding second products to obtain a second accumulated sum corresponding to each exhaust gas component; For each exhaust gas component, a second difference between the corresponding second accumulated sum and a preset threshold is calculated, and the normalized second difference is used as a second weight corresponding to the exhaust gas component.

7. The hybrid power output control method based on emission optimization according to claim 5, characterized in that: Fine-tuning the parameters of the deep neural network model using a supervised learning method based on the updated experience pool includes: For each training sample in the experience pool, obtaining a second predicted Q value for each main engine output power decision based on its corresponding input data and the deep neural network model, and determining a second main engine output power decision corresponding to an optimal second predicted Q value based on each second predicted Q value; Filtering out a first training sample from all the training samples, in which the output power decision of the second main engine is consistent with the output power decision of the first main engine; For each of the first training samples, calculating a third difference between the corresponding actual Q value and the optimal second predicted Q value; The mean of the sum of squares of the third differences is used as a loss value, and the parameters of the deep neural network model are adjusted through back propagation according to the loss value.

8. The hybrid power output control method based on emission optimization according to claim 1, characterized in that: The pre-trained long short-term memory network model obtains the propulsion power requirement for the next time step based on all the propulsion power data, including: collecting first sea state data of the plurality of historical time steps and second sea state data of the next time step, splicing the first sea state data with the propulsion power data according to the time step to obtain a two-dimensional time series feature tensor, and splicing the second sea state data to the tail of the two-dimensional time series feature tensor; extracting propulsion power trend change features from the two-dimensional time series feature tensor through a one-dimensional convolutional neural network; The propulsion power trend change characteristics are input into the long short-term memory network model to obtain the propulsion power demand for the next time step.

9. A hybrid power output control device based on emission optimization, characterized in that: include: A hybrid power output control device based on emission optimization is characterized by comprising: a propulsion power demand prediction module, a combustion state classification module, a feature fusion module, a Q value prediction module and a power distribution module; The propulsion power demand prediction module is used to collect propulsion power data of several historical time steps of the ship, and obtain the propulsion power demand of the next time step based on all the propulsion power data through the pre-trained long short-term memory network model; The combustion state classification module is used to collect first exhaust gas component data of exhaust gas emitted by the ship at the current time step, and determine the first combustion state of the main engine at the current time step through a pre-trained support vector machine based on the first exhaust gas component data; The feature fusion module is used to collect the first load state of the main engine at the current time step, and weightedly fuse the first combustion state and the first load state through the attention mechanism module to obtain a multimodal feature vector; The Q-value prediction module is configured to obtain a pre-trained deep neural network model and, based on the deep neural network model and the multimodal feature vector, obtain a first predicted Q-value for each main engine output power decision within a preset discrete action space; wherein the deep neural network is obtained by constructing a reward function with the goal of minimizing exhaust emissions and training using a deep reinforcement learning algorithm based on the reward function; The power distribution module is used to obtain the first main engine output power decision of the optimal first predicted Q value, and control the hybrid power output system to perform power distribution operations according to the propulsion power demand and the first main engine output power decision.

10. The hybrid power output control device based on emission optimization according to claim 9, characterized in that: The combustion state classification module includes: a support vector machine selection unit and a support vector machine calling unit; the combustion state classification module is used to determine the first combustion state of the main engine at the current time step based on the first exhaust gas composition data using a pre-trained support vector machine, including: The support vector machine selection unit is configured to select a corresponding first support vector machine according to the first load state; wherein the first support vector machine is trained based on second exhaust gas component data of exhaust gas emitted by the ship when the main engine is in different combustion states under the first load state; The support vector machine calling unit is used to determine the first combustion state of the main engine when it is in the first load state at the current time step based on the first support vector machine and the first exhaust gas composition data.

Images