An intelligent micro-grid system of an industrial internet platform

Abstract

The application relates to the technical field of industrial internet, and discloses an intelligent microgrid system of an industrial internet platform, which comprises a photovoltaic power supply module, a cooperative control module, a dispatch execution module and a data storage hub module, and can be additionally provided with a data acquisition module. The cooperative control module generates a future power consumption curve and a photovoltaic power supply prediction curve with a short-term fluctuation range and a long-term trend based on historical power consumption data and climate prediction data through a power consumption prediction model, and controls the photovoltaic power supply module and the dispatch execution module. The dispatch execution module matches power and generates real-time instructions in combination with a dynamic programming algorithm. The data acquisition module acquires photovoltaic data at a preset interval and superimposes a climate correction coefficient, and a state evaluation module generates an energy storage state matrix. A model training module uses a multi-dimensional convolutional recurrent network to update parameters and correct power fluctuations according to real-time data and dispatch results. The system improves energy utilization rate and stability, and is suitable for intelligent management of industrial energy.

Description

Technical Field

[0001] This invention relates to the field of industrial internet technology, specifically to an intelligent microgrid system for an industrial internet platform. Background Technology

[0002] With the rapid development of the Industrial Internet, the limitations of traditional microgrid systems in energy management are becoming increasingly apparent, making it difficult to meet the demands of industrial scenarios for efficient energy utilization, precise control, and stable supply. Traditional microgrid systems typically rely on manually set scheduling strategies or simple threshold controls, which cannot dynamically adapt to complex and changing electricity demands and weather conditions, resulting in low energy utilization, large power fluctuations, and potentially affecting the continuity and stability of industrial production.

[0003] In terms of energy supply, traditional microgrid photovoltaic power supply systems lack the ability to deeply analyze and predict climate data and historical electricity consumption data. Real-time changes in climate factors (such as sunlight intensity and ambient temperature) can significantly affect the power generation efficiency of photovoltaic arrays, and traditional systems often cannot predict these fluctuations in advance and make corresponding adjustments, resulting in poor stability of photovoltaic power supply and difficulty in accurately matching electricity demand.

[0004] In terms of data processing and scheduling execution, traditional microgrid systems rely on limited data acquisition methods, failing to comprehensively and in real-time obtain operating parameters of photovoltaic arrays and energy storage devices. For example, it is difficult to accurately collect key information such as DC voltage and current data of the photovoltaic array and equipment temperature and operating inertia parameters of the energy storage device. This prevents the system from comprehensively assessing its own operating status, thus affecting the scientific validity and effectiveness of the scheduling strategy. Furthermore, traditional scheduling execution modules lack dynamic optimization capabilities, failing to perform rapid and accurate power matching calculations between photovoltaic power supply and electricity demand based on real-time data. This can easily lead to system power imbalances, increasing dependence on the external power grid and reducing energy utilization efficiency.

[0005] In terms of model training and parameter calibration, traditional microgrid systems lack an effective model training mechanism and cannot dynamically optimize the prediction model and scheduling strategy based on actual operating data. When the system's operating conditions change (such as equipment aging or abnormal weather conditions), traditional systems cannot adjust relevant parameters in a timely manner, leading to a gradual increase in prediction errors, a decrease in the adaptability and accuracy of the scheduling strategy, and difficulty in maintaining good operating performance in the long term. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent microgrid system for an industrial internet platform to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent microgrid system for an industrial internet platform, the system comprising:

[0008] The system comprises a photovoltaic power supply module, a collaborative control module, a scheduling execution module, and a data storage center. The collaborative control module is connected to the data storage center, which stores historical electricity consumption data and climate prediction data. The collaborative control module acquires historical electricity consumption data from the data storage center, generates a future electricity consumption curve using an electricity consumption prediction model, and combines this with climate prediction data to generate a photovoltaic power supply prediction curve. The collaborative control module is also connected to the photovoltaic power supply module, sending the photovoltaic power supply prediction curve to it. The photovoltaic power supply module adjusts the inverter output parameters of the photovoltaic array based on the photovoltaic power supply prediction curve transmitted by the collaborative control module. The collaborative control module is also connected to the scheduling execution module, sending the future electricity consumption curve to it and receiving real-time scheduling instructions generated by the scheduling execution module. The scheduling execution module is connected to the data storage center, storing scheduling records therefore for use by the collaborative control module.

[0009] Preferably, it also includes a data acquisition module, a state assessment module, and a model training module; the collaborative control module is connected to the data acquisition module and the state assessment module, and is used to send acquisition instructions to the data acquisition module and the state assessment module. The acquisition instructions include obtaining real-time power supply data of the photovoltaic array and operating parameters of the energy storage device, uploading the acquired real-time data to the data storage center, and the data acquisition module and the state assessment module sending the photovoltaic power supply matrix and the energy storage state matrix to the scheduling execution module; the collaborative control module is connected to the model training module. When the state assessment module determines that the photovoltaic power supply deviation exceeds the limit, the collaborative control module requests a correction parameter set from the model training module and sends it to the scheduling execution module. The system comprises several modules; the coordinated control module generates an index from real-time data during operation and the scheduling results from the scheduling execution module, and transmits it to the model training module; the model training module includes a time series analysis network, receives correction instructions from the coordinated control module, generates a set of correction parameters matching the current operating conditions, and sends it to the coordinated control module; the coordinated control module obtains the correction parameter set and transmits it to the scheduling execution module; the scheduling execution module adjusts the charging and discharging speed of the energy storage device according to the correction parameter set to balance system power fluctuations; at the end of each scheduling cycle, the model training module receives real-time data and scheduling results from the coordinated control module and updates the parameters of the time series analysis network.

[0010] Preferably, the photovoltaic power supply prediction curve includes a short-term power supply fluctuation range and a long-term power supply trend function.

[0011] Preferably, the data acquisition module collects DC voltage and current data of the photovoltaic array at preset time intervals according to the instructions of the coordinated control module, adds climate correction coefficients, stores the data in the data storage center, converts it into photovoltaic power supply matrix data, and sends it to the scheduling execution module through the optical fiber network.

[0012] Preferably, the state assessment module periodically acquires the equipment temperature and operating inertia parameters of the energy storage device, adds the mechanical loss coefficient, stores them in the data storage center, and converts the collected parameters into an energy storage state matrix, which is then sent to the scheduling execution module via an optical fiber network.

[0013] Preferably, at the beginning of the scheduling cycle, the scheduling execution module receives the future electricity consumption curve transmitted by the collaborative control module; during the scheduling process, it receives the photovoltaic power supply matrix and energy storage status matrix transmitted by the data acquisition module and the status assessment module, uses dynamic programming method to perform power matching calculation between photovoltaic power supply and electricity demand, generates real-time scheduling instructions and feeds them back to the collaborative control module, and stores them in the data storage center according to time stamps.

[0014] Preferably, the time series analysis network in the model training module is a multi-dimensional convolutional recurrent architecture. In this architecture, the time convolution kernel of each layer corresponds to climate features at different time scales, and the recurrent unit of each layer is associated with the time dependency of historical electricity consumption data. The feature tensors output by all layers together constitute the correction parameter set. The model training module reads the climate data and electricity consumption records from the current data storage center and inputs them into the multi-dimensional convolutional recurrent architecture to initialize each layer of the network to a state that matches the current system operation. Based on the power balance constraint, the weight coefficients of each layer of the network are adjusted to generate the optimal correction parameter set, which is then sent to the collaborative control module.

[0015] Preferably, it also includes an energy storage device module, which includes an energy storage module, a bearing support module, and a power conversion module. In the multi-dimensional convolutional recurrent architecture, the bottom layer network corresponds to the inertial time parameter of the energy storage module, the middle layer network corresponds to the loss coefficient of the bearing support module, the high layer network corresponds to the response delay parameter of the power conversion module, and the final output layer corresponds to the system power deviation.

[0016] Preferably, in each layer of the multi-dimensional convolutional recurrent architecture, the feature extraction order from front to back is set as climate factors, electricity consumption time characteristics, and equipment aging parameters. Among them, climate factors are normalized vectors of light intensity and ambient temperature data; electricity consumption time characteristics are the periodic decomposition part of historical electricity consumption curves; and equipment aging parameters are a weighted combination of photovoltaic module degradation rate and energy storage device mechanical wear coefficient. According to the classification of climate factors, electricity consumption time characteristics, and equipment aging parameters, the multi-dimensional convolutional recurrent architecture configures a cross-dimensional feature fusion mechanism in the gated convolutional layer and recurrent connection layer of the network.

[0017] Preferably, the electricity consumption prediction model is constructed using a sliding window analysis method, taking the daily and weekly periods of electricity consumption data as input. After wavelet denoising, the model generates a future electricity consumption curve through an adaptive weighted fusion module. The future electricity consumption curve includes the base load portion, the peak-valley regulation portion, and the random fluctuation range. The state assessment module uses a fuzzy comprehensive evaluation method to classify the health status of the energy storage device, mapping the equipment vibration amplitude, operating eccentricity, and insulation resistance value into a health index matrix through a membership function. Subsequently, the energy storage state matrix is ​​generated through a weighted aggregation module.

[0018] Compared with the prior art, the beneficial effects of the present invention are:

[0019] In terms of energy forecasting and regulation, the collaborative regulation module combines historical electricity consumption data and climate prediction data to generate future electricity consumption curves and photovoltaic power supply prediction curves through an electricity consumption prediction model. This mechanism enables the system to predict the changing trends of electricity demand and photovoltaic power supply in advance. The photovoltaic power supply module can adjust the inverter output parameters according to the prediction curves to achieve dynamic optimization of photovoltaic power supply, improve the utilization rate of photovoltaic energy, and reduce the instability of power supply caused by climate fluctuations. At the same time, the scheduling execution module uses dynamic programming to perform power matching calculations based on future electricity consumption curves and real-time collected photovoltaic power supply matrix and energy storage status matrix, generating real-time scheduling instructions. This ensures accurate matching between photovoltaic power supply and electricity demand, effectively balances system power, reduces dependence on the external power grid, and improves the economy and reliability of energy utilization.

[0020] In terms of data acquisition and status assessment, the data acquisition module collects DC voltage and current data of the photovoltaic array at preset time intervals, and stores and transmits the data after adding a climate correction coefficient, ensuring the accuracy and real-time performance of the photovoltaic power supply data. The status assessment module periodically acquires equipment temperature, operating inertia parameters, etc., of the energy storage device, and generates an energy storage status matrix after adding a mechanical loss coefficient, achieving a comprehensive and accurate assessment of the health status of the energy storage device. This data provides the scheduling execution module with rich real-time information, enabling it to formulate more scientific and reasonable scheduling strategies. It also provides strong support for system fault early warning and maintenance, extends the service life of the equipment, and improves the stability and reliability of the system.

[0021] In terms of model training and parameter calibration, the model training module includes a multi-dimensional convolutional recurrent architecture time-series analysis network. This network updates its parameters based on real-time data and scheduling results, generating a set of calibration parameters that matches the current operating status. This architecture incorporates multi-dimensional information such as climate factors, electricity consumption time characteristics, and equipment aging parameters. Through a cross-dimensional feature fusion mechanism, it achieves in-depth analysis and prediction of the system's operating status. When the state assessment module determines that the photovoltaic power supply deviation exceeds the limit, the model training module can promptly generate a set of calibration parameters to adjust the charging and discharging speed of the energy storage device, quickly balancing system power fluctuations. This dynamic optimization mechanism enables the system to continuously adapt to environmental changes and equipment aging, improving the accuracy and adaptability of the prediction model and scheduling strategy, and ensuring the long-term stable and efficient operation of the system.

[0022] In terms of system integration and intelligence, the modules achieve data transmission and collaborative work through efficient communication methods such as fiber optic networks, forming an organic whole. The introduction of the industrial internet platform enables the system to integrate multi-source data, achieving global management and intelligent decision-making for the microgrid system. From data acquisition and predictive analysis to scheduling execution and model optimization, each link is closely connected and collaborates with each other, significantly improving the system's automation and intelligence levels, reducing manual intervention, lowering operation and maintenance costs, and providing an efficient and reliable solution for energy management in industrial scenarios. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the working principle of the intelligent microgrid system of the industrial internet platform described in this invention;

[0024] Figure 2 This is a flowchart of the data processing in the data acquisition module;

[0025] Figure 3 A flowchart for data processing in the status assessment module;

[0026] Figure 4 This is a flowchart for scheduling the execution module. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figures 1-4This invention relates to an intelligent microgrid system for an industrial internet platform. The system includes: a photovoltaic power supply module, a collaborative control module, a scheduling execution module, and a data storage center. The modules interact and transmit commands via data links. The specific implementation steps are as follows:

[0029] The coordinated control module establishes a bidirectional connection with the data storage center, which pre-stores historical electricity consumption data and climate prediction data. The coordinated control module first obtains historical electricity consumption data from the data storage center and generates future electricity consumption curves through a built-in electricity consumption prediction model; at the same time, it combines climate prediction data (such as sunlight intensity and temperature) to generate photovoltaic power supply prediction curves.

[0030] The generated photovoltaic power supply prediction curve is transmitted to the photovoltaic power supply module through a dedicated communication link. The photovoltaic power supply module adjusts the inverter output parameters of the photovoltaic array according to the curve, including but not limited to voltage, current threshold and power regulation coefficient, in order to optimize the real-time output of photovoltaic power.

[0031] A real-time communication channel is established between the coordinated control module and the scheduling execution module. The coordinated control module sends future power consumption curves to the scheduling execution module, which then performs power scheduling based on these curves and real-time collected system operation data. After generating real-time scheduling instructions, the scheduling execution module feeds these instructions back to the coordinated control module and simultaneously stores the scheduling records, timestamped, in the data storage center for later analysis by the coordinated control module. The entire process forms a closed-loop control, achieving dynamic power balance and optimized scheduling of the microgrid system.

[0032] The present invention will be further described below with reference to Examples 1 to 5:

[0033] Example 1:

[0034] The coordinated control module, acting as the central control unit, sends acquisition commands to the data acquisition module and the status assessment module via a standardized communication protocol. The data acquisition module, deployed at the photovoltaic array site and equipped with high-precision sensors and a data preprocessing unit, acquires DC voltage and current data from the photovoltaic array at preset time intervals (e.g., once per minute). After hardware filtering to remove high-frequency noise, the data preprocessing unit calculates climate correction coefficients based on real-time light intensity, ambient temperature, and other climate parameters, and stores the corrected voltage and current values ​​along with their corresponding timestamps in the data storage center. Simultaneously, the data acquisition module arranges the voltage and current data within a certain time window (e.g., 1 hour) in a time sequence, converting it into a two-dimensional photovoltaic power supply matrix containing rows (time points) and columns (electrical parameters). This matrix is ​​then transmitted in encrypted format via a fiber optic network to the scheduling execution module, providing the foundational data for real-time power matching.

[0035] The condition assessment module is connected to the sensor network of the energy storage device, periodically (e.g., hourly) acquiring real-time data such as equipment temperature and operating inertia parameters. Equipment temperature is collected via distributed temperature sensors, covering key heat-generating components of the energy storage device; operating inertia parameters are acquired jointly by vibration sensors and speed monitoring devices, reflecting the operating status of the device's mechanical components. The collected data is converted from analog to digital and then superimposed with mechanical loss coefficients (generated based on preset parameters such as equipment model and operating time) by the condition assessment module, forming intermediate data containing information such as temperature thresholds and inertia deviations, which is stored in the data storage center. Simultaneously, the condition assessment module normalizes these parameters and maps them into multi-dimensional vectors to construct an energy storage condition matrix. This matrix includes dimensions such as equipment health status level and operating efficiency index, and is synchronously transmitted to the scheduling execution module via a fiber optic network for evaluating the real-time availability and power regulation capabilities of the energy storage device.

[0036] A parameter calibration and data feedback link is established between the coordinated control module and the model training module. When the state assessment module determines that the photovoltaic power supply deviation exceeds the limit (such as a set ±5% threshold) through a preset logic algorithm (e.g., comparing the difference between the actual photovoltaic power supply and the predicted value), the coordinated control module sends a calibration request to the model training module containing information such as the current system operating time, climate conditions, and load status. Upon receiving the request, the model training module triggers its built-in time series analysis network to generate a calibration parameter set based on historical calibration records and real-time collected operating data. During the generation of the calibration parameter set, the coordinated control module continuously collects real-time data (such as the photovoltaic power supply matrix and energy storage state matrix) and the scheduling results of the scheduling execution module. It generates a unique index using a hash algorithm, binds the data with the index, and transmits it to the model training module as input samples for subsequent model training.

[0037] At the end of each scheduling cycle (e.g., 24 hours), the model training module initiates a parameter update process. This module first retrieves all real-time data and corresponding scheduling results from the data storage center for the current cycle, organizing them into a training dataset in chronological order. Then, a time series analysis network extracts features from the dataset to identify the correlation between photovoltaic power supply fluctuation patterns, energy storage device response characteristics, and scheduling strategies. During the analysis, the network adjusts the connection weights of its internal neurons to optimize its fitting ability to multi-dimensional data, ultimately updating the model parameters to improve the prediction and correction accuracy for the next cycle.

[0038] After receiving the calibration parameter set generated by the model training module, the scheduling execution module parses the charge / discharge rate adjustment instructions (such as the charging current gain coefficient and the upper limit of the discharge power) in the parameter set and transmits them to the local controller of the energy storage device via hardwired connection or wireless communication. The local controller adjusts the charge / discharge rate of the energy storage device according to the instructions. For example, when the calibration parameter set indicates a need to increase the charging rate, the controller increases the charging current threshold of the battery pack and shortens the charging time; when a need to decrease the discharge rate, it limits the output power of the inverter to prevent over-discharge of the energy storage device. Through this process, the system can respond in real time to photovoltaic power supply deviations, balancing the active and reactive power of the microgrid system through power regulation of the energy storage device, and reducing dependence on the external power grid.

[0039] The collaborative work of the data acquisition module, status assessment module, and model training module forms a closed-loop control chain of "data acquisition - status assessment - model calibration - execution adjustment." Specifically, the data acquisition module provides real-time operational data support for the system, the status assessment module enables quantitative analysis of the status of key equipment, and the model training module improves the system's adaptive capabilities through machine learning algorithms. These three modules work together with the core module to ensure the stable operation and efficient control of the microgrid system under complex operating conditions. The entire process involves no human intervention, achieving data acquisition, analysis, calibration, and execution entirely through automated mechanisms, demonstrating the intelligent and autonomous characteristics of the smart microgrid system under the industrial internet platform.

[0040] Example 2:

[0041] The photovoltaic power supply prediction curve is generated by the collaborative control module and includes two core dimensions: short-term power supply fluctuation range and long-term power supply trend function. These two dimensions serve the real-time adjustment and long-term planning of the microgrid system, respectively.

[0042] The generation of short-term power supply fluctuation ranges is based on the latest real-time climate data and the historical output characteristics of the photovoltaic array. The collaborative control module obtains short-term forecast data pushed by the meteorological service platform through the data interface (such as the predicted light intensity and cloud cover rate for the next 0.5-2 hours), and simultaneously retrieves historical output data of the photovoltaic array under the same climatic conditions from the data storage center (such as voltage and current records every 10 minutes for similar weather conditions in the past month). By analyzing the fluctuation patterns of historical data through a sliding window algorithm and combining this with changes in short-term forecast climate parameters, the power supply fluctuation range for the next short period is generated. For example, when intermittent cloud cover is predicted for the next 30 minutes, the system determines the current short-term power supply fluctuation range based on the photovoltaic output fluctuation range for similar weather conditions in historical data (such as voltage fluctuation ±5% and current fluctuation ±8%), and converts this range into an adjustment threshold for the inverter output parameters. After receiving this threshold, the photovoltaic power supply module dynamically adjusts the voltage and current loop parameters of the inverter through a PID controller to maintain the actual output power within the fluctuation range, avoiding large fluctuations in power supply caused by sudden weather changes.

[0043] The construction of the long-term power supply trend function is based on the periodic characteristics of climate prediction data and historical power supply data. The collaborative control module extracts the daily and weekly periodic components of historical electricity consumption data from the data storage center (such as the photovoltaic power supply demand corresponding to peak electricity consumption periods of 8:00-12:00 and 14:00-18:00 on weekdays), and combines it with the weather trend data for the next day to one week provided by the meteorological platform (such as consecutive sunny days and cloudy / rainy weather forecasts). Through spectral analysis methods such as Fourier transform, the periodic trend components of photovoltaic power supply are decomposed. For example, when a week of consecutive sunny days is predicted, the system analyzes the photovoltaic power supply trend within the sunny day cycle in historical data (such as the linear increase in power supply during the morning's enhanced sunlight phase and the linear decrease during the afternoon's weakened sunlight phase), and fits a polynomial trend function (such as a quadratic or cubic function) with time as the independent variable. This function not only reflects the changing trend of photovoltaic power supply over time, but also includes the rate of power change at different times (such as the derivative term), which is used to guide the photovoltaic power supply module in adjusting macroscopic parameters. Specifically, the photovoltaic power supply module adjusts the tilt angle drive parameters of the photovoltaic array in advance according to the trend function, so that the photovoltaic panels maintain the optimal light-receiving angle during peak sunshine periods. At the same time, it formulates a daily charging and discharging plan based on the remaining capacity of the energy storage device, such as increasing the energy storage charging amount on predicted peak power supply days to reserve energy for possible cloudy and rainy weather in the future.

[0044] Short-term power supply fluctuation range and long-term power supply trend function are transmitted to different control levels of the photovoltaic power supply module through the output interface of the collaborative control module. The short-term fluctuation range directly affects the real-time control loop of the inverter, achieving dynamic adjustment at the minute level; the long-term trend function is input to the global scheduling unit of the photovoltaic array for hourly to daily operation strategy optimization. The combination of the two enables the system to balance real-time performance and planning: the short-term fluctuation range ensures the rapid response of photovoltaic power supply to weather changes, reducing the impact of power surges on the microgrid system; the long-term trend function improves the overall efficiency of renewable energy absorption by pre-planning equipment operating parameters and energy storage strategies. Throughout the process, the collaborative control module continuously monitors the deviation between the actual power supply data and the predicted curve. If the deviation exceeds a preset threshold (e.g., ±10% of the short-term fluctuation range, ±5% of the long-term trend function), the data storage center is triggered to record the deviation event, and the relevant data is included in the input samples of the model training module, providing a basis for the subsequent optimization of the prediction model. This hierarchical prediction mechanism and feedback correction mechanism constitute a complete technical path for photovoltaic power supply prediction, realizing refined management and dynamic optimization of photovoltaic power output.

[0045] Example 3:

[0046] This embodiment describes in detail the working mechanism and data processing flow of the scheduling execution module. As the core of power regulation in the microgrid system, the scheduling execution module achieves dynamic balance between "photovoltaic power supply - energy storage regulation - electricity load" by receiving multi-source data, performing dynamic programming calculations, and generating real-time scheduling instructions.

[0047] At the beginning of each scheduling cycle (e.g., set to 15 minutes), the scheduling execution module obtains the future electricity consumption curve from the collaborative control module via a dedicated communication link. This curve comprises three key parts: the base load component (calculated based on the average level of historical electricity consumption data, such as the stable electricity demand of regular production equipment in a factory), the peak-valley regulation component (reflecting the expected start-up and shutdown periods and power changes of adjustable loads, such as the staggered power consumption arrangements for non-critical production lines), and the random fluctuation range (determined by statistically analyzing the extreme values ​​of fluctuations in historical data, used to cope with sudden electricity demand or power supply anomalies). The scheduling execution module parses the future electricity consumption curve into a set of power demand points on a time series, which serves as the objective function for subsequent power matching.

[0048] Upon entering the real-time scheduling phase, the scheduling execution module continuously receives the photovoltaic power supply matrix transmitted by the data acquisition module and the energy storage status matrix transmitted by the status assessment module via the fiber optic network. The photovoltaic power supply matrix contains real-time output power data of the photovoltaic array sampled at minute intervals (such as the current DC voltage, current, and converted active power values), while the energy storage status matrix contains real-time parameters of the energy storage devices (such as remaining power SOC, charge / discharge efficiency coefficient, and device health status level). The scheduling execution module aligns this real-time data with future electricity consumption curves in time and space to construct a three-dimensional data model that includes "photovoltaic power supply - energy storage status - electricity demand".

[0049] Based on this 3D data model, the scheduling execution module uses dynamic programming to calculate power matching. The objective function of dynamic programming is to minimize the system power deviation (i.e., the sum of the absolute value of photovoltaic power supply + energy storage device charging and discharging power - electricity demand). Constraints include the upper limit of the energy storage device's charging and discharging power, the upper and lower limits of the remaining power, and equipment safety operating parameters. In specific implementation, the scheduling execution module divides the scheduling cycle into multiple time sub-intervals (e.g., each sub-interval is 1 minute). Within each sub-interval, the optimal charging and discharging power of the energy storage device is calculated based on the difference between the current photovoltaic power supply and the electricity demand. For example, when the photovoltaic power supply is greater than the electricity demand and the remaining power of the energy storage device is lower than the capacity limit, the dynamic programming algorithm calculates an adjustment scheme that allows the energy storage device to charge at the maximum allowable charging power within that sub-interval. When the photovoltaic power supply is insufficient and the remaining power of the energy storage device is higher than the discharge lower limit, a scheme is calculated that allows the energy storage device to discharge at the minimum discharge power to meet the power gap. If the energy storage device cannot fully compensate for the power gap, a load grading control strategy is triggered, and some non-critical loads are cut off according to preset priorities (e.g., critical production equipment > lighting system > non-critical auxiliary equipment).

[0050] The real-time scheduling instructions generated by dynamic programming calculations include three categories: energy storage device control instructions, load regulation instructions, and data logging instructions. Energy storage device control instructions directly act on the energy storage converter (PCS), including charge / discharge power setpoints, current and voltage limit parameters, etc., and are transmitted to the local controller of the energy storage device via hardwired connection or Modbus protocol. Load regulation instructions are sent to the intelligent load switch via industrial Ethernet to control the start / stop or power level switching of adjustable loads. Data logging instructions instruct the storage of current scheduling results (such as actual charge / discharge power, load disconnection status, etc.) in a time-stamped manner (accurate to the second level) to the data storage center, forming a complete scheduling log.

[0051] Throughout the scheduling process, the scheduling execution module maintains real-time interaction with the collaborative control module: on the one hand, it feeds back the generated real-time scheduling instructions to the collaborative control module, enabling the collaborative control module to synchronously update the system operating status; on the other hand, it receives the model training module correction parameter set (such as the charge / discharge rate adjustment parameters from Example 1) forwarded by the collaborative control module and uses it as external input parameters for dynamic programming calculations to adjust the power matching strategy for the next stage. For example, when the model training module optimizes the charge / discharge efficiency parameters of the energy storage device based on historical scheduling data, the scheduling execution module updates the energy storage efficiency coefficient in the dynamic programming model in real time, making subsequent calculations closer to the actual operating characteristics of the equipment.

[0052] The scheduling records stored in the data storage hub are not only used for tracing the system's operational status but also serve as crucial input data for the model training module. At the end of each scheduling cycle, the scheduling execution module packages all scheduling records for that cycle and transmits them to the model training module. The model training module analyzes the power adjustment patterns and actual effects in the scheduling records to identify parameter deviations in the dynamic programming model (such as insufficient consideration of energy storage device response delays). It then adjusts the relevant weights of the time series analysis network to continuously optimize the scheduling strategy. The entire process follows a closed-loop flow of "data acquisition - dynamic calculation - command execution - record feedback," ensuring that the scheduling execution module can dynamically adjust power allocation based on real-time operating conditions. This maximizes the utilization of photovoltaic renewable energy and optimizes the charge-discharge cycle life of the energy storage device while ensuring the stable operation of the microgrid system.

[0053] Example 4:

[0054] This embodiment focuses on the time series analysis network of the model training module, detailing its architecture design and working principle. The time series analysis network built into the model training module adopts a multi-dimensional convolutional recurrent architecture. This architecture, through hierarchical design and cross-dimensional feature fusion, achieves in-depth analysis of multi-source data from the microgrid system and generation of correction parameters.

[0055] The multi-dimensional convolutional recurrent architecture consists of alternating temporal convolutional layers and recurrent layers at its bottom layer. Each temporal convolutional kernel is configured to capture climate features at different time scales. For example, the time window of the first convolutional kernel is set to 5 minutes to extract local features from short-term climate data (such as sudden changes in light intensity caused by cloud movement); the time window of the second convolutional kernel is set to 1 hour to analyze phased climate trends (such as the increasing pattern of morning light intensity over time). Each recurrent unit (such as an LSTM or GRU unit) is responsible for associating the temporal dependencies of historical electricity consumption data. Through iterative updates of the hidden states, it memorizes the regular fluctuations in electricity load within daily and weekly cycles (such as peak electricity consumption in the afternoon of a weekday and low load conditions on weekends). The convolutional outputs and recurrent states of each layer are concatenated to form a feature tensor containing information at different temporal granularities, providing multi-dimensional support for the generation of calibration parameter sets.

[0056] During the network initialization phase, the model training module reads the current climate data (such as normalized vectors of real-time light intensity and ambient temperature) and historical electricity consumption records from the data storage center (such as minute-by-minute load data for the past 45 minutes). After standardization, these data are input into the multi-dimensional convolutional recurrent architecture. The weight parameters of each layer of the network are initialized based on the pre-trained model and adapted to the current system operating conditions through a fine-tuning mechanism. For example, if the current system is experiencing continuous rainy weather, the weights of the convolutional kernels at the bottom layer of the network will automatically enhance the response to low light intensity features, while the recurrent units in the middle layer will strengthen their memory of abnormal fluctuation patterns in electricity load during rainy weather.

[0057] During parameter training, the model training module uses power balance constraints as the optimization objective and adjusts the weight coefficients of each layer of the network through the backpropagation algorithm. Specifically, the model uses the deviation between the real-time photovoltaic power supply prediction curve provided by the collaborative control module and the actual power supply data as the loss function, tracing back layer by layer the feature extraction links that cause the deviation. For example, if the photovoltaic power supply prediction deviation in a certain period is mainly caused by sudden changes in light intensity, the algorithm will increase the weight of the light intensity feature corresponding to the bottom convolution kernel to improve the extraction accuracy of such features; if the deviation is related to the non-periodic fluctuations of electricity load, the gating parameters of the middle-layer cyclic unit will be adjusted to enhance the ability to identify abnormal load patterns. After multiple rounds of iterative calculations, the network generates a set of correction parameters that matches the current operating conditions. This set of parameters contains the weighted combination coefficients of the feature tensors at each level, directly corresponding to control commands such as the charging and discharging speed of the energy storage device and the inverter adjustment threshold.

[0058] The parameter mapping relationship between the multi-dimensional convolutional recurrent architecture and the energy storage module is as follows: the output features of the bottom layer network correspond to the inertial time parameter of the energy storage module, which describes the response delay characteristics of the battery pack during charging and discharging (such as the time constant for the charging current to reach the set value); the features of the middle layer network are mapped to the loss coefficient of the bearing support module, which reflects the impact of mechanical component wear on the efficiency of the energy storage device (such as the proportion of energy loss caused by bearing friction); the features of the high layer network are associated with the response delay parameter of the power conversion module, which characterizes the time interval from receiving the command to completing the power adjustment of the inverter; the feature tensor of the final output layer directly corresponds to the system power deviation, and the adjustment coefficient in the correction parameter set is generated through linear transformation.

[0059] When generating the calibration parameter set, the model training module simultaneously considers the combined effects of climate factors, electricity consumption time characteristics, and equipment aging parameters. Climate factors are input into the bottom layer of the network as normalized vectors of light intensity and ambient temperature. Electricity consumption time characteristics are input into the middle layer in the form of Fourier series expansion coefficients of historical electricity consumption curves. Equipment aging parameters (such as photovoltaic module degradation rate and energy storage device mechanical wear coefficient) are input into the upper layer after weighted summation. Through a cross-dimensional feature fusion mechanism in the gated convolutional layer and the recurrent connection layer, the network dynamically adjusts the weight ratio of features in each dimension. For example, when the equipment operation log shows that the mechanical wear coefficient of the energy storage device exceeds the threshold, the fusion mechanism automatically increases the influence weight of the equipment aging parameters, making the calibration parameter set more inclined to compensate for power fluctuations caused by equipment aging.

[0060] The entire time series analysis network operates entirely based on data, eliminating the need for manually setting complex adjustment rules. The model training module continuously updates the network's training dataset by collecting real-time data from the collaborative regulation module and scheduling results from the scheduling execution module. This enables the network to adapt to dynamic operating conditions within the microgrid system, such as photovoltaic array degradation, energy storage device aging, and changes in production load. This intelligent correction mechanism, based on a multi-dimensional convolutional recurrent architecture, provides the microgrid system with cross-level regulation capabilities, from underlying device parameters to macroscopic system power, ensuring that the system maintains high prediction accuracy and regulation efficiency at different operational stages.

[0061] Example 5:

[0062] This embodiment further refines the feature extraction logic of the multi-dimensional convolutional recurrent architecture and the parameter association of the energy storage device module. Each layer of the multi-dimensional convolutional recurrent architecture follows a specific feature extraction order, sequentially processing three core inputs: climate factors, electricity consumption time characteristics, and equipment aging parameters. At the same time, it achieves deep interaction of multi-source data through a cross-dimensional feature fusion mechanism.

[0063] Regarding the feature extraction order, each network layer first receives a normalized vector of climate factors. This vector is generated from real-time monitoring data such as light intensity and ambient temperature through standardization (e.g., eliminating dimensional differences through Z-score standardization). For example, if the light intensity is 800 W / m² and the ambient temperature is 25℃, these are mapped to two-dimensional vectors in the range [-0.5, 0.3], serving as the initial input to the bottom layer network. Subsequently, the network introduces electricity consumption time characteristic data, which is obtained by periodically decomposing historical electricity consumption curves. For example, the daily electricity consumption curve is decomposed into fundamental (24-hour cycle) and harmonics (12-hour, 8-hour, etc. periodic components), and the load fluctuation pattern of each time period is represented in the form of Fourier coefficients. The equipment aging parameters are a weighted combination of the photovoltaic module attenuation rate and the energy storage device mechanical wear coefficient. The photovoltaic module attenuation rate is calculated by the open-circuit voltage attenuation slope, and the energy storage device mechanical wear coefficient is obtained by bearing vibration spectrum analysis. The two are synthesized with a weight of 0.6:0.4 and then input into the upper layers of the network.

[0064] The cross-dimensional feature fusion mechanism is implemented through gating units in gated convolutional layers and recurrent connection layers. The output of the gating unit is a weight coefficient between 0 and 1, used to dynamically adjust the contribution of climate factors, electricity consumption time characteristics, and equipment aging parameters in feature fusion. For example, when a sudden change in real-time light intensity is detected (such as a 30% decrease within 10 minutes), the lower-level gating unit automatically increases the weight of climate factors to 0.7, reducing the impact of electricity consumption time characteristics and equipment aging parameters; when the system enters a periodic equipment maintenance cycle, the higher-level gating unit increases the weight of equipment aging parameters to 0.8, prioritizing the impact of equipment status on the control strategy. This dynamic weight adjustment mechanism enables the network to flexibly adjust the feature fusion strategy according to real-time operating conditions, avoiding control bias caused by single-dimensional data dominance.

[0065] The three sub-modules of the energy storage module (energy storage module, bearing support module, and power conversion module) form a clear mapping relationship with the hierarchy of the multi-dimensional convolutional loop architecture. The inertial time parameter of the energy storage module (such as the charging time constant of the lithium battery pack) corresponds to the output characteristics of the bottom layer network. This parameter is obtained through current step response testing and reflects the dynamic response speed of the energy storage device to charging and discharging commands. The loss coefficient of the bearing support module (such as the friction loss ratio of the ball bearing) is extracted by the middle layer network. This coefficient is obtained by analyzing the acceleration signal collected by the vibration sensor through Fast Fourier Transform (FFT) and is used to evaluate the impact of the mechanical component operating state on energy storage efficiency. The response delay parameter of the power conversion module (such as the time interval between the inverter receiving the command and adjusting the output power) is output by the higher layer network. This parameter is determined by measuring the time difference between the command trigger signal and the power change waveform using an oscilloscope. The final output layer integrates the characteristics of each layer to generate a set of correction parameters directly related to the system power deviation, such as parameters including the charging and discharging power gain coefficient and the inverter adjustment delay compensation value.

[0066] During network training, the model training module optimizes the feature extraction paths and fusion weights at each level through the backpropagation algorithm. For example, when it is found that the charging and discharging speed regulation of the energy storage device lags behind the power deviation, the algorithm traces back to the connection weights of the bottom and middle layers of the network, increasing the weights of the coupling terms of the inertia time parameter and the loss coefficient to strengthen their synergistic effect. If the response delay of the power conversion module causes regulation overshoot, the mapping matrix between the higher layers and the output layer is adjusted, and a differential element is introduced to predict the delay effect. This hierarchical mapping and cross-layer optimization mechanism enables the network to construct a control model based on the physical characteristics of the device, avoiding the problem of insufficient control accuracy caused by ignoring the coupling relationship of the device's bottom-level parameters in traditional control methods.

[0067] In practical applications, the multi-dimensional convolutional recurrent architecture achieves self-evolution by continuously learning from historical data in the data storage hub. For example, during the low-load period of early Sunday morning, the model training module automatically starts deep training mode, using all data from the past week (including climate data, electricity consumption records, and equipment status monitoring data) to globally optimize the network parameters. During the optimization process, the network verifies layer by layer the explanatory power of the feature combination of climate factors, electricity consumption time characteristics, and equipment aging parameters on power deviation, eliminates redundant features, and strengthens key feature links to ensure that the model can maintain stable predictive performance in different seasons and different equipment aging stages. This architecture design, based on a combination of physical models and data-driven approaches, provides microgrid systems with end-to-end control capabilities from the underlying characteristics of equipment to the overall power balance of the system, realizing precise perception and intelligent adjustment of the operating status of energy storage devices.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent microgrid system for an industrial internet platform, characterized in that: It includes a photovoltaic power supply module, a collaborative control module, a scheduling execution module, and a data storage center. The collaborative control module is connected to the data storage center, which stores historical electricity consumption data and climate prediction data. The collaborative control module acquires the historical electricity consumption data from the data storage center, generates a future electricity consumption curve through an electricity consumption prediction model, and combines it with climate prediction data to generate a photovoltaic power supply prediction curve. The collaborative control module is also connected to the photovoltaic power supply module and sends the photovoltaic power supply prediction curve to the photovoltaic power supply module. The photovoltaic power supply module adjusts the inverter output parameters of the photovoltaic array according to the photovoltaic power supply prediction curve transmitted by the collaborative control module; the collaborative control module is connected to the scheduling execution module, sends the future power consumption curve to the scheduling execution module, and receives the real-time scheduling instructions generated by the scheduling execution module; the scheduling execution module is connected to the data storage center, stores the scheduling records in the data storage center, and makes them available for the collaborative control module to call. It also includes a data acquisition module, a status assessment module, and a model training module. The collaborative control module is connected to the data acquisition module and the status assessment module, and is used to send acquisition instructions to the data acquisition module and the status assessment module. The acquisition instructions include obtaining real-time power supply data of the photovoltaic array and operating parameters of the energy storage device, uploading the acquired real-time data to the data storage center, and sending the photovoltaic power supply matrix and energy storage status matrix to the scheduling execution module. The collaborative control module is connected to the model training module. When the status assessment module determines that the photovoltaic power supply deviation exceeds the limit, the collaborative control module requests a correction parameter set from the model training module and sends it to the scheduling execution module. The coordinated control module generates an index from real-time data during operation and the scheduling results from the scheduling execution module, and transmits it to the model training module. The model training module includes a time series analysis network. It receives correction instructions from the coordinated control module, generates a set of correction parameters matching the current operating conditions, and sends it to the coordinated control module. The coordinated control module then transmits the correction parameter set to the scheduling execution module. The scheduling execution module adjusts the charging and discharging speed of the energy storage device according to the correction parameter set to balance system power fluctuations. After each scheduling cycle, the model training module receives real-time data and scheduling results from the coordinated control module and updates the parameters of the time series analysis network. The time series analysis network in the model training module is a multi-dimensional convolutional recurrent architecture. In this architecture, the time convolution kernel of each layer corresponds to climate features at different time scales, and the recurrent unit of each layer is associated with the time dependency of historical electricity consumption data. The feature tensors output by all layers together constitute the correction parameter set. The model training module reads climate data and electricity consumption records from the current data storage center and inputs them into the multi-dimensional convolutional recurrent architecture to initialize each layer of the network to a state that matches the current system operation. Based on the power balance constraint, the weight coefficients of each layer of the network are adjusted to generate the optimal correction parameter set, which is then sent to the collaborative control module. In each layer of the multi-dimensional convolutional recurrent architecture, the feature extraction order from front to back is set as climate factors, electricity consumption time characteristics, and equipment aging parameters. Among them, climate factors are normalized vectors of light intensity and ambient temperature data; electricity consumption time characteristics are the periodic decomposition part of historical electricity consumption curves. The equipment aging parameters are a weighted combination of the photovoltaic module degradation rate and the mechanical wear coefficient of the energy storage device; in the multi-dimensional convolutional recurrent architecture, a cross-dimensional feature fusion mechanism is configured in the gated convolutional layer and the recurrent connection layer of the network according to the classification of climate factors, electricity consumption time characteristics and equipment aging parameters.

2. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that: The photovoltaic power supply prediction curve includes the short-term power supply fluctuation range and the long-term power supply trend function.

3. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that: The data acquisition module collects DC voltage and current data of the photovoltaic array at preset time intervals according to the instructions of the coordinated control module. After superimposing the climate correction coefficient, the data is stored in the data storage center and converted into photovoltaic power supply matrix data, which is then sent to the scheduling execution module through the optical fiber network.

4. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that: The state assessment module periodically acquires the equipment temperature and operating inertia parameters of the energy storage device, adds the mechanical loss coefficient, and stores them in the data storage center. The acquired parameters are then converted into an energy storage state matrix and sent to the scheduling and execution module through the fiber optic network.

5. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that: At the start of the scheduling cycle, the scheduling execution module receives the future electricity consumption curve transmitted by the collaborative control module. During the scheduling process, it receives the photovoltaic power supply matrix and energy storage status matrix transmitted by the data acquisition module and the status assessment module, uses dynamic programming to perform power matching calculations between photovoltaic power supply and electricity demand, generates real-time scheduling instructions and feeds them back to the collaborative control module, and stores them in the data storage center according to time stamps.

6. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that, It also includes an energy storage module, which comprises an energy storage module, a bearing support module, and a power conversion module. In the multi-dimensional convolutional recurrent architecture, the bottom layer network corresponds to the inertial time parameter of the energy storage module, the middle layer network corresponds to the loss coefficient of the bearing support module, the top layer network corresponds to the response delay parameter of the power conversion module, and the final output layer corresponds to the system power deviation.

7. The intelligent microgrid system of an industrial internet platform according to claim 1, characterized in that: The electricity consumption prediction model is constructed using a sliding window analysis method. It takes the daily and weekly portions of electricity consumption data as input, and after wavelet denoising, generates a future electricity consumption curve through an adaptive weighted fusion module. This future electricity consumption curve includes the base load portion, peak-valley regulation portion, and random fluctuation range. The state assessment module uses a fuzzy comprehensive evaluation method to classify the health status of the energy storage device. It maps equipment vibration amplitude, operating eccentricity, and insulation resistance values ​​into a health index matrix through a membership function, and then generates an energy storage state matrix through a weighted aggregation module.

Images