A photovoltaic protocol converter and data acquisition method based on the Internet of Things

Abstract

The present application relates to photovoltaic power generation data acquisition technical field, specifically to a kind of photovoltaic protocol converter and data acquisition method based on Internet of Things, including Internet of Things communication module, protocol self-learning module, protocol conversion module, edge computing and evaluation module, collaborative scheduling module;The method includes capturing the original message when photovoltaic inverter and monitoring equipment communication, extract the numerical value change time series of each register address, generate semantic feature library;Target data item and the register address in semantic feature library are matched with semantics, establish dynamic mapping table, read original data and convert into standard format;According to the historical change rate of each data item, the influence weight of power grid stability and cloud subscription priority, dynamically calculate its real-time value weight, output value weight data;Real-time adjustment protocol conversion module operating parameter, write the adjusted operating parameter into dynamic mapping table again, drive to execute subsequent collection.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation data acquisition technology, specifically to a photovoltaic protocol converter and data acquisition method based on the Internet of Things. Background Technology

[0002] Traditional photovoltaic protocol conversion solutions primarily employ a fixed mapping table approach. Technicians pre-define the inverter's communication protocol documentation, manually configure the mapping between register addresses and data items, and then embed this mapping table into the protocol converter. When the inverter model changes or the protocol version is upgraded, on-site reconfiguration or remote firmware upgrades are required. This approach has the following inherent drawbacks: extensive manual research and configuration work is required before deployment, resulting in a long on-site debugging cycle; it is incompatible with inverters using proprietary or non-public protocols; response time is delayed after protocol changes, leading to high maintenance costs.

[0003] In recent years, some improved solutions have introduced protocol adaptive identification technology. For example, CN113972746A discloses a low-voltage grid-connected distributed photovoltaic monitoring and control method, in which the photovoltaic protocol converter automatically identifies the inverter type through a preset protocol library and completes bidirectional conversion between DL / T698 and Modbus protocols. This solution can reduce some manual configuration work, but it still relies on a pre-established protocol feature library and is powerless against new or proprietary protocols. Another type of solution focuses on optimizing the storage of acquired data. For example, CN117807546A proposes a high-efficiency data storage method based on the rotating gate algorithm, which dynamically adjusts the compression threshold by calculating the specificity of DC current data to improve the accuracy of storing fluctuating data. However, this solution only addresses the data compression stage and does not involve dynamic adjustment of protocol adaptation or acquisition strategies.

[0004] Furthermore, CN119420045A discloses an intelligent data acquisition and control terminal for distributed power sources, which assesses the risk of exceeding limits by collecting operating status parameters and invokes preset control strategies. Although this solution has a certain degree of local intelligence, its risk classification and strategy priority are statically configured, lacking the ability to perceive and adaptively schedule real-time resource conditions such as CPU utilization, communication success rate, and power supply.

[0005] Regarding the above-mentioned technical solutions, existing technologies typically rely on preset libraries for protocol adaptation, lack self-learning capabilities, and cannot automatically identify and adapt to inverters with unknown or proprietary protocols. The acquisition strategies are statically fixed, making it impossible to perceive real-time resource status, which can easily lead to resource overload or delays in critical data acquisition. Furthermore, the lack of data value-driven differentiated acquisition scheduling reduces the adaptability and data acquisition efficiency of protocol converters in complex operating environments. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides an IoT-based photovoltaic protocol converter, comprising an IoT communication module, a protocol self-learning module, a protocol conversion module, an edge computing and evaluation module, and a collaborative scheduling module. The IoT communication module is used to communicate with the cloud platform and receive a data acquisition template containing the target data item set and the basic acquisition cycle of each data item. The protocol self-learning module is used to connect to the power communication interface of the photovoltaic inverter, capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library. The protocol conversion module is used to semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the dynamic mapping table and the effective acquisition cycle, and convert the raw data into a standard format. The edge computing and evaluation module is used to monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, and output monitoring data. Based on the historical change rate of each target data item, the impact weight of power grid stability, and the cloud subscription priority, it dynamically calculates the real-time value weight and outputs the value weight data. The collaborative scheduling module is used to adjust the operating parameters of the protocol conversion module in real time based on monitoring data and value weight data. The operating parameters include the actual acquisition cycle of each data item, conversion timeout, number of retries, and batch read size.

[0007] The protocol self-learning module calculates the physical quantity type corresponding to each register address and generates a semantic feature library. The specific operation is as follows: A fast Fourier transform is performed on the time series of numerical changes extracted from the register address to extract the main frequency component; when the main frequency component falls within the power frequency range and the amplitude fluctuation range conforms to the characteristics of the power grid voltage, the register address is initially marked as AC voltage or current; For register addresses initially labeled as AC voltage or current, calculate the sliding window correlation coefficient between the register value and the output power of the photovoltaic inverter. If the absolute value of the correlation coefficient is greater than the first standard value, it is determined to be current; if the absolute value of the correlation coefficient is less than the second standard value, it is determined to be voltage. For unmarked register addresses, calculate the median of the absolute value of the first difference of the time series of the register values. If the median is less than the preset value and the value drifts in one direction, it is determined to be temperature. The judgment result, along with the corresponding register address, numerical range, and rate of change characteristics, is stored in the semantic feature library, and a timestamp and confidence score are added to each semantic feature.

[0008] The correlation coefficient between the calculated register value and the sliding window of the photovoltaic inverter output power is as follows: For a register address initially marked as AC, let the value sequence of that register be... Simultaneously, the synchronization sequence for obtaining the output power of the photovoltaic inverter is... ,in, ; The Pearson correlation coefficient is calculated segment by segment using a sliding window method. The formula is as follows: in, is the correlation coefficient of the k-th window, with a value range of [-1, 1]; W is the length of the sliding window. Let X be the mean of the register value sequence within the sliding window. The mean of the output power synchronization sequence Y within the sliding window.

[0009] The grid voltage characteristics include the amplitude of the numerical sequence falling within the normal grid voltage range, the waveform distortion rate being lower than the distortion rate threshold, the fluctuation range of the numerical sequence not exceeding ±2% of the rated value, and the deviation of the phase difference between any two phases in the three-phase system from 120° falling within the allowable error range.

[0010] The formula for calculating the real-time value weight in the edge computing and evaluation module is as follows: in, Let be the real-time value weight of the i-th data item at time t; Let be the rate of change of the i-th data item within the sliding window at time t. Let be the historical average rate of change of the i-th data item; Let be the preset weighting coefficient for the impact of the i-th data item on power grid stability, 0 < <1; Let j be the cloud subscription priority coefficient for the i-th data item at time t, j represent the index of all target data items in the current data acquisition template, and N be the total number of target data items in the current data acquisition template. This represents the sum of the cloud subscription priority coefficients for all target data items at time t, used for... Normalization is performed. α β is the first weighting coefficient, β is the second weighting coefficient, and γ is the third weighting coefficient. a+b+c= 1.

[0011] The protocol scheduling module employs a model predictive control algorithm to optimize operating parameters by minimizing the acquisition delay of high-value data items and maximizing the overall data acquisition value.

[0012] The specific operation of the rolling optimization running parameters is as follows: Establish a discrete-time dynamic model, with the inputs being the acquisition period, protocol conversion timeout, and batch read size of each data item at the current moment, and the outputs being the conversion delay and CPU utilization at the next moment; In each control cycle, starting from the current state, multiple sets of candidate operating parameter sequences are enumerated within the preset control time domain. For each candidate sequence, a weighted acquisition delay cost function is calculated based on the value weight data. in, Let i be the value weight of the i-th data item. Let λ be the acquisition latency of the i-th data item, λ be the penalty coefficient, and U be the predicted CPU utilization. This represents the maximum CPU utilization. The candidate sequence with the smallest delay cost function value is selected as the optimal running parameter and output to the protocol conversion module for execution. The optimal running parameter is written into the dynamic mapping table to drive the subsequent acquisition.

[0013] The photovoltaic protocol converter also includes a resource awareness and degradation module, which monitors the power supply voltage and remaining power of the photovoltaic protocol converter. When the remaining power is lower than a first threshold, the protocol conversion degradation mode is automatically triggered.

[0014] The protocol conversion degradation mode is as follows: it instructs the value weight of all data items except for device heartbeat, voltage over-limit, and overcurrent protection to be temporarily set to zero; it instructs the protocol conversion module to suspend the actual collection of data items with temporarily zero value weights, and read the most recent valid value from the historical data buffer or use the linear prediction value as a substitute; it instructs the IoT communication module to switch to non-acknowledgment message mode and reduce the reporting frequency; when the power supply voltage is restored and the remaining power is higher than the second threshold, it automatically exits the degradation mode, restores the data collection and communication functions, and compresses the historical data generated during the degradation period and retransmits it to the cloud platform through the IoT communication module.

[0015] The second aspect also provides a data acquisition method for a photovoltaic protocol converter based on the Internet of Things (IoT), applied to the IoT-based photovoltaic protocol converter described above, the method comprising: Data Acquisition: Communicate with the cloud platform to receive a data acquisition template containing the target data item set and the basic acquisition cycle for each data item; Protocol self-learning: Capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library; Protocol conversion: Semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the original data according to the mapping table and the effective acquisition cycle, and convert it into a standard format; Edge computing and evaluation: Monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, output monitoring data, and dynamically calculate the real-time value weight based on the historical change rate of each data item, the impact weight of power grid stability, and the cloud subscription priority, and output the value weight data. Collaborative scheduling: Based on monitoring data and value weight data, the operating parameters of the protocol conversion module are adjusted in real time, the adjusted operating parameters are rewritten into the dynamic mapping table, driving the subsequent collection and uploading the data to the cloud platform.

[0016] This invention relates to a photovoltaic protocol converter and data acquisition method based on the Internet of Things (IoT). Compared with existing technologies where protocol converters can only rely on preset protocol libraries or fixed mapping tables for data forwarding, the advantages are as follows: (1) This invention can generate a semantic feature library without relying on a preset protocol library or manual configuration, which reduces the dependence on inverter protocol documents, reduces the amount of manual debugging during the on-site deployment stage, and enhances the adaptability to inverters with unknown or private protocols. It also alleviates the maintenance difficulties caused by protocol version upgrades or equipment replacements in traditional solutions. The system can sense changes in its own operating status and data importance, and continuously optimize operating parameters such as acquisition cycle, timeout time, number of retries and batch read size. The optimization results are directly written into the dynamic mapping table through an internal feedback loop. Thus, under the constraints of limited computing and communication resources, the timeliness of high-value data acquisition is improved, and the performance degradation caused by resource competition is alleviated. This setting realizes the intelligence and self-adaptation of the protocol conversion process, which is conducive to improving the self-adaptation capability and data acquisition efficiency of the protocol converter in complex operating environments.

[0017] (2) By using Fast Fourier Transform to extract the main frequency component of the numerical change time series, it is possible to quickly distinguish between power frequency AC quantities and non-power frequency quantities. This frequency domain feature has strong robustness and is not easily affected by instantaneous noise, thereby improving the accuracy of the initial labeling of physical quantities. For register addresses initially labeled as AC, by calculating the sliding window correlation coefficient between its value and the inverter output power, and utilizing the physical law that current and power are highly positively correlated and voltage and power are weakly correlated, voltage and current can be distinguished in a quantitative way, avoiding the risk of misjudgment by simply relying on amplitude or phase. The judgment result, along with the register address, numerical range, and rate of change features, is stored in the semantic feature library, and a timestamp and confidence score are added to each semantic feature. This structured storage method facilitates the multi-candidate sorting and confidence fusion during subsequent semantic matching, alleviates the misjudgment problem that may exist in a single discrimination method, and improves the reliability of semantic mapping.

[0018] (3) By introducing the rate of change term, the data items that change drastically can obtain shorter acquisition delay and higher transmission priority in subsequent collaborative scheduling. Compared with the traditional fixed priority scheme, dynamic value assessment can respond to abnormal power grid events in a timely manner, which alleviates the problem of delayed processing of abnormal data caused by fixed priority. The power grid stability impact weight is used as a preset coefficient to ensure that key protection data, such as overvoltage, overcurrent and island detection, maintain a high value weight under any operating condition, avoiding the risk of low weight of key data under steady state caused by relying entirely on the rate of change. The normalization of the cloud subscription priority coefficient enables the cloud demand of different data items to be compared on the same scale, realizing the collaborative value definition between the edge side and the cloud. Compared with the scheduling method based on fixed priority or single threshold in the existing technology, it improves the rationality of resource allocation.

[0019] (4) By establishing a discrete-time dynamic model of the protocol conversion module, the mapping relationship between controllable input parameters such as acquisition cycle, timeout time, and batch read size and the conversion delay and CPU utilization rate at the next moment can be quantitatively expressed. This can predict the resource consumption and performance under different control strategies in advance, thus alleviating the problem of adjustment lag. In each control cycle, multiple sets of candidate operating parameter sequences are enumerated and optimized in the preset control time domain. This enables the system to adapt to the dynamic changes in the working conditions. Only the first optimal control action is executed each time, and the system is re-optimized in the next cycle, thus balancing real-time performance and optimality. The candidate sequence that minimizes the cost function is selected as the optimal operating parameter and output to the protocol conversion module for execution. Compared with the fixed acquisition cycle in the existing technology, the rolling optimization mechanism can actively reduce the acquisition frequency of low-value data items or increase the batch read size when resources are scarce, thereby improving the real-time performance of high-value data item acquisition. When resources are abundant, high-precision acquisition can be restored, thus alleviating the problem of resource waste.

[0020] (5) By monitoring the power supply voltage and remaining power in real time, the system can sense its own energy status. In outdoor or battery-powered application scenarios, the photovoltaic protocol converter may face unstable power supply. Traditional solutions often directly cut off the power or cause random faults when the power is insufficient, resulting in data loss and communication interruption. Through the active degradation mechanism, when the power is lower than the first threshold, the actual collection of non-critical data items is selectively suspended, and only critical data such as device heartbeat, voltage over-limit, and overcurrent protection are retained. Non-zero value weight data items are replaced with historical values ​​or linear prediction values. Under the premise of sacrificing some data accuracy, the working time of the device is significantly extended, and the risk of complete failure due to power depletion is mitigated. When the power supply voltage is restored and the remaining power is higher than the second threshold, the system automatically exits the degradation mode, restores all collection and communication functions, and compresses and retransmits the historical data generated during the degradation period to the cloud. The retransmission mechanism ensures the integrity of the data and avoids the permanent loss of information due to degradation. The compressed storage reduces the communication traffic and time required for retransmission. Through the combination mechanism of resource awareness, hierarchical degradation and delayed retransmission, the survivability and data integrity of the photovoltaic protocol converter in the unstable power supply environment are improved. Detailed Implementation

[0021] Exemplary embodiments of this disclosure will now be described in more detail.

[0022] Example 1 This embodiment provides a photovoltaic protocol converter based on the Internet of Things (IoT), including an IoT communication module, a protocol self-learning module, a protocol conversion module, an edge computing and evaluation module, and a collaborative scheduling module. The IoT communication module is electrically connected to the protocol self-learning module, the protocol self-learning module is electrically connected to the protocol conversion module, the edge computing and evaluation module is electrically connected to the protocol conversion module and the protocol scheduling module, and the protocol conversion module is electrically connected to the protocol scheduling module.

[0023] The IoT communication module is used to communicate with the cloud platform and receive a data acquisition template that includes the target data item set and the basic acquisition cycle of each data item.

[0024] The protocol self-learning module is used to connect to the power communication interface of the photovoltaic inverter, capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library.

[0025] The protocol conversion module is used to semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the dynamic mapping table and the effective acquisition cycle, and convert the raw data into a standard format.

[0026] The edge computing and evaluation module is used to monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, and output monitoring data. Based on the historical change rate of each target data item, the impact weight of power grid stability, and the cloud subscription priority, it dynamically calculates the real-time value weight and outputs the value weight data.

[0027] The collaborative scheduling module is used to adjust the operating parameters of the protocol conversion module in real time based on monitoring data and value weight data. The operating parameters include the actual acquisition cycle of each data item, conversion timeout, number of retries, and batch read size.

[0028] The protocol self-learning module calculates the physical quantity type corresponding to each register address and generates a semantic feature library. The specific operation is as follows: S101. Perform a fast Fourier transform on the time series of numerical changes extracted from the register address to extract the main frequency component; when the main frequency component falls within the power frequency range and the amplitude fluctuation range conforms to the characteristics of the power grid voltage, initially mark the register address as AC voltage or current. When the main frequency component is around 50Hz or 60Hz, it only indicates that the signal has power frequency periodicity characteristics, but it is not enough to distinguish whether it is a voltage or current signal, nor is it enough to eliminate noise or harmonic interference. Therefore, to conform to the power grid voltage characteristics, the following conditions must be met simultaneously: The amplitude of the numerical sequence falls within the normal grid voltage range. For example, the effective value for a 220V power system is between 198V and 242V. If the amplitude is much lower than the normal voltage range, it may be a control signal or measurement noise, rather than the grid voltage.

[0029] The waveform distortion rate is below the distortion rate threshold. If the harmonic content is too high, it may be due to switching noise or other interference.

[0030] In the absence of load changes or grid disturbances, the fluctuation range of the numerical sequence in steady state does not exceed ±2% of the rated value. If the sequence exhibits a slow unidirectional drift, it is more consistent with the characteristics of a temperature signal than grid voltage.

[0031] In a three-phase system, the deviation of the phase difference between any two phase voltages from 120° must fall within the allowable error range, and the phase difference between the voltage and the corresponding current should be within a reasonable range.

[0032] Therefore, in the initial marking stage of the protocol self-learning module, if the main frequency component is near the power frequency and the amplitude fluctuation range is within ±10% of the rated voltage, it meets the characteristics of the power grid voltage and is initially marked as AC voltage or current.

[0033] S102. For the register address initially marked as AC voltage or current, calculate the sliding window correlation coefficient between the register value and the output power of the photovoltaic inverter. If the absolute value of the correlation coefficient is greater than the first standard value, it is determined to be current; if the absolute value of the correlation coefficient is less than the second standard value, it is determined to be voltage. The correlation coefficient between the calculated register value and the sliding window of the photovoltaic inverter output power is as follows: For a register address initially marked as AC, let the value sequence of that register be... Simultaneously, the synchronization sequence for obtaining the output power of the photovoltaic inverter is... ,in, ; The Pearson correlation coefficient is calculated segment by segment using a sliding window method. The formula is as follows: in, is the correlation coefficient of the kth window, with a value range of [-1, 1]; W is the sliding window length, taking 60 to 120 sampling points, corresponding to 5 to 10 power frequency cycles; Let X be the mean of the register value sequence within the sliding window. The mean of the output power synchronization sequence Y within the sliding window.

[0034] Current and power are highly positively correlated, while voltage and power are weakly correlated, especially when the voltage at the grid connection point is supported by the grid and the change is not obvious. Finally, the median of the absolute values ​​of the correlation coefficients of multiple sliding windows is used for judgment. If the absolute value of the correlation coefficient is greater than the first standard value, it is judged as current. If the absolute value of the correlation coefficient is less than the second standard value, it is judged as voltage. If the absolute value of the correlation coefficient is between the first standard value and the second standard value, it is not judged for the time being, and the sampling window length is increased or more data is waited for analysis.

[0035] S103. For unmarked register addresses, calculate the median of the absolute value of the first difference of the time series of the register values. If the median is less than the rated preset value and the value drifts in one direction, it is determined to be temperature. S104. Store the judgment result and the corresponding register address, numerical range and rate of change features into the semantic feature library, and attach a timestamp and confidence score to each semantic feature.

[0036] The confidence score is used to characterize the reliability of the protocol self-learning module's determination of a single register address physical quantity. The formula for calculating the confidence score is: Where M represents the number of discrimination methods involved in the discrimination process. Let m be the preset weights for the m-th discrimination method. This represents the single-shot confidence level given by the m-th discrimination method, with a value ranging from 0 to 1.

[0037] Examples of the assignment logic for single-shot confidence scores for each discrimination method: FFT frequency analysis: If the frequency component is in the range of 48~52Hz or 58~62Hz, and the fundamental amplitude accounts for more than 95% of the total amplitude, then the value is 0.9; if the frequency exists but the harmonic content is high, then the value is 0.6; otherwise, the value is 0.2.

[0038] Sliding window correlation coefficient: If the judgment result is clear, the value is 0.9; if the absolute value of the correlation coefficient is between the first standard value and the second standard value, the judgment is ambiguous, and the value is 0.4.

[0039] First-order difference median: If the difference median is less than the rated preset value and the value drifts in one direction, it is determined to be temperature and the value is 0.85; if only one condition is met, the value is 0.5.

[0040] Nameplate parameter matching: If the mean of the numerical sequence deviates from the nameplate rated value by less than 10%, the value is 0.8; otherwise, the value is 0.3.

[0041] By extracting the dominant frequency component of the numerical change time series using Fast Fourier Transform (FFT), it is possible to quickly distinguish between power frequency AC quantities and non-power frequency quantities. This frequency domain feature has strong robustness and is not easily affected by transient noise, thus improving the accuracy of the initial labeling of physical quantities. For register addresses initially labeled as AC, by calculating the sliding window correlation coefficient between their values ​​and inverter output power, and utilizing the physical law that current and power are highly positively correlated while voltage and power are weakly correlated, voltage and current can be distinguished in a quantitative manner, avoiding the risk of misjudgment based solely on amplitude or phase. For non-AC quantities, the median of the absolute value of the first-order difference is used as the discrimination criterion. Taking advantage of the slow change and small difference value of temperature signals, it is possible to... This effectively eliminates random noise interference. Furthermore, the judgment result, along with register address, numerical range, and rate of change characteristics, is stored in the semantic feature library. Each semantic feature is appended with a timestamp and confidence score. This structured storage method facilitates multi-candidate sorting and confidence fusion during subsequent semantic matching, mitigating potential misjudgments by a single discrimination method and improving the reliability of semantic mapping. This setup allows multiple discrimination methods to complement each other, identifying features of different types of physical quantities such as AC quantities, current / voltage distinctions, and temperature, collectively enhancing the accuracy and robustness of protocol self-learning. This, in turn, improves the protocol converter's adaptability and data acquisition efficiency in complex operating environments.

[0042] The formula for calculating the real-time value weight in the edge computing and evaluation module is as follows: in, Let be the real-time value weight of the i-th data item at time t; Let be the rate of change of the i-th data item within the sliding window at time t. Let be the historical average rate of change of the i-th data item; Let be the preset weighting coefficient for the impact of the i-th data item on power grid stability, 0 < <1; Let be the cloud subscription priority coefficient for the i-th data item at time t, which is dynamically distributed by the cloud platform according to real-time scheduling needs or user settings. Let j represent the index of all target data items in the current data acquisition template, and N be the total number of target data items in the current data acquisition template. This represents the sum of the cloud subscription priority coefficients for all target data items at time t, used for... Normalization is performed. α β is the first weighting coefficient, β is the second weighting coefficient, and γ is the third weighting coefficient. a+b+c= 1.

[0043] The specific calculation formula for real-time value weight is defined by three weighted components: the first component is the ratio of the current rate of change of a data item to the historical average rate of change, reflecting the dynamic activity of the data; the second component is the preset weight of the data item's impact on power grid stability, reflecting the inherent importance of different data items in power grid security assessment; and the third component is the normalized value of the cloud subscription priority coefficient, representing the real-time demand of upper-layer applications for data items.

[0044] The technical advantages of this calculation formula are reflected in the following aspects: By introducing a rate of change term, data items with drastic changes, such as voltage drops and current surges, automatically acquire higher value weights, thereby achieving shorter acquisition delays and higher transmission priorities in subsequent collaborative scheduling; compared to traditional fixed-priority schemes, this dynamic value assessment can respond promptly to abnormal power grid events, alleviating the problem of delayed processing of abnormal data caused by fixed priorities; the power grid stability impact weight, as a preset coefficient, integrates domain prior knowledge, ensuring that critical protection data, such as overvoltage, overcurrent, and islanding detection, maintain high value weights under most operating conditions, avoiding the risk of excessively low weights for critical data in steady state due to complete reliance on the rate of change; cloud-based subscription priority coefficients... Normalization processing enables the cloud requirements of different data items to be compared on the same scale, and this coefficient can be dynamically issued by the cloud platform according to real-time scheduling needs or user settings, realizing the collaborative value definition between the edge and the cloud. The sum of the three weighted coefficients is 1, which normalizes the overall value weight to the 0-1 range, facilitating the calculation of the cost function in the subsequent model prediction and control algorithm. With this setting, the value weight calculation formula through multi-factor fusion realizes the quantitative expression of data value, providing a more scientific and refined basis for resource allocation of the collaborative scheduling module. Compared with the scheduling methods based on fixed priorities or single thresholds in the existing technology, it improves the rationality of resource allocation, thereby helping to improve the adaptability of the protocol converter and the data acquisition efficiency in complex operating environments.

[0045] The collaborative scheduling module employs a model predictive control algorithm, with the optimization objectives of minimizing the acquisition latency of high-value data items and maximizing the overall data acquisition value. It continuously optimizes operating parameters, specifically through the following steps: S201. Establish a discrete-time dynamic model. The inputs are the acquisition period, protocol conversion timeout time and batch read size of each data item at the current time. The outputs are the conversion delay and CPU utilization rate at the next time. S202. In each control cycle, starting from the current state, enumerate multiple sets of candidate operating parameter sequences within the preset control time domain. For each candidate sequence, calculate the weighted acquisition delay cost function based on the value weight data: in, Let i be the value weight of the i-th data item. Let λ be the acquisition latency of the i-th data item, λ be the penalty coefficient, and U be the predicted CPU utilization. This represents the maximum CPU utilization. S203. Select the candidate sequence with the smallest delay cost function value as the optimal operating parameter and output it to the protocol conversion module for execution. Write the optimal operating parameter into the dynamic mapping table through the internal feedback loop of the protocol conversion module to drive the protocol conversion module to execute a new round of acquisition.

[0046] By establishing a discrete-time dynamic model for the protocol conversion module, the mapping relationship between controllable input parameters such as acquisition period, timeout, and batch read size and the conversion delay and CPU utilization at the next moment is quantified. This allows the system to predict resource consumption and performance under different control strategies in advance, thus taking into account both the current state and future impacts when making optimization decisions. Compared to traditional reactive regulation, such as adjusting only after the delay exceeds the limit, model predictive control has feedforward characteristics, alleviating the problem of regulation lag. In each control cycle, multiple sets of candidate operating parameter sequences are enumerated and optimized in a rolling manner within a preset control time domain. This allows the system to adapt to dynamic changes in operating conditions, executing only the first optimal control action each time, and re-optimizing in the next cycle, thus balancing real-time performance and efficiency. Optimality: In the weighted acquisition latency cost function, the acquisition latency of high-value data items is given a larger weight, while a penalty term is introduced when the CPU utilization exceeds the upper limit. This ensures that the optimization objective simultaneously considers the economy of data acquisition and the safety of equipment operation. This multi-objective trade-off mechanism avoids the risk of CPU overload that may result from simply pursuing low latency. The candidate sequence that minimizes the cost function is selected as the optimal operating parameters and output to the protocol conversion module for execution. Compared with the fixed acquisition cycle in existing technologies, the rolling optimization mechanism can proactively reduce the acquisition frequency of low-value data items or increase the batch read size when resources are scarce, thereby improving the real-time acquisition of high-value data items. When resources are abundant, high-precision acquisition can be restored, alleviating the problem of resource waste. With this setting, the dynamic optimization of protocol conversion operating parameters is achieved through the model predictive control algorithm, improving the system's acquisition performance under resource constraints. This, in turn, helps to improve the adaptability of the protocol converter and the data acquisition efficiency in complex operating environments.

[0047] The photovoltaic protocol converter also includes a resource awareness and degradation module, which monitors the power supply voltage and remaining power of the photovoltaic protocol converter. When the remaining power is lower than a first threshold, the protocol conversion degradation mode is automatically triggered.

[0048] The protocol conversion and downgrade mode is as follows: S301 indicates that the value weight of all data items except for equipment heartbeat, voltage over-limit and overcurrent protection will be temporarily set to zero; When the remaining power falls below the first threshold, the photovoltaic protocol converter enters degrade mode. At this time, not all data acquisition tasks need to be maintained, but the following three types of data items are crucial for grid security and equipment protection and must be collected and reported: Device heartbeats are used to prove to the cloud platform that the protocol converter is still online and running. They include the device ID, timestamp, and running status flags. Losing heartbeats can cause the cloud to mistakenly determine that the device is offline, thus triggering unnecessary alarms or maintenance tickets.

[0049] Voltage exceeding limits: When the voltage at the grid connection point exceeds the normal range, it may cause equipment damage, grid instability, or protection device activation. It needs to be reported in real time so that the cloud can promptly dispatch reactive power compensation or adjust the grid operation mode.

[0050] Overcurrent protection: When the inverter output current exceeds 1.2 to 1.5 times the rated value, it may indicate a line short circuit, grounding fault, or internal inverter fault. It should be reported first so that maintenance personnel can quickly locate the fault and perform protection actions.

[0051] Other data items, such as inverter temperature, daily power generation, and irradiance, are non-critical data in degraded mode. Actual data collection can be temporarily suspended, and historical values ​​or linear prediction values ​​can be used instead to extend the equipment's operating time.

[0052] S302, Instructs the protocol conversion module to pause the actual collection of temporary zero-value weighted data items and instead read the most recent valid value from the historical data buffer or use the linear predicted value as a substitute; S303, instructs the IoT communication module to switch to UDP-based CoAP unacknowledged message mode and reduce the reporting frequency; S304. When the power supply voltage is restored and the remaining power is higher than the second threshold, the degradation mode is automatically exited, the data acquisition and communication functions are restored, and the historical data generated during the degradation period is compressed and then transmitted to the cloud platform through the IoT communication module.

[0053] By adopting the above technical solution, the system can perceive its own energy status by monitoring the supply voltage and remaining power in real time, providing a basis for decision-making. In outdoor or battery-powered application scenarios, photovoltaic protocol converters may face unstable power supply. Traditional solutions often directly cut off power or cause random faults when the power is insufficient, resulting in data loss and communication interruption. Through the active degradation mechanism, when the power is below the first threshold, the actual collection of non-critical data items is selectively suspended, and only critical data such as device heartbeat, voltage over-limit, and overcurrent protection are retained. Non-zero value weight data items are replaced with historical values ​​or linear prediction values. Thus, while sacrificing some data accuracy, the working time of the device is significantly extended, and the risk of complete failure due to power depletion is mitigated. In degradation mode, the IoT communication module is instructed to switch from MQTT long connection mode to UDP-based CoAP non-acknowledgment message mode and reduce the reporting frequency. MQTT long connection requires maintaining TCP connection and heartbeat keep-alive, which consumes power. The CoAP unacknowledged mode is based on UDP, requires no connection maintenance, and has low message header overhead. After switching, it can effectively reduce communication power consumption. This dynamic switching of communication protocols is more conducive to maintaining basic communication capabilities under energy-constrained conditions than using a fixed communication method. When the power supply voltage is restored and the remaining power is higher than the second threshold, the system automatically exits the degradation mode, restores all acquisition and communication functions, and compresses and retransmits the historical data generated during the degradation period to the cloud. The retransmission mechanism ensures the integrity of the data and avoids permanent loss of information due to degradation. The compressed storage reduces the communication traffic and time required for retransmission. With this setting, the combination of resource awareness, hierarchical degradation and delayed retransmission improves the survivability and data integrity of the photovoltaic protocol converter in unstable power supply environments. It alleviates the problem of indiscriminate power outages or data loss in traditional solutions when power is insufficient, thereby improving the adaptability and data acquisition efficiency of the protocol converter in complex operating environments.

[0054] This embodiment also provides a data acquisition method for a photovoltaic protocol converter based on the Internet of Things, applied to the aforementioned photovoltaic protocol converter. The method specifically includes: Data Acquisition: Communicate with the cloud platform to receive a data acquisition template containing the target data item set and the basic acquisition cycle for each data item; Protocol self-learning: Capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library; Protocol conversion: Semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the original data according to the mapping table and the effective acquisition cycle, and convert it into a standard format; Edge computing and evaluation: Monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, output monitoring data, and dynamically calculate the real-time value weight based on the historical change rate of each data item, the impact weight of power grid stability, and the cloud subscription priority, and output the value weight data. Collaborative scheduling: Based on monitoring data and value weight data, the operating parameters of the protocol conversion module are adjusted in real time. The operating parameters include the actual collection cycle, conversion timeout, number of retries, and batch read size for each data item. The adjusted operating parameters are rewritten into the dynamic mapping table to drive subsequent collection and upload the data to the cloud platform.

[0055] Example 2 This embodiment discloses a photovoltaic protocol converter based on the Internet of Things (IoT), including an IoT communication module, a protocol self-learning module, a protocol conversion module, an edge computing and evaluation module, a collaborative scheduling module, and a resource awareness and degradation module.

[0056] In this example, the IoT communication module uses a wireless communication chip that supports both MQTT and CoAP dual protocol stacks, such as the ESP32-C3 or SIM7070G module. This module establishes a secure communication connection with the cloud platform via 4G / 5G or Wi-Fi, and uses TLS / DTLS encryption during the connection.

[0057] After the device powers on and initializes, the IoT communication module sends a registration request to the cloud platform, carrying the device's unique identifier, such as IMEI or a custom ID. Once the cloud platform verifies the request, it issues an initial configuration command, which includes a data acquisition template. The data acquisition template is a JSON-formatted configuration file. After parsing the template, the IoT communication module stores the target data item set and the basic acquisition cycle for each data item in its local memory for use by other modules.

[0058] The protocol self-learning module is used to connect to the power communication interface of the photovoltaic inverter, capture the original messages when it communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library.

[0059] The protocol self-learning module is also used to obtain the nameplate parameters or standard configuration file of the photovoltaic inverter. The nameplate parameters include at least the rated voltage, rated current, rated power and temperature limit. The statistical features of the numerical change time series, such as mean, variance and extreme values, are compared with the nameplate parameters. If the matching degree between the statistical features and the nameplate parameters of a certain physical quantity exceeds a preset threshold, the type of physical quantity corresponding to the register address is inferred. The matching degree is calculated using Euclidean distance or cosine similarity, and the inference result is stored in the semantic feature library.

[0060] In this example, the protocol self-learning module is implemented through a physical connection to the power communication interface of the photovoltaic inverter, such as RS-485, CAN, or PLC. This module contains a high-speed message capture unit that adopts a passive listening mode, so as not to interfere with the normal communication between the inverter and the original monitoring equipment, such as the data acquisition unit.

[0061] In its implementation, the protocol self-learning module uses a microcontroller in conjunction with a high-performance serial peripheral interface and direct memory access technology to continuously capture communication messages. For each complete request-response message pair captured, the module extracts the register address and its corresponding time series of numerical changes. To analyze the patterns of numerical changes, the module incorporates a sliding time window with a default window length of 60 seconds and a sampling interval of 100 milliseconds, storing the latest 600 numerical points for each register address.

[0062] The protocol conversion module is used to semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the mapping table and the effective acquisition cycle, and convert it into a standard format.

[0063] In this example, the core of the protocol conversion module includes a semantic mapper and an executor. The semantic mapper receives data acquisition templates from the IoT communication module and a semantic feature library from the protocol self-learning module, and executes a matching algorithm: Traverse each target data item in the data acquisition template; search the semantic feature library for register addresses whose physical quantity type matches the semantics of the target data item; establish a dynamic mapping table, where each mapping record contains: target data item name, register address, and data conversion formula, such as linear transformation: actual value = original value × coefficient + offset, as well as acquisition period, initially the basic acquisition period.

[0064] The semantic mapper is also used to: parse the implicit data dependencies in the data acquisition template and generate a directed acyclic graph; each node of the directed acyclic graph represents a target data item, and the directed edges represent computational or temporal dependencies between data items; according to the directed acyclic graph, data items with an in-degree of zero are preferentially matched to the registered addresses already identified in the semantic feature library and bound to a group of register addresses that can be read concurrently, so as to reduce the number of serial queries.

[0065] According to the dynamic mapping table and the current valid acquisition cycle, the actuator sends a read command to the inverter through the power communication interface, such as Modbus function code 03, to obtain the raw data. Then, it calculates the data into a standard format according to the conversion formula in the mapping table, unifies it into floating-point numbers, uses the International System of Units (SI), and stores the standard format data into a shared memory buffer.

[0066] To improve concurrent read efficiency, the semantic mapper also parses the implicit data dependencies in the data acquisition template, generating a directed acyclic graph. For example, if power = voltage × current, then power depends on voltage and current. Data items with an in-degree of zero, such as voltage and current, are preferentially bound to register address groups that can be read concurrently, for example, using Modbus's batch read function to read consecutive addresses at once. In this way, the actuator can read multiple independent data items in a single communication transaction, significantly reducing the number of serial lookups.

[0067] The edge computing and evaluation module is used to monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU and memory usage. It outputs monitoring data and dynamically calculates the real-time value weight based on the historical change rate of each data item, the impact weight of power grid stability, and the cloud subscription priority, and outputs the value weight data.

[0068] In this embodiment, the cloud subscription priority coefficient The acquisition method is as follows: the cloud platform dynamically assigns a priority coefficient to each data item based on factors such as the real-time operating status of the power grid, market electricity price signals, and user-defined strategies. For example, during peak grid load periods, the cloud may set the priority of the grid connection point power to 0.9, while setting the internal temperature of the inverter to 0.1. This coefficient is sent to the protocol converter periodically or triggered by events through the IoT communication module.

[0069] In this embodiment, the normalized denominator The method for obtaining this is as follows: Since the priority coefficients of different data items may be distributed in different numerical ranges, the result is obtained by dividing by the sum of the priorities of all data items. The result strictly falls between 0 and 1, and is independent of the number of data items, making it easy to correlate with... and The two items are then summed using a weighted average.

[0070] The edge computing and evaluation module also includes a collaborative controller, which triggers a data dimensionality reduction strategy when the conversion latency exceeds a first threshold or the CPU utilization exceeds a second threshold. Data items in the dynamic mapping table whose real-time value weight is lower than a preset level are marked as frequency reduction items, and their collection cycle is extended. The down-frequency item is temporarily removed from the dynamic mapping table, and an aggregation rule is generated to generate an approximate value of the down-frequency item by linear interpolation or historical trend fitting using the collected high-value data items. The aggregation rules are sent to the semantic mapper so that the approximate value can be directly output during subsequent protocol conversion without actually collecting the original data of the down-frequency term.

[0071] The edge computing and evaluation module runs on the MCU of the protocol converter, utilizing multiple tasks of a lightweight real-time operating system. The edge computing and evaluation module comprises two sub-units: a performance observer and a data value evaluator.

[0072] Performance observer: By reading the task operation logs of the protocol conversion module, it calculates the average conversion latency for each acquisition cycle, i.e., the time from sending the read command to receiving the response and completing the conversion; monitors the communication success rate of the power communication interface, i.e., the number of successful responses / the total number of requests; and obtains CPU utilization and memory utilization through the operating system API, with a sampling period of 1 second.

[0073] Data value evaluator: Calculates the value weight of each data item in real time, enabling the edge computing and evaluation module to output the latest value weight data every control cycle, such as 10 seconds, for use by the collaborative scheduling module.

[0074] Collaborative scheduling module: Used to adjust the operating parameters of the protocol conversion module in real time based on monitoring data and value weight data. The operating parameters include the actual collection cycle of each data item, conversion timeout, number of retries and batch read size. The collaborative scheduling module uses the Model Predictive Control (MPC) algorithm to achieve rolling optimization of operating parameters. The specific implementation steps are as follows: A discrete-time dynamic model was established, historical data was collected, and a simplified model of the protocol conversion module was identified.

[0075] Enumerate candidate sequences, control the time domain to 3 periods, and the optional parameter range for each period is: acquisition period and batch size. Use grid search or genetic algorithm to enumerate a finite number of candidate sequences. The cost function is calculated, optimal parameters are selected, and a candidate sequence that minimizes J is chosen. The acquisition cycle, timeout, number of retries, and batch size are then written into a dynamic mapping table. The next acquisition cycle of the protocol conversion module will be executed according to the new parameters.

[0076] In addition, a data acquisition method for a photovoltaic protocol converter based on the Internet of Things is provided, including data acquisition steps, protocol self-learning steps, protocol conversion steps, edge computing and evaluation steps, collaborative scheduling steps, and resource awareness and degradation steps.

[0077] Data acquisition steps: After the protocol converter is powered on, it first initializes the network connection, registers with the cloud platform, and after successful registration, communicates with the cloud platform to receive a data acquisition template containing the target data item set and the basic acquisition cycle of each data item.

[0078] Protocol self-learning steps: capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, perform frequency domain analysis and time domain feature extraction, calculate the physical quantity type corresponding to each register address, and generate a semantic feature library; Protocol conversion steps: Semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the mapping table and the effective acquisition cycle, and convert it into a standard format.

[0079] Edge computing and evaluation steps: Monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU and memory usage, output monitoring data, dynamically calculate the real-time value weight based on the historical change rate of each data item, the impact weight of power grid stability, and the cloud subscription priority, and output the value weight data. Collaborative scheduling steps: Based on monitoring data and value weight data, adjust the operating parameters of the protocol conversion module in real time. The operating parameters include the actual collection cycle of each data item, conversion timeout, number of retries, and batch read size. Among them, the model predictive control algorithm is adopted to minimize the acquisition delay of high-value data items and maximize the overall data acquisition value as the optimization objectives. The operating parameters are continuously optimized and written into the dynamic mapping table through the internal feedback loop to drive the subsequent acquisition and upload the data to the cloud platform. The specific process of rolling optimization of operating parameters includes: establishing a discrete-time dynamic model of the protocol conversion module; enumerating candidate operating parameter sequences in the current control cycle; calculating the weighted acquisition delay cost function of each group of candidate sequences based on real-time value weight data; selecting the candidate sequence with the smallest cost function as the optimal operating parameter; writing the optimal operating parameter into the dynamic mapping table; and driving the adjustment of acquisition behavior.

Claims

1. An Internet of Things based photovoltaic protocol converter, characterized in that, It includes an IoT communication module, a protocol self-learning module, a protocol conversion module, an edge computing and evaluation module, and a collaborative scheduling module; The IoT communication module is used to communicate with the cloud platform and receive a data acquisition template containing the target data item set and the basic acquisition cycle of each data item. The protocol self-learning module is used to connect to the power communication interface of the photovoltaic inverter, capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library. The protocol conversion module is used to semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the dynamic mapping table and the effective acquisition cycle, and convert the raw data into a standard format. The edge computing and evaluation module is used to monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, and output monitoring data. Based on the historical change rate of each target data item, the impact weight of power grid stability, and the cloud subscription priority, it dynamically calculates the real-time value weight and outputs the value weight data. The collaborative scheduling module is used to adjust the operating parameters of the protocol conversion module in real time based on monitoring data and value weight data. The operating parameters include the actual acquisition cycle of each data item, conversion timeout, number of retries, and batch read size.

2. The IoT-based PV protocol converter of claim 1, wherein, The protocol self-learning module calculates the physical quantity type corresponding to each register address and generates a semantic feature library. The specific operation is as follows: A fast Fourier transform is performed on the time series of numerical changes extracted from the register address to extract the main frequency component; when the main frequency component falls within the power frequency range and the amplitude fluctuation range conforms to the characteristics of the power grid voltage, the register address is initially marked as AC voltage or current; For register addresses initially labeled as AC voltage or current, calculate the sliding window correlation coefficient between the register value and the output power of the photovoltaic inverter. If the absolute value of the correlation coefficient is greater than the first standard value, it is determined to be current; if the absolute value of the correlation coefficient is less than the second standard value, it is determined to be voltage. For unmarked register addresses, calculate the median of the absolute value of the first difference of the time series of the register values. If the median is less than the preset value and the value drifts in one direction, it is determined to be temperature. The judgment result, along with the corresponding register address, numerical range, and rate of change characteristics, is stored in the semantic feature library, and a timestamp and confidence score are added to each semantic feature.

3. The IoT-based PV protocol converter of claim 2, wherein, The correlation coefficient between the calculated register value and the sliding window of the photovoltaic inverter output power is as follows: For a register address initially marked as AC, let the value sequence of that register be... Simultaneously, the synchronization sequence for obtaining the output power of the photovoltaic inverter is... ,in, ; The Pearson correlation coefficient is calculated segment by segment using a sliding window method. The formula is as follows: in, is the correlation coefficient of the k-th window, with a value range of [-1, 1]; W is the length of the sliding window; Let X be the mean of the register value sequence within the sliding window. The mean of the output power synchronization sequence Y within the sliding window.

4. The IoT-based PV protocol converter of claim 2, wherein, The grid voltage characteristics include the amplitude of the numerical sequence falling within the normal grid voltage range, the waveform distortion rate being lower than the distortion rate threshold, the fluctuation range of the numerical sequence not exceeding ±2% of the rated value, and the deviation of the phase difference between any two phases in the three-phase system from 120° falling within the allowable error range.

5. The IoT-based PV protocol converter of claim 1, wherein, The formula for calculating the real-time value weight in the edge computing and evaluation module is as follows: in, Let be the real-time value weight of the i-th data item at time t; Let be the rate of change of the i-th data item within the sliding window at time t. Let be the historical average rate of change of the i-th data item; Let be the preset weighting coefficient for the impact of the i-th data item on power grid stability, 0 < <1; Let j be the cloud subscription priority coefficient for the i-th data item at time t, j represent the index of all target data items in the current data acquisition template, and N be the total number of target data items in the current data acquisition template. This represents the sum of the cloud subscription priority coefficients for all target data items at time t, used for... Normalization is performed. α β is the first weighting coefficient, β is the second weighting coefficient, and γ is the third weighting coefficient. α+β+γ= 1.

6. The IoT-based PV protocol converter of claim 1, wherein, The protocol scheduling module employs a model predictive control algorithm to optimize operating parameters by minimizing the acquisition delay of high-value data items and maximizing the overall data acquisition value.

7. The IoT-based PV protocol converter of claim 6, wherein, The specific operation of the rolling optimization running parameters is as follows: Establish a discrete-time dynamic model, with the inputs being the acquisition period, protocol conversion timeout, and batch read size of each data item at the current moment, and the outputs being the conversion delay and CPU utilization at the next moment; In each control cycle, starting from the current state, multiple sets of candidate operating parameter sequences are enumerated within the preset control time domain. For each candidate sequence, a weighted acquisition delay cost function is calculated based on the value weight data. wherein, is the value weight of the i-th data item, is the collection delay of the i-th data item, λ is a penalty coefficient, and U is the predicted CPU occupancy, is the CPU occupancy upper limit; The candidate sequence with the smallest delay cost function value is selected as the optimal running parameter and output to the protocol conversion module for execution. The optimal running parameter is written into the dynamic mapping table to drive the subsequent acquisition.

8. The IoT-based PV protocol converter of claim 1, wherein, It also includes a resource awareness and degradation module, which monitors the power supply voltage and remaining power of the photovoltaic protocol converter. When the remaining power is lower than the first threshold, the protocol conversion degradation mode is automatically triggered.

9. The IoT-based photovoltaic protocol converter according to claim 8, characterized in that, The protocol conversion degradation mode is as follows: it instructs the value weight of all data items except for device heartbeat, voltage over-limit, and overcurrent protection to be temporarily set to zero; it instructs the protocol conversion module to suspend the actual collection of data items with temporarily zero value weights, and read the most recent valid value from the historical data buffer or use the linear prediction value as a substitute; it instructs the IoT communication module to switch to non-acknowledgment message mode and reduce the reporting frequency; when the power supply voltage is restored and the remaining power is higher than the second threshold, it automatically exits the degradation mode, restores the data collection and communication functions, and compresses the historical data generated during the degradation period and retransmits it to the cloud platform through the IoT communication module.

10. A data acquisition method of an Internet of Things based photovoltaic protocol converter, applied to the Internet of Things based photovoltaic protocol converter of any one of claims 1-9, characterized in that, The method includes: Data Acquisition: Communicate with the cloud platform to receive a data acquisition template containing the target data item set and the basic acquisition cycle for each data item; Protocol self-learning: Capture the original messages when the photovoltaic inverter communicates with the monitoring equipment, extract the time series of numerical changes of each register address, and calculate the physical quantity type corresponding to each register address through frequency domain analysis and time domain feature extraction to generate a semantic feature library; Protocol conversion: Semantically match the target data item with the register address in the semantic feature library, establish a dynamic mapping table, read the raw data according to the mapping table and the effective acquisition cycle, and convert it into a standard format; Edge computing and evaluation: Monitor the conversion latency of the protocol conversion module, the communication success rate of the power communication interface, and the CPU utilization rate, output monitoring data, and dynamically calculate the real-time value weight based on the historical change rate of each data item, the impact weight of power grid stability, and the cloud subscription priority, and output the value weight data. Collaborative scheduling: Based on monitoring data and value weight data, the operating parameters of the protocol conversion module are adjusted in real time, the adjusted operating parameters are rewritten into the dynamic mapping table, driving the subsequent collection and uploading the data to the cloud platform.

Images