A time-space integrated control method for ballast water thermal treatment system

Abstract

The present invention discloses a time-space integrated control method for a ballast water heat treatment system, including obtaining a multi-stage series treatment system of a cylinder water heat source for ballast water heat treatment to simplify obtaining an equivalent plug flow reactor model; obtaining the time-space temperature data of the equivalent plug flow reactor model, and constructing an optimal temperature time-space model for ballast water heat treatment according to the time-space temperature data; and constructing a nonlinear predictive controller for controlling the temperature of heated ballast water according to the optimal temperature time-space model. The present invention solves the problem that the existing heat treatment technology does not effectively take into account practical factors such as cabin space allocation and efficient energy utilization, the existing modeling method has linear assumption limitations, insufficient local structure retention, and cannot fully explore the nonlinear characteristics and uncertainty factors in the ballast water heat treatment process to explore the connection between temperature fluctuations and changes in microbial concentrations, resulting in the inability to effectively achieve the control accuracy and efficiency of the ballast water heat treatment system.

Description

Technical Field

[0001] The present invention relates to the technical field of ballast water heat treatment and distributed parameter system predictive control strategy research, and in particular to a time-space integrated control method for a ballast water heat treatment system. Background Art

[0002] The rapid development of the global shipping industry has led to an increase in the scale of ocean shipping, resulting in a surge in the tonnage of ballast water transfer, which has not only aggravated the pollution of alien species, but also caused damage to coastal ecosystems. It is estimated that more than 1 billion tons of ballast water carry more than a thousand species of marine organisms each year, and these alien species pose a serious threat to local ecosystems [1,2]. Specifically, the flow nature of ballast water makes it the main route for species transmission. The rapid reproduction and spread of alien organisms pose a direct threat to the living environment of local species, resulting in the loss of fishery resources and economic damage. The international community has listed the problem of alien organism invasion brought by ship ballast water as one of the main threats facing the ocean. Designing a safe, efficient, energy-saving and reliable ballast water treatment method has always been a key issue of concern to the shipping and shipbuilding industries.

[0003] Existing ballast water treatment methods mainly include mechanical, physical and chemical methods, which have played a role in controlling the invasion of foreign organisms to a certain extent [3]. However, these methods have some problems in engineering practice. Mechanical methods achieve solid-liquid separation through filtration, centrifugal separation and other technologies, but may encounter problems such as filter blockage and increased discharge resistance of ballast pumps, and are not effective against small pathogens. Physical methods destroy the living conditions of microorganisms through ultraviolet rays, ultrasound and other means, which are harmless to the environment, but may have limited effects on certain microorganisms [4]. In addition, ultraviolet treatment is greatly affected by the turbidity of water, and ultrasound treatment may affect the hull structure and the health of people on board. Chemical methods kill organisms in the water by adding chemicals such as chlorine and ozone. Although the killing effect can be achieved in a relatively short time, the storage and management of active substances is difficult, which may affect pipelines, ballast tank structures and coatings. In addition, ballast water management systems using chemical methods may generate disinfection by-products while killing aquatic organisms, posing a threat to the environment. Therefore, it is particularly important to explore safer, more efficient, energy-saving and reliable treatment methods. Ballast water thermal treatment technology is a specific application of pasteurization in the engineering field. Based on the principle that high temperature can kill microorganisms, heat treatment is a very effective means of removing microorganisms and pathogens from ballast water[5]. This method does not require the use of chemically active substances, reduces potential environmental impacts, and does not have the problem of chemical residues. Secondly, heat treatment is non-corrosive to ballast water tanks, pipelines, etc., has low maintenance costs, and is easy to operate. The temperature control of heat treatment can be accurately controlled within the range of 38°C-45°C, which is sufficient to kill most marine organisms after a certain period of time. It is a reliable option for solving the problem of alien biological invasion caused by ballast water transfer[6].

[0004] Ballast water thermal treatment involves the distribution and transfer of heat in the treatment device. Parameters such as temperature and microbial concentration vary at different locations and at different time points. The changes in these parameters are crucial to the effect of the entire thermal treatment process, so it can be described as a distributed parameter system. Research on distributed parameter systems mainly includes two parts: modeling and control. Among them, the modeling research of distributed parameter systems is mainly based on data-driven methods [7]. Among the existing ballast water thermal treatment methods, they mainly include boiler heating method and jacket water heating method. Boiler heating method refers to pumping ballast water to the boiler and heating it to a specified temperature (such as 60°C), and then pumping it back to the ballast tank. The implementation of this technology is relatively simple, and only three factors need to be considered: effective heating time, effective treatment temperature and ballast water inlet temperature. The jacket water heating method uses waste heat and a heat storage cabinet to treat harmful microorganisms in ballast water. The specific process includes the ballast water being heated to a value slightly higher than the ambient temperature in the preheating heat exchanger, and then the temperature is heated to the treatment range of 38°C-45°C through the main engine heat exchanger, and finally sealed in the heat storage cabinet for a period of time to complete the treatment. This method can recycle waste heat to a certain extent, saving energy while avoiding repeated pollution. However, the existing heat treatment technology does not effectively take into account practical factors such as cabin space allocation and efficient use of energy. Based on the above analysis, the existing ballast water heat treatment method has the following defects:

[0005] 1. Space configuration: The boiler heating method requires the installation of a separate boiler and the configuration of a fuel pipeline system. In particular, a larger capacity boiler is required for a heating system with high processing capacity requirements, which directly affects the layout and use of the cabin space. Although the jacket water heating method does not require a separate heating system, the configuration of a heat storage cabinet will increase the space occupied by the cabin.

[0006] 2. Sustainable development: The boiler heating method requires additional fuel consumption and produces exhaust gas, which is not very economical and environmentally friendly. It is also prone to thermal expansion and contraction, posing a threat to the safety of the ship. The jacket water heating method utilizes the waste heat of the ship's main engine to a certain extent, but there is a problem of low utilization rate of the main engine's waste heat. At the same time, the configuration of the heat storage cabinet will bring costs for subsequent equipment maintenance.

[0007] The study of exploring the distribution of thermal energy and changes in microbial concentration during the thermal treatment of ballast water can be regarded as a modeling study of distributed parameter systems. Existing distributed parameter system modeling techniques are mostly based on data-driven modeling. Compared with the mechanism modeling method, data-driven modeling does not require an in-depth understanding of the complex internal mechanisms of the system. It only needs to fit the nonlinear characteristics of the system through input / output data, and significantly saves computing costs in the subsequent control algorithm design. The thermal treatment process of ballast water is often accompanied by spatiotemporal coupling, strong nonlinearity, energy exchange and the influence of unknown factors. Therefore, using data-driven modeling to explore its dynamic characteristics has significant advantages. However, the existing modeling methods have problems such as linear assumption limitations and insufficient local structure preservation. Combined with the summary of the technical route, the existing distributed parameter system modeling methods mainly have the following problems:

[0008] 1. Spatiotemporal decomposition: Existing spatiotemporal decomposition methods are mainly based on linear assumptions and focus on capturing the global characteristics of data. They cannot effectively handle nonlinear relationships and complex data structures, and cannot effectively maintain the local structural information of data. There is a possibility of ignoring local spatiotemporal characteristics, resulting in the loss of important nonlinear characteristics of the data after spatiotemporal decomposition.

[0009] 2. Time series identification: The existing time series identification methods mainly have the problem of being unable to effectively balance the training speed and nonlinear feature fitting ability. For example, the least squares support vector machine can effectively improve the accuracy of the time series model through parameter optimization, but the training speed is slow; although the extreme learning machine significantly improves the modeling speed, there is a problem of decreased generalization due to the random assignment of hidden layer neurons, which in turn affects the fitting effect. Summary of the invention

[0010] The present invention provides a time-space integrated control method for a ballast water heat treatment system to overcome the above technical problems.

[0011] In order to achieve the above object, the technical solution of the present invention is:

[0012] A time-space integrated control method for a ballast water heat treatment system specifically comprises the following steps:

[0013] S1: Construct a multi-stage series treatment system of cylinder water heat source for ballast water thermal treatment, and simplify it to obtain an equivalent plug flow reactor model;

[0014] S2: Collect and obtain the spatiotemporal temperature data of the equivalent plug flow reactor model, and construct the optimal spatiotemporal temperature model for ballast water thermal treatment based on the spatiotemporal temperature data;

[0015] And the construction method of the optimal temperature spatiotemporal model includes

[0016] S21: Perform kernel principal component analysis on spatiotemporal temperature data to construct spatial basis functions;

[0017] S22: Obtaining a solution time coefficient sequence of the spatiotemporal temperature data according to the spatial basis function;

[0018] Obtain historical sequence data based on the solution time coefficient sequence combined with the plug flow reactor model;

[0019] S23: defining a sequence data pair according to the solved time coefficient sequence and the historical sequence data;

[0020] Divide the sequence data pairs into a data training set and a data test set according to a preset ratio;

[0021] S24: Based on the sparrow optimization algorithm, a time model based on the kernel extreme learning machine is constructed according to the data training set;

[0022] S25: Obtaining a trained temperature spatiotemporal model according to the spatial basis function and in combination with the constructed time model;

[0023] S26: Based on the relative error and root mean square error function, the trained temperature spatiotemporal model is evaluated according to the data validation set, and the trained temperature spatiotemporal model that meets the preset evaluation index is used as the optimal temperature spatiotemporal model;

[0024] For the trained temperature spatiotemporal model that does not meet the preset evaluation index, the model weights are adaptively updated based on the back propagation method, and S24 to S25 are repeatedly executed;

[0025] S3: A nonlinear predictive controller for controlling the temperature of heated ballast water is constructed based on the optimal temperature spatiotemporal model to achieve spatiotemporal integrated control of the ballast water thermal treatment system.

[0026] Furthermore, the method for obtaining the equivalent plug flow reactor model in S1 is:

[0027] Assuming that the structural characteristics of the multi-stage series treatment system of the jacket water heat source include

[0028] Main engine jacket water pipeline, multiple tubular heat exchangers connected in parallel, three-way valves and ballast water pipeline;

[0029] And the tubular heat exchanger includes a tube side part and a shell side part;

[0030] The pipe section is connected in parallel with the main engine cylinder jacket water pipeline;

[0031] The shell side portion is connected in series with the ballast water pipeline;

[0032] The three-way valve is used to achieve heat treatment of ballast water in the ballast water pipeline by adjusting the flow distribution of the jacket water heat source applied to each tubular heat exchanger;

[0033] The multi-stage tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is equivalent to a multi-stage series heater equivalent module;

[0034] The inlet position of the first shell side of the tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is regarded as the ballast water inlet in the heater equivalent module, and the outlet position of the last shell side of the tubular heat exchanger is regarded as the ballast water outlet in the equivalent model, so as to serve as an equivalent plug flow reactor model.

[0035] Furthermore, the S21 specifically includes the following steps:

[0036] S211: Perform nonlinear mapping on the spatiotemporal temperature data and combine the kernel function to obtain the feature mapping space vector, which is expressed as

[0037] , where: is the kernel function, indicating and The result of the inner product operation in the feature space; Represents the vector in the input space The vector after mapping to the feature space; Represents the vector in the input space The vector after mapping to the feature space, and ; Represents a nonlinear mapping; represents the acquired spatiotemporal temperature data and , ; Indicates The sensor at the time The temperature response is the temperature of the heater equivalent module collected; Indicates The sensor at the time Temperature response; Indicates the total time; It represents the number of sensors equally distributed in the axial direction of the multi-stage series heater equivalent module;

[0038] S212: Obtaining a covariance matrix of the feature mapping space according to the feature mapping space vector;

[0039] The formula for obtaining the covariance matrix is:

[0040] ;

[0041] S213: Define the covariance matrix The characteristic value of With the eigenvector , the solution formula for obtaining the covariance matrix and eigenvector is ;

[0042] Will Treat it as a linear operator and obtain the solution formula based on the linear operator The equivalent formula of is

[0043] , , where: Represents the feature vector The weight of the linear correlation coefficient;

[0044] S214: According to the covariance matrix and the equivalent formula, the intermediate process formula is obtained as follows:

[0045] ;

[0046] S215: Define kernel function With weight vector ,and , , and according to the kernel function With the weight vector , rewrite the intermediate process formula as follows:

[0047] , where: Represents the kernel function A simplified form of

[0048] S216: Define the eigenvalues ​​of the covariance matrix as and , based on the kernel function The symmetry characteristics of the rewritten intermediate process formula can be transformed to obtain ,

[0049] By considering the eigenvector The unit length of the weight vector Perform normalization processing, and the expression of the normalization processing is

[0050] , where: Representation Matrix The number of non-zero eigenvalues ​​in ; represents the covariance matrix eigenvalues; express The abbreviation of Representation Matrix The total number of eigenvalues ​​in ;

[0051] Pair Matrix In The eigenvalues ​​are arranged in descending order to obtain a feature sequence list;

[0052] And through the cumulative contribution rate calculation formula, according to the feature sequence table, obtain the preset cumulative threshold Before p The cumulative contribution rate of eigenvalues Right now ;

[0053] And the cumulative contribution rate calculation formula is:

[0054] ,

[0055] Before The eigenvalues ​​are used as the data variance of kernel principal component analysis to determine the spatial basis function.

[0056] And the space basis function The expression is

[0057] , ,

[0058] Where: Indicates Moment The weight vector of the sensors; Indicates The spatial basis functions of the sensors and ;

[0059] Furthermore, the S22 specifically includes the following steps:

[0060] S221: Obtaining a solution time coefficient sequence of spatiotemporal temperature data according to a spatial basis function;

[0061] The solution time coefficient sequence expression is:

[0062] , , where: Represents the solution time coefficient sequence; represents the inner product operation;

[0063] S222: Acquire historical sequence data according to the solved time coefficient sequence and the plug flow reactor model;

[0064] The expression of the historical sequence data is:

[0065] , where: Represents historical sequence data; denote the hysteresis orders of the output and input of the plug flow reactor model respectively and ; represents the plug flow reactor model input at time The hysteresis result is , represents the number of heaters in the equivalent model, i.e., the plug flow reactor model; represents the output of the plug flow reactor model at time The hysteresis result;

[0066] Further, the S24 specifically includes the following steps:

[0067] S241: Based on sequence data , define and obtain the functional relationship for kernel extreme learning machine identification, and the expression of the functional relationship is

[0068] , where: It represents the mapping relationship between the time coefficient sequence and the time-space sequence data based on the kernel extreme learning machine identification;

[0069] S242: constructing a regression function model for solving strong nonlinear data based on the functional relationship for the data sequence of the data training set;

[0070] The expression of the regression function model is:

[0071] And meet ;

[0072] Where: Represents the functional relationship between the input data and the hidden layer of the kernel extreme learning machine network, ; Represents the output weights between the output layer and the hidden layer of the kernel extreme learning machine network; Represents the output error; represents the penalty factor; Indicates the total time before The sequence data pairs corresponding to each moment;

[0073] S243: Based on the Karush-Kuhn-Tucker condition, obtain the output weights of the extreme learning machine network kernel extreme learning machine according to the data training set , to obtain the network model of the kernel extreme learning machine;

[0074] , , , where: represents the mapping matrix of the input data and ; express A simplified form of represents the identity matrix; for A simplified form of , which represents the kernel matrix set according to the Mercer condition and ; Represents the output of the kernel extreme learning machine network model;

[0075] The model hyperparameters are determined by introducing the sparrow optimization algorithm into the time series identification based on the kernel extreme learning machine network model. The model hyperparameters include the penalty factor With the kernel matrix Kernel parameters;

[0076] The temperature spatiotemporal model is obtained by reconstructing the kernel extreme learning machine network model and combining the spatial basis function;

[0077] The expression of the temperature space-time model is:

[0078] , where: represents the output response of the temperature space-time model; express A simplified form of

[0079] Furthermore, the method for constructing a nonlinear predictive controller for controlling the temperature of heated ballast water according to the optimal temperature spatiotemporal model described in S3 specifically comprises the following steps:

[0080] S31: Defining the control domain of the optimal temperature spatiotemporal model M With prediction time domain P and , and set the desired control trajectory of the nonlinear predictive controller;

[0081] And soften the expected control trajectory to obtain a reference state signal;

[0082] The formula for obtaining the reference state signal is:

[0083] , where: Indicates from Time has passed The expected control trajectory at the time after sampling; Indicates that the softening process obtains the reference state signal and ; represents a vector of all 1s; Represents a softening matrix with all elements between [0,1];

[0084] S32: defined in Moment The output of the optimal temperature spatiotemporal model for sub-sampling is expressed as

[0085] , where: express Relationship function with plug flow reactor model input; Indicated in Time plug flow reactor model input;

[0086] S33: constructing a performance indicator target model for ensuring that the spatiotemporal model effectively tracks the reference state according to the reference state signal and the output of the optimal temperature spatiotemporal model;

[0087] And the performance indicator target model includes an objective function and constraint conditions;

[0088] The expression of the objective function is

[0089] ,

[0090] The constraint condition is expressed as

[0091] , , , where: and represents the weight matrix, represents an integer parameter and ; represents the minimum value of the control input of the nonlinear predictive controller; represents the maximum value of the control input of the nonlinear predictive controller; Represents the minimum value of the control increment of the nonlinear predictive controller; It represents the maximum value of the control increment of the nonlinear predictive controller;

[0092] The performance index target model is solved to obtain the optimal control sequence of the nonlinear predictive controller, and the first element in the optimal control sequence is selected as the input of the ballast water thermal treatment system to achieve the time-space integrated control of the ballast water thermal treatment system;

[0093] The expression of the optimal control sequence is:

[0094] , , where: represents the optimal control sequence of the nonlinear predictive controller; represents the elements of the optimal control sequence; Indicated in is the optimal control input of the nonlinear predictive controller.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] 1. The present invention constructs a multi-stage series treatment system of a cylinder jacket water heat source for ballast water heat treatment. Through the main engine cylinder jacket water pipeline, the waste heat generated during the operation of the ship's main engine is used for ballast water heat treatment without the need to set up other additional heating equipment. This solves the defects of low waste heat recycling rate and high space occupancy rate of traditional ballast water heat treatment systems. The design of continuously treating ballast water through a multi-tube heat exchanger not only saves cabin space, but also significantly improves the recovery and utilization of main engine waste heat, reduces energy consumption and further promotes green navigation.

[0097] 2. The present invention obtains the spatiotemporal temperature data of the equivalent plug flow reactor model, and constructs the optimal temperature spatiotemporal model of ballast water heat treatment based on the spatiotemporal temperature data, which solves the defects that the general spatiotemporal decomposition technology cannot effectively handle nonlinear relationships and complex data structures, and there is a possible defect of ignoring local spatiotemporal characteristics, and the problem that the model training speed and nonlinear feature fitting ability cannot be effectively balanced in the existing time series identification method. By simplifying the proposed multi-stage series treatment system of the cylinder jacket water heat source into an equivalent model of the plug flow reactor, the relationship between the heating temperature and the microorganism concentration is intuitively explored, and the optimal spatiotemporal model is constructed for the temperature data using a kernel learning-based modeling method, so that the nonlinear dynamics and unknown disturbances of the system can be effectively described;

[0098] 3. The present invention designs a nonlinear predictive controller for controlling the temperature of heated ballast water through a spatiotemporal model based on temperature data. It solves the problem of difficulty in directly monitoring and controlling the concentration of microorganisms during heat treatment by indirectly reducing the concentration of harmful microorganisms, which has practical significance for achieving effective killing of harmful microorganisms in ballast water. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0100] Figure 1 It is a flow chart of the time-space integrated control method of the ballast water heat treatment system of the present invention;

[0101] Figure 2 This is a schematic diagram of the structure of the multi-stage series treatment system of the cylinder water heat source in this embodiment;

[0102] Figure 3 Schematic diagram of an equivalent plug flow reactor model in this embodiment;

[0103] Figure 4A flowchart of constructing a time series model of a core extreme learning machine based on the sparrow optimization algorithm in this embodiment;

[0104] Figure 5 It is a flowchart of the time-space integrated control technology based on the equivalent model in this embodiment;

[0105] Figure 6 This is a temperature curve diagram of the situation that is not effectively treated in this embodiment;

[0106] Figure 7 is a microorganism concentration curve diagram of the situation where no effective treatment is performed in this embodiment;

[0107] Figure 8 This is a comparison chart of the fitting effect of the time series model on the time series in this embodiment;

[0108] Fig. 9 is the absolute relative error of the optimal temperature spatiotemporal model in this embodiment;

[0109] Fig.10 is a graph of the controlled temperature response in this embodiment;

[0110] Fig.11 Graph showing the response of the controlled microorganism concentration in this example. DETAILED DESCRIPTION

[0111] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0112] This embodiment provides a time-space integrated control method for a ballast water thermal treatment system. Figure 1 As shown, the specific steps include:

[0113] S1: Construct a multi-stage series treatment system of cylinder water heat source for ballast water thermal treatment, and simplify it to obtain an equivalent plug flow reactor model;

[0114] The S1 specifically includes the following steps:

[0115] Assuming that the structural characteristics of the multi-stage series treatment system of the jacket water heat source include

[0116] Main engine jacket water pipeline, multiple tubular heat exchangers connected in parallel, three-way valves and ballast water pipeline;

[0117] And the tubular heat exchanger includes a tube side part and a shell side part;

[0118] The pipe section is connected in parallel with the main engine cylinder jacket water pipeline;

[0119] The shell side portion is connected in series with the ballast water pipeline;

[0120] The three-way valve is used to achieve heat treatment of ballast water in the ballast water pipeline by adjusting the flow distribution of the jacket water heat source applied to each tubular heat exchanger;

[0121] The multi-stage tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is equivalent to a multi-stage series heater equivalent module;

[0122] The inlet position of the first shell side of the tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is regarded as the ballast water inlet in the heater equivalent module, and the outlet position of the last shell side of the tubular heat exchanger is regarded as the ballast water outlet in the equivalent model, so as to serve as an equivalent plug flow reactor model.

[0123] like Figure 2 As shown, in order to avoid the problem of high space occupancy caused by the existing ballast water heat treatment methods, which require the installation of a separate boiler and the configuration of a fuel pipeline system or the addition of a heat storage cabinet, a multi-stage series treatment system for the jacket water heat source is constructed, and it is composed of a ship main engine, a fresh water pump, a plurality of tubular heat exchangers, and matching three-way valves and pipelines; in this embodiment, the waste heat generated during the operation of the ship main engine is transferred to the jacket water in the main engine jacket water pipeline, and the tube side part of the tubular heat exchanger is connected in parallel with the main engine jacket water pipeline, and the jacket water after heat transfer is transported to the tube side part of the tubular heat exchanger through the pipeline, while the shell side part is connected in series with the ballast water pipeline, and then the heat is transferred to the shell side part through the tube side part of the tubular heat exchanger, so as to achieve heat treatment of the ballast water in the ballast water pipeline. In addition, the multi-stage distribution of the flow rate from the main engine jacket water pipeline into each tubular heat exchanger can be changed by automatically adjusting the opening of the three-way valve, thereby effectively controlling its average heating temperature and ensuring that the ballast water obtains a continuous and effective treatment temperature. Figure 3 As shown, the multi-stage series treatment system of the jacket water heat source can be further simplified into an equivalent model of a plug flow reactor, in which the tubular reactor can be regarded as a heater in the equivalent model, the inlet position of the first tubular heat exchanger shell in the treatment system can be regarded as the ballast water inlet in the equivalent model, and the outlet position of the last tubular heat exchanger shell can be regarded as the ballast water outlet in the equivalent model. This simplified treatment of the treatment system is carried out under the assumption that the heat distribution of each tubular heat exchanger is consistent and the microorganisms in the ballast water are uniform and constant. By simplifying the model of the treatment system, the relationship between the heat distribution and the concentration of harmful microorganisms during the treatment process can be intuitively studied;

[0124] S2: Collect and obtain the spatiotemporal temperature data of the equivalent plug flow reactor model, and construct the optimal spatiotemporal temperature model for ballast water thermal treatment based on the spatiotemporal temperature data;

[0125] In this embodiment, temperature data is collected for an equivalent plug flow reactor model, wherein the temperature data Is The data is collected by sensors that are evenly distributed in the axial direction. Indicates time The input of the multi-stage series treatment system of the cylinder water heat source, Indicates the sensor at time Temperature response; Indicates the total time; The number of heaters representing the equivalent plug flow reactor model; the collected temperature data are modeled using a kernel learning-based method Construct temperature space-time model;

[0126] And the construction method of the optimal temperature spatiotemporal model includes

[0127] The temperature spatiotemporal modeling method based on kernel learning includes the spatiotemporal decomposition process based on kernel principal component analysis and the time series identification process based on kernel extreme learning machine. The spatiotemporal decomposition based on kernel principal component analysis is achieved through nonlinear mapping. The input space is mapped to the feature space, and then the spatial basis function is calculated. The following describes the derivation process of the spatial basis function based on kernel principal component analysis. For the convenience of derivation and calculation, , by nonlinear mapping , can be further rewritten as ;

[0128] S21: Perform kernel principal component analysis on spatiotemporal temperature data to construct spatial basis functions;

[0129] Due to the spatial dimension and the complexity of the mapping, nonlinear mapping cannot be determined directly, so the mapping function is explicitly described by using the kernel trick so that the temperature data The kernel function can be used to explicitly map the result, which includes the following steps:

[0130] S211: Perform nonlinear mapping on the spatiotemporal temperature data and combine the kernel function to obtain the feature mapping space vector, which is expressed as

[0131] , where: is the kernel function, indicating and The result of the inner product operation in the feature space; Represents the vector in the input space The vector after mapping to the feature space; Represents the vector in the input space The vector after mapping to the feature space, and ; Represents a nonlinear mapping; represents the acquired spatiotemporal temperature data and ; Indicates The sensor at the time The temperature response is the temperature of the heater equivalent module collected; Indicates The sensor at the time Temperature response; Indicates the total time; It represents the number of sensors equally distributed in the axial direction of the multi-stage series heater equivalent module;

[0132] S212: Assuming temperature response in input space Transformed into feature map space vector , and have , then the covariance matrix of the feature mapping space is obtained according to the feature mapping space vector;

[0133] The formula for obtaining the covariance matrix is:

[0134] ;

[0135] S213: Define the covariance matrix The characteristic value of With the eigenvector , the solution formula for obtaining the covariance matrix and eigenvector is ; where all eigenvectors are in the mapping vector In the space formed, the mapping vector satisfies the centralization condition, that is, the mapping vector is centered at the origin in the feature space, and the mean of the mapping vector is 0;

[0136] If the feature space is infinite-dimensional, Treat it as a linear operator and obtain the solution formula based on the linear operator The equivalent formula of is

[0137] ,

[0138] And for the eigenvector, there exists

[0139] , where: Represents the feature vector The weights of a set of linear correlation coefficients;

[0140] S214: According to the covariance matrix and the equivalent formula, the intermediate process formula is obtained as follows:

[0141] ;

[0142] S215: Define kernel function With weight vector ,and , , and according to the kernel function With the weight vector , rewrite the intermediate process formula as follows:

[0143] , where: Represents the kernel function A simplified form of

[0144] S216: Define the eigenvalues ​​of the covariance matrix as and , based on the kernel function The symmetry characteristics of the rewritten intermediate process formula can be transformed to obtain

[0145] ,

[0146] By considering the eigenvector The unit length of the weight Perform normalization processing, and the expression of the normalization processing is

[0147] , where: Representation Matrix The number of non-zero eigenvalues ​​in ; represents the covariance matrix eigenvalues; express The abbreviation of Representation Matrix The total number of eigenvalues ​​in ; Indicates The weight vector corresponding to the non-zero eigenvalues; Indicates non-zero eigenvalues;

[0148] Pair Matrix In The eigenvalues ​​are arranged in descending order to obtain a feature sequence list;

[0149] And through the cumulative contribution rate calculation formula, according to the feature sequence table, obtain the preset cumulative threshold Before The cumulative contribution rate of eigenvalues Right now ;

[0150] And the cumulative contribution rate calculation formula is:

[0151] ,

[0152] The above formula is expressed as The amount of data variance explained by the principal component corresponding to the eigenvalue is is the preset cumulative threshold. The corresponding eigenvalues When the original temperature data is n Dimensionality The cumulative threshold is set by engineering practice needs and is usually greater than 95%;

[0153] Before The eigenvalues ​​are used as the data variance of kernel principal component analysis to determine the spatial basis function.

[0154] And the space basis function The expression is

[0155]

[0156] Where: Indicates Moment The weight vector of the sensors; Indicates The spatial basis functions of the sensors and ;

[0157] S22: Obtaining a solution time coefficient sequence of the spatiotemporal temperature data according to the spatial basis function;

[0158] Obtain historical sequence data based on the solution time coefficient sequence combined with the plug flow reactor model;

[0159] The specific steps include:

[0160] S221: Obtaining a solution time coefficient sequence of spatiotemporal temperature data according to a spatial basis function;

[0161] The solution time coefficient sequence expression is:

[0162] , , where: Represents the solution time coefficient sequence; represents the inner product operation;

[0163] S222: By introducing the idea of ​​time series prediction, historical sequence data is obtained by solving the time coefficient sequence and combining the plug flow reactor model;

[0164] The expression of the historical sequence data is:

[0165] , where: Represents historical sequence data; denote the hysteresis orders of the output and input of the plug flow reactor model respectively and ; represents the plug flow reactor model input at time The hysteresis result is , represents the number of heaters in the equivalent model, i.e., the plug flow reactor model; represents the output of the plug flow reactor model at time The hysteresis result;

[0166] S23: defining a sequence data pair according to the solved time coefficient sequence and the historical sequence data;

[0167] Divide the sequence data pairs into a data training set and a data test set according to a preset ratio; wherein the preset ratio of the data training set to the data test set does not exceed 8:2;

[0168] S24: Based on the sparrow optimization algorithm, a time model based on a kernel extreme learning machine is constructed according to the data training set, which specifically includes the following steps

[0169] S241: Based on sequence data , define and obtain the functional relationship for kernel extreme learning machine identification, and the expression of the functional relationship is

[0170] , where: It represents the mapping relationship between the time coefficient sequence and the time-space sequence data based on the kernel extreme learning machine identification;

[0171] S242: constructing a regression function model for solving strong nonlinear data based on the functional relationship for the data sequence of the data training set;

[0172] The expression of the regression function model is:

[0173] And meet ;

[0174] Where: Represents the functional relationship between the input data and the hidden layer of the kernel extreme learning machine network, ; Represents the output weights between the output layer and the hidden layer of the kernel extreme learning machine network; Represents the output error; represents the penalty factor; Indicates the total time before The sequence data pairs corresponding to each moment;

[0175] S243: Based on the Karush-Kuhn-Tucker condition, the output weight of the extreme learning machine network core extreme learning machine is obtained according to the data training set to obtain the network model of the core extreme learning machine;

[0176] , , , where: represents the mapping matrix of the input data and ; express A simplified form of ; I represents the identity matrix; for A simplified form of , which represents the kernel matrix set according to the Mercer condition and ; Represents the output of the kernel extreme learning machine network model;

[0177] The model hyperparameters are determined by introducing the sparrow optimization algorithm into the time series identification based on the kernel extreme learning machine network model. The model hyperparameters include the penalty factor With the kernel matrix Kernel parameters;

[0178] The temperature spatiotemporal model is obtained by reconstructing the kernel extreme learning machine network model and combining the spatial basis function;

[0179] In a specific embodiment, the method of introducing the sparrow optimization algorithm into the time series identification based on the kernel extreme learning machine network model to determine the model parameter number is specifically as follows:

[0180] Assume that the number of sparrows is And the location of the sparrow individual represents a set of model hyperparameters, and the dimension of the model hyperparameters is , the solution space of model hyperparameters can be described as ; and select the mean square error or the determination coefficient as the fitness function to obtain the fitness value of the number of sparrows in the sparrow optimization algorithm, and sort the fitness values ​​in the sparrow population, and update the population by combining the position update formula of the sparrow population, wherein the position update formula for population update includes the position update formula of the discoverer, the position update formula of the joiner and the position update formula of the vigilant in sequence;

[0181] The finder's position update formula is:

[0182] ,in, It is A sparrow in the The location of the dimension; Indicates the current iteration number; is a column vector of all 1s; is the maximum number of iterations; , Represent the warning value and safety value respectively; and are all random numbers, ,in Satisfy normal distribution; It means that the finder has a larger search area, but it means that there is danger. Need to change the search location to another area;

[0183] The position update formula of the joiner is:

[0184] in, Represent the best and worst positions of the finder respectively, L represents a one-dimensional ±1 randomly distributed matrix, represents the generalized inverse matrix of L. When When The joiners will be moved to other areas;

[0185] The position update formula of the sentinel is:

[0186] ,in, represents the global optimal position; represents the step length, represents a random number in [-1,1], is a small constant; Indicates the best and worst fitness values ​​of the current global position; Indicates The individual fitness of a sparrow; When , the process of searching for hyperparameters by the sparrow optimization algorithm ends, and then the parameters of the time series model based on the kernel extreme learning machine are determined;

[0187] The expression of the temperature space-time model is:

[0188] , where: represents the output response of the temperature space-time model; express A simplified form of

[0189] S25: Obtaining a trained temperature spatiotemporal model according to the spatial basis function and in combination with the constructed time model;

[0190] S26: Based on the relative error and root mean square error function, the trained temperature spatiotemporal model is evaluated according to the data validation set, and the trained temperature spatiotemporal model that meets the preset evaluation index is used as the optimal temperature spatiotemporal model;

[0191] For the trained temperature spatiotemporal model that does not meet the preset evaluation index, the model weights are adaptively updated based on the back propagation method, and S24 to S25 are repeatedly executed;

[0192] S3: Based on the optimal temperature spatiotemporal model, a nonlinear predictive controller for controlling the temperature of heated ballast water is constructed to achieve spatiotemporal integrated control of the ballast water thermal treatment system;

[0193] In this embodiment, after completing the spatiotemporal decomposition and time series modeling of the equivalent model, a nonlinear predictive controller is designed based on the optimal temperature spatiotemporal model based on kernel learning, thereby achieving the purpose of indirectly regulating the concentration of harmful microorganisms by controlling the temperature of the heated ballast water;

[0194] The method for constructing a nonlinear predictive controller for controlling the temperature of heated ballast water according to the optimal temperature spatiotemporal model specifically comprises the following steps:

[0195] S31: Defining the control domain of the optimal temperature spatiotemporal model M With prediction time domain P and , and set the desired control trajectory of the nonlinear predictive controller;

[0196] And soften the expected control trajectory to obtain a reference state signal;

[0197] The formula for obtaining the reference state signal is:

[0198] , where: Indicates from Time has passed The expected control trajectory at the time after sampling; Indicates that the softening process obtains the reference state signal and ; represents a vector of all 1s; Represents a softening matrix whose elements are between [0,1]; subscript Indicates from Time has passed The time after sampling; subscript Indicated in The predicted state at the time given by the time information, ;

[0199] S32: Defined at time l The output of the optimal temperature spatiotemporal model for sub-sampling is expressed as

[0200] , where: express Relationship function with plug flow reactor model input; Indicated in Time plug flow reactor model input;

[0201] S33: constructing a performance indicator target model for ensuring that the spatiotemporal model effectively tracks the reference state according to the reference state signal and the output of the optimal temperature spatiotemporal model;

[0202] And the performance indicator target model includes an objective function and constraint conditions;

[0203] The expression of the objective function is

[0204] ,

[0205] The constraint condition is expressed as

[0206] , , , where: and represents the weight matrix, represents an integer parameter and ; represents the minimum value of the control input of the nonlinear predictive controller; represents the maximum value of the control input of the nonlinear predictive controller; Represents the minimum value of the control increment of the nonlinear predictive controller; It represents the maximum value of the control increment of the nonlinear predictive controller;

[0207] The performance index target model is solved to obtain the optimal control sequence of the nonlinear predictive controller, and the first element in the optimal control sequence is selected as the input of the ballast water thermal treatment system to achieve the time-space integrated control of the ballast water thermal treatment system;

[0208] The performance index target model is solved by using traditional common solution methods to solve control problems, such as sequential quadratic programming or optimal control based on reinforcement learning.

[0209] The expression of the optimal control sequence is:

[0210] , , where: represents the optimal control sequence of the nonlinear predictive controller; represents the elements of the optimal control sequence; Indicated in is the optimal control input of the nonlinear predictive controller.

[0211] In summary, the time-space integrated control method of the ballast water heat treatment system proposed in this embodiment has the following beneficial effects:

[0212] 1. This embodiment solves the defects of low waste heat recycling rate and high space occupancy rate of traditional ballast water heat treatment system by designing a multi-stage series treatment system of cylinder jacket water heat source for ballast water heat treatment. The design of continuous treatment of ballast water by multi-tube heat exchanger not only saves cabin space, but also significantly improves the recovery and utilization of main engine waste heat, reduces energy consumption and further promotes green navigation.

[0213] 2. This embodiment obtains the spatiotemporal temperature data of the equivalent plug flow reactor model, and constructs the optimal temperature spatiotemporal model of ballast water thermal treatment based on the spatiotemporal temperature data, which solves the defects that the general spatiotemporal decomposition technology cannot effectively handle nonlinear relationships and complex data structures, and there is a possible defect of ignoring local spatiotemporal characteristics, and the problem that the model training speed and nonlinear feature fitting ability cannot be effectively balanced in the existing time series identification method. By simplifying the proposed multi-stage series treatment system of the cylinder water heat source into an equivalent model of the plug flow reactor, the relationship between the heating temperature and the microorganism concentration is intuitively explored, and the optimal spatiotemporal model is constructed for the temperature data using a kernel learning-based modeling method, so that the nonlinear dynamics and unknown disturbances of the system can be effectively described;

[0214] 3. This embodiment designs a nonlinear predictive controller for controlling the temperature of heated ballast water through a spatiotemporal model based on temperature data. It solves the problem of difficulty in directly monitoring and controlling the concentration of microorganisms during the heat treatment process by indirectly reducing the concentration of harmful microorganisms, which is of practical significance for achieving effective killing of harmful microorganisms in ballast water.

[0215] In order to verify the effectiveness of the ballast water heat treatment system and the time-space integrated control method proposed in this embodiment, a set of dimensionless partial differential equations will be introduced in this embodiment to describe the situation where harmful microorganisms in the multi-stage series treatment system of the cylinder water heat source are not effectively treated, and numerical simulation experiments will be carried out using the MATLAB platform. First, a time-space model is constructed for the temperature data for solving the dimensionless partial differential equation, and then a predictive controller is designed to adjust the temperature to indirectly control the concentration of harmful microorganisms at the ballast water outlet. The following partial differential equation is introduced:

[0216] ,

[0217] Satisfy the boundary conditions:

[0218] ,

[0219] And the initial conditions:

[0220] , where: and denote the dimensionless spatiotemporal temperature and microbial concentration, respectively. represents a coupling term, represents the simulation coefficient; represents an input term related to the heater position, and its expression is:

[0221] , where: represents the Heaviside function, Indicates control input.

[0222] In this embodiment, the first The control inputs are:

[0223] , where: is a uniformly distributed random number satisfying (0,1). Above The number of samples, heaters and sensors are 5 and 16 respectively, and the simulation coefficients Take 0.4, such as Figures 4 to 5 The temperature data and microbial concentration curves obtained by solving the partial differential equation are shown. Figures 5 and 6 It shows that under actual treatment conditions, due to the heat loss in the multi-stage series treatment system of the jacket water heat source, the temperature in the ballast water heat treatment system changes dramatically, which makes it impossible to completely remove some microorganisms at the outlet. Therefore, the ideal approach is to adjust the heating temperature of the treatment system to a constant value (or a small range) so that the corresponding concentration data can drop to zero at the end of the model, thereby achieving the goal of eliminating harmful microorganisms through heat treatment. Figure 6 The corresponding temperature data is modeled and a predictive controller is designed based on this. At any time, the expected trajectory is set as:

[0224] ;

[0225] The trace is related to the effective process temperature and is sampled at 0.01 intervals to obtain 1500 sample points in the time domain. Figure 6 to Figure 7, the temperature and microbial concentration change rapidly in the initial stage, which can be interpreted as an incomplete treatment stage, so the first 300 sampling points are discarded. Of the remaining 1200 sampling points, the first 900 are used to train the spatiotemporal model, and the last 300 are used for testing. In the kernel learning modeling method, the kernel function type of the kernel principal component analysis used for spatiotemporal decomposition and the kernel extreme learning machine used for time series identification are both selected as radial basis function kernels. The kernel parameters of the kernel principal component analysis are selected as, and the hyperparameters of the kernel extreme learning machine are determined by the sparrow optimization algorithm. Since the selected kernel function type is the radial basis function kernel, the sparrow optimization algorithm only needs to determine the penalty factor of the kernel extreme learning machine. and kernel parameters , and the parameter space dimension of the sparrow optimization algorithm is 2. In addition, define the number of sparrows is 50, the total number of iterations is 100, and the hysteresis orders of output and input are 3 and 2 respectively. Kernel principal component analysis selects the principal component corresponding to the cumulative contribution rate greater than 98%, then the attached Figure 6 The corresponding temperature data is converted into a 2D representation, and the corresponding two sets of time series are used for subsequent time model identification. Definition error To evaluate the identification effect of the time series model, the absolute relative error (ARE) and root mean square error (RMSE) are introduced to evaluate the modeling effect of the spatiotemporal model. The expression is:

[0226] , ,

[0227] like Figures 8 to 9 The modeling effects of the time series model and the temperature spatiotemporal model are shown in the test set. Figures 8 to 9 , the fitting error of the time series model to the time series is within the range, the maximum absolute relative error of the spatiotemporal model does not exceed 4%, and the temperature spatiotemporal model of this embodiment significantly improves the RMSE by about 15%. Figure 10 to Figure 11 The temperature and microbial concentration output response curves of the control design based on the spatiotemporal model of kernel learning are shown. The design of the predictive controller is carried out for the time corresponding to the training model. Among them, the control time domain and the prediction time domain are both selected as 4, and the weight matrices and are 0.5 times the unit matrix and 0.1 times the unit matrix respectively. , , , After about the 500th moment, the controlled temperature output stabilizes near the desired value, which indirectly causes the microbial concentration at the outlet to approach zero, achieving the purpose of effective thermal treatment of the ballast water.

[0228] The related documents also involved in this embodiment are as follows:

[0229] [1] Lin Xiaofen. Review of ship ballast water treatment methods[J]. China Water Transport, 2012,12(04):9-11.

[0230] [2] Yuan Xitao, Yuan Chuan, Wu Haibo. Heating-electrolysis mixed treatment method for ship ballast water[J]. World Shipping, 2011, 34(05): 41-43.

[0231] [3] Wang Wencheng, Gong Fan, Zheng Yu, et al. Review of ship ballast water treatment[J]. Journal of Shanghai Ship and Shipping Research Institute, 2013, 36(04):11-14.

[0232] [4] Thach ND, Hung P V. Optimal UV quantity for a ballastwatertreatment system for compliance with IMO standards[J]. PolishMaritimeResearch, 2023, 30(4):31-42.

[0233] [5] Mesbahi E, Norman RA, Vourdachas A, et al. Design of high-temperature thermal ballast water treatment system[J]. Proceedings of the Institution of Mechanical Engineers, Part M: Journal of Engineering for the Maritime Environment, 2007, 221(1):31-42.

[0234] [6] Hao Junli. Research on ballast water treatment using waste heat from ship diesel engines[D]. Dalian Maritime University, 2004.

[0235] [7] Li HX, Qi C. Modeling of distributed parameter systems for applications-A synthesized review from time-space separation[J]. Journal of Process Control, 2010, 20(8):891-901.

[0236] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A time-space integrated control method for a ballast water heat treatment system, characterized in that: The specific steps include: S1: Construct a multi-stage series treatment system of cylinder water heat source for ballast water thermal treatment, and simplify it to obtain an equivalent plug flow reactor model; The method for obtaining an equivalent plug flow reactor model in S1 is: Assume that the structural features of the multi-stage series treatment system of the cylinder water heat source include Main engine jacket water pipeline, multiple tubular heat exchangers connected in parallel, three-way valves and ballast water pipeline; And the tubular heat exchanger includes a tube side part and a shell side part; The pipe section is connected in parallel with the main engine cylinder jacket water pipeline; The shell side portion is connected in series with the ballast water pipeline; The three-way valve is used to achieve heat treatment of ballast water in the ballast water pipeline by adjusting the flow distribution of the jacket water heat source applied to each tubular heat exchanger; The multi-stage tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is equivalent to a multi-stage series heater equivalent module; The inlet position of the first shell side of the tubular heat exchanger in the multi-stage series treatment system of the cylinder water heat source is regarded as the ballast water inlet in the heater equivalent module, and the outlet position of the last shell side of the tubular heat exchanger is regarded as the ballast water outlet in the equivalent model, so as to serve as an equivalent plug flow reactor model; S2: Collect and obtain the spatiotemporal temperature data of the equivalent plug flow reactor model, and construct the optimal spatiotemporal temperature model for ballast water thermal treatment based on the spatiotemporal temperature data; And the construction method of the optimal temperature spatiotemporal model includes S21: Perform kernel principal component analysis on spatiotemporal temperature data to construct spatial basis functions; S22: Obtaining a solution time coefficient sequence of the spatiotemporal temperature data according to the spatial basis function; Obtain historical sequence data based on the solution time coefficient sequence combined with the plug flow reactor model; S23: defining a sequence data pair according to the solved time coefficient sequence and the historical sequence data; Divide the sequence data pairs into a data training set and a data test set according to a preset ratio; S24: Based on the sparrow optimization algorithm, a time model based on the kernel extreme learning machine is constructed according to the data training set; S25: Obtaining a trained temperature spatiotemporal model according to the spatial basis function and in combination with the constructed time model; S26: Based on the relative error and root mean square error function, the trained temperature spatiotemporal model is evaluated according to the data validation set, and the trained temperature spatiotemporal model that meets the preset evaluation index is used as the optimal temperature spatiotemporal model; For the trained temperature spatiotemporal model that does not meet the preset evaluation index, the model weights are adaptively updated based on the back propagation method, and S24 to S25 are repeatedly executed; S3: A nonlinear predictive controller for controlling the temperature of heated ballast water is constructed based on the optimal temperature spatiotemporal model to achieve spatiotemporal integrated control of the ballast water thermal treatment system.

2. A time-space integrated control method for a ballast water heat treatment system according to claim 1, characterized in that: The S21 specifically includes the following steps S211: Perform nonlinear mapping on the spatiotemporal temperature data and obtain the feature mapping space vector by combining the kernel function, which is expressed as κ1(T j , T k )= (T j ) T (T k ) Where: κ1(T j , T k ) is the kernel function, indicating T j With T k The result of the inner product operation in the feature space; (T j ) represents the vector T in the input space j The vector after mapping to the feature space; (T k ) represents the vector T in the input space k The vector after being mapped to the feature space, with k = 1, ..., n; (·) indicates nonlinear mapping; T j represents the acquired spatiotemporal temperature data and T(t i , x j ) represents the temperature response of the jth sensor at time i, i.e., the temperature of the heater equivalent module collected; T k represents the temperature response of the kth sensor at time i; L represents the total number of moments; n represents the number of sensors equally distributed in the axial direction of the multi-stage series heater equivalent module; S212: Obtaining a covariance matrix of the feature mapping space according to the feature mapping space vector; The formula for obtaining the covariance matrix is: S213: Define the eigenvalues ​​of the covariance matrix C as λ and the eigenvectors as The solution formula for obtaining the covariance matrix and eigenvector is Will (T k ) T (T k ) is regarded as a linear operator, and the solution formula is obtained based on the linear operator The equivalent formula of is Where: Represents the feature vector The weight of the linear correlation coefficient; S214: According to the covariance matrix and the equivalent formula, the intermediate process formula is obtained as follows: S215: Define kernel function K[j, k] and weight vector α, and K[j, k] = κ1(T j , T k ),α=[α1,α2,...,α n ] T , and according to the kernel function K[j, k] and the weight vector α, the intermediate process formula is rewritten as: nλKα=K 2 a Where: K represents the simplified form of the kernel function K[j, k]; S216: Define the eigenvalues ​​of the covariance matrix as and Based on the symmetry characteristics of the kernel function K, the rewritten intermediate process formula is transformed to obtain By considering the eigenvector The weight vector α is normalized by the unit length of , and the expression of the normalization is: Where: q represents the number of non-zero eigenvalues ​​in matrix k and q≤n; represents the lth eigenvalue of the covariance matrix; α represents α l is the abbreviation of ; n represents the total number of eigenvalues ​​in the matrix K; Arrange the n eigenvalues ​​in the matrix K in descending order to obtain a characteristic sequence list; And through the cumulative contribution rate calculation formula, according to the feature sequence table, obtain the cumulative threshold μ set The cumulative contribution rate μ of the first p eigenvalues p That is μ p >μ set ; And the cumulative contribution rate calculation formula is: The first p eigenvalues ​​are used as the data variance of kernel principal component analysis to determine the spatial basis function; And the expression of the spatial basis function Φ(x) is Φ(x)=|Φ1(x),Φ2(x),...,Φ p (x)] T Where: α ji represents the weight vector of the jth sensor at the i-th time; Φ j (x) represents the spatial basis function of the jth sensor and j=1,…,p,p<n.

3. A time-space integrated control method for a ballast water heat treatment system according to claim 2, characterized in that: The S22 specifically includes the following steps S221: Obtaining a solution time coefficient sequence of spatiotemporal temperature data according to a spatial basis function; The solution time coefficient sequence expression is: Y(t)=[Y1(t),Y2(t),…,Y p (t)] T Y j (t)=<Φ j (x),T(t,x)> Where: Y(t) represents the solution time coefficient sequence; <·, ·> represents the inner product operation; S222: Acquire historical sequence data according to the solved time coefficient sequence and the plug flow reactor model; The expression of the historical sequence data is: Where: X j (t i ) represents historical sequence data; d y , d u denote the hysteresis orders of the output and input of the plug flow reactor model respectively and j=1,...,p; represents the plug flow reactor model input u(t i ) is the hysteresis result, and u(t i )=[u1(t i ), u2(t i ), ..., u m (t i )] T , m represents the number of heaters in the equivalent model, i.e., the plug flow reactor model; represents the output Y of the plug flow reactor model at time i j (t) hysteresis result.

4. A time-space integrated control method for a ballast water heat treatment system according to claim 3, characterized in that: The S24 specifically includes the following steps S241: According to the sequence data {X j (t i ),Y j (t i )}, define and obtain the functional relationship for kernel extreme learning machine identification, and the expression of the functional relationship is Y j (t i )=Γ(X j (t i )) Where: Γ represents the mapping relationship between the time coefficient sequence and the space-time sequence data based on kernel extreme learning machine identification; S242: constructing a regression function model for solving strong nonlinear data based on the functional relationship for the data sequence of the data training set; The expression of the regression function model is: And meet Where: h(X j (t i )) represents the functional relationship between the input data and the hidden layer of the kernel extreme learning machine network. β j Represents the output weights between the output layer and the hidden layer of the kernel extreme learning machine network; represents the output error; C represents the penalty factor; L tr Indicates the total time before L tr The sequence data pairs corresponding to each moment; S243: Based on the Karush-Kuhn-Tucker condition, obtain the output weight β of the extreme learning machine network kernel extreme learning machine according to the data training set j , to obtain the network model of the kernel extreme learning machine; Ω j [i,k]=h(X j (t i )·h T (X j (t k ))=κ2(X j (t i ),X j (t k )) Where: H j represents the mapping matrix of the input data and Y j Represents Y j (t i ) is a simplified form; I represents the unit matrix; Ω j Ω j The simplified form of [i, k] represents the kernel matrix set according to the Mercer condition and i, k = 1, ..., L tr ; Represents the output of the kernel extreme learning machine network model; The model hyperparameters are determined by introducing the sparrow optimization algorithm into the time series identification based on the kernel extreme learning machine network model. The model hyperparameters include the penalty factor C and the kernel matrix Ω. j Kernel parameters; The temperature spatiotemporal model is obtained by reconstructing the kernel extreme learning machine network model and combining the spatial basis function; The expression of the temperature space-time model is: Where: represents the output response of the temperature space-time model; express A simplified form of .

5. A time-space integrated control method for a ballast water heat treatment system according to claim 4, characterized in that: The method for constructing a nonlinear predictive controller for controlling the temperature of heated ballast water according to the optimal temperature spatiotemporal model described in S3 specifically comprises the following steps S31: define the control time domain M and the prediction time domain P of the optimal temperature space-time model and M≤P, and set the expected control trajectory of the nonlinear predictive controller; And soften the expected control trajectory to obtain a reference state signal; The formula for obtaining the reference state signal is: R(t i+l ,x)=AR(t i+l-1 ,x)+(1 n×1 -A)W(t i+l ,x) Where: W(t i+l , x) represents the i The expected control trajectory after l samplings from time t; R( i+l , x) represents the reference state signal obtained by softening processing and 1 n×1 represents a vector of all 1s; Represents a softening matrix with all elements between [0,1]; S32: defined at t i The output of the optimal temperature spatiotemporal model at the lth sampling time is expressed as Where: express Relationship function with plug flow reactor model input; u(t i+l|i ) indicates that at t i+l Time plug flow reactor model input; S33: constructing a performance indicator target model for ensuring that the spatiotemporal model effectively tracks the reference state according to the reference state signal and the output of the optimal temperature spatiotemporal model; And the performance indicator target model includes an objective function and constraint conditions; The expression of the objective function is The constraint condition is expressed as u min ≤u(t i+l )≤u max Thu min ≤Δu(t i+h )≤Δu max u(t i+l )=u(t i+M ),when l>M Where: M1 and M2 represent weight matrices, h represents an integer parameter and h = 0, 1, ..., M-1; u min represents the minimum value of the control input of the nonlinear predictive controller; u max represents the maximum value of the control input of the nonlinear predictive controller; Δu min Indicates the minimum value of the control increment of the nonlinear predictive controller; Δu max It represents the maximum value of the control increment of the nonlinear predictive controller; The performance index target model is solved to obtain the optimal control sequence of the nonlinear predictive controller, and the first element in the optimal control sequence is selected as the input of the ballast water thermal treatment system to achieve the time-space integrated control of the ballast water thermal treatment system; The expression of the optimal control sequence is: U * (t i )=[u * (t i+1 ),…,u * (t i+P )] T Where: U * (t i ) represents the optimal control sequence of the nonlinear predictive controller; u * (t i+P ) represents the element of the optimal control sequence; u * (t i+M ) indicates that at t i+M is the optimal control input of the nonlinear predictive controller.

Images