Deep-sea aquaculture power energy load intelligent analysis method based on deep learning

Abstract

The present application relates to the technical field of power energy load analysis, and particularly relates to a deep-sea aquaculture power energy load intelligent analysis method based on deep learning. The method comprises the following steps: acquiring power multi-energy equipment data and user power consumption data of a deep-sea aquaculture power energy system; establishing an optimized operation light-diesel-storage power energy integration model based on the power multi-energy equipment data and the user power consumption data to collect power multi-energy instant operation load data; acquiring multi-modal operation perception element data of the deep-sea aquaculture power energy system; establishing an operation load transformation trend prediction model based on the multi-modal operation perception element data to generate the operation load transformation trend prediction model; and transmitting the power multi-energy instant operation load data to the operation load transformation trend prediction model to analyze power energy operation load trend characteristic data. The present application realizes accurate prediction of the operation load of the deep-sea aquaculture power energy system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power energy load analysis, and particularly relates to a deep-sea aquaculture electric power energy load intelligent analysis method based on deep learning. BACKGROUND

[0002] In the current development process of the energy field, the application scope of deep learning technology is continuously expanding, and it plays an increasingly important role in energy resource management, energy consumption prediction, intelligent scheduling and many other key aspects. Among the many aspects of energy management, load prediction and intelligent scheduling are the most important, and they have a decisive significance for improving energy utilization, reducing energy consumption, promoting energy conservation and achieving environmental protection and other goals. In the coordinated operation of multiple devices and multiple energy forms in deep-sea aquaculture electric power energy management, more precise load management is needed. With the help of algorithm model to dynamically predict the load change trend, optimize energy scheduling, and early warning of load peak or abnormal fluctuation, the development of deep-sea aquaculture towards intelligence and green is promoted. However, in the existing deep-sea aquaculture electric power energy load intelligent analysis technology, due to the strong coupling of each subsystem in the platform energy management system, the strong nonlinear characteristics of electric power energy operation are ignored, and the differences in feeding, fishing and other working conditions of the aquaculture platform are not considered, which cannot reflect the high variability of the load, resulting in insufficient precision of the aquaculture electric power energy load prediction. SUMMARY

[0003] Therefore, the present application provides a deep-sea aquaculture electric power energy load intelligent analysis method based on deep learning to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a deep-sea aquaculture electric power energy load intelligent analysis method based on deep learning comprises the following steps:

[0005] Step S1: Obtain electric power multi-energy device data and user power consumption data of the deep-sea aquaculture electric power energy system; analyze the user power consumption characteristics according to the user power consumption data to generate user power consumption characteristic data;

[0006] Step S2: Establish a demand target operation mapping relationship specific to the electric power multi-energy device based on the electric power multi-energy device data, generate a demand target operation electric power energy integrated model; based on the user power consumption characteristic data and the demand target operation electric power energy integrated model, establish an optimized operation light-diesel-storage electric power energy integrated model based on the user power consumption characteristic, and generate an optimized operation light-diesel-storage electric power energy integrated model;

[0007] Step S3: Obtain multi-modal operation perception element data of the deep-sea aquaculture power energy system; perform power multi-energy operation load influence factor characteristic analysis according to the multi-modal operation perception element data to generate power multi-energy operation load influence factor characteristic data; and establish an operation load transformation trend prediction model based on the power multi-energy operation load influence factor characteristic data to generate the operation load transformation trend prediction model;

[0008] Step S4: Collect power multi-energy real-time operation load data by optimizing the operation of the light-diesel-storage power energy integrated model; transmit the power multi-energy real-time operation load data to the operation load transformation trend prediction model to perform power energy operation load trend characteristic analysis and generate power energy operation load trend characteristic data; and perform power energy load collaborative control operations for deep-sea aquaculture based on the power energy operation load trend characteristic data.

[0009] Further, step S1 includes the following steps:

[0010] Step S11: Obtain power multi-energy equipment data and user electricity data of the deep-sea aquaculture power energy system;

[0011] Step S12: Perform electricity period analysis according to the user electricity data to generate electricity period data, and perform period difference electricity load time series slicing processing on the user electricity data through the electricity period data to generate time-period user electricity load time series slicing data;

[0012] Step S13: Perform load time series characteristic extraction according to the time-period user electricity load time series slicing data to obtain time-period user electricity load time series characteristic data;

[0013] Step S14: Perform user electricity behavior clustering analysis according to the time-period user electricity load time series characteristic data to generate user electricity behavior clustering data;

[0014] Step S15: Perform electricity mode association characteristic analysis according to the user electricity behavior clustering data to generate electricity mode association characteristic data;

[0015] Step S16: Perform user electricity characteristic analysis on the time-period user electricity load time series characteristic data through the electricity mode association characteristic data to generate user electricity characteristic data.

[0016] Further, the power multi-energy equipment data of step S11 includes photovoltaic power energy equipment data, diesel power energy equipment data, and energy storage power energy equipment data.

[0017] Further, step S2 includes the following steps:

[0018] Step S21: Analyze the power multi-energy topology structure according to the power multi-energy equipment data, and generate power multi-energy topology structure data;

[0019] Step S22: Analyze the power multi-energy connection relationship of the power multi-energy equipment data, and generate power multi-energy connection relationship data;

[0020] Step S23: Perform network topology modeling processing of the power multi-energy connection according to the power multi-energy connection relationship data and the power multi-energy topology structure data, and generate a power multi-energy connection network topology model;

[0021] Step S24: Perform energy-specific mechanism operation analysis according to the power multi-energy connection network topology model, and generate energy-specific mechanism operation data;

[0022] Step S25: Perform mechanism operation constraint identification on the energy-specific mechanism operation data, and generate constraint energy-specific mechanism operation data;

[0023] Step S26: Perform power multi-energy optimization operation objective function analysis according to the power multi-energy connection network topology model, and generate a power multi-energy optimization operation objective function;

[0024] Step S27: Transfer the constraint energy-specific mechanism operation data and the power multi-energy optimization operation objective function to the power multi-energy connection network topology model to perform mapping processing of the operation constraint condition and the optimization operation objective function, and generate a demand target operation power energy integration model;

[0025] Step S28: Perform power consumption multi-scenario boundary condition analysis according to user power consumption characteristic data, and generate power consumption multi-scenario boundary condition data;

[0026] Step S29: Transfer the power consumption multi-scenario boundary condition data to the demand target operation power energy integration model to perform intelligent optimization processing of the operation parameters of the light-diesel-storage power energy integration, and generate an optimized operation light-diesel-storage power energy integration model.

[0027] Further, step S29 includes the following steps:

[0028] Transfer the power consumption multi-scenario boundary condition data to the demand target operation power energy integration model to perform power energy integration simulation operation characteristic analysis of each scenario, and generate power energy integration simulation operation characteristic data;

[0029] According to the power energy integration simulation operation characteristic data, the power energy integration optimization configuration parameters of each scenario are analyzed to generate the power energy integration optimization configuration parameters. The power energy integration optimization configuration parameters are used to intelligently optimize the operation parameters of the solar-diesel-storage power energy integration of the demand target operation power energy integration model to generate the optimized operation solar-diesel-storage power energy integration model.

[0030] Furthermore, step S3 includes the following steps:

[0031] Step S31: Acquire multimodal operation perception element data of the deep-sea aquaculture power energy system, wherein the multimodal operation perception element data includes environmental perception data, power multi-energy operation load data, and power multi-energy equipment status data;

[0032] Step S32: preprocessing the standard multimodal operation perception element data to obtain the standard multimodal operation perception element data;

[0033] Step S33: performing load correlation data analysis of electric multi-energy operation based on the standard multi-modal operation perception element data to generate electric multi-energy operation load correlation data, and performing electric multi-energy operation load influencing factor characteristic analysis on the electric multi-energy operation load correlation data to generate electric multi-energy operation load influencing factor characteristic data;

[0034] Step S34: performing an operation load cognitive rule feature analysis based on the preset power multi-energy load operation cognitive architecture and the power multi-energy load operation influencing factor feature data to generate operation load cognitive rule feature data;

[0035] Step S35: performing Gaussian mixture analysis on the operating load cognitive rule feature data to generate the optimal number of operating load components, and establishing a Gaussian mixture model architecture of the operating load characteristics by configuring a preset Gaussian mixture algorithm based on the optimal number of operating load components to obtain a Gaussian mixture operating load characteristic model architecture;

[0036] Step S36: Based on the characteristic data of the power multi-energy operation load influencing factors and the characteristic data of the operation load cognitive rules, a trend prediction model of the operation load transformation is established for the Gaussian mixture operation load characteristic model architecture to generate an operation load transformation trend prediction model.

[0037] Furthermore, step S36 includes the following steps:

[0038] Step S361: extracting the operating load transformation time series hierarchical features from the operating load recognition rule feature data to obtain the operating load transformation time series hierarchical feature data;

[0039] Step S362: Based on the power multi-energy operation load influence factor characteristic data and the operation load transformation time sequence level characteristic data, operation load transformation trend fuzzy characteristic analysis and training optimization processing are performed to generate optimized operation load transformation trend fuzzy characteristic data.

[0040] Step S363: The optimized operation load transformation trend fuzzy characteristic data is transmitted to the Gaussian mixed operation load characteristic model architecture to establish a trend prediction model for operation load transformation, and an operation load transformation trend prediction model is generated.

[0041] Further, step S361 includes the following steps:

[0042] According to the power multi-energy operation load influence factor characteristic data, local time sequence level specificity analysis of the load influence factor is performed to generate load influence factor local time sequence level specificity data, and the load influence factor local time sequence level specificity data is used to design operation load time sequence level memory extraction network rules;

[0043] The operation load time sequence level memory extraction network rules are used to extract operation load transformation time sequence level characteristic data from operation load cognitive rule characteristic data.

[0044] Further, step S362 includes the following steps:

[0045] The power multi-energy operation load influence factor characteristic data is used as input data and the corresponding operation load transformation time sequence level characteristic data is used as output data to perform fuzzy characteristic logic conversion of operation load transformation trend, to obtain operation load transformation trend fuzzy characteristic data;

[0046] The operation load transformation fuzzy trend characteristic data is processed by the preset operation load transformation fuzzy logic rule base to generate membership mapping operation load transformation trend fuzzy characteristic data, and the membership mapping operation load transformation trend fuzzy characteristic data is trained and optimized by the back propagation algorithm to generate optimized operation load transformation trend fuzzy characteristic data.

[0047] Further, step S4 includes:

[0048] Based on the optimized operation light-diesel-storage power energy integration model, the deep sea aquaculture power energy system is executed for power multi-energy operation optimization, and the power multi-energy operation optimization is used to collect power multi-energy real-time operation load data.

[0049] The application has the beneficial effects that: the application realizes accurate characterization of user electricity characteristics by systematically obtaining power multi-energy equipment data and user electricity data, combining multi-dimensional analysis means, and carrying out electricity period analysis in a targeted manner. The user electricity data is time-sliced according to different periods such as feeding, daily operation and fishing, and the differentiated characteristics of the load in each period are accurately captured, solving the problem of fuzzy load description caused by neglecting the special working condition of the breeding platform in analysis.

[0050] Through load time sequence feature extraction and user electricity behavior clustering, the electricity mode correlation characteristics of different user groups are mined, realizing the transformation from raw data to rule cognition, and providing fine user demand basis for subsequent energy system modeling. The multi-energy equipment data such as photovoltaic, diesel and energy storage are clearly included, covering the typical energy composition of deep-sea aquaculture platforms, ensuring the comprehensiveness of the analysis, and the user electricity characteristic data generated finally can truly reflect the high variability of the load, laying a data foundation for the optimization matching of the energy system, and effectively improving the adaptation accuracy of energy supply and user demand.

[0051] Focusing on the modeling and optimization of integrated photovoltaic-diesel-energy storage energy system, through multi-level topology analysis, mechanism analysis and parameter optimization, an energy integration model that fits the actual deep-sea aquaculture platform is constructed, with outstanding technical features and advantages. In the model construction stage, from power multi-energy topology structure analysis to network topology modeling, the connection relationship and coupling characteristics of photovoltaic, diesel, energy storage and other subsystems are completely restored, breaking through the limitations of traditional linear models that ignore strong nonlinear characteristics, and more truly reflecting the dynamic behavior of the system.

[0052] Through energy-specific mechanism operation analysis and constraint identification, the operation boundaries and interaction rules of each energy equipment are clearly defined, and combined with the optimization objective function, the demand target operation model generated has multi-constraint and multi-objective optimization capability. Further combined with user electricity characteristic data for multi-scenario boundary condition analysis, through simulation operation and parameter optimization, the final optimization operation model can adapt to the load demand in different electricity scenarios, realizing the optimization of the energy system from static design to dynamic adaptation, and providing core technical support for solving the problems of deep-sea aquaculture platform energy regulation not being timely and low supply-demand matching precision.

[0053] By multi-dimensional data fusion and intelligent algorithm application, a high-precision trend prediction model is constructed, with distinctive technical features. At the data level, environmental perception data (such as light, wind speed, etc.), power multi-energy operation load data and equipment state data are explicitly included to form a multi-modal operation perception element system, fully covering the key factors affecting load changes. In the data processing link, data quality is ensured through preprocessing, and load correlation data is deeply analyzed to extract influencing factor characteristics, providing a solid foundation for model construction. In the model architecture, the pre-set power multi-energy load operation cognitive architecture and Gaussian mixture model are combined, with time sequence hierarchical memory extraction network rules and fuzzy feature logic conversion, realizing multi-level analysis and modeling of load transformation trends. Among them, the time sequence hierarchical feature extraction is designed for the local time sequence specificity of load influencing factors, and the fuzzy feature analysis is optimized through membership mapping and back propagation algorithm, enhancing the model's adaptability to nonlinear and highly variable loads. The finally constructed operation load transformation trend prediction model can accurately capture the dynamic load change law of the aquaculture platform under different working conditions.

[0054] The introduction of multi-modal operation perception element data fully considers the influence of environmental, equipment state and other factors on the load, avoids the deviation caused by neglecting key variables in prediction, and improves the comprehensiveness of prediction. Data preprocessing and influencing factor characteristic analysis ensure the reliability and relevance of input data, providing a high-quality training basis for the model and reducing noise interference. The modeling method combining time sequence hierarchical memory extraction, fuzzy logic and Gaussian mixture model can deeply mine the dynamic evolution mechanism of the load, especially for the nonlinear load change characteristics of the aquaculture platform in different periods such as feeding and fishing, realizing more accurate trend prediction and solving the problem of inability to reflect the high variability of the load. The optimized prediction model can provide accurate load prediction for the energy management system and data support for subsequent energy collaborative control, helping to improve the timeliness and accuracy of energy regulation and ensuring the stable and efficient operation of the aquaculture platform energy system.

[0055] Based on the optimized operation of the light-diesel-storage power energy integration model, the characteristics of different energies such as photovoltaic, diesel and energy storage can be fully utilized, and the reasonable allocation of energy can be realized through the coupling relationship. By collecting real-time operation load data and combining the operation load transformation trend prediction model for trend analysis, the dynamic changes of the load can be mastered in real time, providing accurate basis for subsequent regulation and control, solving the problem of untimely energy regulation caused by load prediction lag, and executing load collaborative control based on trend characteristic data, which can realize the dynamic matching of the energy supply side and the load side. The load side can adjust the electricity plan according to its own characteristics and dispatching information to achieve flexible control effect, effectively cope with the fluctuations caused by system strong nonlinearity and high penetration of new energy, ensure the safe and stable operation of the aquaculture platform power energy system, and improve the overall economic and environmental benefits.

[0056] Therefore, in the deep-sea aquaculture power energy load intelligent analysis method based on deep learning, the coupling relationship and strong nonlinear characteristics of each energy subsystem are fully analyzed, the limitations of traditional linear models ignoring strong nonlinear characteristics are broken through, the model is more suitable for actual operation scenarios, and through analyzing the user power mode, the power period of feeding, fishing and other aquaculture platform specific working conditions is analyzed, the load difference under different working conditions is accurately captured, and the problem that the load high variability cannot be reflected due to not considering the working condition difference is solved. Combined with deep learning technology, the load dynamic evolution mechanism is deeply mined, the load prediction accuracy is improved, and the problem of insufficient prediction accuracy caused by not fully considering system characteristics and working condition differences in the prior art is effectively solved. Through the combination of the optimized operation model and the trend prediction model, collaborative control is realized, the load change can be responded in real time, the fluctuation caused by the strong nonlinearity of the system is coped with, and the stable and efficient operation of the energy system is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is a step flow schematic diagram of the deep-sea aquaculture power energy load intelligent analysis method based on deep learning.

[0058] Figure 2 is Figure 1 is a detailed implementation step flow schematic diagram of step S1 in the deep-sea aquaculture power energy load intelligent analysis method based on deep learning.

[0059] Figure 3 is Figure 1 is a detailed implementation step flow schematic diagram of step S2 in the deep-sea aquaculture power energy load intelligent analysis method based on deep learning.

[0060] Figure 4 is Figure 1 is a detailed implementation step flow schematic diagram of step S3 in the deep-sea aquaculture power energy load intelligent analysis method based on deep learning.

[0061] The realization of the object, functional characteristics and advantages of the present application will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0062] The technical method of the present application will be described clearly and completely below in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the present application.

[0063] In addition, the accompanying drawings are only schematic and are non-drawing to scale unless otherwise specified, and are merely meant to aid in understanding the present application and are a part of this disclosure. Like reference numbers in the drawings are intended to represent the same or similar elements unless otherwise stated. Certain repetitive description of these elements can be omitted for sake of brevity. Some of the blocks in the drawings are functional entities that may not necessarily have a corresponding physical or logical entity in an implantation, and can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods. The term "and / or" as used herein includes any and all combinations of one or more of the associated associated items.

[0064] To achieve the above object, the present application provides a deep-sea aquaculture power energy load intelligent analysis method based on deep learning, and in the embodiment of the present application, please refer to Figures 1 to 4 The present application provides a deep-sea aquaculture power energy load intelligent analysis method based on deep learning, and in the embodiment of the present application, please refer to Figure 1 The present application provides a deep-sea aquaculture power energy load intelligent analysis method based on deep learning, and in the embodiment of the present application, please refer to

[0065] Based on this, the present application provides a deep-sea aquaculture power energy load intelligent analysis method based on deep learning to solve at least one of the above technical problems.

[0066] To achieve the above object, the present application provides a deep-sea aquaculture power energy load intelligent analysis method based on deep learning, and in the embodiment of the present application, please refer to

[0067] Step S1: obtaining the power multi-energy equipment data and user power consumption data of the deep-sea aquaculture power energy system; analyzing the user power consumption characteristics according to the user power consumption data to generate user power consumption characteristic data;

[0068] In the embodiment of the present application, a sensor network is deployed on a deep-sea aquaculture platform to collect real-time power multi-energy equipment data of the photovoltaic panel, such as power generation voltage, current and conversion efficiency, output power, fuel consumption and operating temperature of the diesel generator, and charging and discharging current, residual capacity and cycle number of the energy storage battery; and real-time user power consumption data of the user power consumption equipment in the aquaculture platform, such as the bait feeder, oxygenation pump, monitoring equipment and lighting system, such as real-time power, operating time and start-stop time. The user power consumption data is divided into power consumption periods according to the feeding period, daily operation period and fishing period to generate power consumption period data. According to the power consumption period data, the sliding time window method is used to slice the user power consumption data in time sequence in units of 10 minutes to obtain time-period user power consumption load time slice data. The Fourier transform is used to extract the periodic characteristics of the load, and the wavelet transform is used to extract the mutation characteristics, to obtain time-period user power consumption load time sequence characteristic data. The K-means clustering algorithm is used to cluster the time sequence characteristic data, and the devices with similar power consumption modes are classified into a category to generate user power consumption behavior clustering data. The power consumption association between different clustering data is analyzed by the association rule mining algorithm, such as the change rule of the oxygenation pump power when the bait feeder is running, to generate power consumption mode association characteristic data. Finally, the time-period time sequence characteristic data is comprehensively analyzed in combination with the association characteristic data to determine the power consumption rule of each period and each type of device, and user power consumption characteristic data is generated.

[0069] Step S2: establishing a demand target operation mapping relationship specific to the power multi-energy equipment based on the power multi-energy equipment data to generate a demand target operation power energy integration model; and establishing a user power consumption characteristic optimization operation light-diesel-storage power energy integration model based on the user power consumption characteristic data and the demand target operation power energy integration model to generate an optimized operation light-diesel-storage power energy integration model;

[0070] In the embodiment of the present application, according to the collected electric power multi-energy equipment data, the connection mode and positional relationship of devices such as photovoltaic array, diesel generator, energy storage battery, inverter and distribution box are analyzed by using a topological graph drawing method, and electric power multi-energy topology structure data is generated. The power transmission direction, wire type and rated carrying capacity between devices are determined through circuit principle analysis, and electric power multi-energy connection relationship data is generated. Based on the two kinds of data, a graph theory modeling method is used to construct an electric power multi-energy connection network topology model containing nodes (devices) and edges (connection relationship). For the photovoltaic device in the model, the relationship between light intensity and output power is analyzed according to the photoelectric conversion principle; for the diesel generator, the relationship between speed and output power is analyzed according to the working principle of internal combustion engine; for the energy storage battery, the relationship between charge and discharge rate and capacity change is analyzed according to the electrochemical principle, and energy-specific mechanism operation data is generated. According to the device nameplate parameters, the power upper limit, efficiency range, working temperature interval and other constraint conditions of each device are marked, and constraint energy-specific mechanism operation data is generated. Taking minimizing fuel consumption, maximizing photovoltaic utilization rate and prolonging energy storage battery life as the target, a multi-objective optimization function is established, and an electric power multi-energy optimization operation target function is generated. The constraint data and target function are input into the network topology model, the association between the constraint conditions and the target function is established through a mapping algorithm, and a demand target operation electric power energy integration model is generated. According to the user power consumption characteristic data, the maximum load, duration, voltage stability requirement and other boundary conditions in the feeding, daily operation and fishing scenes are analyzed, and power consumption multi-scene boundary condition data is generated. The boundary condition data is input into the demand target operation model, the energy flow in each scene is simulated, the optimal output ratio of photovoltaic, diesel and energy storage is analyzed, the optimization parameters such as inverter conversion efficiency and energy storage charge and discharge threshold are determined, the model is adjusted, and an optimized operation light-diesel-storage electric power energy integration model is generated.

[0071] Step S3: acquiring multi-modal operation perception element data of the deep sea aquaculture electric power energy system; performing electric power multi-energy operation load influence factor characteristic analysis according to the multi-modal operation perception element data to generate electric power multi-energy operation load influence factor characteristic data; and establishing an operation load transformation trend prediction model based on the electric power multi-energy operation load influence factor characteristic data to generate an operation load transformation trend prediction model;

[0072] In the embodiment of the present application, the meteorological sensor is installed on the breeding platform to collect environmental perception data such as light intensity, wind speed, temperature, humidity, etc.; the power monitoring device is used to collect real-time active power, reactive power, voltage, current and other power multi-energy operation load data; the equipment state monitoring module is used to collect photovoltaic panel cleanliness, diesel generator fault code, energy storage battery internal resistance and other power multi-energy equipment state data, to form multi-modal operation perception element data. The collected data is subjected to the Relyda criterion to remove outliers, the linear interpolation method is used to fill in missing values, and the maximum-minimum standardization method is used to map the data to the [0, 1] interval, to obtain standard multi-modal operation perception element data. The Pearson correlation coefficient method is used to analyze the correlation between environmental factors, equipment state and operation load in the standard data, such as the correlation coefficient between light intensity and photovoltaic output, the correlation coefficient between wind speed and fan (if any) output, to generate power multi-energy operation load correlation data. The principal component analysis method is used on the correlation data to extract key factors affecting the load, such as light intensity, equipment operating temperature, bait machine power, etc., to generate power multi-energy operation load influence factor feature data. Based on the preset power multi-energy load operation cognitive architecture including perception layer, fusion layer and decision layer, the influence factor feature data is input into the fusion layer for feature fusion, and the decision layer is used to determine the rules such as “diesel generator load increases at high temperature” and “photovoltaic output increases at strong light”, to generate operation load cognitive rule feature data. The Bayesian information criterion is used to determine the optimal number of components of the Gaussian mixture model, and the mean and covariance parameters of the Gaussian mixture algorithm are configured according to the number of components, to construct a Gaussian mixture operation load feature model architecture. According to the influence factor feature data, the change law of different factors at the hour, day and week time scales is analyzed, and the hierarchical structure of the time sequence hierarchical memory extraction network is designed, such as the input layer receiving the original data, the hidden layer extracting the hour-level features, and the output layer extracting the day-level features. The network is used to process the operation load cognitive rule feature data, to obtain operation load transformed time sequence hierarchical feature data. The influence factor feature data is taken as the input, and the time sequence hierarchical feature data is taken as the output, which is subjected to fuzzy logic conversion through the Takagi-Sugeno-Kang type fuzzy device, to establish fuzzy rules such as “if the light intensity is high and the temperature is suitable, the load will show an upward trend”, to generate operation load transformed trend fuzzy feature data. The membership function parameters of the fuzzy rules are adjusted through the back propagation algorithm to minimize the output error, to generate optimized operation load transformed trend fuzzy feature data, which is input into the Gaussian mixture model architecture to complete the construction of the trend prediction model of the operation load transformation.

[0073] Step S4: Collect power multi-energy instant operation load data by optimizing the operation of the light-diesel-storage power energy integration model; transmit the power multi-energy instant operation load data to the operation load transformation trend prediction model to analyze the power energy operation load trend characteristics and generate power energy operation load trend characteristic data; and perform power energy load collaborative control operation for deep sea aquaculture based on the power energy operation load trend characteristic data.

[0074] In the embodiment of the present application, based on the optimized operation of the light-diesel-storage power energy integration model, the MPPT (maximum power point tracking) strategy of the photovoltaic inverter, the start-stop threshold of the diesel generator, and the charge-discharge strategy of the energy storage battery are adjusted in real time by the energy management controller to perform power multi-energy operation optimization. During the optimization process, the real-time output of photovoltaic, the real-time power of diesel generator, and the real-time charge-discharge power of energy storage battery are collected by the Hall sensors installed at the output ends of the energy devices to obtain power multi-energy instant operation load data. The instant operation load data is input into the operation load transformation trend prediction model. The model analyzes the matching degree of the data and the historical load mode, combines the current environmental perception data (such as the light intensity change rate and the wind speed fluctuation value), and outputs the load rising / falling trend, fluctuation amplitude, peak occurrence time, and other power energy operation load trend characteristic data within 1 hour, 3 hours, and 24 hours. According to the trend characteristic data, if the load is predicted to rise within 1 hour and the photovoltaic output is sufficient, the energy management controller issues an instruction to make the energy storage battery enter the floating charge state and the diesel generator maintain standby; if the load is predicted to drop and the photovoltaic output is excessive, the energy storage battery is controlled to switch to the charging state and the inverter output voltage is adjusted to store the excess power efficiently; if the load is predicted to rise sharply and the photovoltaic output is insufficient, the diesel generator is started immediately, and the energy storage battery releases power to ensure that the total output matches the load demand. The feeding machine, oxygen pump, and other devices on the load side receive the dispatching signal from the control center and adjust the working time according to their own operation cycle, such as reducing the sampling frequency of the monitoring devices during non-emergency periods, to realize collaborative control of the power energy load.

[0075] Further, as an embodiment of the present application, referring to FIG. 1, which is a detailed step flowchart of step S1 in the embodiment, step S1 includes the following steps: Figure 2 Figure 1 In the embodiment, step S1 includes the following steps:

[0076] Step S11: Obtain power multi-energy device data and user power consumption data of the deep sea aquaculture power energy system;

[0077] ​In the embodiment of the present application, sensors are installed on the power multi-energy equipment of the deep-sea aquaculture platform, such as photovoltaic arrays, diesel generators, energy storage batteries, etc., wherein voltage sensors and current sensors are arranged at the photovoltaic panels to collect output voltage, current and power generation efficiency in real time; power sensors and fuel flow sensors are installed on the diesel generators to record output active power, reactive power and hourly fuel consumption; the energy storage battery pack is connected to an electric quantity monitoring module to collect charging and discharging current, residual capacity and cycle number; the above data collectively constitute the power multi-energy equipment data. At the same time, metering devices are installed on the power supply lines of user electrical equipment such as feeders, oxygenation pumps, water quality monitors and lighting equipment to collect the start time, running time, real-time power and stop time of each device, forming user electricity consumption data. All data are collected to a data acquisition terminal through wired transmission mode, and the transmission frequency is set to once per minute to ensure the continuity and timeliness of the data, providing complete raw data support for subsequent analysis.

[0078] Step S12: analyzing the electricity consumption period according to the user electricity consumption data, generating electricity consumption period data, and performing period difference electricity load time sequence slicing processing on the user electricity consumption data through the electricity consumption period data to generate period-difference user electricity load time sequence slicing data;

[0079] In the embodiment of the present application, according to the running time mark of each device in the user electricity consumption data, the electricity consumption period is divided into three categories: feeding period, daily operation period and fishing period. The feeding period is determined according to the fixed working period of the feeder; the daily operation period is the period excluding the feeding period and the fishing period; and the fishing period is set according to the aquaculture plan as a specific time period in the last three days of each month. Based on the electricity consumption period data, the user electricity consumption data is processed by using the fixed time interval segmentation method, taking 5 minutes as a time slicing unit, and the electricity consumption data of the feeding period, the daily operation period and the fishing period is sliced, respectively. Each slice contains the power data of all electrical equipment in the period, and period-difference user electricity load time sequence slicing data is generated. Through this period-difference slicing processing, the load changes in different aquaculture operation stages can be clearly distinguished, and the feature ambiguity caused by mixing of data in different periods is avoided.

[0080] Step S13: performing load time sequence feature extraction according to the period-difference user electricity load time sequence slicing data to obtain period-difference user electricity load time sequence feature data;

[0081] In the embodiment of the present application, for the time-of-period user power load time sequence slice data, a time series analysis method is used to extract the load time sequence characteristics. For the slice data of the feeding period, the maximum value, the minimum value, the average value and the power change rate in each slice are calculated, wherein the power change rate is obtained by the ratio of the power difference of the adjacent two slices to the time interval, reflecting the load fluctuation when the baiting machine starts and stops; for the slice data of the daily operation period, the periodic characteristics of the power data are analyzed, and the operation period of the lighting equipment, the circulating water pump and other equipment is determined by calculating the similarity of the power curve in the adjacent period; for the slice data of the fishing period, the number of times and the duration of the power peak value are counted to capture the load characteristics when the high-power equipment such as the net machine is running. The maximum value, the minimum value, the average value, the change rate, the periodicity parameter, the number of peak values and the duration are summarized to form the time-of-period user power load time sequence characteristic data, and the dynamic change law of the load in different periods is completely presented.

[0082] Step S14: performing user power behavior clustering analysis according to the time-of-period user power load time sequence characteristic data to generate user power behavior clustering data;

[0083] In the embodiment of the present application, according to the time-of-period user power load time sequence characteristic data, a clustering analysis method is used to classify the user power behavior. The power average value, the change rate and the peak duration in each period are selected as the clustering characteristic indexes, and the devices with close distances are classified into one category by calculating the Euclidean distance of different power devices on these indexes. For example, the baiting machine and the net machine show the characteristics of high power peak value and fast change rate when working, and they are classified into one category; the lighting equipment and the small monitoring equipment have stable power and slow change, and they are classified into another category; the oxygen pump and the circulating water pump have medium power and long running time, and they are classified into a third category. Through this clustering processing, the user power behavior clustering data is generated, the behavior modes of different types of power devices are distinguished, and the foundation for subsequent analysis of the power correlation of various devices is laid.

[0084] Step S15: performing power mode correlation characteristic analysis according to the user power behavior clustering data to generate power mode correlation characteristic data;

[0085] In an embodiment of the present invention, based on the clustering data of user electricity consumption behavior, an association rule analysis method is used to mine the associated characteristics of electricity consumption patterns between different clusters. The clustering data of the feeding period is analyzed, and the power changes of the aerator when the bait feeder is started are counted. The probability and amplitude of the increase in the power of the aerator after the bait feeder is started are calculated to determine the collaborative operation relationship between the two in the feeding operation; the clustering data of the daily operation period is analyzed to observe the correlation between the lighting equipment and the power of the monitoring equipment, such as the law that the power of the monitoring equipment is stable due to the improvement of the image acquisition clarity after the lighting equipment is turned on; the clustering data of the fishing period is analyzed to record the changes in the discharge power of the energy storage battery when the net machine is running, and to clarify the impact of the operation of high-power equipment on the energy storage system. These associations are presented in the form of characteristic parameters, such as the probability of collaborative operation, the ratio of power change amplitude, etc., to generate electricity consumption pattern association feature data, revealing the interactive rules of electricity consumption of different types of equipment in different periods.

[0086] Step S16: performing a user power consumption characteristic analysis on the time series characteristic data of the user power load in different periods by using the power consumption pattern association characteristic data to generate user power consumption characteristic data.

[0087] In an embodiment of the present invention, combined with the power consumption pattern correlation characteristic data, a comprehensive analysis is performed on the user power load time series characteristic data in different periods to generate user power consumption characteristic data. During the feeding period, based on the correlation characteristics of the bait feeding machine and the aeration pump, the power change rate in the time series characteristics of the different periods is corrected, and the synergistic operation coefficient of the two is added to more accurately reflect the load fluctuation characteristics of the feeding operation; during the daily operation period, based on the correlation law of the lighting equipment and the monitoring equipment, the periodic parameters in the time series characteristics are adjusted and the linkage cycle between the equipment is incorporated; during the fishing period, based on the correlation characteristics of the net lifting machine and the energy storage system, the corresponding relationship between the peak duration and the energy storage discharge rate in the time series characteristics is supplemented. Through this integrated analysis, the load time series characteristics of different periods are combined with the correlation characteristics between the equipment to form user power consumption characteristic data containing parameters such as the load mean value, fluctuation range, equipment synergy coefficient, and energy storage dependence of each period, which comprehensively depicts the power consumption patterns and characteristics of the deep-sea aquaculture platform in different operation stages.

[0088] Furthermore, the electric multi-energy equipment data in step S11 includes photovoltaic electric energy equipment data, diesel electric energy equipment data and energy storage electric energy equipment data.

[0089] Further, as an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment, step S2 includes the following steps:

[0090] Step S21: According to the power multi-energy equipment data, the power multi-energy topology structure is analyzed to generate power multi-energy topology structure data;

[0091] In the embodiment of the present application, according to the installation position, physical parameters and function identification of photovoltaic panels, diesel generators, energy storage batteries and inverters in the power multi-energy equipment data, a topological structure analysis method is used to draw a system physical connection framework. For a photovoltaic array, the arrangement order of each group of photovoltaic panels is marked according to the series-parallel combination mode, the cable direction and fixing mode between the photovoltaic panels and the combiner box are measured and recorded, and the installation distance and connection port type of the combiner box and the photovoltaic inverter are determined; for a diesel generator, the relative position of the generator and the main distribution cabinet is marked, and the cable laying path and support structure from the generator output to the distribution cabinet are recorded; for an energy storage battery pack, the arrangement layout in the battery cabin is marked according to the module grouping, and the cable length and interface specification between the battery pack and the bidirectional converter are measured. The spatial coordinates of each device are determined by a laser range finder, and the power multi-energy topology structure data including device number, coordinate position, connection port type and cable specification are generated by combining the model parameters on the device nameplate, which completely presents the physical distribution and connection framework of the three types of energy equipment on the breeding platform, and lays a spatial foundation for subsequent electrical relationship analysis.

[0092] Step S22: The power multi-energy equipment data is analyzed to generate power multi-energy connection relationship data;

[0093] In the embodiment of the present application, the electrical parameters and operation records in the power multi-energy equipment data are analyzed to determine the connection relationship between the devices. For a photovoltaic system, the direct current voltage level of the photovoltaic panel output and the combiner box input is measured, the continuity of the circuit is detected by a multimeter, the one-way transmission relationship of the current from the photovoltaic panel to the combiner box is determined, and the maximum transmission current between the combiner box and the inverter is recorded; for a diesel generator, the three-phase connection phase of the generator output and the main circuit breaker is determined by a phase detector, the line impedance between the generator and the bus is measured, and the capacity limitation of the power transmission from the generator to the bus is determined; for an energy storage system, the current flow direction between the energy storage battery and the bidirectional converter is detected by a charge-discharge tester, the bidirectional transmission characteristics of the current from the converter to the battery during charging and from the battery to the converter during discharging are determined, and the voltage conversion range of the converter and the DC bus is recorded. These electrical relationships are quantified as power multi-energy connection relationship data including transmission direction, voltage level, current limit and phase matching, which clearly reflects the interaction rules of each device at the electrical level and ensures the safety and matching of energy transmission.

[0094] Step S23: According to the power multi-energy connection relationship data and the power multi-energy topology structure data, the network topology modeling of the power multi-energy connection is processed to generate a power multi-energy connection network topology model;

[0095] In the embodiment of the present application, the power multi-energy connection relationship data and the topological structure data are combined, and a network topological model is constructed by using a graph theory modeling method. Devices are taken as nodes, the connections between devices are taken as edges, each node is given attributes such as device type (photovoltaic / diesel / energy storage), rated power, and operating state, and each edge is given attributes such as transmission direction, electrical parameter, and loss coefficient. For a photovoltaic node, the edge attribute between the photovoltaic node and the junction box node is defined as “direct current transmission, voltage XXV, and maximum current XXA”; for a diesel generator node, the edge attribute between the diesel generator node and the main distribution cabinet node is defined as “three-phase alternating current, voltage XXXV, and frequency XXHz”; and for an energy storage node, the edge attribute between the energy storage node and the bidirectional converter node is defined as “bidirectional direct current and conversion efficiency XX%”. Through the association mapping of nodes and edges, an overall network structure including a photovoltaic subnetwork (photovoltaic panel-junction box-inverter), a diesel subnetwork (generator-main circuit breaker-bus), and an energy storage subnetwork (battery-bidirectional converter-bus) is constructed, and a power multi-energy connection network topological model is formed. The model visually presents the electrical connection paths and parameter restrictions of each subsystem, can intuitively reflect the transmission path and interaction law of energy in the system, and provides a structured model basis for subsequent mechanism analysis.

[0096] Step S24: performing energy-specific mechanism operation analysis according to the power multi-energy connection network topological model, and generating energy-specific mechanism operation data;

[0097] In the embodiment of the present application, based on the power multi-energy connection network topological model, energy-specific mechanism operation analysis is performed on photovoltaic, diesel, and energy storage devices. For a photovoltaic system, according to the photoelectric conversion principle, light intensity data in different time periods are continuously collected through a light intensity sensor, the output power of the photovoltaic panel is recorded synchronously, a relationship curve of light intensity and output power is drawn, the influence law of temperature on conversion efficiency is analyzed, and the voltage-current characteristic and power output curve of the photovoltaic array under different light and temperature conditions are determined; for a diesel generator, according to the working principle of an internal combustion engine, the output power is adjusted by changing a load resistance, the fuel consumption, speed, and exhaust temperature under different powers are recorded, a relationship curve of power and fuel consumption rate is drawn, the influence of speed fluctuation on output voltage stability is analyzed, and the power response characteristic and efficiency variation law of the generator are determined; and for an energy storage system, according to the electrochemical principle, different charging and discharging currents are set through a charging and discharging tester, the battery voltage, capacity, and internal resistance variation are recorded, a relationship curve of charging and discharging depth and capacity retention rate is drawn, the influence of cycle number on battery performance attenuation is analyzed, and the charging and discharging characteristic and life variation law of the energy storage battery are determined. These analysis results are quantified as characteristic parameters, such as a variation equation of photovoltaic conversion efficiency with light and temperature, a diesel generator fuel consumption characteristic equation, and an energy storage battery capacity attenuation coefficient, and energy-specific mechanism operation data are generated, so that the internal operation mechanism of each type of energy device is completely presented.

[0098] Step S25: mechanism operation constraint identification is performed on the energy-specific mechanism operation data to generate constrained energy-specific mechanism operation data;

[0099] In the embodiments of the present application, the energy-specific mechanism operation data is subjected to mechanism operation constraint identification in combination with the physical characteristics of the equipment and the safe operation specifications. For example, for a photovoltaic system involved, the maximum output power limit is marked according to the component nameplate parameters, the working temperature range (-25℃ to 65℃) is marked according to the environmental adaptation standard, and the voltage fluctuation threshold (±3% rated value) is marked according to the circuit protection requirement, so as to ensure that the photovoltaic array operates within a safe range; for a diesel generator, the minimum stable operation power (not less than 25% of the rated power) is marked according to the unit operation manual, the continuous operation time (not more than 8 hours) is marked according to the mechanical fatigue limit, and the start interval time (not less than 10 minutes) is marked according to the start system characteristics, so as to prevent the generator from being damaged due to low load or frequent start-stop; for an energy storage system, the maximum charge-discharge current (not more than 1.5 times of the rated current) is marked according to the battery safety standard, the SOC (state of charge) range (15%-90%) is marked according to the capacity retention requirement, and the maximum cycle number (not more than 3000 times) is marked according to the cycle life design, so as to avoid performance degradation of the energy storage battery due to overcharge, overdischarge or excessive cycling. The constraint conditions are associated with the characteristic parameters in the energy-specific mechanism operation data to generate constrained energy-specific mechanism operation data, which clearly defines the boundary conditions that must be followed by various energy equipment during operation, and provides a constraint basis for subsequent energy system optimization.

[0100] Step S26: power multi-energy optimization operation objective function analysis is performed according to the power multi-energy connection network topology model to generate a power multi-energy optimization operation objective function;

[0101] In the embodiment of the present application, according to the power multi-energy connection network topology model, combined with the operation demand of the deep sea aquaculture platform, the power multi-energy optimization operation objective function is constructed. Taking reducing energy cost as the core target, the fuel consumption of the diesel generator, the utilization rate of photovoltaic, and the maintenance cost of the energy storage battery are included in the objective function, the weight coefficients of each parameter are determined through cost accounting to form a cost optimization sub-function; taking improving system stability as an important target, the voltage fluctuation amplitude, frequency deviation, and power supply interruption times are included in the objective function, the allowed range of each parameter is determined according to the power system operation standard to form a stability optimization sub-function; taking prolonging the service life of the equipment as an auxiliary target, the cycle number of the energy storage battery and the start-stop number of the diesel generator are included in the objective function, the influence coefficient of each parameter is determined according to the equipment life evaluation model to form a life optimization sub-function. The three sub-functions are integrated into a comprehensive optimization objective function through weighted summation, the function variables are set as photovoltaic output, diesel generator output power, and energy storage charging and discharging power, and the value range of each variable is determined (based on the constraint energy specificity mechanism operation data). The generated power multi-energy optimization operation objective function can quantitatively reflect the comprehensive optimization direction of the energy system in terms of cost, stability, and equipment life, and provide a quantitative basis for subsequent energy dispatching strategy making. The power multi-energy optimization operation objective function can be adaptively adjusted according to the demand of the management personnel.

[0102] Step S27: The constraint energy specificity mechanism operation data and the power multi-energy optimization operation objective function are transmitted to the power multi-energy connection network topology model for mapping processing of the operation constraint condition and the optimization operation objective function, to generate a demand target operation power energy integrated model;

[0103] In the embodiment of the present application, the constraint energy specificity mechanism operation data and the power multi-energy optimization operation objective function are input into the power multi-energy connection network topology model, and the dynamic association of the constraint condition and the objective function is established through a mapping algorithm. In the model, a constraint module is set, the constraint conditions such as the maximum output power of photovoltaic, the minimum stable power of the diesel generator, and the SOC range of the energy storage are converted into mathematical inequalities, and embedded into the solution domain of the objective function; an optimization module is set, and the gradient descent method is used to solve the objective function, and in the solving process, it is checked in real time whether the variables meet the constraint condition, if the photovoltaic output calculated value exceeds the maximum limit, it is automatically reduced, and if the diesel generator output is lower than the minimum power, the adjustment mechanism is triggered. Through the iterative matching of variables and constraint conditions, the optimal solution of the objective function always falls within the constraint range, forming a demand target operation power energy integrated model containing the cooperative operation rules of photovoltaic, diesel, and energy storage. The model can output an energy distribution scheme that meets the lowest cost and the highest stability under the constraint condition, such as the cooperative strategy of photovoltaic output priority utilization, diesel generator supplement peak regulation, and energy storage fluctuation suppression, and completely presents the operation logic of the energy system under the dual action of constraint and optimization.

[0104] Step S28: according to the user electricity characteristic data, performing electricity multi-scene boundary condition analysis to generate electricity multi-scene boundary condition data;

[0105] In the embodiment of the application, according to the load characteristics of different periods in the user electricity characteristic data, the scene analysis method is used to determine the electricity multi-scene boundary condition. For the feeding scene, the maximum load value and the load duration in minutes of the scene are determined according to the load peak value and the duration of the baiting machine during operation, and the allowable range of voltage fluctuation (not more than ±2% of the rated value) is set in combination with the cooperative operation requirement of the oxygenation pump; for the daily operation scene, the average load level and the minimum power supply duration in hours are determined according to the continuous operation characteristics of lighting, monitoring and other equipment, and the frequency deviation range (±0.5 Hz) is set according to the voltage tolerance of the equipment; for the fishing scene, the peak load multiple and the peak duration in seconds are determined according to the instantaneous starting characteristics of high-power equipment such as the netting machine, and the standby power supply response time (not more than 5 seconds) is set in combination with the safety operation requirement. At the same time, combined with the marine environment data, the light intensity range and the wind speed limit in different scenes are supplemented as environmental boundary conditions, such as the photovoltaic output in the night scene is treated as zero value, and the diesel generator standby start in the strong wind scene. These parameters are summarized as electricity multi-scene boundary condition data containing load parameters, time parameters, electrical parameters and environmental parameters, and the running threshold value of the energy system to be met in each scene is clearly defined.

[0106] Step S29: transmitting the electricity multi-scene boundary condition data to the demand target operation power energy integration model to perform intelligent optimization processing of the operation parameters of the light-diesel-storage power energy integration, and generating an optimized operation light-diesel-storage power energy integration model.

[0107] In the embodiment of the present application, the multi-scene boundary condition data of electricity use is input into the demand target operation power energy integration model, and the simulation optimization method is used to adjust the operation parameters of photovoltaic-diesel-storage. For the feeding scene, the model simulates the energy flow in the load peak period, adjusts the photovoltaic inverter MPPT tracking accuracy, diesel generator start-up delay and energy storage discharge rate, so that the total output of the three is accurately matched with the load peak, while ensuring that the number of diesel generator start-up is minimized; for the daily operation scene, the model simulates the energy distribution under stable load, stores the excess photovoltaic power by optimizing the energy storage charging threshold, and keeps the diesel generator in standby state, and only starts when the photovoltaic output is insufficient, thereby reducing fuel consumption; for the fishing scene, the model simulates the instantaneous peak impact, adjusts the energy storage battery discharge rate and diesel generator fast response coefficient, so that the energy storage preferentially releases the power to suppress the peak, and the diesel generator synchronously follows up to supplement, thereby ensuring the voltage stability. Through the parameter iteration of multi-scene simulation, the optimal combination of photovoltaic utilization rate, diesel power generation efficiency and energy storage charging and discharging depth in each scene is determined, and the optimized operation photovoltaic-diesel-storage power energy integration model is generated. The model can automatically switch the parameter configuration according to the scene, such as increasing the priority of energy storage discharge during feeding, and increasing the weight of photovoltaic utilization during daily operation, thereby realizing the efficient adaptation of the energy system in different scenes.

[0108] Further, step S29 comprises the following steps:

[0109] The multi-scene boundary condition data of electricity use is transmitted to the demand target operation power energy integration model for power energy integration simulation operation characteristic analysis of each scene, and power energy integration simulation operation characteristic data is generated.

[0110] According to the power energy integration simulation operation characteristic data, the power energy integration optimization configuration parameter analysis of each scene is performed, the power energy integration optimization configuration parameter is generated, and the intelligent optimization processing of the photovoltaic-diesel-storage power energy integration operation parameters is performed on the demand target operation power energy integration model through the power energy integration optimization configuration parameter, thereby generating the optimized operation photovoltaic-diesel-storage power energy integration model.

[0111] In the embodiment of the present application, the power consumption multi-scene boundary condition data is input into the demand target operation power energy integration model, and the time sequence simulation method is used to perform power energy integration simulation operation on each scene. For the feeding scene, according to the set feeding machine working period, power peak value and duration in the boundary condition, the curve of the photovoltaic output changing with the illumination in the period, the starting response process of the diesel generator, and the charging and discharging state switching of the energy storage battery are simulated in the model, and the actual utilization rate of photovoltaic every 10 minutes, the fuel consumption rate of the diesel generator, the SOC variation amplitude of the energy storage, and the system bus voltage fluctuation value are recorded. For the daily operation scene, according to the average load level and the duration, the coordinated operation of photovoltaic, diesel and energy storage within 24 hours is simulated, and the running efficiency of the diesel generator, the depth of discharge of the energy storage and the load matching degree in the period without illumination at night are recorded. For the fishing scene, according to the instantaneous peak load and the duration, the energy impact response when the high-power equipment starts is simulated, and the maximum discharge current of the energy storage battery, the power ramping rate of the diesel generator and the system frequency stability time are recorded. Through multiple simulations, the energy output curve, the equipment operation parameter and the system stability index under each scene are summarized to generate the power energy integration simulation operation characteristic data, and the dynamic operation characteristics of the light-diesel-storage system under different scenes are completely presented. Based on the power energy integration simulation operation characteristic data, the parameter optimization method is used to analyze the power energy integration optimization configuration parameters of each scene. For the feeding scene, the photovoltaic utilization rate under different photovoltaic inverter MPPT tracking frequencies, the total fuel consumption under different diesel generator starting thresholds, and the voltage stability under different energy storage discharge initial SOCs are compared to determine the inverter tracking frequency, the generator starting threshold and the energy storage initial SOC that make the photovoltaic utilization rate highest, the fuel consumption least and the voltage fluctuation smallest. For the daily operation scene, the energy storage capacity utilization rate corresponding to different energy storage charging cutoff voltages and the energy loss corresponding to different diesel generator standby powers are analyzed to determine the charging cutoff voltage and standby power that take into account the energy storage life and diesel energy consumption. For the fishing scene, the peak suppression effect under different energy storage discharge rates and the frequency deviation corresponding to different diesel generator response delays are evaluated to determine the discharge rate and response delay parameters that can quickly suppress the peak and ensure the frequency stability. These parameters are arranged as the power energy integration optimization configuration parameters including the photovoltaic equipment adjustment parameters, the diesel generator control parameters and the energy storage system operation parameters, which are substituted into the demand target operation power energy integration model to replace the default parameters in the original model, the intelligent optimization processing of the operation parameters of the light-diesel-storage power energy integration is completed, the optimized operation light-diesel-storage power energy integration model is generated, and the balance between energy efficient utilization and system stable operation of the model under each scene is ensured.

[0112] Further, as an embodiment of the present application, referring to Figure 4 , a Figure 1The detailed step flowchart of step S3 is shown in the figure. In this embodiment, step S3 includes the following steps:

[0113] Step S31: Obtain multi-modal operation perception element data of the deep-sea aquaculture power energy system, wherein the multi-modal operation perception element data includes environmental perception data, power multi-energy operation load data, and power multi-energy equipment state data.

[0114] In the embodiment of the present application, a multi-dimensional perception system is built on the deep-sea aquaculture platform to collect multi-modal operation perception element data. The environmental perception data is obtained by installing meteorological monitoring equipment on the upper layer of the platform, including light intensity, sea wind speed, sea water temperature and air humidity every 10 minutes. The light intensity is captured in real time by a photodiode sensor, the wind speed is measured by a propeller type wind speed sensor, and the temperature and humidity are recorded synchronously by a waterproof temperature and humidity sensor. The power multi-energy operation load data is collected by monitoring devices deployed in the power distribution hub, covering total power, phase voltage and current every 2 minutes. The power is measured by an electromagnetic induction type mutual inductor to ensure data accuracy. The power multi-energy equipment state data is extracted by the built-in monitoring unit of the equipment. The photovoltaic panel state includes panel temperature and conversion efficiency attenuation value. The diesel generator state includes cylinder temperature, fuel pressure and running time. The energy storage battery state includes total voltage and charge-discharge cycle number. The collection interval is every 1 minute. All data is transmitted to the local storage unit through shielded cable to form multi-modal operation perception element data covering environmental parameters, load parameters and equipment state parameters, and fully covers the key variables affecting the operation of the energy system.

[0115] Step S32: Data preprocessing is performed on the standard multi-modal operation perception element data to obtain the standard multi-modal operation perception element data.

[0116] In the embodiment of the present application, the standardization preprocessing procedure is performed on the multi-modal operation perception element data to generate standard data. First, the abnormal value is eliminated. For the wind speed and light intensity in the environmental data, the Dixon test method is used to identify the data points significantly deviating from the group, and the average value of the three effective data before and after the period is replaced. For the voltage fluctuation value in the power load data, the sliding window method (window size is set to 5 collection periods) is used to identify the mutation data, and the median value in the window is replaced. Secondly, the missing value is filled. For single period missing data, the linear interpolation result of the adjacent two effective data is filled; for missing data within three consecutive periods, the arithmetic mean of the same type of data at the same period in the past is filled. Finally, the data is normalized. The light intensity in the environmental data is converted according to the maximum measurement range, the power in the power load data is scaled according to the rated power ratio, and the temperature, pressure and other parameters in the equipment state data are standardized according to the safe operation range, so that all data are uniformly mapped to the interval [0, 1], and the standard multi-modal operation perception element data is generated, and the dimensional difference of different types of data is eliminated.

[0117] Step S33: performing load correlation data analysis of power multi-energy operation according to the standard multi-modal operation perception element data, generating power multi-energy operation load correlation data, and performing power multi-energy operation load influence factor feature analysis on the power multi-energy operation load correlation data, generating power multi-energy operation load influence factor feature data;

[0118] In the embodiment of the present application, based on the standard multi-modal operation perception element data, the method combining statistical analysis and mechanism derivation is adopted to carry out load correlation analysis and influence factor feature extraction. For the light intensity in the environmental perception data and the photovoltaic output in the power multi-energy operation load data, the linear correlation degree between the two is determined by drawing a scatter plot and calculating the Pearson correlation coefficient, and at the same time, the change rate of photovoltaic output under different light intervals is derived combined with the photoelectric conversion efficiency curve; for the wind speed data and the output of the wind turbine (if configured), the segmented fitting method is adopted to analyze the output response law in the low wind speed, medium wind speed and high wind speed intervals, and the influence coefficient of wind speed on the load of the wind turbine is determined. For the diesel generator speed and output power in the power multi-energy equipment state data, according to the internal combustion engine power formula, a nonlinear mapping relationship between speed and power is established, and by continuously measuring the power value at different speeds, the correlation curve of the two is generated; for the SOC of the energy storage battery and the charging and discharging power, combined with the electrochemical characteristics, the charging and discharging efficiency difference of SOC in the intervals of 20%-50%, 50%-80% and 80%-90% is analyzed, and the constraint relationship of SOC on the energy storage load is determined, thereby generating the correlation data of the power multi-energy operation load. On this basis, the influence factor feature extraction is carried out on the correlation data: for the light intensity, the daily variation amplitude, the peak value duration and the lag response time of the photovoltaic output are calculated; for the wind speed, the instantaneous fluctuation frequency and the interference coefficient of direction change on the output of the wind turbine are extracted; for the diesel generator, the correlation characteristics of speed fluctuation rate, cylinder temperature rising rate and power fluctuation are analyzed; for the energy storage battery, the corresponding relationship between SOC change rate, charging and discharging cycle number and capacity attenuation is extracted. These characteristics are quantified as specific indexes, such as the photovoltaic output adjustment amount when the light intensity changes 10000lux per hour, the power deviation value corresponding to 1% speed fluctuation of the diesel generator, and the power multi-energy operation load influence factor feature data is generated, which accurately reflects the driving mechanism of each factor on the load change.

[0119] Step S34: performing operation load cognition rule feature analysis through the preset power multi-energy load operation cognition architecture and the power multi-energy operation load influence factor feature data, and generating operation load cognition rule feature data;

[0120] In the embodiment of the present application, the preset power multi-energy load operation cognition architecture includes three levels of perception layer, fusion layer and decision layer, and the operation load cognition rule feature analysis is carried out in combination with the power multi-energy operation load influence factor characteristic data. The perception layer receives the influence factor characteristic data such as light intensity, equipment temperature and energy storage SOC, divides the continuous data into discrete states according to the threshold value, such as dividing the light intensity into "weak (less than 30000 lux), medium (30000-70000 lux) and strong (more than 70000 lux)", and dividing the energy storage SOC into "low (less than 30%), medium (30%-70%) and high (more than 70%)". The fusion layer combines the discrete states to form the associated modes such as "high light intensity + high energy storage SOC" and "high temperature + large diesel generator load", and counts the load change direction (up / down / stable) and amplitude when each mode appears. The decision layer generates the cognition rule based on the mode statistical result of the fusion layer, for example, when the "high light intensity and in the feeding period" mode appears, the probability and average rising amplitude of the load rising are recorded to form the rule that "high light intensity + feeding period" leads to "load rising, amplitude is a set value"; when the "wind speed > 10 m / s and at night" mode appears, the diesel generator starting frequency is counted to form the rule that "high wind speed + at night" leads to "diesel generator load increase". These rules are quantified as operation load cognition rule feature data including premise condition, load change conclusion and confidence, and the associated law of influence factor combination and load change is clear.

[0121] Step S35: Gaussian mixed operation load optimal component number analysis is performed on the operation load cognition rule feature data, the operation load optimal component number is generated, and the preset Gaussian mixed algorithm is configured by the operation load optimal component number to establish the Gaussian mixed model architecture of the operation load feature, so as to obtain the Gaussian mixed operation load feature model architecture;

[0122] In the embodiment of the present application, the optimal component number of the Gaussian mixture model is determined by statistical test method for operating load cognitive rule feature data. The rule feature data is divided into three independent data sets of rising, falling and stable according to the load change type, and each data set contains rule confidence, impact factor strength and other characteristic parameters. For each data set, the component number is set from 1 to 8 in turn, the Gaussian mixture model parameters (mean, covariance, weight) under different component numbers are estimated by using the maximum likelihood algorithm, and the Bayesian information criterion value of each model is calculated. The value considers the model fitting degree and complexity, and the smaller the value, the better the model. By comparing the Bayesian information criterion values corresponding to different component numbers, the component number corresponding to the minimum value is selected as the optimal component number of the operating load, for example, the optimal component number of the load rising data set is 3, the falling data set is 4, and the stable data set is 2. According to the optimal component number, the Gaussian mixture algorithm is configured, the probability distribution of each component is allocated, and the Gaussian mixture operating load feature model architecture including the input layer (receiving rule feature data), the mixed layer (corresponding to the optimal component number) and the output layer (outputting load change probability) is constructed, so that the model can accurately capture the probability distribution characteristics of different load change types.

[0123] Step S36: Based on the power multi-energy operating load impact factor feature data and the operating load cognitive rule feature data, the trend prediction model of the operating load transformation of the Gaussian mixture operating load feature model architecture is established, and the operating load transformation trend prediction model is generated.

[0124] In the embodiment of the present application, based on the power multi-energy operating load impact factor feature data and the operating load cognitive rule feature data, the Gaussian mixture operating load feature model architecture is trained to generate a trend prediction model. The real-time parameters such as the illumination change rate and the equipment temperature fluctuation value in the impact factor feature data are used as model inputs together with the rule confidence in the cognitive rule feature data, and the future 30-minute load change trend (rising / falling / stable) is used as the output. The maximum likelihood algorithm is used to iteratively optimize the model parameters, the posterior probability of each sample belonging to different Gaussian components is calculated, and the mean and covariance of the components are adjusted, so that the prediction error of the model for historical data gradually decreases. During the training process, the model accuracy is verified once every 50 iterations, and the training is stopped when the accuracy fluctuation is less than 1% for three consecutive verifications. The finally generated operating load transformation trend prediction model can output the probability distribution of load rising, falling and stable in the next 30 minutes according to the real-time input of the impact factor and the rule feature, for example, outputting “rising probability 58%, stable 32%, falling 10%”, and can distinguish the load change law in different scenes such as feeding period and fishing period, and provide accurate trend basis for energy collaborative control.

[0125] Further, step S36 includes the following steps:

[0126] Step S361: running load transformation time sequence level feature extraction is performed on the running load cognition rule feature data to obtain running load transformation time sequence level feature data;

[0127] In the embodiment of the present application, the running load transformation time sequence level features are extracted from the running load cognition rule feature data by using the time sequence level decomposition method. First, the time scale characteristics of each factor in the power multi-energy running load influencing factor feature data are analyzed. The environmental factors such as light intensity and wind speed change at the hourly level, and the device temperature and energy storage SOC change at the minute level. The local time sequence level specificity of the load influencing factors is determined, and time sequence level division data containing hourly, minute and second levels is generated. Based on the data, a running load time sequence level memory extraction network rule is designed. The network is divided into three layers: the input layer receives the running load cognition rule feature data, the hidden layer extracts the features of the corresponding level according to the three-level division of hours, minutes and seconds, and the output layer integrates the features of each level. For the rule that "light intensity + feeding period" leads to "load rise" in the rule feature data, the starting period of load rise at the hourly level, the rise rate change at the minute level, and the moment of instantaneous peak value at the second level are extracted. These features are quantified into parameters such as time stamp, rate gradient and peak value duration by the network rule, and running load transformation time sequence level feature data is generated, which fully presents the dynamic change characteristics of the load at different time scales.

[0128] Step S362: based on the power multi-energy running load influencing factor feature data and the running load transformation time sequence level feature data, running load transformation trend fuzzy feature analysis and training optimization processing are performed to generate optimized running load transformation trend fuzzy feature data;

[0129] In the embodiment of the present application, the power multi-energy operation load influence factor characteristic data is taken as input, the operation load transformation time sequence level characteristic data is taken as output, and the fuzzy logic conversion method is used for fuzzy characteristic analysis of operation load transformation trend. A three-dimensional fuzzy logic rule library is established, the input variables are the illumination intensity change rate, the diesel generator load fluctuation value and the energy storage charge and discharge rate, and the output variable is the fuzzy category of the load transformation trend (rapid rise, slow rise, stable, slow decline, rapid decline). The input data is fuzzified, the illumination intensity change rate is divided into five fuzzy sets of "negative large, negative small, zero, positive small and positive large", and different membership functions are correspondingly used. Through the rules in the rule library, such as "if the illumination intensity change rate is positive large and the diesel generator load fluctuation value is positive small, then the load trend is slow rise", fuzzy reasoning is performed on the input data to generate fuzzy characteristic data of the operation load transformation trend. The membership function parameters of the fuzzy rules are optimized by using the back propagation algorithm, the error between the reasoning result and the actual time sequence level characteristic data is calculated, the center value and the width of the membership function are adjusted so that the error is controlled within 5%, and the optimized fuzzy characteristic data of the operation load transformation trend is generated to improve the matching degree of the characteristic and the actual load change.

[0130] Step S363: The optimized fuzzy characteristic data of the operation load transformation trend is transmitted to the Gaussian mixed operation load characteristic model architecture to establish a trend prediction model of the operation load transformation, and the trend prediction model of the operation load transformation is generated.

[0131] In the embodiment of the present application, the optimized fuzzy characteristic data of the operation load transformation trend is input into the Gaussian mixed operation load characteristic model architecture, and a probability distribution fitting method is used to establish a trend prediction model of the operation load transformation. The input layer of the model receives the membership values in the fuzzy characteristic data, the mixed layer allocates the probability weight of each fuzzy characteristic according to the determined optimal number of components (for example, 3 components for load rise), and the output layer calculates the posterior probability of each load trend category. For the "rapid rise" category fuzzy characteristic, the probability of belonging to each Gaussian component is calculated by the model to determine the typical load change curve corresponding to this category of characteristics (for example, rising by 20% within 10 minutes); for the "stable" category fuzzy characteristic, the probability distribution under each component is fitted to determine the load fluctuation range (for example, within ±3%). The mean and covariance parameters of the model are adjusted through iteration to make the probability distribution output by the model consistent with the actual distribution of the optimized fuzzy characteristic data, and finally the trend prediction model of the operation load transformation is generated. According to the input fuzzy characteristic data, the model can output the specific probability that the load is in the states of rapid rise, slow rise, stability, slow decline and rapid decline within the next 15 minutes, and provide clear trend guidance for energy regulation.

[0132] Further, step S361 includes the following steps:

[0133] According to the power multi-energy operation load influence factor characteristic data, a local time sequence level specificity analysis of the load influence factor is performed to generate local time sequence level specificity data of the load influence factor, and the local time sequence level specificity data of the load influence factor is used to design operation load time sequence level memory extraction network rules;

[0134] The operation load time sequence level memory extraction network rules are used to perform operation load transformation time sequence level feature extraction on the operation load cognitive rule characteristic data to obtain operation load transformation time sequence level feature data.

[0135] In the embodiment of the present application, according to the power multi-energy operation load influence factor characteristic data, the time series decomposition method is used to carry out the local time sequence level specificity analysis of the load influence factor. For the light intensity factor, the levels are divided according to the time intervals of 1 hour, 30 minutes and 10 minutes, the light intensity change rate at different levels is calculated, and the specificity that it presents hour-level slow fluctuation during the day and 30-minute-level stepwise rise during the feeding period is determined. For the diesel generator load factor, the levels are divided according to the time intervals of 5 minutes, 1 minute and 30 seconds, and the specificity that it presents 30-second-level sharp rise in the starting stage and 5-minute-level gentle fluctuation in the stable operation stage is analyzed. For the energy storage SOC factor, the levels are divided according to the time intervals of 2 hours, 1 hour and 20 minutes, and the specificity that it presents 2-hour-level continuous rise in the charging stage and 1-hour-level uniform decline in the discharging stage is determined. These analysis results are quantified as parameters containing time interval, change amplitude and duration to generate the local time sequence level specificity data of the load influence factor. Based on the data, the operation load time sequence level memory extraction network rule is designed, which includes level division standard (such as hour-level corresponding to ≥60-minute data window), feature extraction index (such as change rate, peak value occurrence frequency), and level association logic (such as minute-level feature needs to match hour-level trend), to ensure that the network can extract the load features of the corresponding level according to the factor specificity. The operation load time sequence level memory extraction network rule is used to perform hierarchical extraction on the operation load cognitive rule characteristic data to obtain the operation load transformation time sequence level feature data. For the rule characteristic data of “light intensity + feeding period” leading to “load rise”, according to the hour-level standard in the network rule, the load daily cumulative rise amount and the hour interval of peak value occurrence in the rule effective period are extracted; according to the 30-minute-level standard, the load rise rate and fluctuation amplitude every 30 minutes are extracted; according to the 10-minute-level standard, the load sharp rise peak value and duration at the instant of bait machine start are extracted. For the rule characteristic data of “high wind speed + night → diesel generator load increase”, according to the 5-minute-level standard, the load ramping rate after the start of the diesel generator is extracted; according to the 1-minute-level standard, the load instantaneous fluctuation value at the wind speed mutation is extracted; according to the 30-second-level standard, the load oscillation frequency before voltage stabilization is extracted. The feature parameters extracted at each level are summarized to generate the operation load transformation time sequence level feature data, which fully presents the detailed features of load transformation at different time scales.

[0136] Further, step S362 includes the following steps:

[0137] The power multi-energy operation load influence factor characteristic data is taken as input data, and the corresponding operation load transformation time sequence level feature data is taken as output data to perform fuzzy feature logic conversion on the operation load transformation trend, so as to obtain operation load transformation trend fuzzy feature data;

[0138] The fuzzy rule membership mapping processing is performed on the fuzzy trend characteristic data of the operating load conversion by using the preset operating load conversion fuzzy logic rule base, the membership mapping operating load conversion trend fuzzy characteristic data is generated, and the training optimization is performed on the membership mapping operating load conversion trend fuzzy characteristic data by using the back propagation algorithm, thereby generating the optimized operating load conversion trend fuzzy characteristic data.

[0139] In the embodiments of the present application, the illumination intensity change rate, diesel generator power fluctuation value and energy storage battery charge / discharge rate in the power multi-energy operation load influence factor feature data are taken as input data, the hourly load change amount and minute-level rising / falling rate in the corresponding operation load transformation time sequence level feature data are taken as output data, and the fuzzy logic conversion method is used for fuzzy feature extraction of the operation load transformation trend. For example, the fuzzy division standards of the input variables are set as follows: the illumination intensity change rate is divided into "negative large (≤-20% / h), negative small (-20% / h to-5% / h), zero (-5% / h to 5% / h), positive small (5% / h to 20% / h) and positive large (≥20% / h)"; the diesel generator power fluctuation value is divided into "severe fluctuation (≥±15%), medium fluctuation (±5% to ±15%) and slight fluctuation (≤±5%)"; and the energy storage charge / discharge rate is divided into "fast charging (≥10% / h), slow charging (3% / h to 10% / h), static (-3% / h to 3% / h), slow discharging (-10% / h to-3% / h) and fast discharging (≤-10% / h)". The output variables are divided into five fuzzy categories, i.e., "sharp rise (≥15% / h), slow rise (5% / h to 15% / h), stability (-5% / h to 5% / h), slow decline (-15% / h to-5% / h) and sharp decline (≤-15% / h)". The input data is converted into the corresponding output fuzzy categories through fuzzy logic operation, the operation load transformation trend fuzzy feature data is generated, and the fuzzy association between the influence factor and the load trend is realized. It is assumed that the preset operation load transformation fuzzy logic rule base contains 125 rules (5×3×5 input combinations), such as "if the illumination intensity change rate is positive large, the diesel generator power fluctuation is slight fluctuation and the energy storage charge / discharge rate is fast charging, then the load trend is sharp rise". The operation load transformation trend fuzzy feature data is input into the rule base, the membership degree value (range 0-1) of each input data belonging to the corresponding fuzzy set is calculated through the membership function, for example, the illumination intensity change rate is 25% / h at a certain moment, the membership degree of "positive large" is 1, and the membership degrees of other fuzzy sets are 0. According to the fuzzy reasoning of the rule base, the input membership degree is multiplied by the rule confidence (preset as 0.8-1.0) to obtain the output membership degree of each rule. The maximum-minimum synthesis method is used to aggregate the outputs of all rules to generate the membership mapping operation load transformation trend fuzzy feature data, which contains the membership degree distribution of each output fuzzy category. For example, the Takagi-Sugeno-Kang type fuzzy device is used, and the rules in the fuzzy rule base are as follows: for: ; wherein and are the linguistic variables in the antecedent and the consequent of the rule, respectively; is the input variable linguistic value. is the output exact variable. For a given input x=\left [ {{x}^{*}_{1},...,{x}^{*}_{n}} \right ]^{T} , the exact variable of its output is: {q}^{*}=\frac {{\Sigma}^{l}_{j=1}\left [ {{\Pi}^{n}_{i=1}{u}_{{A}^{j}_{i}}({x}^{*}_{i})\cdot {f}_{j}({x}^{*}_{1},...,{x}^{*}_{n})} \right ]} {{\Sigma}^{l}_{j=1}{\Pi}^{n}_{i=1}{\mu}_{{A}^{j}_{i}}({x}^{*}_{i})}\cdot {x}^{*}_{1},...,{x}^{*}_{n} Represents the actual measured values ​​or specific numerical values ​​of the first to nth input variables, including real-time monitoring data of the rate of change of light intensity in the characteristic data of the influencing factors of the multi-energy operation load, the actual calculated value of the power fluctuation of the diesel generator, and the specific measurement results of the charge and discharge rate of the energy storage battery; It represents the final precise output result of fuzzy reasoning, corresponding to the quantitative value of the operating load transformation trend (such as the load change rate). l represents the total number of rules in the fuzzy rule base, reflecting all possible logical relationships between input variable combinations and output results. j represents the index of the rule, ranging from 1 to 1, and is used to traverse each rule for calculation. n represents the number of input variables, corresponding to the number of influencing factors such as the light intensity change rate and diesel generator power fluctuation value. i represents the index of the input variable, ranging from 1 to n, and is used to distinguish different input variables. It represents the linguistic value corresponding to the i-th input variable in the j-th fuzzy rule. This linguistic value is a qualitative description of the characteristics of the input variable, such as "positive" or "slightly fluctuating". It is used to establish the logical association between the input variable and the output result in the fuzzy reasoning process. Represents the actual value of the i-th input variable Belongs to the corresponding language value in the jth rule The membership degree ranges from 0 to 1, reflecting the matching degree between the input value and the language description. Represents the output function of the jth rule. A backpropagation algorithm is used to train and optimize the membership mapping data. The error between the mapping result and the actual time series hierarchical feature data is calculated. The center value and width of the membership function are adjusted to reduce the error. This generates fuzzy feature data that optimizes the operating load transformation trend, improving the matching accuracy between the fuzzy feature and the actual load changes.

[0140] Further, the step S4 includes collecting the power multi-energy real-time operation load data by optimizing the operation of the photovoltaic-diesel-storage power energy integrated model.

[0141] Based on the optimized operation of the photovoltaic-diesel-storage power energy integrated model, the power multi-energy operation optimization operation is performed on the deep-sea aquaculture power energy system, and the power multi-energy real-time operation load data is collected according to the power multi-energy operation optimization operation.

[0142] In the embodiment of the present application, based on the optimized operation of the photovoltaic-diesel-storage power energy integrated model, the photovoltaic device, diesel generator and energy storage battery of the deep-sea aquaculture power energy system are executed for collaborative operation optimization operation. The optimized operation of the photovoltaic-diesel-storage power energy integrated model receives the latest user power consumption data and performs corresponding power consumption optimization. After receiving the power consumption optimization, the power operation load data in the deep-sea aquaculture power energy system is received, and through the power sensors installed on the photovoltaic combiner box, the output end of the diesel generator, the energy storage battery charging and discharging circuit and the system main bus, the photovoltaic real-time output, the diesel generator real-time output power, the energy storage battery real-time charging and discharging power and the system total active power, reactive power, voltage, current and other data are collected at a frequency of once per second. These data collectively constitute the power multi-energy real-time operation load data, which fully reflects the real-time load state of each energy device in the optimization operation.

[0143] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application being defined by the attached claims rather than the above description, therefore all variations falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the present application.

[0144] The above description is only a specific implementation of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A deep-sea aquaculture power energy load intelligent analysis method based on deep learning, characterized by, The method comprises the following steps: Step S1: obtaining power multi-energy equipment data and user power consumption data of the deep-sea aquaculture power energy system; According to the user power consumption data, the user power consumption characteristics are analyzed, and user power consumption characteristic data is generated; Step S2: based on the power multi-energy equipment data, a demand target operation mapping relationship specific to the power multi-energy equipment is established, and a demand target operation power energy integration model is generated; Based on the user power consumption characteristic data and the demand target operation power energy integration model, an optimized operation light-diesel-storage power energy integration model is established, and an optimized operation light-diesel-storage power energy integration model is generated; Step S3: obtaining multi-modal operation perception element data of the deep-sea aquaculture power energy system; according to the multi-modal operation perception element data, the characteristics of the power multi-energy operation load influencing factor are analyzed, and power multi-energy operation load influencing factor characteristic data is generated; Based on the power multi-energy operation load influencing factor characteristic data, a trend prediction model for operation load transformation is established, and an operation load transformation trend prediction model is generated; Step S3 comprises the following steps: Step S31: obtaining multi-modal operation perception element data of the deep-sea aquaculture power energy system, wherein the multi-modal operation perception element data comprises environmental perception data, power multi-energy operation load data, and power multi-energy equipment state data; Step S32: data preprocessing is performed on the standard multi-modal operation perception element data to obtain standard multi-modal operation perception element data; Step S33: according to the standard multi-modal operation perception element data, the load correlation data analysis of the power multi-energy operation is performed, the power multi-energy operation load correlation data is generated, and the power multi-energy operation load correlation data is analyzed to obtain the power multi-energy operation load influencing factor characteristic data; Step S34: through the preset power multi-energy load operation cognitive architecture and the power multi-energy operation load influencing factor characteristic data, the operation load cognitive rule characteristic analysis is performed, and the operation load cognitive rule characteristic data is generated; Step S35: Gaussian mixed operation load optimal component number analysis is performed on the operation load cognitive rule characteristic data, the operation load optimal component number is generated, and the preset Gaussian mixed algorithm is configured through the operation load optimal component number to establish a Gaussian mixed model architecture of the operation load characteristic, so as to obtain a Gaussian mixed operation load characteristic model architecture; Step S36: based on the power multi-energy operation load influencing factor characteristic data and the operation load cognitive rule characteristic data, the Gaussian mixed operation load characteristic model architecture is established, and a trend prediction model for operation load transformation is established, and an operation load transformation trend prediction model is generated; Step S4: Collect power multi-energy instant operation load data by optimizing the operation of the light-diesel-storage power energy integrated model; transmit the power multi-energy instant operation load data to the operation load transformation trend prediction model for power energy operation load trend characteristic analysis to generate power energy operation load trend characteristic data; and perform power energy load collaborative control operation for deep sea aquaculture based on the power energy operation load trend characteristic data. 2.The deep learning-based intelligent analysis method for power energy load of deep-sea farming according to claim 1, wherein, Step S1 includes the following steps: Step S11: Obtain power multi-energy equipment data and user electricity consumption data of the deep sea aquaculture power energy system; Step S12: Analyze electricity consumption period based on the user electricity consumption data to generate electricity consumption period data, and perform period difference electricity load time sequence slicing processing on the user electricity consumption data based on the electricity consumption period data to generate time-period user electricity load time sequence slicing data; Step S13: Extract load time sequence characteristics based on the time-period user electricity load time sequence slicing data to obtain time-period user electricity load time sequence characteristic data; Step S14: Perform user electricity consumption behavior clustering analysis based on the time-period user electricity load time sequence characteristic data to generate user electricity consumption behavior clustering data; Step S15: Perform electricity consumption mode correlation characteristic analysis based on the user electricity consumption behavior clustering data to generate electricity consumption mode correlation characteristic data; Step S16: Perform user electricity consumption characteristic analysis on the time-period user electricity load time sequence characteristic data based on the electricity consumption mode correlation characteristic data to generate user electricity consumption characteristic data. 3.The deep learning-based intelligent analysis method of power energy load for deep-sea farming according to claim 2, wherein, The power multi-energy equipment data in step S11 includes photovoltaic power energy equipment data, diesel power energy equipment data, and energy storage power energy equipment data. 4.The deep learning-based intelligent analysis method of power energy load for deep-sea farming according to claim 1, wherein, Step S2 includes the following steps: Step S21: Analyze the power multi-energy topology structure based on the power multi-energy equipment data to generate power multi-energy topology structure data; Step S22: Analyze the connection relationship of the power multi-energy equipment data to generate power multi-energy connection relationship data; Step S23: Perform network topology modeling processing of the power multi-energy connection based on the power multi-energy connection relationship data and the power multi-energy topology structure data to generate a power multi-energy connection network topology model; Step S24: Perform energy-specific mechanism operation analysis based on the power multi-energy connection network topology model to generate energy-specific mechanism operation data; Step S25: Identify mechanism operation constraints based on the energy-specific mechanism operation data to generate constraint energy-specific mechanism operation data; Step S26: Perform power multi-energy optimization operation objective function analysis based on the power multi-energy connection network topology model to generate a power multi-energy optimization operation objective function; Step S27: Transmit the constraint energy-specific mechanism operation data and the power multi-energy optimization operation objective function to the power multi-energy connection network topology model for mapping processing of the operation constraint conditions and the optimization operation objective function to generate a demand target operation power energy integrated model; Step S28: Perform multi-scenario boundary condition analysis of electricity consumption based on the user electricity consumption characteristic data to generate electricity consumption multi-scenario boundary condition data; Step S29: transmit the electricity multi-scene boundary condition data to the demand target operation power energy integration model for intelligent optimization processing of the operation parameters of the light-diesel-storage power energy integration to generate an optimized operation light-diesel-storage power energy integration model. 5.The deep learning-based intelligent analysis method for power energy load of deep-sea farming according to claim 4, characterized in that, Step S29 includes the following steps: transmit the electricity multi-scene boundary condition data to the demand target operation power energy integration model for simulation operation characteristic analysis of the power energy integration of each scene to generate power energy integration simulation operation characteristic data; analyze the power energy integration optimization configuration parameters of each scene according to the power energy integration simulation operation characteristic data, generate power energy integration optimization configuration parameters, and perform intelligent optimization processing of the operation parameters of the light-diesel-storage power energy integration on the demand target operation power energy integration model through the power energy integration optimization configuration parameters, to generate an optimized operation light-diesel-storage power energy integration model. 6.The deep learning-based intelligent analysis method of power energy load for deep-sea farming according to claim 1, wherein, Step S36 includes the following steps: Step S361: extract the operation load transformation time sequence level feature data from the operation load cognitive rule feature data to obtain the operation load transformation time sequence level feature data; Step S362: perform fuzzy feature analysis and training optimization processing on the operation load transformation trend based on the power multi-energy operation load influence factor feature data and the operation load transformation time sequence level feature data to generate optimized operation load transformation trend fuzzy feature data; Step S363: transmit the optimized operation load transformation trend fuzzy feature data to the Gaussian mixed operation load feature model architecture to establish a trend prediction model for operation load transformation, and generate an operation load transformation trend prediction model.

7. The deep learning-based intelligent analysis method of power energy load for deep sea farming according to claim 6, characterized in that, Step S361 includes the following steps: perform local time sequence level specificity analysis of the load influence factor based on the power multi-energy operation load influence factor feature data to generate load influence factor local time sequence level specificity data, and use the load influence factor local time sequence level specificity data to design an operation load time sequence level memory extraction network rule; extract the operation load transformation time sequence level feature data from the operation load cognitive rule feature data using the operation load time sequence level memory extraction network rule to obtain the operation load transformation time sequence level feature data. 8.The deep learning-based intelligent analysis method of power energy load for deep-sea farming according to claim 6, characterized in that, Step S362 includes the following steps: perform fuzzy feature logic conversion of the operation load transformation trend using the power multi-energy operation load influence factor feature data as input data and the corresponding operation load transformation time sequence level feature data as output data to obtain operation load transformation trend fuzzy feature data; perform fuzzy rule membership mapping processing on the operation load transformation fuzzy trend feature data through a pre-set operation load transformation fuzzy logic rule library to generate membership-mapped operation load transformation trend fuzzy feature data, and perform training optimization on the membership-mapped operation load transformation trend fuzzy feature data through a back propagation algorithm to generate optimized operation load transformation trend fuzzy feature data. 9.The deep learning-based intelligent analysis method of power energy load for deep-sea farming according to claim 1, wherein, The step S4 includes: Based on the optimized operation of light-diesel-storage power energy integration model, the power multi-energy operation optimization operation is performed on the deep sea aquaculture power energy system, and the power multi-energy real-time operation load data is collected according to the power multi-energy operation optimization operation.

Images