Conflux box communication system based on high-reliability communication module

Abstract

The invention relates to a combiner box communication system based on a high-reliability communication module, and the system comprises the communication module which is used for continuously collecting interference pulse time sequence data and signal attenuation factor time sequence data after a photovoltaic DC combiner box is connected with a monitoring system according to a target communication protocol; the signal coding module is used for determining an interference type corresponding to the interference pulse time sequence data and calling a corresponding pulse coding strategy according to the interference type so as to determine an effective pulse coding sequence of the target pulse signal in the next time period; and the signal compensation module is used for determining the signal intensity value of the target pulse signal at each sampling moment in the next time period, and compensating the signal intensity value of the target pulse signal at each sampling moment in the next time period by adopting a preset pulse signal compensation strategy. According to the application, the problems of weak interference resistance, long-distance attenuation, poor protocol compatibility and the like of a traditional combiner box communication system are effectively solved, and the communication reliability is remarkably improved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic communication, and in particular to a combiner box communication system based on a high-reliability communication module. Background Technology

[0002] Traditional photovoltaic (PV) DC combiner box communication systems connect to the backend monitoring system via a bus, requiring manual on-site configuration, such as manually setting the baud rate and address code. Due to the complex environment of PV areas, PV DC combiner box communication systems are susceptible to high temperatures and strong electromagnetic interference. Furthermore, the dispersed distribution of PV DC combiner boxes, with distances reaching hundreds of meters, makes general-purpose communication modules prone to interference and communication interruptions. Additionally, communication protocols differ between monitoring systems from different brands, resulting in poor module compatibility and requiring customized development, increasing debugging costs. Traditional PV DC combiner box communication systems suffer from the following technical drawbacks: Firstly, traditional photovoltaic DC combiner box communication systems have weak anti-interference capabilities and are prone to data packet loss or disconnection in strong electromagnetic environments, leading to frequent alarms ("screen refresh") in the background monitoring system and affecting operation and maintenance efficiency.

[0003] Secondly, the signal attenuation is severe during long-distance transmission, and the module needs to be manually restarted after the line is disconnected. The recovery time is long, usually more than 30 minutes, resulting in missing power generation statistics.

[0004] Thirdly, the traditional photovoltaic DC combiner box communication system requires manual parameter modification for protocol adaptation with monitoring systems from different brands, resulting in poor compatibility and increased installation and debugging time costs.

[0005] In summary, existing photovoltaic DC combiner box communication systems have weak anti-interference capabilities, are prone to data packet loss or disconnection in strong electromagnetic environments, and require manual parameter modification for protocol adaptation with different brands of monitoring systems, resulting in poor compatibility. The applicant has made corresponding explorations to address these issues. Summary of the Invention

[0006] The purpose of this application is to solve the above problems by providing a combiner box communication system based on a high-reliability communication module.

[0007] To achieve the above objectives, the present application adopts the following technical solution: A combiner box communication system based on a high-reliability communication module includes: a protocol configuration module, which is used to acquire a handshake request pulse sequence of multiple target pulse signal transmission requests initiated by the monitoring system to the photovoltaic DC combiner box, transmit the handshake request pulse sequence to the first cyclic pulse memory sub-network in the communication scheduling control model that has been trained to a converged state, so as to extract the timing feature vector of the protocol to be matched, and match the corresponding target communication protocol in the communication protocol library according to the timing feature vector of the protocol to be matched, wherein the target pulse signal includes a voltage pulse signal and its corresponding current pulse signal; The communication module is used to establish a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol, and then continuously collect the interference pulse timing data and signal attenuation factor timing data corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period. The signal encoding module is used to transmit the interference pulse timing data to the second cyclic pulse memory sub-network in the communication scheduling control model that has been trained to convergence, so as to determine the interference type corresponding to the interference pulse timing data, and call the corresponding pulse encoding strategy according to the interference type to determine the effective pulse encoding sequence of the target pulse signal in the next time period. The signal compensation module is used to transmit the signal attenuation factor timing data and the effective pulse coding sequence in the next time period to the third cyclic pulse memory sub-network in the communication scheduling and control model that has been trained to convergence state, so as to determine the signal strength value of the target pulse signal at each sampling moment in the next time period, and to compensate the signal strength value of the target pulse signal at each sampling moment in the next time period using a preset pulse signal compensation strategy, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding.

[0008] Optionally, the handshake request pulse sequence includes a start bit, an address code, and a check code, and the timing feature vector of the protocol to be matched includes the duration of the start bit pulse, the time difference between adjacent address code pulses, and the generation time difference of the check code relative to the address code. The communication module includes an electromagnetic shielding unit, and the interference pulse timing data characterizes the timing data constructed by the occurrence time of inverter noise data and high-voltage cable radiation data, the phase of the target pulse signal, and the amplitude of the target pulse signal. The signal attenuation factor time series data characterizes the time series data constructed from the transmission distance between the photovoltaic DC combiner box and the monitoring system, the cable impedance, and the ambient temperature of the photovoltaic DC combiner box within the current time period. The types of interference include random interference, periodic interference, or sudden interference; The basic network architecture of the communication scheduling and control model is an improved recurrent spiking neural network. The improved recurrent spiking neural network includes a pulse input layer, a pulse timing coding layer, a recurrent pulse memory layer, and a backpropagation layer. The recurrent pulse memory layer includes a first recurrent pulse memory subnetwork, a second recurrent pulse memory subnetwork, and a third recurrent pulse memory subnetwork. The basic network architecture of the first recurrent pulse memory subnetwork is a hybrid LSTM-CNN structure, and the basic network architectures of the second and third recurrent pulse memory subnetworks are hybrid LSTM-SNN structures.

[0009] Optionally, the handshake request pulse sequence is transmitted to the first cyclic pulse memory subnetwork in the communication scheduling control model that has been trained to a convergent state to extract the timing feature vector of the protocol to be matched, and the corresponding target communication protocol is matched in the communication protocol library according to the timing feature vector of the protocol to be matched, including: The first LSTM unit in the first cyclic pulse memory sub-network is used to perform time-series feature mapping on the generation time difference between the check code and the address code in the handshake request pulse sequence to determine the first time-series associated feature, so as to construct the first time-series associated feature vector. The first CNN unit in the first cyclic pulse memory sub-network extracts the duration of the start bit pulse from the handshake request pulse sequence to determine the first discrete feature and construct the first discrete feature vector, and extracts the time difference between adjacent address code pulses from the handshake request pulse sequence to determine the second discrete feature and construct the second discrete feature vector. The first discrete feature vector, the second discrete feature vector, and the first temporal correlation feature vector after dimensional standardization are concatenated to construct the temporal feature vector of the protocol to be matched. The cosine similarity between the timing feature vector of the protocol to be matched and the timing feature vector of each standard communication protocol in the communication protocol library in the protocol configuration module is calculated. If the cosine similarity is greater than a preset similarity threshold, the standard communication protocol is used as the target communication protocol for establishing a connection between the monitoring system and the photovoltaic DC combiner box.

[0010] Optionally, establishing a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol includes: The protocol configuration module generates a pulse control command containing baud rate, parity method and data bit length according to the target communication protocol, and sends the pulse control command to the communication module to configure the communication parameters in the communication module; A handshake response pulse sequence is generated based on the communication parameters and returned to the monitoring system, so that the photovoltaic DC combiner box establishes a communication connection with the monitoring system.

[0011] Optionally, the interference pulse timing data is transmitted to the second cyclic pulse memory subnetwork in the communication scheduling and control model that has been trained to a convergent state, in order to determine the interference type corresponding to the interference pulse timing data, including: The electromagnetic shielding unit in the communication module collects the timing data of the interference pulses corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period; The second LSTM unit in the second cyclic pulse memory sub-network is used to perform feature mapping on the long-term dependency relationship between the occurrence time, phase and amplitude of the interference pulse time series data, and to determine the second time series correlation feature to construct the second time series correlation feature vector. The second SNN unit in the second cyclic pulse memory sub-network captures the phase difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the third discrete feature to construct the third discrete feature vector, and extracts the amplitude difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the fourth discrete feature to construct the fourth discrete feature vector. The third discrete feature vector, the fourth discrete feature vector, and the second time-series correlation feature vector, after dimensionality standardization, are concatenated to construct the interference feature vector. The interference feature vector is then matched with a preset interference type feature library to determine the interference type corresponding to the interference pulse time-series data.

[0012] Optionally, based on the interference type, the corresponding pulse coding strategy is invoked to determine the effective pulse coding sequence of the target pulse signal in the next time period, including: If the interference type corresponding to the interference pulse timing data is random interference, then the time-phase joint coding strategy in the communication scheduling control model is invoked to fix the transmission time of the effective pulse signal in the next time period, and the phase of the effective pulse signal is locked within a preset phase threshold range to determine the effective pulse coding sequence in the next time period. The preset phase threshold range is within an interval of no more than 5° above or below 0°. If the interference type corresponding to the interference pulse timing data is periodic interference, then the timing interference avoidance coding strategy in the communication scheduling control model is invoked to offset the emission time of the effective pulse signal in the next time period by half of the interference period, so as to determine the effective pulse coding sequence in the next time period. The effective pulse coding sequence includes the pulse emission time, phase and frequency of the pulse signal. If the interference type corresponding to the interference pulse timing data is sudden interference, then the triple redundancy coding strategy in the communication scheduling control model is invoked to generate multiple pulse signals of the same duration with a preset time offset, so as to determine the effective pulse coding sequence in the next time period.

[0013] Optionally, the signal attenuation factor time-series data and the effective pulse coding sequence in the next time period are transmitted to the third cyclic pulse memory subnetwork in the communication scheduling control model that has been trained to convergence, in order to determine the signal strength value of the target pulse signal at each sampling time in the next time period, including: The third LSTM unit in the third cyclic pulse memory sub-network is used to perform time-series correlation feature mapping on the long-term dependency between the signal attenuation factor time-series data and the effective pulse coding sequence to determine the third time-series correlation feature, so as to construct the third time-series correlation feature vector. The fifth discrete feature is determined by capturing the discrete feature changes of the effective pulse coding sequence under different signal attenuation factor time series data combinations through the third SNN unit in the third cyclic pulse memory sub-network, so as to construct the fifth discrete feature vector. The third temporal correlation feature and the fifth discrete feature vector, after dimensionality standardization, are concatenated to construct a fused feature vector. The fused feature vector is then transmitted to the fully connected layer in the third cyclic pulse memory sub-network. After nonlinear transformation by the Sigmoid activation function, it is then denormalized and mapped to map the fused feature vector to the signal strength values ​​at each sampling time in the next time period.

[0014] Optionally, a preset pulse signal compensation strategy is used to compensate the signal strength values ​​of the target pulse signal at each sampling moment in the next time period, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding, including: A preset pulse signal compensation strategy is invoked to detect whether the signal strength value of the target pulse signal at each sampling moment in the next time period falls within the preset signal strength range. If it does, the current transmission power of the photovoltaic DC combiner box is gradually increased to the target transmission power corresponding to the preset signal strength range, and the corresponding resistance parameters are matched according to the transmission distance between the photovoltaic DC combiner box and the monitoring system.

[0015] Optionally, the standard communication protocol includes the Modbus-RTU communication protocol and the DL / T 645 communication protocol.

[0016] Optionally, the protocol configuration module includes an industrial-grade STM32F407 main control chip, an integrated pulse signal acquisition interface, a Flash protocol storage unit, and a UART communication interface.

[0017] Compared to existing technologies, this application addresses the shortcomings of traditional photovoltaic DC combiner box communication systems, such as weak anti-interference capabilities, susceptibility to data packet loss or disconnection in strong electromagnetic environments, and poor compatibility due to the need for manual parameter modification for protocol adaptation with different brands of monitoring systems. This application addresses these pain points through a comprehensive innovative design encompassing intelligent protocol adaptation, accurate interference identification, dynamic encoding optimization, and pre-compensation for signals. The benefits of this application include, but are not limited to, the following: Firstly, traditional photovoltaic DC combiner box communication systems suffer from differences in communication protocols among different brands of monitoring systems, such as baud rate, address code format, and verification rules. This necessitates manual on-site parameter modification, resulting in excessively long debugging times, poor compatibility, and high installation and debugging costs. This application achieves automatic protocol adaptation through handshake pulse parsing and feature vector matching. Based on an LSTM-CNN hybrid structure, the first cyclic pulse memory subnetwork extracts discrete features such as the duration of the start bit and the time difference between adjacent address codes from the handshake request pulse sequence of the monitoring system through CNN units. The LSTM unit captures the temporal features of the generation time difference between the check code and the address code, constructing a unique temporal feature vector for the protocol to be matched. This temporal feature vector is equivalent to the temporal fingerprint of the standard communication protocol, comprehensively characterizing the core parameters of the protocol. By calculating the cosine similarity between the vector to be matched and the standard vector in the communication protocol library, the target communication protocol is automatically locked, significantly shortening the communication protocol adaptation time. No manual parameter configuration is required, and it is compatible with multiple mainstream brand monitoring systems, reducing the installation and debugging time of a single combiner box from hours to minutes. This significantly reduces the cost of project implementation.

[0018] Secondly, traditional photovoltaic DC combiner box communication systems are prone to data packet loss under strong electromagnetic interference such as inverter noise and high-voltage cable radiation, leading to "screen-filling alarms" in the monitoring system and severely impacting operation and maintenance efficiency. The electromagnetic shielding unit of the communication module in this application collects interference pulse timing data, simultaneously recording the occurrence time, target pulse signal phase, and target pulse signal amplitude, providing high-quality data support for subsequent identification. Based on a second cyclic pulse memory sub-network with an LSTM-SNN hybrid structure, the LSTM unit captures the long-term temporal correlation features of interference pulses, while the SNN unit captures the discrete jump features of phase and amplitude, achieving accurate differentiation between random, periodic, and sudden interference. Dedicated coding strategies are applied for different interference types: random interference uses a time-phase joint coding strategy to lock the phase; periodic interference uses a time-series interference avoidance coding strategy to offset the interference period by half; and sudden interference uses a triple-redundancy coding strategy to generate time-off redundant pulses. This changes the passive mode of traditional single coding to deal with all interferences, significantly reducing the packet loss rate in strong electromagnetic environments and fundamentally solving the problem of screen-filling alarms in monitoring systems.

[0019] Thirdly, traditional photovoltaic DC combiner box communication systems often experience communication interruptions over transmission distances of several hundred meters (over 500 meters) due to severe signal attenuation. After an interruption, manual on-site module restart is required, with recovery times exceeding 30 minutes, leading to missing power generation statistics. This application achieves a breakthrough in long-distance transmission reliability through advance prediction and precise compensation mechanisms. It utilizes a third cyclic pulse memory subnetwork based on an LSTM-SNN hybrid structure, fusing signal attenuation factor time-series data with effective pulse coding sequences. The LSTM unit learns the long-term dependency relationship between the attenuation factor and coding parameters, while the SNN unit captures discrete feature changes caused by parameter mutations, ultimately outputting predicted signal strength values ​​for each sampling moment in the next time period. The signal compensation module performs dual-dimensional compensation based on the predicted signal strength values ​​for each sampling moment in the next time period. If the strength is lower than a preset range, the transmission power is gradually increased, and the terminal resistor is matched according to the transmission distance, significantly reducing attenuation. Meanwhile, after compensation, a 0.5dBm level fine-tuning is triggered by a ±2dBm fluctuation threshold to ensure that the signal remains stable within the effective range, achieving "pre-compensation rather than post-disconnection repair," completely eliminating the cost of manual restart and ensuring the integrity of power generation statistics.

[0020] Fourth, this application upgrades from passive emergency repair to proactive early warning. Traditional photovoltaic DC combiner box communication systems rely on manual inspection and troubleshooting, resulting in delayed fault detection and low processing efficiency. This application constructs a new operation and maintenance model based on state prediction and proactive intervention through full-link data collection and analysis of the communication scheduling and control model. Interference type identification results can provide early warning of electromagnetic environment deterioration (e.g., a continuous increase in periodic interference intensity), guiding operation and maintenance personnel to specifically investigate potential problems with inverters or high-voltage cables. Signal strength prediction data can identify transmission link attenuation trends in advance (e.g., a monthly increase in cable impedance), enabling preventative replacement of aging cables. Protocol adaptation logs can automatically record cross-brand connection parameters, providing data support for subsequent system expansion.

[0021] This model reduces operation and maintenance response time from hours to minutes, significantly lowering operation and maintenance costs and further highlighting the practical value of high-reliability communication.

[0022] In summary, this application effectively solves the technical pain points of traditional systems, such as weak anti-interference, long-distance attenuation, and poor protocol compatibility, through a closed-loop technology of intelligent communication protocol adaptation, accurate interference identification, and early signal compensation. It significantly improves communication reliability and achieves several times the improvement in operation and maintenance efficiency, which fully meets the actual needs of photovoltaic power plants. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is an exemplary system block diagram of a combiner box communication system based on a high-reliability communication module in the embodiments of this application; Figure 2 This is a flowchart illustrating the process of matching the target communication protocol in the communication protocol library based on the timing feature vector of the protocol to be matched, as described in this application embodiment. Figure 3 This is a flowchart illustrating the process of establishing a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol in this embodiment of the application. Figure 4 This is a flowchart illustrating the process of determining the interference type corresponding to the timing data of the interference pulse in this embodiment of the application. Figure 5 This is a flowchart illustrating the process of determining the effective pulse code sequence of the target pulse signal in the next time period in an embodiment of this application. Figure 6 This is a flowchart illustrating the process of determining the signal strength value of the target pulse signal at each sampling moment in the next time period, as described in an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0025] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0026] Traditional photovoltaic (PV) DC combiner box communication systems connect to the backend monitoring system via a bus, requiring manual on-site configuration, such as manually setting the baud rate and address code. Due to the complex environment of PV areas, PV DC combiner box communication systems are susceptible to high temperatures and strong electromagnetic interference. Furthermore, the dispersed distribution of PV DC combiner boxes, with distances reaching hundreds of meters, makes general-purpose communication modules prone to interference and communication interruptions. Additionally, communication protocols differ between monitoring systems from different brands, resulting in poor module compatibility and requiring customized development, increasing debugging costs. Traditional PV DC combiner box communication systems suffer from the following technical drawbacks: Firstly, traditional photovoltaic DC combiner box communication systems have weak anti-interference capabilities and are prone to data packet loss or disconnection in strong electromagnetic environments, leading to frequent alarms ("screen refresh") in the background monitoring system and affecting operation and maintenance efficiency.

[0027] Secondly, the signal attenuation is severe when transmitting over long distances (more than 500 meters), and the module needs to be manually restarted after a disconnection, which takes a long time to recover (usually more than 30 minutes), resulting in missing power generation statistics.

[0028] Third, the traditional photovoltaic DC combiner box communication system requires manual parameter modification for protocol adaptation with monitoring systems of different brands, resulting in poor compatibility and increased installation and debugging time costs (debugging a single combiner box takes about 1 hour).

[0029] Based on the above exemplary scenario, please refer to Figure 1 In one embodiment of the combiner box communication system based on a high-reliability communication module of this application, the system includes: The protocol configuration module 1100 is used to acquire the handshake request pulse sequence of multiple target pulse signal transmission requests initiated by the monitoring system to the photovoltaic DC combiner box, and transmit the handshake request pulse sequence to the first cyclic pulse memory sub-network in the communication scheduling control model that has been trained to convergence state, so as to extract the timing feature vector of the protocol to be matched, and match the corresponding target communication protocol in the communication protocol library according to the timing feature vector of the protocol to be matched. The target pulse signal includes a voltage pulse signal and its corresponding current pulse signal. The basic network architecture of the communication scheduling control model is an improved cyclic spiking neural network, wherein the improved cyclic spiking neural network includes a pulse input layer, a pulse timing coding layer, a cyclic pulse memory layer and a backpropagation layer. The cyclic pulse memory layer includes a first cyclic pulse memory sub-network, a second cyclic pulse memory sub-network and a third cyclic pulse memory sub-network. The basic network architecture of the first cyclic pulse memory sub-network is an LSTM-CNN hybrid structure, and the basic network architecture of the second cyclic pulse memory sub-network and the third cyclic pulse memory sub-network is an LSTM-SNN hybrid structure.

[0030] In some embodiments, the handshake request pulse sequence includes a start bit, an address code, and a check code; the timing feature vector of the protocol to be matched includes the duration of the start bit pulse, the time difference between adjacent address code pulses, and the generation time difference of the check code relative to the address code; the communication module includes an electromagnetic shielding unit; and the interference pulse timing data characterizes the timing data constructed from the occurrence time of inverter noise data and high-voltage cable radiation data, the phase of the target pulse signal, and the amplitude of the target pulse signal. The signal attenuation factor time-series data characterizes the time-series data constructed from the transmission distance between the photovoltaic DC combiner box and the monitoring system, the cable impedance, and the ambient temperature of the photovoltaic DC combiner box within the current time period; the interference types include random interference, periodic interference, or sudden interference.

[0031] In some embodiments, the protocol configuration module in the combiner box communication system of this application can adopt an industrial-grade STM32F407 main control chip, an integrated pulse signal acquisition interface, a Flash protocol storage unit, and a UART communication interface, with an operating temperature adapted to extreme photovoltaic environments ranging from -40℃ to 85℃. The integrated pulse signal acquisition interface supports voltage pulse detection from 0 to 10V and current pulse detection from 0 to 20mA; the Flash protocol storage unit can have a capacity of 128MB or 256MB, etc., and pre-stores standard timing feature vector libraries for multiple standard communication protocols such as Modbus-RTU and DL / T 645; the UART communication interface is used to interact with the communication module in the combiner box communication system to exchange configuration commands.

[0032] Traditional photovoltaic DC combiner box communication systems require manual parameter modification for protocol adaptation with monitoring systems from different brands, resulting in poor compatibility and increased installation and debugging time costs, with debugging a single combiner box taking several hours or more. Since the handshake request pulse sequence is the temporal fingerprint of the communication protocol between the combiner box communication system and the monitoring system, and different standard communication protocols have different temporal characteristics, this provides a data foundation for subsequent identification. The handshake request pulse sequence, including the start bit, address code, and check bit, initiated by the monitoring system is collected. This handshake request pulse sequence is then transmitted to the first cyclic pulse memory sub-network in the communication scheduling and control model, which has been trained to convergence, to extract the temporal feature vector of the protocol to be matched. The temporal feature vector of the protocol to be matched contains discrete features and temporal correlation features. Discrete features include the duration of the start bit pulse and the time difference between adjacent address code pulses, while temporal correlation features include the generation time difference of the check bit relative to the address code, etc. For example, in the DL / T 645 protocol, the check bit is generated 1ms after the data bits.

[0033] For further embodiments, please refer to Figure 2 The handshake request pulse sequence is transmitted to the first cyclic pulse memory subnetwork in the communication scheduling control model that has been trained to convergence, in order to extract the timing feature vector of the protocol to be matched, and to match the corresponding target communication protocol in the communication protocol library according to the timing feature vector of the protocol to be matched, including: Step S101: Using the first LSTM unit in the first cyclic pulse memory sub-network, perform time-series feature mapping on the generation time difference between the check code and the address code in the handshake request pulse sequence to determine the first time-series associated feature, so as to construct the first time-series associated feature vector; Step S102: Using the first CNN unit in the first cyclic pulse memory sub-network, extract the duration of the start bit pulse from the handshake request pulse sequence to determine the first discrete feature and construct the first discrete feature vector; and extract the time difference between adjacent address code pulses from the handshake request pulse sequence to determine the second discrete feature and construct the second discrete feature vector. Step S103: Concatenate the first discrete feature vector, the second discrete feature vector, and the first temporal correlation feature vector after dimensional standardization to construct the temporal feature vector of the protocol to be matched. Step S104: Calculate the cosine similarity between the timing feature vector of the protocol to be matched and the timing feature vector of each standard communication protocol in the communication protocol library in the protocol configuration module. If the cosine similarity is greater than the preset similarity threshold, then the standard communication protocol is used as the target communication protocol for establishing a connection between the monitoring system and the photovoltaic DC combiner box.

[0034] Specifically, when the monitoring system initiates a target pulse signal transmission request to the photovoltaic DC combiner box, the protocol configuration module captures multiple handshake request pulse sequences in real time through the pulse acquisition interface. Each handshake request pulse sequence includes a start bit, address code, and check code. The sampling interval is set to 1ms, and the peak timing data of the voltage pulse signal and the corresponding peak timing data of the current pulse signal are recorded synchronously. The peak value of the voltage pulse signal can be 3V±0.5V, etc., and the peak value of the current pulse signal can be 15mA±2mA, etc.

[0035] The collected multiple handshake request pulse sequences are fused into a standardized time-series dataset and input into the first recurrent pulse memory sub-network of the communication scheduling control model that has been trained to convergence. The basic network architecture of the first recurrent pulse memory sub-network is an LSTM-CNN hybrid structure, which includes a first LSTM unit and a first CNN unit. The first LSTM unit contains 64 LSTM neurons and the first CNN unit contains 32 convolutional kernels. The first recurrent pulse memory sub-network extracts the duration of the start bit pulse in each handshake request pulse sequence through the first CNN unit to determine the first discrete feature and construct the first discrete feature vector. It extracts the time difference between adjacent address code pulses in each handshake request pulse sequence to determine the second discrete feature and construct the second discrete feature vector. It performs time-series feature mapping on the generation time difference between the check code and the address code in the handshake request pulse sequence to determine the first time-series associated feature and construct the first time-series associated feature vector. For example, the check bit of the DL / T 645 protocol is generated 1ms after the data bit.

[0036] The first discrete feature vector, the second discrete feature vector, and the first temporal correlation feature vector after dimensionality standardization are concatenated to output a 1×128 dimensional temporal feature vector of the protocol to be matched. The protocol configuration module 1100 calls the communication protocol library in the Flash storage unit and compares the feature vector to be matched with the protocol timing feature vector in the communication protocol library using a cosine similarity algorithm. This determines the cosine similarity between the timing feature vector of the protocol to be matched and the timing feature vector of each standard communication protocol in the communication protocol library of the protocol configuration module. If the cosine similarity is greater than or equal to 0.90 or 0.95, then the standard communication protocol is used as the target communication protocol for establishing a connection between the monitoring system and the photovoltaic DC combiner box. The standard communication protocols include Modbus-RTU communication protocol, DL / T 645 communication protocol, etc.

[0037] Subsequently, a handshake response pulse sequence is sent to the monitoring system, which includes voltage-current corresponding pulse codes. After receiving the response without error, the target communication protocol and corresponding communication parameters are transmitted to the communication module through the UART interface. For example, the target communication protocol can be the DL / T 645 communication protocol, and its communication parameters include a baud rate of 9600bps, even parity, and 8 data bits.

[0038] Through steps S101 to S104, the protocol adaptation time for a single combiner box is reduced from several hours of traditional manual debugging to tens of seconds, and the success rate of communication protocol adaptation is greatly improved. There is no need for on-site personnel to manually modify parameters, which greatly saves manpower and resources. Through the model-driven feature extraction and matching process, the identification and determination of the target communication protocol are automatically completed, replacing the inefficient mode of traditional manual debugging. This significantly reduces the deployment and maintenance costs of the photovoltaic DC combiner box communication system, while improving the compatibility of different brands of monitoring equipment with combiner boxes.

[0039] The communication module 1200 is used to establish a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol, and then continuously collect the interference pulse timing data and signal attenuation factor timing data corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period. In some embodiments, please refer to Figure 3 Establishing a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol includes: Step S201: The protocol configuration module generates a pulse control command containing baud rate, parity method and data bit length according to the target communication protocol, and sends the pulse control command to the communication module to configure the communication parameters in the communication module; Step S202: Generate a handshake response pulse sequence according to the communication parameters and return it to the monitoring system so that the photovoltaic DC combiner box establishes a communication connection with the monitoring system.

[0040] Specifically, the communication module in the combiner box communication system of this application is based on an RS485 communication chip, and integrates an electromagnetic shielding unit, a signal amplification circuit with an adjustable gain range of 10dB to 20dB, a high-precision data acquisition unit, and a LoRa wireless backup channel. The hardware interface is compatible with the UART interface specification of traditional combiner boxes. The electromagnetic shielding unit adopts a double shielding of nickel-zinc ferrite shielding layer and copper mesh.

[0041] In the combiner box communication system of this application, the protocol configuration module 1100 generates a pulse control command containing baud rate, parity method, and data bit length according to the target communication protocol, and sends the pulse control command to the communication module to configure the communication parameters in the communication module; the communication module 1200 receives the communication parameters corresponding to the DL / T 645 communication protocol output by the protocol configuration module 1100, configures the communication parameters through the MAX485 chip, and can complete the RS485 bus connection between the photovoltaic DC combiner box and the monitoring system in a short time. After the link is established, a target pulse signal containing a voltage pulse signal and its corresponding current pulse signal is output to the monitoring system, wherein the target pulse signal can be a voltage of 3V, a current of 15mA, and a duration of 100ms.

[0042] Furthermore, the system continuously collects the interference pulse timing data and signal attenuation factor timing data corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period; the system can also collect the interference pulse data of the target pulse signal once every 1ms through the pulse detector built into the electromagnetic shielding unit, record the occurrence time, phase, and amplitude of the interference pulse data, and form an interference pulse timing dataset with 1000 sampling moments per second, where the current time period can be 10s, etc. The system uses integrated distance and temperature sensors to collect ambient temperature data of the transmission distance (e.g., 800 meters) and the photovoltaic DC combiner box every 100ms, and stores this data in association with the signal strength value of the target pulse signal collected simultaneously. Interference pulse timing data and signal attenuation factor timing data are transmitted via an internal bus to the signal encoding module and signal compensation module, respectively. Phase encryption is used during transmission to prevent data tampering.

[0043] The signal encoding module 1300 is used to transmit the interference pulse timing data to the second cyclic pulse memory sub-network in the communication scheduling control model that has been trained to the convergence state, so as to determine the interference type corresponding to the interference pulse timing data, and call the corresponding pulse coding strategy according to the interference type to determine the effective pulse coding sequence of the target pulse signal in the next time period. The signal encoding module 1300 can use an FPGA chip to implement parallel data processing, and integrates a pulse code generation unit, an interference feature storage unit, and an interface unit with the communication scheduling and control model.

[0044] In some embodiments, please refer to Figure 4 The interference pulse timing data is transmitted to the second cyclic pulse memory subnetwork in the communication scheduling and control model that has been trained to a convergent state, in order to determine the interference type corresponding to the interference pulse timing data, including: Step S301: Collect the interference pulse timing data corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period through the electromagnetic shielding unit in the communication module; When the photovoltaic DC combiner box is in normal operation, after the communication module and the monitoring system establish an initial connection via RS485 bus, the electromagnetic shielding unit automatically starts interference pulse data acquisition. For the interference pulse data corresponding to the target pulse signal, it continuously acquires interference pulse data within the current time period at 1ms intervals, simultaneously recording multiple core parameters such as the occurrence time of the interference pulse data, the phase of the target pulse signal, and the amplitude of the target pulse signal. The current time period can be 10s, and the occurrence time can be accurate to 1ms, such as the 100ms or 200ms. The phase can be measured by a phase detection circuit, such as 90° or 180°. The voltage amplitude accuracy is 0.01V, and the current amplitude accuracy is 0.1mA, such as 2.2V or 12mA.

[0045] After the collected data is filtered by hardware to remove high-frequency noise, it is stored as an interference pulse time series dataset with 10,000 records in the format of "time-phase-amplitude" triplet, and transmitted to the input buffer of the communication scheduling control model through the SPI interface.

[0046] Step S302: Through the second LSTM unit in the second cyclic pulse memory sub-network, perform feature mapping on the long-term dependency relationship between the occurrence time, phase and amplitude of the interference pulse time series data, determine the second time series correlation feature, and construct the second time series correlation feature vector. The second LSTM unit in the second recurrent pulse memory subnetwork can adopt a 3-layer stacked structure, with each layer containing 64 LSTM neurons. The input sequence length is set to 100, which corresponds to 100ms of time-series data. The activation function in the hidden layer is the tanh activation function, and the output layer dimension is set to 64.

[0047] The interference pulse time series dataset collected in step S301 above is divided into sliding window segments with a window size of 100ms (containing 100 sampling times) and a step size of 50ms, generating 991 input subsequences. Each subsequence contains the associated data of "time interval-phase-amplitude".

[0048] The input subsequence is fed into the second LSTM unit, which learns long-term dependencies through a gating mechanism. For periodic interference subsequences, such as an interference pulse with a 90° phase and a 2V amplitude appearing every 100ms, the second LSTM unit captures the correlation pattern with a fixed time interval of 100ms, phase fluctuation ≤5°, and amplitude fluctuation ≤0.2V, and the output feature values ​​are concentrated in the range of 0.8 to 1.0. For random interference subsequences, such as those with no fixed time interval, phase that changes randomly from 0° to 360°, and amplitude that fluctuates from 1 to 3V, the second LSTM unit identifies the feature of irregular time intervals and no significant correlation between phase and amplitude, and the output feature values ​​are scattered in the range of 0.1 to 0.3.

[0049] The output features of each subsequence are extracted and concatenated in chronological order to form a 1×64-dimensional second temporal correlation feature vector. The vector element values ​​are quantized to reflect the correlation strength of the interference parameters at different times.

[0050] Step S303: Using the second SNN unit in the second cyclic pulse memory sub-network, capture the phase difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the third discrete feature to construct the third discrete feature vector, and extract the amplitude difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the fourth discrete feature to construct the fourth discrete feature vector. The second SNN unit in the second recurrent spiking memory subnetwork adopts a 2-layer spiking neuron structure. The input layer contains 32 spiking neurons and the output layer contains 64 spiking neurons. The spiking threshold is set to 0.5, and the time step is consistent with the sampling interval, which is 1ms.

[0051] Using the raw interference pulse timing data acquired in step S301 as input, the second SNN unit calculates features point by point according to the sampling time: Calculate the phase difference of the target pulse signal at adjacent sampling times, such as 0° phase at time t1 and 90° phase at time t2, with a jump value of 90°; if the jump value is >10°, the spiking neuron is triggered to fire and output 1, otherwise it outputs 0, generating a third discrete feature vector containing 10,000 binary values. Calculate the difference in amplitude of the target pulse signal between adjacent sampling times, with a threshold of 0.2V for voltage and 2mA for current. For example, if the amplitude is 2V at time t3 and 5V at time t4, the change value of 3V > 0.2V, triggering the firing of the spiking neuron, outputting 1, and generating the fourth discrete feature vector.

[0052] The third and fourth discrete feature vectors are sparsely encoded to retain the pulse firing position information and compressed into 1×64-dimensional standardized third and fourth discrete feature vectors.

[0053] Step S304: The third discrete feature vector, the fourth discrete feature vector, and the second time-series correlation feature vector after dimensionality standardization are concatenated to construct the interference feature vector. The interference feature vector is then matched with a preset interference type feature library to determine the interference type corresponding to the interference pulse time-series data.

[0054] The interference type feature library defines the interference pulse time series data as follows: if the phase jump value of the third discrete feature vector is ≤5 and its proportion is >90%, the amplitude change value of the fourth discrete feature vector is ≤0.2V and its proportion is >90%, and the correlation strength value of the second time series correlation feature vector is >0.8 and its proportion is >85%, then the interference pulse time series data is defined as periodic interference; if the phase jump value of the third discrete feature vector is irregularly distributed, the amplitude change value of the fourth discrete feature vector is irregularly distributed, and the correlation strength value of the second time series correlation feature vector is <0.3 and its proportion is >90%, then the interference pulse time series data is defined as random interference; if the single phase jump value of the third discrete feature vector is >90% and its proportion is ≥10%, the single amplitude change value of the fourth discrete feature vector is >1V and its proportion is ≥10%, and the correlation strength value of the second time series correlation feature vector rises and falls sharply, then the interference pulse time series data is defined as sudden interference.

[0055] The third discrete feature vector, the fourth discrete feature vector, and the second time-series correlation feature vector, after dimensionality standardization, are concatenated to construct the interference feature vector. The interference feature vector is then matched with a preset interference type feature library to determine the interference type corresponding to the interference pulse time-series data.

[0056] As demonstrated in the above embodiments, by matching the interference feature vector with a preset feature library, algorithms such as cosine similarity can be used to quickly identify the interference type. The matching process requires no manual intervention, has a fast response speed, and high accuracy. Accurate interference type identification can directly provide a decision-making basis for subsequent coding strategy invocation, realizing a closed-loop linkage between interference identification and anti-interference adaptation, significantly reducing the signal packet loss rate caused by interference, and significantly improving the communication reliability between the photovoltaic DC combiner box and the monitoring system.

[0057] For further embodiments, please refer to Figure 5 Based on the interference type, the corresponding pulse coding strategy is invoked to determine the effective pulse coding sequence of the target pulse signal in the next time period, including: Step S3001: If the interference type corresponding to the interference pulse timing data is random interference, then the time-phase joint coding strategy in the communication scheduling control model is invoked to fix the transmission time of the effective pulse signal in the next time period, and the phase of the effective pulse signal is locked within a preset phase threshold range to determine the effective pulse coding sequence in the next time period. The preset phase threshold range is within an interval of no more than 5° above or below 0°. The next time period can be set to 10 seconds. The effective pulse signal transmission frequency is calculated to be 960 times / second according to the preset communication rate (9600bps). The transmission time interval of each pulse is fixed at 1.0417ms (1 / 960 seconds), that is, the pulses are transmitted sequentially at 1.0417ms, 2.0834ms...10000ms, with the time deviation controlled within ±0.01ms.

[0058] The phase control unit is invoked to lock the phase of the effective target pulse signal containing the voltage pulse signal and its corresponding current pulse signal within the range of 0°±5°. The phase closed-loop adjustment circuit corrects it in real time: if a pulse phase shift to 6° is detected, a fine-tuning command is immediately triggered to pull the phase back to the 0°-5° range. The phase adjustment response time is <10μs.

[0059] A valid pulse code sequence is generated according to the "emission time-phase-frequency" triplet format, containing a total of 9600 records. Examples include 1.0417ms-2°-960Hz and 2.0834ms-4°-960Hz, with the frequency fixed at 960Hz to match the communication rate. The sequence is transmitted to the signal compensation module via the internal bus.

[0060] Step S3002: If the interference type corresponding to the interference pulse timing data is periodic interference, then the timing interference avoidance coding strategy in the communication scheduling control model is invoked to offset the emission time of the effective pulse signal in the next time period by half of the interference period, so as to determine the effective pulse coding sequence in the next time period. The effective pulse coding sequence includes the pulse emission time, phase and frequency of the pulse signal. The communication scheduling and control model re-verifies the timing data of the interference pulses using a sliding window algorithm, confirming that the interference period is stable at 100ms and the interference pulses are concentrated at 100ms, 200ms...10000ms.

[0061] The offset is calculated to be 1 / 2 of the interference period. The emission time of the target pulse signal in the next time period (10 seconds) is adjusted from the original plan of "100ms, 200ms..." to "150ms, 250ms...10050ms", to ensure that the target pulse signal and the interference pulse signal are completely staggered on the time axis, with a time offset accuracy of ≤1ms.

[0062] The phase and frequency parameters are combined to generate an encoding sequence. The threshold is further tightened to enhance anti-interference. The phase is locked at 0°±3°, the frequency is set to 480Hz according to the communication protocol requirements, the transmission interval is 2.0833ms, and the effective pulse encoding sequence contains 4800 records. Each record clearly indicates information such as "150ms-1°-480Hz" and "250ms-3°-480Hz". The interference period and offset parameters are stored synchronously for traceability.

[0063] Step S3003: If the interference type corresponding to the interference pulse timing data is sudden interference, then the triple redundancy coding strategy in the communication scheduling control model is invoked to generate multiple pulse signals of the same duration with a preset time offset, so as to determine the effective pulse coding sequence in the next time period.

[0064] For each original valid pulse in the next time period (10 seconds), such as the 3V voltage pulse and 15mA current pulse at 50ms, two redundant pulses with the same content are generated at the same time. The time offsets of the three pulses are 0ms, 1ms and 2ms respectively. That is, the original pulse is issued at 50ms, the first redundant pulse is issued at 51ms and the second redundant pulse is issued at 52ms.

[0065] The original pulses and redundant pulses are integrated into an effective pulse coding sequence in chronological order. The sequence length is three times the original plan. For example, if the original plan was 4800 records, the integrated sequence will have 14400 records. Each record is labeled with the attributes and corresponding release times of the original pulse, the first redundant pulse, and the second redundant pulse. The phase is uniformly locked at 0°±5°, and the frequency remains unchanged at 480Hz.

[0066] A redundant identifier field is added to the header of the encoded sequence. The monitoring system adopts a majority voting decoding rule. When the phase and amplitude of at least 20 pulses out of 30 pulses match, it is determined to be a valid pulse encoded sequence. Individual pulses affected by interference can be ignored.

[0067] The signal compensation module 1400 is used to transmit the signal attenuation factor timing data and the effective pulse code sequence in the next time period to the third cyclic pulse memory sub-network in the communication scheduling control model that has been trained to convergence state, so as to determine the signal strength value of the target pulse signal at each sampling moment in the next time period, and to compensate the signal strength value of the target pulse signal at each sampling moment in the next time period using a preset pulse signal compensation strategy, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding.

[0068] In some embodiments, please refer to Figure 6 The signal attenuation factor time-series data and the effective pulse coding sequence in the next time period are transmitted to the third cyclic pulse memory subnetwork in the communication scheduling control model that has been trained to convergence, in order to determine the signal strength value of the target pulse signal at each sampling time in the next time period, including: Step S401: Through the third LSTM unit in the third cyclic pulse memory sub-network, perform time-series correlation feature mapping on the long-term dependency relationship between the signal attenuation factor time-series data and the effective pulse coding sequence to determine the third time-series correlation feature, so as to construct the third time-series correlation feature vector; The signal attenuation factor timing data and the effective pulse coding sequence are input into the third cyclic pulse memory sub-network. The signal attenuation factor timing data can be collected once every 100ms within the current 10-second period, including data such as transmission distance of 800m, ambient temperature of 45℃, and cable impedance of 120Ω, for a total of 100 sets of data. The effective pulse coding sequence is generated by the signal coding module using a timing interference avoidance coding strategy for periodic interference, including the emission time of 4800 pulses within the next 10-second time period, phase of 0°±3°, and frequency of 480Hz. The third LSTM unit of the third cyclic pulse memory sub-network adopts a 3-layer stacked structure, with each layer containing 64 LSTM neurons, an input sequence length of 100, and an output dimension of 64.

[0069] The signal attenuation factor timing data is normalized, that is, the transmission distance from 0 to 1500m is mapped to the interval of 0 to 1, for example, 800m corresponds to 0.53; the ambient temperature from -40℃ to 85℃ is mapped to the interval of 0 to 1, for example, 45℃ corresponds to 0.65; the cable impedance from 80 to 150Ω is mapped to the interval of 0 to 1, for example, 120Ω corresponds to 0.57; the core parameters of the effective pulse code sequence, namely the firing time interval of 2.0833ms, phase deviation value ≤3, and frequency of 480Hz, are extracted and converted into 100 sets of timing subsequences aligned with the attenuation factor data.

[0070] The fused "attenuation factor-encoding parameter" subsequence is input into the third LSTM unit, and long-term dependencies are learned through the gating mechanism. For example, it captures the temporal correlation features such as "the theoretical intensity of the encoded pulse decreases by 0.5dBm for every 100m increase in transmission distance" and "the attenuation rate increases by 10% for every 10℃ increase in temperature". Targeting the characteristics of a fixed frequency of 480Hz and stable phase in an effective pulse-coded sequence, an association pattern is identified where, when the coding parameters are stable, the intensity attenuation is determined solely by the attenuation factor, with output feature values ​​concentrated in the range of 0.7 to 0.9. The output features of each subsequence are extracted and concatenated in chronological order to form a 1×64-dimensional third temporal association feature vector. The vector element values ​​quantify the dynamic association strength between the attenuation factor and the coding sequence.

[0071] Step S402: Through the third SNN unit in the third cyclic pulse memory sub-network, capture the discrete feature changes of the effective pulse coding sequence under different signal attenuation factor time series data combinations to determine the fifth discrete feature, so as to construct the fifth discrete feature vector; The third SNN unit of the third recurrent spiking memory subnetwork adopts a 2-layer spiking neuron structure, with 32 spiking neurons in the input layer and 64 spiking neurons in the output layer. The spiking threshold is set to 0.6, and the time step is consistent with the sampling interval of 100ms, which can capture the intensity jump features caused by changes in parameter combinations.

[0072] Using the "attenuation factor-encoding sequence" data preprocessed in step S401 as input, the third SNN unit analyzes the discrete feature changes point by point according to the sampling time. When the signal attenuation factor combination changes from "800m-45℃-120Ω" to "850m-46℃-122Ω", it captures the discrete feature changes of "0.25dBm intensity attenuation due to increased distance and 0.5% increase in attenuation rate due to increased temperature", triggering the firing of the spiking neuron and outputting the feature value 1. When the effective pulse code sequence parameters are stable (no change in firing time, phase, or frequency) and the attenuation factor fluctuation is ≤5%, it is determined that there is no significant discrete change, and the output characteristic value is 0.

[0073] The output feature values ​​at 100 sampling times are arranged in chronological order to form a fifth discrete feature vector of 1×64 dimensions. The sparsity of the vector reflects the degree of discrete change of parameter combinations at different times. During sudden decay, the proportion of 1 values ​​increases, while during stable decay, the proportion of 0 values ​​is >90%.

[0074] Step S403: Perform vector concatenation operation on the third temporal correlation feature and the fifth discrete feature vector after dimensionality standardization to construct a fused feature vector. Transmit the fused feature vector to the fully connected layer in the third cyclic pulse memory sub-network. After nonlinear transformation by the Sigmoid activation function, perform inverse normalization mapping to map the fused feature vector to the signal intensity values ​​of each sampling time in the next time period.

[0075] The fully connected layer of the third recurrent pulse memory subnetwork contains two hidden layers, each containing 128 neurons. The output layer dimension is set to 100, which corresponds to the signal strength values ​​of 100 sampling moments within the next time period of 10 seconds.

[0076] The third temporal correlation feature vector and the fifth discrete feature vector are subjected to Min-Max standardization to eliminate the difference in dimensions. Then, they are horizontally spliced ​​to construct a 1×128-dimensional fusion feature vector, which simultaneously covers the dynamic correlation pattern and the details of discrete changes.

[0077] The fused feature vector is input into a fully connected layer, and a sigmoid activation function is used for non-linear mapping to compress the high-dimensional features into 100 output values ​​in the range of 0 to 1. Then, inverse normalization mapping is performed to obtain the predicted signal strength values ​​at each sampling time. For example, the predicted value at 100ms is -58dBm, and the predicted value at 200ms is -58.5dBm.

[0078] The signal strength values ​​at 100 sampling times are arranged in chronological order to form the signal strength time sequence for the next time period, and then output to the compensation strategy execution unit of the signal compensation module.

[0079] Through the above steps, the signal strength values ​​at each sampling moment of the next time period can provide accurate prediction basis for the signal compensation module in advance. If the predicted strength is lower than the preset range, compensation strategies such as power adjustment and impedance matching can be activated in advance to avoid signal loss caused by "attenuation before compensation". At the same time, the millisecond-level prediction response speed is adapted to the dynamic interference scenario of photovoltaic communication, ensuring that the compensation strategy can adapt to environmental changes in real time and further reduce the signal packet loss rate.

[0080] In some embodiments, a preset pulse signal compensation strategy is used to compensate the signal strength values ​​of the target pulse signal at each sampling time in the next time period, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding, including: A preset pulse signal compensation strategy is invoked to detect whether the signal strength value of the target pulse signal at each sampling moment in the next time period falls within the preset signal strength range. If it does, the current transmission power of the photovoltaic DC combiner box is gradually increased to the target transmission power corresponding to the preset signal strength range, and the corresponding resistance parameters are matched according to the transmission distance between the photovoltaic DC combiner box and the monitoring system.

[0081] Specifically, a preset pulse signal compensation strategy is used to compensate the signal strength values ​​of the target pulse signal at each sampling moment within the next time period (10s). The intensity detection unit of the signal compensation module verifies the predicted signal strength values ​​at 100 sampling moments against the preset signal strength range at 100ms intervals, wherein the preset signal strength range is between -50dBm and -60dBm. When all predicted signal strength values ​​are detected to be in the range of -66dBm to -69dBm and do not fall within the preset signal strength range, the mean of all predicted signal strength values ​​is calculated to be -67.5dBm. Based on the mapping table, the target transmission power is determined to be 22dBm, which corresponds to the range of -50dBm to -55dBm, with a 5dBm compensation redundancy reserved.

[0082] The current transmission power of the photovoltaic DC combiner box is read as 12dBm. The difference between the target transmission power and the current transmission power is calculated to be 10dBm. A pulse signal compensation strategy of increasing by 1dBm every 200ms is adopted to make the power amplifier execute the pulse signal compensation command: From 0ms to 200ms, the transmission power was maintained at 12dBm, and the actual signal strength value was synchronously collected as -67.5dBm. At 200ms, the current transmission power is increased to 13dBm, and the actual signal strength value rises to -66.5dBm without overshoot. This incremental adjustment continues until 2000ms, when the power stabilizes at 22dBm, corresponding to an actual strength of -52dBm at the sampling time, falling within the preset signal strength range. For the remaining 8000ms, the power is maintained at 22dBm. If the strength fluctuation exceeds ±2dBm (e.g., drops to -57dBm), a 0.5dBm fine-tuning is triggered to ensure stability.

[0083] Based on a transmission distance of 900 meters, the target resistance parameter is determined to be 125Ω by querying the matching table. The current resistance is read as 100Ω. An adjustment command is sent via serial communication to smoothly switch the resistance from 100Ω to 125Ω within 30ms, avoiding signal reflection caused by abrupt changes.

[0084] As can be seen from the above embodiments, compared with the prior art, this application addresses the problems of weak anti-interference capability, easy data packet loss or disconnection in traditional photovoltaic DC combiner box communication systems in the prior art, as well as the need for manual parameter modification for protocol adaptation with different brands of monitoring systems, resulting in poor compatibility. This application solves the pain points of the prior art through a full-link innovative design including intelligent protocol adaptation, accurate interference identification, dynamic encoding optimization, and signal pre-compensation. This application includes, but is not limited to, the following beneficial effects: Firstly, traditional photovoltaic DC combiner box communication systems suffer from differences in communication protocols among different brands of monitoring systems, such as baud rate, address code format, and verification rules. This necessitates manual on-site parameter modification, resulting in excessively long debugging times, poor compatibility, and high installation and debugging costs. This application achieves automatic protocol adaptation through handshake pulse parsing and feature vector matching. Based on an LSTM-CNN hybrid structure, the first cyclic pulse memory subnetwork extracts discrete features such as the duration of the start bit and the time difference between adjacent address codes from the handshake request pulse sequence of the monitoring system through CNN units. The LSTM unit captures the temporal features of the generation time difference between the check code and the address code, constructing a unique temporal feature vector for the protocol to be matched. This temporal feature vector is equivalent to the temporal fingerprint of the standard communication protocol, comprehensively characterizing the core parameters of the protocol. By calculating the cosine similarity between the vector to be matched and the standard vector in the communication protocol library, the target communication protocol is automatically locked, significantly shortening the communication protocol adaptation time. No manual parameter configuration is required, and it is compatible with multiple mainstream brand monitoring systems, reducing the installation and debugging time of a single combiner box from hours to minutes. This significantly reduces the cost of project implementation.

[0085] Secondly, traditional photovoltaic DC combiner box communication systems are prone to data packet loss under strong electromagnetic interference such as inverter noise and high-voltage cable radiation, leading to "screen-filling alarms" in the monitoring system and severely impacting operation and maintenance efficiency. The electromagnetic shielding unit of the communication module in this application collects interference pulse timing data, simultaneously recording the occurrence time, target pulse signal phase, and target pulse signal amplitude, providing high-quality data support for subsequent identification. Based on a second cyclic pulse memory sub-network with an LSTM-SNN hybrid structure, the LSTM unit captures the long-term temporal correlation features of interference pulses, while the SNN unit captures the discrete jump features of phase and amplitude, achieving accurate differentiation between random, periodic, and sudden interference. Dedicated coding strategies are applied for different interference types: random interference uses a time-phase joint coding strategy to lock the phase; periodic interference uses a time-series interference avoidance coding strategy to offset the interference period by half; and sudden interference uses a triple-redundancy coding strategy to generate time-off redundant pulses. This changes the passive mode of traditional single coding to deal with all interferences, significantly reducing the packet loss rate in strong electromagnetic environments and fundamentally solving the problem of screen-filling alarms in monitoring systems.

[0086] Thirdly, traditional photovoltaic DC combiner box communication systems often experience communication interruptions over transmission distances of several hundred meters (over 500 meters) due to severe signal attenuation. After an interruption, manual on-site module restart is required, with recovery times exceeding 30 minutes, leading to missing power generation statistics. This application achieves a breakthrough in long-distance transmission reliability through advance prediction and precise compensation mechanisms. It utilizes a third cyclic pulse memory subnetwork based on an LSTM-SNN hybrid structure, fusing signal attenuation factor time-series data with effective pulse coding sequences. The LSTM unit learns the long-term dependency relationship between the attenuation factor and coding parameters, while the SNN unit captures discrete feature changes caused by parameter mutations, ultimately outputting predicted signal strength values ​​for each sampling moment in the next time period. The signal compensation module performs dual-dimensional compensation based on the predicted signal strength values ​​for each sampling moment in the next time period. If the strength is lower than a preset range, the transmission power is gradually increased, and the terminal resistor is matched according to the transmission distance, significantly reducing attenuation. Meanwhile, after compensation, a 0.5dBm level fine-tuning is triggered by a ±2dBm fluctuation threshold to ensure that the signal remains stable within the effective range, achieving "pre-compensation rather than post-disconnection repair," completely eliminating the cost of manual restart and ensuring the integrity of power generation statistics.

[0087] Fourth, this application upgrades from passive emergency repair to proactive early warning. Traditional photovoltaic DC combiner box communication systems rely on manual inspection and troubleshooting, resulting in delayed fault detection and low processing efficiency. This application constructs a new operation and maintenance model based on state prediction and proactive intervention through full-link data collection and analysis of the communication scheduling and control model. Interference type identification results can provide early warning of electromagnetic environment deterioration (e.g., a continuous increase in periodic interference intensity), guiding operation and maintenance personnel to specifically investigate potential problems with inverters or high-voltage cables. Signal strength prediction data can identify transmission link attenuation trends in advance (e.g., a monthly increase in cable impedance), enabling preventative replacement of aging cables. Protocol adaptation logs can automatically record cross-brand connection parameters, providing data support for subsequent system expansion.

[0088] This model reduces operation and maintenance response time from hours to minutes, significantly lowering operation and maintenance costs and further highlighting the practical value of high-reliability communication.

[0089] In summary, this application effectively solves the technical pain points of traditional systems, such as weak anti-interference, long-distance attenuation, and poor protocol compatibility, through a closed-loop technology of intelligent communication protocol adaptation, accurate interference identification, and early signal compensation. It significantly improves communication reliability and achieves several times the improvement in operation and maintenance efficiency, which fully meets the actual needs of photovoltaic power plants.

[0090] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A combiner box communication system based on a high-reliability communication module, characterized in that, include: The protocol configuration module is used to acquire the handshake request pulse sequence of multiple target pulse signal transmission requests initiated by the monitoring system to the photovoltaic DC combiner box, and transmit the handshake request pulse sequence to the first cyclic pulse memory sub-network in the communication scheduling control model that has been trained to the convergence state, so as to extract the timing feature vector of the protocol to be matched, and match the corresponding target communication protocol in the communication protocol library according to the timing feature vector of the protocol to be matched, wherein the target pulse signal includes a voltage pulse signal and its corresponding current pulse signal; The communication module is used to establish a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol, and then continuously collect the interference pulse timing data and signal attenuation factor timing data corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period. The signal encoding module is used to transmit the interference pulse timing data to the second cyclic pulse memory sub-network in the communication scheduling control model that has been trained to convergence, so as to determine the interference type corresponding to the interference pulse timing data, and call the corresponding pulse encoding strategy according to the interference type to determine the effective pulse encoding sequence of the target pulse signal in the next time period. The signal compensation module is used to transmit the signal attenuation factor timing data and the effective pulse coding sequence in the next time period to the third cyclic pulse memory sub-network in the communication scheduling and control model that has been trained to convergence state, so as to determine the signal strength value of the target pulse signal at each sampling moment in the next time period, and to compensate the signal strength value of the target pulse signal at each sampling moment in the next time period using a preset pulse signal compensation strategy, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding.

2. The combiner box communication system based on a high-reliability communication module according to claim 1, characterized in that... The handshake request pulse sequence includes a start bit, an address code, and a check code. The timing feature vector of the protocol to be matched includes the duration of the start bit pulse, the time difference between adjacent address code pulses, and the generation time difference of the check code relative to the address code. The communication module includes an electromagnetic shielding unit, and the interference pulse timing data characterizes the timing data constructed by the occurrence time of inverter noise data and high-voltage cable radiation data, the phase of the target pulse signal, and the amplitude of the target pulse signal. The signal attenuation factor time series data characterizes the time series data constructed from the transmission distance between the photovoltaic DC combiner box and the monitoring system, the cable impedance, and the ambient temperature of the photovoltaic DC combiner box within the current time period. The types of interference include random interference, periodic interference, or sudden interference; The basic network architecture of the communication scheduling and control model is an improved recurrent spiking neural network. The improved recurrent spiking neural network includes a pulse input layer, a pulse timing coding layer, a recurrent pulse memory layer, and a backpropagation layer. The recurrent pulse memory layer includes a first recurrent pulse memory subnetwork, a second recurrent pulse memory subnetwork, and a third recurrent pulse memory subnetwork. The basic network architecture of the first recurrent pulse memory subnetwork is a hybrid LSTM-CNN structure, and the basic network architectures of the second and third recurrent pulse memory subnetworks are hybrid LSTM-SNN structures.

3. The combiner box communication system based on a high-reliability communication module according to claim 2, characterized in that, The handshake request pulse sequence is transmitted to the first cyclic pulse memory subnetwork in the communication scheduling control model that has been trained to convergence, in order to extract the timing feature vector of the protocol to be matched, and to match the corresponding target communication protocol in the communication protocol library according to the timing feature vector of the protocol to be matched, including: The first LSTM unit in the first cyclic pulse memory sub-network is used to perform time-series feature mapping on the generation time difference between the check code and the address code in the handshake request pulse sequence to determine the first time-series associated feature, so as to construct the first time-series associated feature vector. The first CNN unit in the first cyclic pulse memory sub-network extracts the duration of the start bit pulse from the handshake request pulse sequence to determine the first discrete feature and construct the first discrete feature vector, and extracts the time difference between adjacent address code pulses from the handshake request pulse sequence to determine the second discrete feature and construct the second discrete feature vector. The first discrete feature vector, the second discrete feature vector, and the first temporal correlation feature vector after dimensional standardization are concatenated to construct the temporal feature vector of the protocol to be matched. The cosine similarity between the timing feature vector of the protocol to be matched and the timing feature vector of each standard communication protocol in the communication protocol library in the protocol configuration module is calculated. If the cosine similarity is greater than a preset similarity threshold, the standard communication protocol is used as the target communication protocol for establishing a connection between the monitoring system and the photovoltaic DC combiner box.

4. The combiner box communication system based on a high-reliability communication module according to claim 3, characterized in that, Establishing a connection between the photovoltaic DC combiner box and the monitoring system according to the target communication protocol includes: The protocol configuration module generates a pulse control command containing baud rate, parity method and data bit length according to the target communication protocol, and sends the pulse control command to the communication module to configure the communication parameters in the communication module; A handshake response pulse sequence is generated based on the communication parameters and returned to the monitoring system, so that the photovoltaic DC combiner box establishes a communication connection with the monitoring system.

5. The combiner box communication system based on a high-reliability communication module according to claim 2, characterized in that, The interference pulse timing data is transmitted to the second cyclic pulse memory subnetwork in the communication scheduling and control model that has been trained to a convergent state, in order to determine the interference type corresponding to the interference pulse timing data, including: The electromagnetic shielding unit in the communication module collects the timing data of the interference pulses corresponding to the target pulse signal in the photovoltaic DC combiner box at preset intervals within the current time period; The second LSTM unit in the second cyclic pulse memory sub-network is used to perform feature mapping on the long-term dependency relationship between the occurrence time, phase and amplitude of the interference pulse time series data, and to determine the second time series correlation feature to construct the second time series correlation feature vector. The second SNN unit in the second cyclic pulse memory sub-network captures the phase difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the third discrete feature to construct the third discrete feature vector, and extracts the amplitude difference of the target pulse signal at adjacent sampling times in the interference pulse time series data to determine the fourth discrete feature to construct the fourth discrete feature vector. The third discrete feature vector, the fourth discrete feature vector, and the second time-series correlation feature vector, after dimensionality standardization, are concatenated to construct the interference feature vector. The interference feature vector is then matched with a preset interference type feature library to determine the interference type corresponding to the interference pulse time-series data.

6. The combiner box communication system based on a high-reliability communication module according to claim 2, characterized in that, Based on the interference type, the corresponding pulse coding strategy is invoked to determine the effective pulse coding sequence of the target pulse signal in the next time period, including: If the interference type corresponding to the interference pulse timing data is random interference, then the time-phase joint coding strategy in the communication scheduling control model is invoked to fix the transmission time of the effective pulse signal in the next time period, and the phase of the effective pulse signal is locked within a preset phase threshold range to determine the effective pulse coding sequence in the next time period. The preset phase threshold range is within an interval of no more than 5° above or below 0°. If the interference type corresponding to the interference pulse timing data is periodic interference, then the timing interference avoidance coding strategy in the communication scheduling control model is invoked to offset the emission time of the effective pulse signal in the next time period by half of the interference period, so as to determine the effective pulse coding sequence in the next time period. The effective pulse coding sequence includes the pulse emission time, phase and frequency of the pulse signal. If the interference type corresponding to the interference pulse timing data is sudden interference, then the triple redundancy coding strategy in the communication scheduling control model is invoked to generate multiple pulse signals of the same duration with a preset time offset, so as to determine the effective pulse coding sequence in the next time period.

7. The combiner box communication system based on a high-reliability communication module according to claim 2, characterized in that, The signal attenuation factor time-series data and the effective pulse-coded sequence for the next time period are transmitted to the third cyclic pulse memory subnetwork in the communication scheduling and control model that has been trained to convergence, in order to determine the signal strength value of the target pulse signal at each sampling time in the next time period, including: The third LSTM unit in the third cyclic pulse memory sub-network is used to perform time-series correlation feature mapping on the long-term dependency between the signal attenuation factor time-series data and the effective pulse coding sequence to determine the third time-series correlation feature, so as to construct the third time-series correlation feature vector. The fifth discrete feature is determined by capturing the discrete feature changes of the effective pulse coding sequence under different signal attenuation factor time series data combinations through the third SNN unit in the third cyclic pulse memory sub-network, so as to construct the fifth discrete feature vector. The third temporal correlation feature and the fifth discrete feature vector, after dimensionality standardization, are concatenated to construct a fused feature vector. The fused feature vector is then transmitted to the fully connected layer in the third cyclic pulse memory sub-network. After nonlinear transformation by the Sigmoid activation function, it is then denormalized and mapped to map the fused feature vector to the signal strength values ​​at each sampling time in the next time period.

8. The combiner box communication system based on a high-reliability communication module according to claim 1, characterized in that, A preset pulse signal compensation strategy is used to compensate the signal strength values ​​of the target pulse signal at each sampling moment in the next time period, so as to determine the compensated signal strength value and transmit it to the monitoring system for decoding, including: A preset pulse signal compensation strategy is invoked to detect whether the signal strength value of the target pulse signal at each sampling moment in the next time period falls within the preset signal strength range. If it does, the current transmission power of the photovoltaic DC combiner box is gradually increased to the target transmission power corresponding to the preset signal strength range, and the corresponding resistance parameters are matched according to the transmission distance between the photovoltaic DC combiner box and the monitoring system.

9. The combiner box communication system based on a high-reliability communication module according to claim 1, characterized in that, The standard communication protocols include Modbus-RTU communication protocol and DL / T 645 communication protocol.

10. The combiner box communication system based on a high-reliability communication module according to any one of claims 1 to 9, characterized in that, The protocol configuration module includes an industrial-grade STM32F407 main control chip, an integrated pulse signal acquisition interface, a Flash protocol storage unit, and a UART communication interface.

Images