A photovoltaic support energy consumption feature modeling and control method based on industrial cloud computing

Abstract

This invention relates to the fields of photovoltaic new energy and industrial intelligent control technology, specifically a photovoltaic support system energy consumption modeling and control method based on industrial cloud computing. This method collects real-time operational and environmental data of the photovoltaic support cluster at the edge, uploads it to an industrial cloud platform to construct an energy consumption model that integrates static attributes and dynamic operating conditions, achieving accurate energy consumption prediction. With the dual objectives of maximizing power generation and minimizing operational energy consumption, cluster-level global optimization is completed in the cloud, generating an optimal tracking strategy and distributing it to the edge controller for execution. Through real-time data feedback, online iteration of the energy consumption model and closed-loop optimization of the strategy are achieved. This invention solves the problems of inaccurate energy consumption modeling, high energy consumption due to excessive tracking, and weak cluster collaboration capabilities in existing photovoltaic support system control, effectively improving the comprehensive power generation efficiency and full life-cycle management capabilities of photovoltaic power plants. It is applicable to complex application scenarios such as large-scale ground-mounted, mountainous, and distributed photovoltaic power plants.

Description

Technical Field

[0001] This invention belongs to the fields of photovoltaic power generation technology, industrial cloud computing technology and intelligent control technology, and in particular relates to a method for modeling and controlling the energy consumption of photovoltaic brackets based on industrial cloud computing. Background Technology

[0002] With the rapid development of the global new energy industry, photovoltaic (PV) power generation, with its advantages of being clean, renewable, and flexible in deployment, has become a core component of energy structure transformation. Tracking PV systems can adjust the orientation of PV modules in real time to maximize their solar irradiance, increasing power generation by 10%-30% compared to traditional fixed systems. This has made it the mainstream choice for large-scale ground-mounted PV power plants. Especially in the solar-rich Northwest region, the application rate of tracking PV systems has exceeded 60%, significantly improving the power generation revenue of PV power plants.

[0003] However, in the practical application of tracking photovoltaic systems, there are still many problems that urgently need to be solved, which seriously restrict the further improvement of photovoltaic power generation efficiency: Low accuracy in energy consumption modeling makes it impossible to accurately characterize the energy consumption characteristics of photovoltaic (PV) brackets. Existing energy consumption estimation methods for PV brackets mostly employ static estimation, considering only the product of motor rated power and operating time, neglecting the coupled effects of multiple factors such as environmental conditions (wind speed, temperature, irradiance), equipment status (aging degree, fault conditions), and operating modes (tracking frequency, angle adjustment range). Because the energy consumption of PV brackets exhibits strong nonlinearity and time-varying characteristics, the error of static estimation methods is typically above 10%, and even exceeds 30% under some extreme conditions, failing to provide reliable quantitative basis for control strategy optimization. For example, in windy weather, wind load significantly increases the drive energy consumption of the motor, but traditional static models cannot accurately estimate this additional energy consumption, still adjusting the angle according to conventional strategies, leading not only to energy waste but also the risk of equipment damage. After 3-5 years of operation, motor efficiency decreases and transmission mechanism wear intensifies, increasing energy consumption by 15%-20%, but static models cannot adapt to this change in a timely manner, resulting in a continuous increase in energy consumption estimation errors.

[0004] The development of industrial cloud computing technology provides a new idea for solving the above problems. It has the ability to store massive amounts of data, enabling the unified storage of all the operation data and environmental data of a photovoltaic support cluster. It has a super-strong computing power scheduling ability, which can support the construction of a high-precision energy consumption model and the rapid solution of a global optimization controller. It has the ability of cloud-edge collaboration, which can achieve the organic combination of global optimization in the cloud and local execution at the edge. However, there is currently no mature technical solution in the industry that deeply integrates industrial cloud computing with photovoltaic support energy consumption modeling and intelligent control. Existing technologies either only achieve cloud storage of data without high-precision modeling and optimization control, or only achieve local energy consumption modeling of a single support without cluster-level collaborative optimization and online iteration of the model, unable to meet the refined and intelligent management requirements of large-scale photovoltaic power stations, restricting the high-quality development of the photovoltaic power generation industry. Therefore, it is of great practical significance and application value to develop a photovoltaic support energy consumption special modeling and control method based on industrial cloud computing to solve the pain points in existing technologies and improve the net power generation revenue and intelligent management level of photovoltaic power stations. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art, and propose a photovoltaic support energy consumption special modeling and control method based on industrial cloud computing. Through the computing power and data capabilities of the industrial cloud computing platform, a high-precision multi-factor coupling energy consumption special model of photovoltaic supports is constructed. At the same time, with power generation gain and energy consumption management and control as dual objectives, cluster-level global optimization control is achieved, effectively balancing the power generation gain of photovoltaic tracking and the operation energy consumption of the support, breaking the data island of single support control, realizing the full-life cycle iterative optimization of the model, and ultimately improving the comprehensive net power generation revenue and intelligent management level of photovoltaic power stations.

[0006] To achieve the above objective, the embodiments of the present invention provide a photovoltaic support energy consumption special modeling and control method based on industrial cloud computing, and the technical solution is as follows: The photovoltaic support energy consumption special modeling and control method based on industrial cloud computing includes the following steps: S1: Data collection and preprocessing. Through the edge collection unit supporting the photovoltaic support, all the operation data of the support cluster and the field environment data are obtained. After being preprocessed by standardization at the edge, they are uploaded to the industrial cloud computing platform through the cloud-edge collaboration channel. The edge collection unit adopts a distributed deployment mode, and each photovoltaic support is equipped with an independent edge collection module, supporting the access of multi-protocol sensors. The cloud-edge collaboration channel adopts the AES-256 end-to-end encryption and breakpoint resumption mechanism, and the data upload success rate is not less than 99.9%; S2: Cloud-based energy consumption modeling. In the industrial cloud computing platform, based on the uploaded full historical and real-time data, a static basic energy consumption model and a dynamic operating condition energy consumption model of the photovoltaic support are constructed respectively. The two types of models are fused to obtain a complete energy consumption model. The static basic energy consumption model is constructed based on the equipment mechanism and factory test data, while the dynamic operating condition energy consumption model is constructed based on LSTM-Attention hybrid neural network training. The average absolute percentage error (MAPE) of the fused model is ≤2.5%, and the root mean square error (RMSE) is ≤0.08kWh. S3: Dual-objective optimization control strategy generation. Based on the completed energy consumption characteristic model and combined with photovoltaic power generation prediction data, a global optimization controller is constructed with the dual optimization objectives of maximizing the power generation gain of the photovoltaic system and minimizing the operating energy consumption of the photovoltaic support as the dual optimization objectives. The solution is completed on the industrial cloud computing platform to generate the optimal tracking control strategy of the photovoltaic support cluster. The global optimization controller adopts an improved particle swarm optimization algorithm that introduces adaptive inertia weights and mutation operations, which improves the solution accuracy by more than 15% and the solution speed by more than 20%. S4: Cloud-edge collaborative control execution. The industrial cloud computing platform will send the generated optimal tracking control strategy to the edge controller of the corresponding bracket. After the edge controller completes the local security verification of the control strategy, it will drive the bracket to perform the corresponding angle tracking action. At the same time, it will collect the operation feedback data after the action is executed in real time. The edge controller has more than 1GB of local cache and offline execution capability. The local security verification adopts a three-level verification mechanism with a verification response time of ≤100ms. S5: Online model iteration and closed-loop optimization. The edge controller periodically uploads operational feedback data to the industrial cloud computing platform. The platform performs incremental learning and parameter updates on the energy consumption model based on the new data. At the same time, it optimizes the solution parameters of the global optimization controller based on the feedback results, completing the closed-loop optimization of the control strategy. Incremental learning is only trained on new data, reducing computing power consumption by more than 60%. Meanwhile, aging parameters are updated synchronously based on the cumulative running time of the equipment and fault records, realizing full life cycle iteration of the model.

[0007] As a further aspect of the present invention, in step S1, the full-scale operation data includes the operating parameters of the photovoltaic bracket drive motor, the real-time tracking angle of the bracket, the equipment running time, fault alarm data, and the power generation data of the corresponding photovoltaic string. The site environment data includes the real-time irradiance of the site, the solar altitude angle and azimuth angle, wind speed and direction, ambient temperature, photovoltaic module temperature, and bracket installation terrain data, thereby achieving full coverage collection of all dimensions of data related to bracket energy consumption.

[0008] As a further aspect of the present invention, in step S1, edge-side standardization preprocessing is performed to remove outliers, fill in missing data, unify data format and align with the time axis on the collected raw data. Then, after downsampling the data according to a preset time interval, the data is packaged and uploaded, which reduces the bandwidth consumption of cloud-edge collaboration while ensuring data validity.

[0009] As a further aspect of the present invention, in step S2, the static basic energy consumption model is used to characterize the basic energy consumption corresponding to the inherent attributes of the photovoltaic support equipment. The inherent parameters of the photovoltaic support equipment are collected, including the rated parameters of the drive motor, the transmission efficiency of the reducer, the friction coefficient of the transmission mechanism, the self-weight and rotational inertia of the support. Combined with the equipment's factory test data and no-load operation data, the calculation relationship of the support's standby energy consumption, basic drive energy consumption, and transmission friction energy consumption is obtained by fitting, thus completing the construction of the static basic energy consumption model and realizing the accurate characterization of the inherent basic energy consumption of the support.

[0010] As a further aspect of the present invention, in step S2, the dynamic operating condition energy consumption model is used to characterize the dynamic energy consumption increment of the photovoltaic support under the influence of external operating conditions and equipment status during actual operation. The core characteristic variables affecting the dynamic energy consumption of the support are screened, and the preprocessed historical operating data is used as the training set. The static basic energy consumption is used as the benchmark value. The initial dynamic operating condition energy consumption model is trained by machine learning algorithm. After testing and verification that it meets the accuracy requirements, it is fused with the static basic energy consumption model to obtain the final energy consumption characteristic model. The core characteristic variables affecting the dynamic energy consumption of the support include the difference between the target angle and the real-time angle of the support, real-time wind speed, ambient temperature, surface irradiance of the components, cumulative operating time of the equipment, and cumulative number of actions, covering all dimensions of external environmental and internal state factors that affect the energy consumption of the support.

[0011] As a further aspect of the present invention, in step S3, during the construction of the global optimization controller, constraints are simultaneously set. These constraints include mechanical limit constraints on the tracking angle of the support, power constraints on the drive motor, angle adjustment rate constraints, and safety wind avoidance constraints under high wind conditions, to ensure the safety and executability of the control strategy.

[0012] As a further aspect of the present invention, in step S3, when generating the optimal tracking control strategy, for the photovoltaic support cluster, the differences in installation terrain, equipment status, and shading effects of different supports are combined to complete the cluster-level collaborative optimization, and an independent optimal tracking control strategy with a corresponding time resolution is generated for each support, thereby realizing the combination of refined management and control of a single support and global optimization of the cluster.

[0013] As a further aspect of the present invention, in step S4, the local security verification content of the edge controller includes mechanical limit verification of the target angle, safety condition verification of real-time wind speed, and fault verification of equipment operating status. After all verifications pass, the control command is executed; if the verification fails, the local security protection strategy is triggered, and the abnormal information is uploaded to the industrial cloud computing platform. Through the dual protection of cloud-edge collaboration, the safety of the support operation is improved.

[0014] As a further aspect of the present invention, in step S5, the incremental learning and parameter update of the energy consumption feature model are performed according to a fixed cycle. At the same time, based on the cumulative running time and fault records of the equipment, the equipment aging parameters in the static basic energy consumption model are updated synchronously to realize the full life cycle iterative optimization of the energy consumption feature model and ensure the long-term stability of the model accuracy.

[0015] The beneficial effects of the technical solutions provided by the embodiments of the present invention include: In this invention, the cloud-edge collaborative management and control architecture for photovoltaic brackets based on industrial cloud computing solves the problems of insufficient computing power and inability to achieve high-precision modeling and global optimization of traditional bracket local controllers by leveraging the massive computing power and storage capabilities of the cloud. At the same time, it breaks down the data silos of independent control of single brackets and realizes the unified aggregation and value mining of power station-level bracket operation data. In this invention, a photovoltaic support energy consumption model is developed that integrates static basic energy consumption and dynamic operating condition energy consumption. The static model covers the basic energy consumption caused by the inherent attributes of the equipment, while the dynamic model couples the influence of environmental conditions and equipment status on energy consumption through machine learning algorithms. Compared with the traditional static energy consumption estimation method, this significantly improves the accuracy of energy consumption modeling and provides a reliable quantitative basis for the optimization of control strategies. This invention proposes a dual-objective optimization control strategy that maximizes power generation gain and minimizes operating energy consumption. This breaks the limitation of traditional control that only maximizes irradiance reception as the sole objective. The optimal tracking strategy is obtained by solving the global optimization controller, which fundamentally avoids the energy waste caused by excessive tracking. It achieves the optimal balance between photovoltaic tracking power generation gain and operating energy consumption, and significantly improves the net power generation revenue of the photovoltaic system. In this invention, the energy consumption characteristic model is optimized online throughout its entire lifecycle. Based on real-time operation feedback data of the support structure, the model is incrementally learned and its parameters are updated. At the same time, it adapts to the changes in energy consumption characteristics caused by equipment aging, ensuring the long-term stability of model accuracy and control effect, and reducing the operation and maintenance cost of photovoltaic support structures throughout their entire lifecycle. In this invention, a cloud-edge collaborative control mode is adopted. The cloud is responsible for global modeling and optimization, while the edge is responsible for local execution and security verification. This ensures the global optimality of the control strategy and guarantees the safety and reliability of equipment operation through local security verification, making it perfectly suited for the large-scale application scenarios of large-scale ground-mounted photovoltaic power plants. Brief Description of the Drawings

[0016] Figure 1 This is the overall flowchart of the modeling and control method for the energy consumption characteristics of a photovoltaic support based on industrial cloud computing according to the present invention. Detailed Embodiments

[0017] The following will describe the technical solutions in the present invention in conjunction with the accompanying drawings.

[0018] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or more advantageous than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two. <----

[0019] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments.

[0020] The present invention provides a technical solution: a modeling and control method for the energy consumption characteristics of a photovoltaic support based on industrial cloud computing, including the following steps: S1: Data collection and preprocessing. Through the edge collection unit supporting the photovoltaic support, all running data of the support cluster and field environment data are obtained. After being preprocessed by standardization at the edge end, they are uploaded to the industrial cloud computing platform through the cloud-edge collaboration channel; Through the edge collection unit deployed on the photovoltaic support, all running data of the photovoltaic support cluster and field environment data are collected. After being preprocessed by standardization at the edge end, they are uploaded to the industrial cloud computing platform through the cloud-edge collaboration channel.

[0021] S101: Full-dimensional data collection. The edge collection unit adopts a distributed deployment mode. Each photovoltaic support is equipped with an independent edge collection module. The module adopts a low-power design, with a working power consumption ≤ 5W. It can be assisted by photovoltaic modules for power supply. The working temperature range is -40°C to 85°C. It has the characteristics of resistance to high and low temperatures, electromagnetic interference, waterproof and dustproof, and can operate stably in harsh outdoor environments. The edge collection module supports multiple communication interfaces such as RS485, Modbus, CAN, and Ethernet, and can adapt to the data collection requirements of different types of sensors such as voltage sensors, current sensors, angle sensors, temperature sensors, wind speed and direction sensors, and irradiance sensors.

[0022] The collected full-scale operating data includes the driving motor operating parameters, including voltage (0-500V, accuracy ±0.5%), current (0-50A, accuracy ±0.5%), power (0-20kW, accuracy ±1%), speed (0-1500r / min, accuracy ±1r / min), power factor (0.5-1.0, accuracy ±0.01), and winding temperature (0-150℃, accuracy ±1℃) of the horizontal and pitch axis motors, with a sampling frequency of 1 time / second. The bracket tracks the following angles: horizontal axis angle (0°360°, accuracy ±0.1°) and pitch axis angle (0°90°, accuracy ±0.1°), with a sampling frequency of 1 time / second. Equipment status data, including cumulative runtime (accurate to minutes), cumulative number of actions (accurate to times), and fault alarm data (including fault type, occurrence time, location, and level), with a fault detection response time ≤100ms; Photovoltaic string power generation data includes string output voltage (0-1000V, accuracy ±0.5%), current (0-20A, accuracy ±0.5%), power (0-20kW, accuracy ±1%), and power generation (cumulative accuracy ±0.1kWh), with a sampling frequency of 1 time / minute.

[0023] Collected site environmental data, irradiance data, horizontal irradiance, and inclined plane irradiance (0-2000W / m²). 2 (accuracy ±2%), sampling frequency 1 time / minute; Astronomical data, including solar altitude angle and azimuth angle (calculated using astronomical algorithms combined with the latitude and longitude of the field area and real-time time, with an accuracy of ±0.5°). Meteorological data, including wind speed (0-50m / s, accuracy ±0.3m / s), wind direction (0°360°, accuracy ±5°), ambient temperature (-40℃ to 85℃, accuracy ±0.5℃), and photovoltaic module backsheet temperature (-40℃ to 85℃, accuracy ±0.5℃), were sampled at a frequency of 1 time per minute. Topographic data, including the installation elevation (accuracy ±0.1m), slope (0°30°, accuracy ±0.1°), and orientation (0°360°, accuracy ±1°) of each support, were obtained through GPS positioning and topographic surveying technology.

[0024] S102: The edge acquisition module performs standardized preprocessing on the raw data. The specific steps are as follows: Outlier removal employs a combination of the 3σ criterion and the Isolation Forest algorithm. First, the 3σ criterion is used to remove obvious outliers: for a dataset of a given variable, the mean μ and standard deviation σ are calculated, and data exceeding the range [μ-3σ, μ+3σ] are identified and removed. Then, an anomaly detection model is built using the Isolation Forest algorithm to detect the remaining data, identifying gradual anomalies caused by sensor drift, achieving an anomaly detection accuracy of at least 98%. For missing data completion, for continuous data such as voltage, current, temperature, and irradiance, linear interpolation is used when the missing rate is ≤5%, with a completion error not exceeding 3%; when 5% < missing rate ≤10%, cubic spline interpolation is used, with a completion error not exceeding 5%. For discrete data such as fault type and number of actions, mode completion or nearest neighbor completion methods are used. The data format is standardized, converting all data to the JSON standard format, clearly defining data fields, types, units, collection time, and bracket number information. The data format for motor voltage is {"field":"motor_voltage","type":"float","unit":"V","value":380.5,"time":"2026-04-09 10:00:00","bracket_id":"Bracket_001"}; Time axis alignment is achieved by using the NTP time synchronization protocol to synchronize the time of all edge acquisition modules with the standard time of the industrial cloud computing platform, with a timestamp error of no more than 10ms, thus avoiding data misalignment caused by time inconsistency. Data downsampling reduces the original 1-second / sample data to 1-minute / sample data using a mean sampling method, where the average of all sampled data within a minute is taken as the representative data for that minute. This reduces the data volume by more than 60 times while preserving the core characteristics of the data. For scenarios with drastic changes in operating conditions, such as strong winds, the downsampling interval can be adjusted to 30 seconds / sample.

[0025] S103: Cloud-edge collaborative data transmission. Pre-processed data is uploaded to the industrial cloud computing platform via a 5G / 4G cloud-edge collaborative channel. The transmission process uses the AES-256 end-to-end encryption protocol to ensure the security and integrity of data transmission and prevent data leakage and tampering. It also employs a breakpoint resume mechanism. When the network is interrupted, the edge acquisition module caches the data locally (cache capacity ≥ 1GB), and resumes uploading the unfinished data after the network is restored. The data upload success rate is no less than 99.9%.

[0026] S2: Cloud-based energy consumption modeling. In the industrial cloud computing platform, based on the uploaded full historical and real-time data, a static basic energy consumption model and a dynamic operating condition energy consumption model of the photovoltaic bracket are constructed respectively. The two types of models are then merged to obtain a complete energy consumption model. In the industrial cloud computing platform, based on the uploaded full historical and real-time data, a static basic energy consumption model and a dynamic operating condition energy consumption model are constructed respectively, and then merged to obtain a complete photovoltaic support energy consumption characteristic model. The industrial cloud computing platform adopts a distributed architecture, equipped with 2×Intel Xeon Gold 6248R CPUs, 256GB memory, and 10TB SSD storage. It adopts an HDFS distributed storage architecture, supports PB-level data storage, and uses a load balancing algorithm for computing power scheduling. The computing latency is ≤100ms, which can simultaneously support the modeling and control needs of tens of thousands of photovoltaic supports.

[0027] S201: Static base energy consumption is used to characterize the energy consumption determined by the inherent properties of the equipment. It includes three parts: standby energy consumption, base drive energy consumption, and transmission friction energy consumption. The calculation formula is as follows:

[0028] in: Standby power consumption is determined by the rated power of sensors, controllers, and auxiliary equipment, and is calculated using the following formula: , This refers to the total power of auxiliary equipment (usually 50W). Runtime; The basic drive energy consumption is proportional to the motor's rated power and angle adjustment range, and the calculation formula is as follows: , This refers to the rated power of the motor. For angle adjustment amount, To maximize the adjustable angle, For driving time; The energy consumption due to transmission friction is proportional to the friction coefficient of the transmission mechanism, the moment of inertia of the support, and the angle adjustment rate. The calculation formula is as follows: , The coefficient of friction, For rotational inertia, Adjust the rate for the angle; The inherent parameters of each support frame were collected: horizontal axis motor rated power 1.5kW, pitch axis motor rated power 0.75kW, reducer transmission ratio 1:150, transmission efficiency 0.92, friction coefficient 0.08, support frame weight 800kg, and moment of inertia 1200kg·m. 2 Based on the factory no-load test data of the equipment (72 hours of testing at a constant temperature of 25℃±2℃) and the on-site no-load operation data, the least squares method and nonlinear regression algorithm were used to fit and obtain the calculation coefficients of energy consumption of each part, and the static basic energy consumption model was completed. The model fitting error did not exceed 3%.

[0029] S202: Construction of dynamic operating condition energy consumption model. Dynamic operating condition energy consumption is used to characterize the energy consumption increment caused by changes in external environment and equipment status. It is constructed using LSTM-Attention hybrid neural network. The core feature variables were selected using a combination of Pearson correlation coefficient and random forest importance ranking, resulting in six core feature variables: the difference between the target angle and the real-time angle Δθ, real-time wind speed v, ambient temperature T, component surface irradiance G, cumulative equipment runtime t, and cumulative number of actions n. Among these, Δθ had the highest correlation coefficient with dynamic energy consumption, reaching 0.82. The dataset was divided into 12 months of historical operating data, and divided into training set, validation set and test set according to the ratio of 70%:20%:10%. The data covers different weather conditions such as sunny day, cloudy day, rainy day and windy day, as well as different states of equipment such as new investment, mid-term and aging. Model structure and training: The structure of the LSTM-Attention hybrid neural network is as follows: The input layer has 6 neurons, corresponding to 6 core feature variables; the LSTM feature extraction layer has 2 stacked layers, each with 128 neurons, using the ReLU activation function; the attention mechanism layer has an 8-head attention mechanism that adaptively assigns weights to different features; and the fully connected output layer has 1 neuron that outputs dynamic energy consumption increments.

[0030] The training process used the Adam optimizer with an initial learning rate of 0.001, a batch size of 64, and 200 training epochs. Early stopping was employed to prevent overfitting. After training, the model accuracy was validated using a test set. The model was tested when MAPE ≤ 2.5%, RMSE ≤ 0.08kWh, and R0.08. 2 When the value is ≥0.98, the model is considered to meet the accuracy requirements.

[0031] S203: Model fusion and accuracy verification. The output of the static basic energy consumption model is added to the output of the dynamic operating condition energy consumption model to obtain the final energy consumption prediction value.

[0032] The accuracy of the fused model was verified by field measurement data. The results showed that the model had an average absolute percentage error of 2.17%, a root mean square error of 0.076 kWh, and a coefficient of determination of 0.985, which is much higher than the accuracy of the traditional static model (MAPE=8.62%) and the single LSTM model (MAPE=4.35%), and can meet the needs of high-precision energy consumption prediction.

[0033] S3: Dual-objective optimization control strategy generation. Based on the completed energy consumption characteristic model and combined with photovoltaic power generation prediction data, a global optimization controller is constructed with the dual optimization objectives of maximizing the power generation gain of the photovoltaic system and minimizing the operating energy consumption of the photovoltaic support as the dual optimization objectives. The solution is completed on the industrial cloud computing platform to generate the optimal tracking control strategy for the photovoltaic support cluster. Based on the completed energy consumption characteristic model and combined with photovoltaic power generation prediction data, a global optimization controller is constructed to generate the optimal tracking control strategy for the photovoltaic support cluster.

[0034] S301: High-precision photovoltaic (PV) power generation forecasting. This method employs multi-source data fusion, integrating weather forecast data (irradiance, wind speed, and temperature for the next 24 hours) with an accuracy ≥85%. Historical power generation data (past 12 months) and real-time irradiance data are also used. A weighted fusion algorithm yields the PV power generation forecast, with a prediction error not exceeding 5%. The formula for calculating PV power generation is: ; in, Rated power of photovoltaic modules This represents the actual irradiance. Standard irradiance (1000 W / m) 2 ), This refers to the component temperature.

[0035] S302: Construction of a global optimization controller, with the optimization objective of maximizing the net power generation of the photovoltaic system, i.e., maximizing power generation gain and minimizing the energy consumption of the support structure. The objective function is:

[0036] in, To predict the time domain over 24 hours, with a step size of 15 minutes, Let be the photovoltaic power generation at time k. Let K be the energy consumption of the support system at time k. Let be the support tracking angle at time k.

[0037] Set constraints: Mechanical limit constraints , Motor power constraint Angle adjustment rate constraint Strong winds and wind shelter restrictions, when hour, ; An improved particle swarm optimization algorithm is used to solve the above optimization problem. This algorithm introduces adaptive inertia weights and mutation operations. Adaptive inertial weights are dynamically adjusted based on particle fitness values. When the fitness is good, the weights are reduced to speed up convergence, and when the fitness is poor, the weights are increased to improve global search capabilities. The weight range is 0.4 to 0.9. The mutation operation randomly mutates the positions of some particles with a probability of 0.1, increasing population diversity and avoiding getting trapped in local optima. The improved algorithm increases the solution accuracy by more than 15% and the solution speed by more than 20%, with a solution time of ≤10 seconds for a single control cycle, which can meet the needs of real-time control.

[0038] S303: Cluster-level collaborative optimization solution. For photovoltaic support clusters, cluster-level collaborative optimization is performed by considering the differences in installation terrain, equipment status, and shading effects of different supports. Due to terrain differences, for supports with higher elevations and gentler slopes, the tracking accuracy should be appropriately increased; for supports with lower elevations and steeper slopes, the tracking accuracy should be appropriately decreased to balance energy consumption and power generation gain. Depending on the condition of the equipment, for severely aged supports, reduce the tracking frequency and the number of motor actions; for newly put into operation supports, adopt a high-precision tracking strategy. Front and rear row obstruction is addressed by calculating the obstruction angle based on the installation spacing and height difference, adjusting the tracking angle of the front row brackets to avoid obstructing the rear row brackets, and ensuring the maximum overall power generation benefit of the entire cluster.

[0039] An independent optimal tracking angle sequence is generated for each support, with a time resolution of 15 minutes / time, which is adjusted to 5 minutes / time when the working conditions are severe.

[0040] S4: Cloud-edge collaborative control execution. The industrial cloud computing platform will send the generated optimal tracking control strategy to the edge controller of the corresponding bracket. After the edge controller completes the local security verification of the control strategy, it will drive the bracket to perform the corresponding angle tracking action and collect the operation feedback data after the action is executed in real time. The industrial cloud computing platform will send the generated optimal tracking and control strategy to the edge controller of the corresponding bracket. After the edge controller completes the local security verification, it will execute the control action and collect feedback data.

[0041] S401: Control strategy distribution and local caching. The industrial cloud computing platform distributes the optimal tracking angle sequence for the next hour to the edge controller. The edge controller has a local cache of more than 1GB, which can cache the control strategy for the next 24 hours. When the cloud-edge collaboration channel is interrupted, the edge controller continues to execute the control strategy based on the cache to avoid equipment downtime; after the channel is restored, it automatically synchronizes the latest control strategy and operating data to ensure control continuity.

[0042] S402: Local multi-level security verification; the edge controller adopts a three-level security verification mechanism with a verification response time ≤100ms. Target angle mechanical limit verification: compare the target angle with the mechanical limit range. If it exceeds the range, refuse to execute and generate angle over-limit exception information. Real-time wind speed safety check: compare the real-time wind speed with the strong wind threshold (12m / s). If the wind speed exceeds the limit, the wind protection will be triggered and the support will be adjusted to a horizontal state. Equipment operation status verification involves detecting the operating status of motors, sensors, and controllers. If a fault is found, corresponding measures are taken according to the fault level: minor faults reduce tracking accuracy, general faults stop tracking and adjust to a safe angle, and severe faults cut off the motor power supply. After all the checks are passed, the edge controller drives the horizontal axis and pitch axis motors to perform angle adjustment actions, with a control accuracy of no less than ±0.1°.

[0043] S403: Operational feedback data acquisition. The edge controller collects motor operating parameters, actual bracket angle, and power generation feedback data in real time after the action is executed. The feedback data is uploaded to the industrial cloud computing platform every 15 minutes to provide a basis for model iteration and strategy optimization.

[0044] S5: Online model iteration and closed-loop optimization. The edge controller periodically uploads the operation feedback data to the industrial cloud computing platform. The platform incrementally learns and updates the parameters of the energy consumption model based on the new data. At the same time, it optimizes the solution parameters of the global optimization controller based on the feedback results, thus completing the closed-loop optimization of the control strategy.

[0045] Based on operational feedback data, the energy consumption model is iteratively optimized online on the industrial cloud computing platform, and the parameters of the global optimization controller are adjusted to form a complete closed-loop optimization system.

[0046] S501: Incremental learning and parameter update. An incremental training algorithm is used to update the energy consumption model, training only on the newly added 15 minutes of running data, eliminating the need to repeatedly train on historical data, thus reducing computing power consumption by over 60%. The incremental learning cycle is 15 minutes per cycle, primarily updating the weight coefficients of the dynamic operating condition energy consumption model and the weight coefficients of the dynamic weighted fusion. A comprehensive parameter update is performed every three months, usually during low-power-generation periods at night, to optimize and adjust all parameters of the model and ensure long-term stability of model accuracy.

[0047] S502: Adaptive adjustment of equipment aging parameters, based on the equipment's cumulative operating time and fault records, synchronously updating the aging parameters in the static basic energy consumption model: The motor efficiency attenuation coefficient is adjusted every 1000 hours, and the attenuation curve is fitted according to the change of motor no-load loss; the reducer transmission efficiency attenuation coefficient is adjusted according to the wear of the transmission mechanism; the transmission friction coefficient increment is adjusted according to the cumulative number of actions and running time; through the adaptive adjustment of aging parameters, the model can accurately reflect the energy consumption characteristics of the equipment after aging and avoid model mismatch.

[0048] S503: Closed-loop optimization of control strategy. Based on the deviation between actual net power generation and predicted value, and the deviation between actual energy consumption and predicted energy consumption, the solution parameters of the global optimization controller are optimized, including the inertia weight range, mutation probability, and number of iterations, to continuously improve the optimality of the control strategy. At the same time, a model performance evaluation mechanism is established to comprehensively evaluate the model accuracy every month. If MAPE>2.5%, the model reconstruction process is triggered to re-select feature variables, adjust the model structure, and retrain. Experimental verification was conducted at a 10MW distributed photovoltaic power station. The power station installed 500 dual-axis tracking photovoltaic brackets, each equipped with an edge acquisition unit and a PLC controller. The industrial cloud computing platform was deployed on Huawei Cloud Stack. The experiment lasted for 3 months. The effects of the method of this invention and traditional astronomical algorithm control were compared, and the results are shown in Table 1. Table 1. Comparison of experimental data between the method of this invention and traditional control methods.

[0049] Experimental results show that the method of the present invention can effectively reduce the energy consumption of photovoltaic bracket operation, avoid excessive tracking, and significantly improve the net power generation revenue and equipment reliability of photovoltaic system. At the same time, in windy weather with wind speed ≥12m / s, the method of the present invention can trigger the wind avoidance strategy within 1 second without any equipment damage. When the power grid issues a power curtailment command, it can quickly adjust the tracking angle of each bracket to achieve precise control of photovoltaic power generation and meet the grid dispatching requirements.

[0050] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for modeling and controlling the energy consumption characteristics of photovoltaic brackets based on industrial cloud computing, characterized in that, It includes the following steps: S1: Data collection and preprocessing. Through the edge collection unit supporting the photovoltaic support, the full-scale operation data of the support cluster and the field environment data are obtained. After being preprocessed by edge-side standardization, they are uploaded to the industrial cloud computing platform through the cloud-edge collaborative channel. The edge collection unit adopts a distributed deployment mode, and each photovoltaic support is equipped with an independent edge collection module, supporting the access of multi-protocol sensors. The cloud-edge collaborative channel adopts the AES-256 end-to-end encryption and breakpoint resumption mechanism, and the data upload success rate is not less than 99.9%; S2: Cloud-side energy consumption model building. In the industrial cloud computing platform, based on the uploaded full-scale historical and real-time data, the static basic energy consumption model and the dynamic working condition energy consumption model of the photovoltaic support are respectively built, and the two types of models are fused to obtain a complete energy consumption model. The static basic energy consumption model is built based on the device mechanism and factory test data, and the dynamic working condition energy consumption model is built based on the training of the LSTM-Attention hybrid neural network. The mean absolute percentage error (MAPE) of the fused model ≤ 2.5%, and the root mean square error (RMSE) ≤ 0.08 kWh; S3: Generation of double-objective optimization control strategy. Based on the built energy consumption model, combined with the photovoltaic power generation prediction data, with the maximization of the power generation gain of the photovoltaic system and the minimization of the operation energy consumption of the photovoltaic support as the two optimization objectives, a global optimization controller is built and solved in the industrial cloud computing platform to generate the optimal tracking control strategy for the photovoltaic support cluster. The global optimization controller adopts an improved particle swarm optimization algorithm introducing an adaptive inertia weight and mutation operation, and the solution accuracy is improved by more than 15%, and the solution speed is improved by more than 20%; S4: Cloud-edge collaborative control execution. The industrial cloud computing platform sends the generated optimal tracking control strategy to the edge controller of the corresponding support. After the edge controller completes the local security verification of the control strategy, it drives the support to execute the corresponding angle tracking action, and at the same time, it collects the operation feedback data after the action is executed in real time. The edge controller has a local cache of more than 1GB and an offline execution ability. The local security verification adopts a three-level verification mechanism, and the verification response time ≤ 100 ms; S5: Online iteration and closed-loop optimization of the model. The edge controller periodically uploads the operation feedback data to the industrial cloud computing platform. The platform performs incremental learning and parameter update on the energy consumption model based on the new data, and at the same time optimizes the solution parameters of the global optimization controller based on the feedback results to complete the closed-loop optimization of the control strategy. The incremental learning only trains the new data, and the computing power consumption is reduced by more than 60%. At the same time, the aging parameters are updated synchronously based on the cumulative operation duration and fault records of the device to realize the iteration of the full life cycle of the model.

2. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S1, the full operational data includes the operating parameters of the photovoltaic support drive motor, the real-time tracking angle of the support, the equipment operating time, fault alarm data, and the power generation data of the corresponding photovoltaic string. The site environment data includes the real-time irradiance of the site, the solar altitude angle and azimuth angle, wind speed and direction, ambient temperature, photovoltaic module temperature, and support installation terrain data. The sampling frequency of the drive motor operating parameters is no less than 1 time / second, including voltage (0-500V, accuracy ±0.5%), current (0-50A, accuracy ±0.5%), power (0-20kW, accuracy ±1%), speed (0-1500r / min, accuracy ±1r / min), and winding temperature (0-150℃, accuracy ±1℃). The measurement accuracy of the support tracking angle is no less than ±0.1°, with a horizontal axis range of 0°360° and a pitch axis range of 0°90°. The site irradiance measurement range is 0-2000W / m. 2 Accuracy ±2%, wind speed measurement range 0-50m / s, accuracy ±0.3m / s.

3. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S1, the edge-end standardization preprocessing specifically involves removing outliers, completing missing data, standardizing data format, and aligning the time axis on the collected raw data. After downsampling the data at preset time intervals, the data is packaged and uploaded. Outlier removal uses a combination of the 3σ criterion and the isolated forest algorithm, with an anomaly detection accuracy of no less than 98%. For missing data completion, linear interpolation or cubic spline interpolation is used for continuous data, with a completion error of no more than 5%. For discrete data, mode or nearest neighbor value completion is used. The data format is standardized to the JSON standard format. The time axis alignment uses the NTP protocol, with a timestamp error of no more than 10ms. The preset time interval is usually 1 minute / time. Downsampling uses the mean sampling method, reducing the data volume by more than 60 times.

4. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S2, the static basic energy consumption model is used to characterize the basic energy consumption corresponding to the inherent attributes of the photovoltaic support equipment. The construction process is as follows: the inherent parameters of the photovoltaic support equipment are collected, and the standby energy consumption, basic drive energy consumption, and transmission friction energy consumption of the support are obtained by combining the equipment factory test data and no-load operation data. The construction of the static basic energy consumption model is completed. The inherent parameters of the equipment include the rated power of the drive motor, the transmission ratio of the reducer, the friction coefficient of the transmission mechanism, the self-weight and rotational inertia of the support, and the parameter measurement error does not exceed 2%. The factory test data includes the no-load loss of the motor, the no-load loss of the reducer, and the no-load operation energy consumption of the support. The test duration is not less than 72 hours. The fitting process adopts a combination of least squares method and nonlinear regression algorithm, and the model fitting error does not exceed 3%.

5. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 4, characterized in that: In step S2, the dynamic operating condition energy consumption model is used to characterize the dynamic energy consumption increment of the photovoltaic support under the influence of external operating conditions and equipment status during actual operation. The construction process specifically involves: screening core characteristic variables affecting the dynamic energy consumption of the support; using preprocessed historical operating data as the training set and static basic energy consumption as the baseline value; training an initial dynamic operating condition energy consumption model using a machine learning algorithm; and after testing and verification to meet accuracy requirements, fusing it with the static basic energy consumption model to obtain the final energy consumption characteristic model. The historical operating data duration is no less than 12 months, covering different seasons, weather conditions, and equipment statuses. The ratio of training set, validation set, and test set is 70%:20%:10%. The machine learning algorithm uses an LSTM-Attention hybrid neural network, with a structure including an input layer, two LSTM feature extraction layers with 128 neurons, an eight-head attention mechanism layer, and a fully connected output layer. The model testing and verification indicators are MAPE ≤ 2.5%, RMSE ≤ 0.08kWh, and R0. 2 ≥0.

98.

6. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 5, characterized in that: The core characteristic variables affecting the dynamic energy consumption of the support system include the difference between the target angle and the real-time angle, real-time wind speed, ambient temperature, component surface irradiance, cumulative equipment runtime, and cumulative number of actions. The difference between the target angle and the real-time angle is the most significant influencing factor, with a correlation coefficient of 0.

82. When the real-time wind speed is ≥12m / s, the impact of wind load on energy consumption increases significantly. Energy consumption is lowest at ambient temperatures of 20℃ and 30℃, while it increases by 5% and 15% respectively below 0℃ and above 40℃. For every 10,000-hour increase in cumulative equipment runtime, energy consumption increases by 5% and 8% respectively. For every 10,000-time increase in cumulative actions, energy consumption increases by 3% and 5% respectively.

7. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S3, during the construction of the global optimization controller, constraints are simultaneously set. These constraints include mechanical limit constraints on the support tracking angle, drive motor power constraints, angle adjustment rate constraints, and wind safety constraints for high wind conditions. The mechanical limit constraints are 0° and 360° for the horizontal axis and 0° and 90° for the pitch axis. The drive motor power constraint is not exceeding 110% of the rated power. The angle adjustment rate constraint is not exceeding 5° / min. The wind safety constraint is that when the wind speed is ≥12m / s, the support is forcibly adjusted to a horizontal position with the pitch axis at 0°.

8. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S3, when generating the optimal tracking control strategy, for the photovoltaic support cluster, the cluster-level collaborative optimization is completed by combining the differences in installation terrain, equipment status, and shading effects of different supports. An independent optimal tracking control strategy with corresponding time resolution is generated for each support. Terrain differences include differences in elevation, slope, and orientation. Equipment status differences include differences in aging and fault status. The shading effects of the front and rear rows are calculated by the installation spacing and height difference. The time resolution is usually 15 minutes / time, which can be adjusted to 5 minutes / time when the working conditions are severe. After the cluster collaborative optimization, the net power generation revenue of the support cluster can be increased by 4% to 8%, and energy consumption can be reduced by more than 25%.

9. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S4, the local security verification of the edge controller includes mechanical limit verification of the target angle, safety condition verification of real-time wind speed, and fault verification of equipment operating status. After all verifications pass, the control command is executed. If the verification fails, the local security protection strategy is triggered and the abnormal information is uploaded to the industrial cloud computing platform. Mechanical limit verification is used to prevent damage to the mechanical structure of the support. Safety condition verification is used for equipment protection in windy weather. Fault verification is used to detect the operating status of motors, sensors, and controllers. Local safety protection strategies include stopping angle adjustment, adjusting to a safe angle, and emergency power cut-off of the motor; abnormal information includes verification type, abnormal value, abnormal time, and bracket number.

10. The photovoltaic support energy consumption modeling and control method based on industrial cloud computing according to claim 1, characterized in that: In step S5, incremental learning and parameter updates of the energy consumption model are performed according to a fixed cycle. Simultaneously, based on the cumulative operating time and fault records of the equipment, the equipment aging parameters in the static basic energy consumption model are updated synchronously to achieve full lifecycle iterative optimization of the energy consumption model. The incremental learning cycle is 15 minutes / time, and the comprehensive parameter update cycle is 3 months / time, usually performed during low power generation periods at night. The equipment aging parameters include the motor efficiency decay coefficient, the reducer transmission efficiency decay coefficient, and the transmission friction coefficient increment, which are adjusted every 1000 hours. A model performance evaluation mechanism is established to comprehensively evaluate the model accuracy every month. If the requirements are not met, the model reconstruction process is triggered.

Images