A photovoltaic thermoelectric power generation energy collection method based on deep learning

Abstract

This invention discloses a deep learning-based photovoltaic thermoelectric power generation energy harvesting method, comprising the following steps: acquiring multi-source thermoelectric power generation data of photovoltaic modules and performing standardization processing; extracting module regional temperature difference, temperature difference change rate, regional temperature difference gradient, thermoelectric output state, energy storage state, and load state to generate thermoelectric power generation characteristic data; constructing a regional thermoelectric power generation characteristic sequence; inputting the regional thermoelectric power generation characteristic sequence into an improved thermal inertia liquid time constant (LTC) energy harvesting model to obtain regional harvestable energy prediction results; generating a regional energy harvesting strategy based on the prediction results, and adjusting the series-parallel switching mode, energy storage access state, and load matching parameters; collecting actual energy harvesting data and generating feedback correction data. This invention, based on thermal inertia liquid time constant (LTC) modeling technology, realizes regional prediction, adaptive harvesting, and closed-loop correction of photovoltaic thermoelectric power generation energy.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic energy recovery and thermoelectric power generation control technology, specifically to a photovoltaic thermoelectric power generation energy harvesting method based on deep learning. Background Technology

[0002] Photovoltaic modules convert solar radiation into electricity during operation. However, due to limitations in photoelectric conversion efficiency, some solar energy accumulates as heat on the module surface and backsheet, causing significant temperature differences between the front and back sides, regional temperature differences, and thermal inertia over time. Thermoelectric power generation technology can utilize the residual heat generated during photovoltaic module operation to further convert thermal energy into electricity, thereby improving the overall utilization rate of solar energy resources in the photovoltaic system. However, the photovoltaic thermoelectric power generation process is affected by many factors, including irradiance, ambient temperature, wind speed, module installation angle, backsheet heat dissipation conditions, thermoelectric unit connection method, energy storage status, and load demand. Significant differences exist in the output voltage, current, and harvestable energy of thermoelectric units in different regions.

[0003] Existing photovoltaic thermoelectric power generation energy harvesting methods mostly rely on fixed thresholds, simple temperature difference judgments, or traditional maximum power point tracking strategies for control. They typically determine the connection status of thermoelectric power generation units based on the current temperature difference or current output voltage, lacking comprehensive analysis of temperature difference trends, thermal inertia hysteresis, and the matching relationship with energy storage loads. When photovoltaic modules are in the heating, stable power generation, or cooling degradation phases, the harvestable energy of the thermoelectric power generation unit is not entirely determined by the instantaneous temperature difference. If a fixed connection strategy is still adopted, problems such as ineffective connection of low-temperature difference units, insufficient harvesting of high-temperature difference units, unreasonable series-parallel switching, and mismatch of energy storage side connection can easily occur, leading to reduced energy harvesting efficiency.

[0004] Furthermore, photovoltaic thermoelectric power generation data is characterized by multi-source heterogeneity, asynchronous sampling, local missing data, and significant temporal fluctuations. Traditional rule-based control methods struggle to establish stable correlations between multi-regional temperature differences, environmental disturbances, electrical output, energy storage status, and load status. While ordinary deep learning prediction models can fit time-series data, they typically lack specific modeling for thermal inertia time constants, temperature difference state evolution, and thermoelectric conversion constraints, making it difficult to directly serve zoned access, series-parallel switching, energy storage access, and load matching control.

[0005] Therefore, how to provide a photovoltaic thermoelectric energy harvesting method based on deep learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a deep learning-based photovoltaic thermoelectric power generation energy harvesting method. This invention fully utilizes photovoltaic thermoelectric power generation data acquisition, regional thermoelectric feature construction, an improved thermal inertia liquid time constant (LTC) energy harvesting model, and energy harvesting control technology. It details the complete energy harvesting process from multi-source thermoelectric power generation data standardization processing, thermoelectric power generation feature generation, regional thermoelectric power generation feature sequence construction, regional harvestable energy prediction, to regional access, series-parallel switching, energy storage access, and load matching control. It has the advantages of strong dynamic adaptability to temperature differences, high energy harvesting control accuracy, good energy storage load matching capability, and high photovoltaic waste heat utilization rate.

[0007] A photovoltaic thermoelectric power harvesting method based on deep learning according to an embodiment of the present invention includes the following steps:

[0008] S1. Obtain multi-source thermoelectric power generation data of photovoltaic modules during the energy collection cycle, and perform time alignment, anomaly removal, missing data filling and normalization on the multi-source thermoelectric power generation data to obtain standardized thermoelectric power generation data.

[0009] S2. Based on standardized thermoelectric power generation data, extract component area temperature difference, temperature difference change rate, area temperature difference gradient, thermoelectric output status, energy storage status and load status to generate thermoelectric power generation characteristic data;

[0010] S3. Based on the location of photovoltaic modules and the connection relationship of thermoelectric power generation units, perform partition mapping and channel association processing on the thermoelectric power generation characteristic data to obtain the partition thermoelectric power generation characteristic sequence.

[0011] S4. Input the regional thermoelectric power generation characteristic sequence into the improved thermal inertia liquid time constant (LTC) energy harvesting model. After thermal inertia time constant generation, thermoelectric state evolution, thermoelectric conversion constraint gating, and regional energy mapping processing, the predicted results of regional harvestable energy are obtained.

[0012] S5. Based on the predicted energy harvestable by zone and the energy storage status, prioritize the access of different thermoelectric power generation units and match the collection channels to generate a zoned energy harvesting strategy.

[0013] S6. Adjust the series-parallel switching mode, energy storage access status and load matching parameters of the thermoelectric power generation unit according to the zoned energy harvesting strategy to obtain the energy harvesting control result;

[0014] S7. Collect the actual collected electrical energy, actual output voltage and actual output current corresponding to the energy harvesting control results, calculate the deviation between the predicted energy harvestable data and the actual energy harvesting data of the zone, and obtain feedback correction data.

[0015] Optionally, step S1 includes the following steps:

[0016] S11. Acquire temperature data, environmental data, thermoelectric output data, energy storage data and load data of photovoltaic modules during the energy collection cycle, and collect them according to the collection time and area number to obtain the original thermoelectric power generation dataset;

[0017] S12. Based on the sampling time of the energy harvesting cycle, perform timestamp correction, sampling interval unification, and area number matching on the original thermoelectric power generation dataset to obtain time-aligned thermoelectric power generation data.

[0018] S13. Based on the preset temperature range, voltage range, current range and load access records, perform anomaly identification, anomaly removal and missing data filling on the time-aligned thermoelectric power generation data to obtain complete thermoelectric power generation data.

[0019] S14. Perform dimension unification, normalization, and time-series processing on the temperature data, environmental data, electrical output data, energy storage data, and load data in the complete thermoelectric power generation data to obtain standardized thermoelectric power generation data.

[0020] Optionally, step S2 includes the following steps:

[0021] S21. Based on the front temperature data and back sheet temperature data in the standardized thermoelectric power generation data, calculate the front and back temperature difference according to the area number to obtain the module area temperature difference data.

[0022] S22. Based on the changes in temperature difference data of the component area between continuous sampling times, calculate the temperature difference increase, temperature difference decrease, and temperature difference hold-up to obtain the temperature difference change rate data.

[0023] S23. Based on the spatial adjacency relationship between different regions of the photovoltaic module, perform horizontal and vertical differential processing on the temperature difference data of the module region to obtain the regional temperature difference gradient data.

[0024] S24. Based on the output voltage, output current, energy storage voltage, energy storage current and load access records in the standardized thermoelectric power generation data, perform joint encoding to obtain thermoelectric output status data, energy storage operation status data and load operation status data.

[0025] S25. The component area temperature difference data, temperature difference change rate data, area temperature difference gradient data, thermoelectric output status, energy storage status and load status are spliced ​​together according to the sampling time to generate thermoelectric power generation characteristic data.

[0026] Optionally, step S3 includes the following steps:

[0027] S31. Based on the row and column positions of photovoltaic modules, the installation positions of thermoelectric power generation units, and the positions of temperature sampling points, establish a mapping relationship between module area numbers and thermoelectric power generation unit numbers to obtain an area unit mapping table.

[0028] S32. Based on the series connection relationship, parallel connection relationship and switching switch connection relationship between the thermoelectric power generation units, establish the channel association relationship between the thermoelectric power generation units and the energy harvesting channel to obtain the unit channel association table.

[0029] S33. Based on the regional unit mapping table, the thermoelectric power generation characteristic data is divided into regions. The temperature difference characteristics, thermoelectric output characteristics, energy storage status characteristics and load status characteristics in the same region are merged to obtain the regional thermoelectric power generation characteristics.

[0030] S34. Based on the unit channel association table, channel marking and connection status marking are performed on the partitioned thermoelectric power generation characteristics to obtain a partitioned thermoelectric power generation characteristic sequence with region number, unit number and channel number.

[0031] Optionally, step S4 includes the following steps:

[0032] S41. Input the temperature difference amplitude, temperature difference rate of change, regional temperature difference gradient, irradiance change and wind speed disturbance in the regional thermoelectric power generation characteristic sequence into the thermal inertia time constant generation unit to generate liquid time constant parameters corresponding to different thermoelectric power generation units.

[0033] S42. Input the partitioned thermoelectric power generation characteristic sequence and liquid time constant parameter into the liquid time constant state evolution layer, and continuously update the hot end state, cold end state, voltage output state and current output state to obtain the dynamic characteristics of thermoelectric power generation.

[0034] S43. Input the thermoelectric power generation dynamic characteristics, output voltage data, output current data, energy storage voltage data and load state data into the thermoelectric conversion constraint gating unit, perform gating coefficient calculation and feature fusion, and obtain the energy harvesting state characteristics.

[0035] S44. Input the energy harvesting status characteristics into the partitioned energy mapping layer, and perform mapping processing on the energy output data of different regions and different thermoelectric power generation units to obtain the partitioned harvestable energy prediction results.

[0036] Optionally, step S41 includes the following steps:

[0037] S411. Extract the temperature difference amplitude sequence, temperature difference rate of change sequence, and regional temperature difference gradient sequence of the same region at continuous sampling time from the regional thermoelectric power generation characteristic sequence to form the thermoelectric thermal inertia input sequence.

[0038] S412. Extract the changes in irradiance intensity, wind speed, and ambient temperature from the regional thermoelectric power generation characteristic sequence to form an environmental disturbance input sequence;

[0039] S413. Input the temperature difference thermal inertia input sequence and the environmental disturbance input sequence into the thermal inertia weight mapping layer, and assign weights to the heating response, steady state maintenance and cooling decay of different regions to obtain thermal inertia weight parameters.

[0040] S414. Associate the thermal inertia weight parameter with the thermoelectric power generation unit number, region number, and channel number to generate liquid time constant parameters for each region, unit, and channel.

[0041] S415. Configure the liquid time constant parameter to the state update term of the liquid time constant state evolution layer to obtain the state evolution input of the improved thermal inertial liquid time constant LTC energy harvesting model.

[0042] Optionally, step S43 includes the following steps:

[0043] S431. Generate thermoelectric state fusion characteristics based on the hot end state characteristics, cold end state characteristics, voltage output state characteristics and current output state characteristics in the dynamic characteristics of thermoelectric power generation.

[0044] S432. Based on the output voltage data, output current data, energy storage voltage data, and load status data, calculate the electrical matching characteristics between the thermoelectric generator unit and the energy storage side and the load side.

[0045] S433. Input the thermoelectric state fusion characteristics and electrical matching characteristics into the gating mapping layer to calculate the thermoelectric conversion gating coefficient of each thermoelectric power generation unit under the corresponding energy harvesting channel.

[0046] S434. The dynamic characteristics of thermoelectric power generation are weighted according to the thermoelectric conversion gate coefficient to obtain the gated dynamic characteristics of thermoelectric power generation.

[0047] S435. The dynamic characteristics of the temperature difference power generation after gating are spliced ​​and mapped with the energy storage state characteristics and the load state characteristics to obtain the energy harvesting state characteristics.

[0048] Optionally, step S5 includes the following steps:

[0049] S51. Based on the predicted energy value, predicted voltage value and predicted current value of different regions in the next energy collection cycle, extract the predicted energy value, predicted voltage value and predicted current value of different regions according to the predicted energy of the region, and obtain the region prediction output data.

[0050] S52. Based on the energy storage voltage, remaining capacity and allowable charging current in the energy storage state, perform access condition matching on the partition prediction output data to obtain a set of candidate thermoelectric generator units that meet the energy storage access conditions.

[0051] S53. Based on the predicted energy value, predicted voltage value, predicted current value and channel connection status in the candidate thermoelectric power generation unit set, prioritize the candidate thermoelectric power generation units to obtain the access order of the thermoelectric power generation units.

[0052] S54. Based on the access sequence of the thermoelectric power generation units and the current occupancy status of the energy collection channels, the energy collection channels are divided into main collection channels, maintenance collection channels and channels to be accessed, and corresponding channels are matched for different thermoelectric power generation units to obtain the collection channel matching results.

[0053] S55. Combine the connection sequence of the thermoelectric generator units and the matching results of the collection channels to generate a zoned energy collection strategy.

[0054] Optionally, step S6 includes the following steps:

[0055] S61. Read the access sequence of thermoelectric power generation units, the matching result of collection channels and the access conditions of energy storage in the partitioned energy harvesting strategy to form energy harvesting control input data;

[0056] S62. Based on the predicted voltage value, predicted current value and channel connection status in the energy harvesting control input data, select the series connection method, parallel connection method or group connection method of the thermoelectric power generation unit, and obtain the series-parallel switching command.

[0057] S63. Based on the energy storage voltage, remaining capacity and allowable charging current in the energy harvesting control input data, select the energy storage unit access state, energy storage unit disconnect state or energy storage bypass state, and obtain the energy storage access command.

[0058] S64. Based on the load status, predicted energy value, and energy harvesting channel status in the energy harvesting control input data, adjust the load matching impedance and converter duty cycle to obtain the load matching parameters.

[0059] S65. Send the series-parallel switching command, energy storage access command and load matching parameters to the energy harvesting control circuit to obtain the energy harvesting control result.

[0060] Optionally, step S7 includes the following steps:

[0061] S71. Collect the actual collected electrical energy, actual output voltage, actual output current, energy storage voltage change and load connection status after the energy harvesting control results are executed, and obtain the actual energy harvesting data;

[0062] S72. Match the actual energy collection data with the predicted energy collection results of the zone according to the zone number, thermoelectric power generation unit number, channel number and sampling time to obtain the predicted actual corresponding data;

[0063] S73. Calculate the predicted energy deviation, predicted voltage deviation, and predicted current deviation based on the actual predicted data to obtain energy harvesting deviation data;

[0064] S74. Based on the energy harvesting deviation data and the temperature difference power generation characteristic data at the corresponding sampling time, calculate the correction amount of the thermal inertia weight parameter, the correction amount of the thermoelectric conversion gate coefficient, and the correction amount of the partitioned energy mapping layer parameter to obtain the model correction parameters.

[0065] S75. Combine the model correction parameters, energy harvesting deviation data, and control parameter correction amounts for the next energy harvesting cycle to obtain feedback correction data.

[0066] The beneficial effects of this invention are as follows: This invention acquires temperature data, environmental data, thermoelectric output data, energy storage data, and load data of photovoltaic modules during the energy harvesting cycle, and performs time alignment, anomaly removal, missing data filling, and normalization processing on multi-source thermoelectric power generation data. This enables data from different acquisition sources, different sampling periods, and different regional locations to form unified standardized thermoelectric power generation data, reducing the impact of sensor anomalies, asynchronous sampling, and local missing data on subsequent prediction and control processes.

[0067] This invention extracts the temperature difference, temperature change rate, regional temperature gradient, thermoelectric output status, energy storage status, and load status of the component area based on standardized thermoelectric power generation data. It also constructs a regional thermoelectric power generation characteristic sequence by combining the regional location of the photovoltaic module and the connection relationship of the thermoelectric power generation unit. This allows the hot-end changes, cold-end changes, spatial distribution differences, and electrical output status of the thermoelectric power generation unit to correspond with the actual energy collection channel, avoiding the collection judgment deviation caused by controlling based only on a single point temperature difference or instantaneous output.

[0068] This invention inputs the characteristic sequence of zoned thermoelectric power generation into an improved thermal inertia liquid time constant (LTC) energy harvesting model. Through thermal inertia time constant generation, temperature difference state evolution, thermoelectric conversion constraint gating, and zoned energy mapping, it models the dynamic changes of different thermoelectric power generation units during heating, steady-state, and cooling processes, enabling more accurate prediction of the harvestable energy of each zone in the next energy harvesting cycle. This model incorporates temperature difference variation patterns, environmental disturbances, and electrical matching relationships into the state evolution process, improving its adaptability to the nonlinearity, hysteresis, and volatility of photovoltaic thermoelectric power generation.

[0069] This invention prioritizes thermoelectric power generation units (TEGs) and matches them with collection channels based on the predicted energy harvestability and storage status of each zone. It further adjusts the series-parallel switching method, storage access status, and load matching parameters to ensure that high-value TEGs are preferentially connected to suitable channels, reducing inefficient access and invalid switching. By calculating the deviation of the predicted results using actual harvested energy, actual output voltage, and actual output current, and generating feedback correction data, the invention can correct model and control parameters during continuous energy harvesting, improving the stability and continuous optimization capability of the photovoltaic thermoelectric power generation energy harvesting process. Attached Figure Description

[0070] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0071] Figure 1 This is an overall flowchart of a photovoltaic thermoelectric power harvesting method based on deep learning proposed in this invention;

[0072] Figure 2 This is a schematic diagram illustrating the process of multi-source thermoelectric power generation data processing, thermoelectric power generation feature generation, and partitioned thermoelectric power generation feature sequence construction in this invention.

[0073] Figure 3 This is a schematic diagram of the improved thermal inertia liquid time constant (LTC) energy harvesting model, the generation of the partitioned energy harvesting strategy, and the feedback correction process in this invention. Detailed Implementation

[0074] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0075] refer to Figures 1 to 3 A deep learning-based photovoltaic thermoelectric power harvesting method includes the following steps:

[0076] S1. Obtain multi-source thermoelectric power generation data of photovoltaic modules during the energy collection cycle, and perform time alignment, anomaly removal, missing data filling and normalization on the multi-source thermoelectric power generation data to obtain standardized thermoelectric power generation data.

[0077] S2. Based on standardized thermoelectric power generation data, extract component area temperature difference, temperature difference change rate, area temperature difference gradient, thermoelectric output status, energy storage status and load status to generate thermoelectric power generation characteristic data;

[0078] S3. Based on the location of photovoltaic modules and the connection relationship of thermoelectric power generation units, perform partition mapping and channel association processing on the thermoelectric power generation characteristic data to obtain the partition thermoelectric power generation characteristic sequence.

[0079] S4. Input the regional thermoelectric power generation characteristic sequence into the improved thermal inertia liquid time constant (LTC) energy harvesting model. After thermal inertia time constant generation, thermoelectric state evolution, thermoelectric conversion constraint gating, and regional energy mapping processing, the predicted results of regional harvestable energy are obtained.

[0080] S5. Based on the predicted energy harvestable by zone and the energy storage status, prioritize the access of different thermoelectric power generation units and match the collection channels to generate a zoned energy harvesting strategy.

[0081] S6. Adjust the series-parallel switching mode, energy storage access status and load matching parameters of the thermoelectric power generation unit according to the zoned energy harvesting strategy to obtain the energy harvesting control result;

[0082] S7. Collect the actual collected electrical energy, actual output voltage and actual output current corresponding to the energy harvesting control results, calculate the deviation between the predicted energy harvestable data and the actual energy harvesting data of the zone, and obtain feedback correction data.

[0083] refer to Figure 1 and Figure 2 In this embodiment, S1 is the standardization processing step for multi-source thermoelectric power generation data.

[0084] Temperature data, environmental data, thermoelectric output data, energy storage data, and load data of photovoltaic modules during the energy harvesting cycle are acquired and aggregated according to the collection time and region number to obtain the original thermoelectric power generation dataset.

[0085] Temperature data includes the front temperature of the photovoltaic module, the backsheet temperature of the photovoltaic module, and the edge temperature of the module; environmental data includes ambient temperature, irradiance, and wind speed.

[0086] Thermoelectric output data includes the output voltage, output current, and connection status of the thermoelectric generator unit;

[0087] Energy storage data includes energy storage voltage, energy storage current, remaining capacity, and allowable charging current;

[0088] Load data includes load connection status, load change records, and load impedance status.

[0089] Different data are associated with each other according to the sampling device number, sampling time, component area number and thermoelectric power generation unit number, forming a raw thermoelectric power generation dataset that can be processed later.

[0090] Based on the sampling time of the energy harvesting cycle, the original thermoelectric power generation dataset is corrected with timestamps, the sampling interval is unified, and the area number is matched to obtain time-aligned thermoelectric power generation data.

[0091] Specifically, the system unifies the data uploaded by the photovoltaic module controller, temperature collector, environmental collector, thermoelectric power generation output collector, and energy storage controller to the same time base, and corrects data whose timestamp offset exceeds the preset range;

[0092] Data generated at different sampling frequencies are resampled to ensure that temperature data, environmental data, thermoelectric output data, energy storage data, and load data within the same energy harvesting cycle are arranged according to a unified sampling time.

[0093] Temperature sampling points, thermoelectric power generation units, and energy harvesting channels within the same photovoltaic module area are matched with regional numbers so that various types of data can be read according to the same regional index.

[0094] Based on preset temperature range, voltage range, current range, and load access records, anomaly identification, anomaly removal, and missing data filling are performed on time-aligned thermoelectric power generation data to obtain complete thermoelectric power generation data.

[0095] Specifically, the system identifies anomalies in data that exceed the reasonable temperature variation range, data with a sudden change amplitude significantly greater than that of adjacent sampling times, and data with sudden changes in output voltage or output current but mismatched temperature difference.

[0096] Mark data gaps caused by communication packet loss, sampling interruption, and missing load access status;

[0097] For short-term missing data, data from adjacent sampling times and historical trends of the same area are used to fill the gaps. For data segments with consecutive missing times exceeding a preset threshold, missing markers are retained and the weights of subsequent features are reduced, thereby obtaining complete thermoelectric power generation data.

[0098] Standardized thermoelectric power generation data is obtained by unifying, normalizing, and time-series processing the temperature data, environmental data, electrical output data, energy storage data, and load data in the complete thermoelectric power generation data.

[0099] Specifically, temperature data is scaled uniformly according to the Celsius temperature range, irradiance is scaled uniformly according to the rated measurement range of the acquisition device, output voltage and output current are scaled uniformly according to the rated output range of the thermoelectric generator, energy storage voltage and remaining capacity are scaled uniformly according to the operating range of the energy storage device, and load connection status is organized according to the discrete state coding method.

[0100] The standardized thermoelectric power generation data are arranged sequentially according to the sampling time. Each sampling time corresponds to a set of temperature characteristics, environmental characteristics, electrical output characteristics, energy storage characteristics and load characteristics, and the area number, unit number and channel number are retained.

[0101] refer to Figure 2 In this embodiment, S2 is the step of generating characteristic data of thermoelectric power generation.

[0102] Based on the front and back temperature data in the standardized thermoelectric power generation data, the temperature difference between the front and back sides is calculated according to the area number to obtain the module area temperature difference data.

[0103] When there are multiple temperature sampling points in the same area, the system first performs a weighted average of the front temperature sampling values ​​in the same area, and then performs a weighted average of the back panel temperature sampling values ​​in the same area. Finally, it calculates the difference between the average front temperature and the average back panel temperature to form component area temperature difference data corresponding to the area number.

[0104] Based on the changes in temperature difference data in the component area between continuous sampling times, the magnitude of temperature difference increase, the magnitude of temperature difference decrease, and the magnitude of temperature difference hold-up are calculated to obtain the temperature difference change rate data.

[0105] Specifically, when the regional temperature difference at the next sampling time is greater than the regional temperature difference at the previous sampling time, the system records the increase in temperature difference.

[0106] When the regional temperature difference at the next sampling time is less than the regional temperature difference at the previous sampling time, the system records the magnitude of the temperature difference decrease.

[0107] When the temperature difference between consecutive sampling times is within a preset small fluctuation range, the system records the temperature difference holding amplitude.

[0108] Temperature difference change rate data can reflect the dynamic changes of photovoltaic modules during the heating, stabilization and cooling stages.

[0109] Based on the spatial adjacency relationship between different regions of the photovoltaic module, the temperature difference data of the module region is processed by horizontal and vertical differential to obtain the regional temperature difference gradient data.

[0110] Photovoltaic module areas can be divided into rows and columns, with different heat dissipation conditions existing between the central area, edge area, and corner area.

[0111] The system calculates the temperature difference between horizontally adjacent areas and the temperature difference between vertically adjacent areas based on the area numbering relationship between adjacent areas, forming regional temperature gradient data.

[0112] Regional temperature gradient data can reflect the degree of uneven heat distribution between the component surface and the backsheet, and provide spatial difference information for subsequent zoned energy harvesting.

[0113] By jointly encoding the output voltage, output current, energy storage voltage, energy storage current, and load access records in the standardized thermoelectric power generation data, we can obtain thermoelectric output status data, energy storage operation status data, and load operation status data.

[0114] The thermoelectric output status data includes the voltage level, current level, and output fluctuation status of the thermoelectric generator at the current sampling time;

[0115] Energy storage operation status data includes energy storage voltage range, energy storage capacity range, and charging reception status;

[0116] Load operating status data includes load connection strength, load impedance change status, and load switching status.

[0117] The system encodes the above states according to a unified sampling time, forming operating characteristics synchronized with the temperature difference data.

[0118] The thermoelectric output status, energy storage status, and load status are spliced ​​together according to the sampling time to generate thermoelectric power generation characteristic data.

[0119] The thermoelectric power generation characteristic data is indexed sequentially by sampling time and located by region number and thermoelectric power generation unit number. Each sampling time corresponds to multiple characteristic items, including regional temperature difference, temperature difference change rate, regional temperature difference gradient, output voltage, output current, energy storage voltage, energy storage capacity, load status, and connection status.

[0120] Through this process, the system integrates thermal characteristics, electrical characteristics, and control state characteristics into a unified data object.

[0121] refer to Figure 2 In this embodiment, S3 is the step of constructing the regional thermoelectric power generation feature sequence.

[0122] Based on the row and column positions of photovoltaic modules, the installation positions of thermoelectric power generation units, and the positions of temperature sampling points, a mapping relationship between module area numbers and thermoelectric power generation unit numbers is established to obtain an area unit mapping table.

[0123] For each photovoltaic module, the system divides the module into a central area, an upper edge area, a lower edge area, a left edge area, and a right edge area based on the installation area of ​​the thermoelectric power generation unit on the module back panel, and establishes a correspondence between each area and one or more thermoelectric power generation units.

[0124] Based on the series connection, parallel connection, and switching connection relationships between the thermoelectric power generation units, a channel association relationship between the thermoelectric power generation units and the energy harvesting channels is established, resulting in a unit channel association table.

[0125] The unit channel association table records the main collection channel, maintenance collection channel, and pending access channel that each thermoelectric power generation unit can connect to, as well as the corresponding switching switch, energy storage port, and load matching circuit.

[0126] This association table enables subsequent prediction results to be directly converted into access order and channel matching results.

[0127] Based on the regional unit mapping table, the thermoelectric power generation characteristic data is divided into regions. The temperature difference characteristics, thermoelectric output characteristics, energy storage status characteristics and load status characteristics within the same region are merged to obtain the regional thermoelectric power generation characteristics.

[0128] The system reads the component area temperature difference data, temperature difference change rate data, and area temperature difference gradient data according to the area number, and reads the output voltage, output current, and connection status of the thermoelectric power generation unit in the same area. Then it matches them with the energy storage status and load status so that each partition corresponds to a complete set of thermal and electrical status.

[0129] Based on the unit channel association table, channel marking and connection status marking are performed on the regional thermoelectric power generation characteristics to obtain a regional thermoelectric power generation characteristic sequence with region number, unit number and channel number.

[0130] The sequence is arranged continuously according to the sampling time within the energy harvesting cycle. Each sequence segment corresponds to a region or a thermoelectric power generation unit. The sequence segment includes temperature difference change, thermoelectric output, energy storage status, load status and channel connection status.

[0131] The characteristic sequence of zoned thermoelectric power generation is used as input data for the improved thermal inertia liquid time constant (LTC) energy harvesting model.

[0132] refer to Figure 3 In this embodiment, S4 is the prediction step of the improved thermal inertia liquid time constant (LTC) energy harvesting model.

[0133] The model includes an input embedding layer, a thermal inertia time constant generation unit, a liquid time constant state evolution layer, a thermoelectric conversion constraint gating unit, a partitioned energy mapping layer, and a prediction output layer.

[0134] The input embedding layer receives the feature sequence of regional thermoelectric power generation and maps temperature-related features, environmental features, electrical output features, energy storage features, and load-related features to the same latent space.

[0135] The thermal inertia time constant generation unit generates liquid time constant parameters based on temperature difference amplitude, temperature difference change rate, regional temperature difference gradient, irradiance change, and wind speed disturbance.

[0136] The liquid time constant state evolution layer continuously updates the hidden state based on the liquid time constant parameter;

[0137] The thermoelectric conversion constraint gating unit constrains the hidden features of the model based on the output voltage, output current, energy storage voltage, and load state;

[0138] The partitioned energy mapping layer outputs the predicted harvestable energy for different regions and different thermoelectric power generation units.

[0139] The temperature difference amplitude, temperature difference rate of change, regional temperature difference gradient, irradiance change and wind speed disturbance in the regional thermoelectric power generation characteristic sequence are input into the thermal inertia time constant generation unit to generate liquid time constant parameters corresponding to different thermoelectric power generation units.

[0140] Specifically, the temperature difference amplitude sequence, temperature difference rate of change sequence, and regional temperature difference gradient sequence of the same region at continuous sampling time are extracted from the regional thermoelectric power generation characteristic sequence to form the thermoelectric thermal inertia input sequence.

[0141] The changes in irradiance, wind speed, and ambient temperature are extracted from the regional thermoelectric power generation characteristic sequence to form an environmental disturbance input sequence;

[0142] The thermal inertia input sequence and the environmental disturbance input sequence are input into the thermal inertia weight mapping layer to assign weights to the heating response, steady-state maintenance and cooling decay of different regions, and obtain the thermal inertia weight parameters.

[0143] The thermal inertia weighting parameter is associated with the thermoelectric power generation unit number, region number, and channel number to generate liquid time constant parameters for each region, unit, and channel.

[0144] By configuring the liquid time constant parameter into the state update term of the liquid time constant state evolution layer, the state evolution input of the improved thermal inertial liquid time constant LTC energy harvesting model is obtained.

[0145] In this embodiment, the thermal inertia time constant generation unit is composed of a feature normalization sublayer, a thermal inertia weight mapping sublayer, and a parameter constraint sublayer connected in sequence.

[0146] The feature normalization sublayer receives temperature difference thermal inertia input sequences and environmental disturbance input sequences, bringing features of different dimensions into the same numerical range;

[0147] The thermal inertia weighted mapping sublayer consists of two levels of linear mapping and nonlinear gating, generating different weights according to the three stages of heating, steady state and cooling;

[0148] The parameter constraint sublayer limits the range of the liquid time constant parameter to avoid the state evolution rate being too fast or too slow.

[0149] Compared with the ordinary LTC model, the liquid time constant in this embodiment is not generated solely by the hidden state, but is generated by combining the thermal inertia of the photovoltaic module, environmental disturbances, and regional temperature gradient, so that the model can fit the heat accumulation and heat decay phenomena in the photovoltaic module temperature difference power generation process.

[0150] The characteristic sequence of thermoelectric power generation in different zones and the liquid time constant parameter are input into the liquid time constant state evolution layer. The hot end state, cold end state, voltage output state and current output state are continuously updated to obtain the dynamic characteristics of thermoelectric power generation.

[0151] The liquid time constant state evolution layer includes a hot end state channel, a cold end state channel, a voltage output state channel, and a current output state channel.

[0152] The hot-end status channel receives frontal temperature, irradiation changes, and regional temperature gradients.

[0153] The cold end status channel receives backplane temperature, ambient temperature, and wind speed disturbance.

[0154] The voltage output status channel receives the output voltage and connection status of the thermoelectric generator unit;

[0155] The current output status channel receives the output current, load status, and energy storage access status.

[0156] The hidden states of the four state channels are interactively fused at each sampling time to form the dynamic characteristics of thermoelectric power generation that reflect thermal and electrical output changes.

[0157] In this embodiment, the liquid time constant state evolution layer adopts the concept of continuous time state update to adapt to different sampling intervals and different temperature difference change rates.

[0158] For regions where the temperature difference rises rapidly, the liquid time constant parameter enhances the influence of the current temperature difference change on the hidden state in the state evolution layer.

[0159] For regions where temperature differences are maintained, the liquid time constant parameter allows the state evolution layer to retain more historical states.

[0160] In regions where the temperature difference decreases rapidly, the liquid time constant parameter enhances the response of the state evolution layer to the cooling decay trend.

[0161] This structure enables the model to distinguish between regions with high instantaneous temperature differences but rapid decay, and regions with slightly lower temperature differences but longer durations of temperature variation, thus providing more stable dynamic characteristics for subsequent regional energy harvesting prediction.

[0162] The dynamic characteristics of thermoelectric power generation, output voltage data, output current data, energy storage voltage data, and load state data are input into the thermoelectric conversion constraint gating unit to calculate the gating coefficient and fuse the features to obtain the energy harvesting state characteristics.

[0163] Specifically, based on the hot-end state characteristics, cold-end state characteristics, voltage output state characteristics, and current output state characteristics in the dynamic characteristics of thermoelectric power generation, thermoelectric state fusion characteristics are generated.

[0164] Based on the output voltage data, output current data, energy storage voltage data, and load status data, the electrical matching characteristics between the thermoelectric generator unit and the energy storage side and load side are calculated.

[0165] The thermoelectric state fusion features and electrical matching features are input into the gating mapping layer to calculate the thermoelectric conversion gating coefficient of each thermoelectric power generation unit under the corresponding energy harvesting channel.

[0166] The dynamic characteristics of thermoelectric power generation are weighted according to the thermoelectric conversion gate coefficient to obtain the gated dynamic characteristics of thermoelectric power generation.

[0167] By splicing and mapping the gated thermoelectric power generation dynamic characteristics with the energy storage state characteristics and load state characteristics, the energy harvesting state characteristics are obtained.

[0168] In this embodiment, the thermoelectric conversion constraint gating unit includes a thermoelectric state fusion sublayer, an electrical matching sublayer, a gating coefficient generation sublayer, and a gating fusion sublayer.

[0169] The thermoelectric state fusion sublayer maps the hot-end state, cold-end state, voltage output state, and current output state to the same hidden space.

[0170] The electrical matching sublayer generates electrical matching characteristics based on the matching relationship between the output voltage and the energy storage voltage, the matching relationship between the output current and the allowable charging current, and the matching relationship between the load impedance and the output state of the thermoelectric generator unit.

[0171] The gating coefficient generation sublayer generates gating coefficients based on thermoelectric state fusion characteristics and electrical matching characteristics;

[0172] The gated fusion sublayer applies a gate coefficient to the dynamic characteristics of thermoelectric power generation.

[0173] With this structure, the model can predict harvestable energy by considering not only the magnitude and variation of the temperature difference, but also the receiving conditions on the energy storage side and the access conditions on the load side.

[0174] In this embodiment, the energy harvesting status characteristics are input into the partitioned energy mapping layer, and the energy output data of different regions and different thermoelectric power generation units are mapped to obtain the partitioned harvestable energy prediction results.

[0175] The partitioned energy mapping layer consists of a region mapping sublayer, a unit mapping sublayer, and an output prediction sublayer.

[0176] The region mapping sublayer generates region-level harvestable energy predictions based on the component region number;

[0177] The unit mapping sublayer generates predicted unit-level collectable energy values ​​according to the thermoelectric power generation unit number;

[0178] The output prediction sublayer simultaneously outputs the predicted energy value, predicted voltage value, and predicted current value for the next energy harvesting cycle.

[0179] The partitioning can collect energy prediction results and retain the area number, unit number, channel number and prediction time, which is convenient for subsequent generation of access priority and control instructions.

[0180] In this embodiment, the training data for the improved thermal inertia liquid time constant (LTC) energy harvesting model comes from the historical operation records of photovoltaic modules, the historical output records of thermoelectric generators, the charging and discharging records of energy storage devices, and the load connection records.

[0181] The training samples are divided according to the energy harvesting cycle, and each sample includes a historical observation window and a prediction control window.

[0182] The data within the historical observation window includes temperature, environment, thermoelectric output, energy storage, and load characteristics;

[0183] The labels in the predictive control window include actual collected energy, actual output voltage, actual output current, and channel access status.

[0184] The actual collected electrical energy is recorded by the energy metering module, the actual output voltage and actual output current are recorded by the thermoelectric generator output acquisition interface, and the channel access status is recorded by the energy harvesting controller.

[0185] After anomaly removal and time alignment, the training data is divided into training set, validation set and test set. The training set accounts for 70% of the historical samples, the validation set accounts for 15% and the test set accounts for 15%.

[0186] In this embodiment, the model training uses a weighted combination of prediction error, state constraint error, and gating stability error to form a loss function.

[0187] The prediction error is calculated based on the difference between the predicted harvestable energy, predicted voltage, predicted current and the actual recorded values.

[0188] The state constraint error is calculated based on the physical consistency between the hot-end state, the cold-end state, and the thermoelectric output state.

[0189] The gating stability error is calculated based on the degree of abrupt change in the gating coefficient within adjacent energy harvesting cycles.

[0190] The loss function can be expressed in the following simple form: Total loss = Energy prediction loss + Voltage and current prediction loss + State constraint loss + Gated stability loss.

[0191] The model is trained using the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, and 100 training epochs.

[0192] Training is stopped and model parameters are saved when the validation set loss decreases below a preset threshold for 10 consecutive rounds, or when the average error of the validation set prediction energy reaches a preset requirement.

[0193] refer to Figure 3 In this embodiment, S5 is the step of generating the partitioned energy harvesting strategy.

[0194] Based on the energy harvestable prediction results of each region, the predicted energy value, predicted voltage value, and predicted current value of each region in the next energy harvesting cycle are extracted to obtain the region prediction output data.

[0195] The partitioned prediction output data is arranged by region number, unit number, and channel number. Each record includes predicted harvestable energy, predicted output voltage, predicted output current, and predicted stability flags.

[0196] The system determines whether the thermoelectric generator unit is suitable for connection to the corresponding channel in the next energy harvesting cycle based on the predicted stability marker.

[0197] Based on the energy storage voltage, remaining capacity, and allowable charging current in the energy storage state, the access conditions are matched on the zone prediction output data to obtain a set of candidate thermoelectric generator units that meet the energy storage access conditions.

[0198] The system compares the predicted output voltage with the energy storage voltage, the predicted output current with the allowable charging current, and the predicted energy value with the minimum access energy threshold to select thermoelectric generator units that can stably supply energy to the energy storage side or the load side.

[0199] For thermoelectric generator units with low predicted energy, output voltage lower than the energy storage access conditions, or large output current fluctuations, the system will not include them in the candidate thermoelectric generator unit set for the time being.

[0200] Based on the predicted energy value, predicted voltage value, predicted current value, and channel connection status of the candidate thermoelectric generator units, the candidate thermoelectric generator units are prioritized to obtain the access order of the thermoelectric generator units.

[0201] The sorting process takes into account factors such as the amount of energy that can be collected, the stability of the output voltage, the stability of the output current, the cost of channel switching, and the receiving conditions on the energy storage side.

[0202] Thermoelectric generators with higher predicted energy, stable output, and lower switching costs are given higher priority.

[0203] Thermoelectric generators with low predicted energy or large output fluctuations are ranked with lower priority.

[0204] Based on the access sequence of the thermoelectric generator units and the current occupancy status of the energy collection channels, the energy collection channels are divided into main collection channels, maintenance collection channels, and channels to be accessed. Corresponding channels are matched for different thermoelectric generator units to obtain the collection channel matching results.

[0205] The main collection channel receives high predicted energy, has stable output, and is suitable for energy storage integration;

[0206] Thermoelectric generator unit that maintains moderate predicted energy reception, small output fluctuations, and is suitable for continuous low-power collection;

[0207] The waiting access channel receives thermoelectric generator units that do not yet meet the access conditions but may meet the access conditions in a later cycle.

[0208] The connection sequence of thermoelectric generators and the matching results of collection channels are combined to generate a zoned energy harvesting strategy.

[0209] refer to Figure 3 In this embodiment, S6 is the energy harvesting control result generation step.

[0210] Read the access sequence of thermoelectric generator units, the matching results of collection channels, and the access conditions of energy storage in the partitioned energy harvesting strategy to form energy harvesting control input data.

[0211] The energy harvesting control input data includes the thermoelectric generator unit number, area number, channel number, predicted energy value, predicted voltage value, predicted current value, energy storage access conditions, load status, and current switch status.

[0212] The energy harvesting controller generates corresponding circuit control commands based on the above data.

[0213] Based on the predicted voltage, predicted current and channel connection status in the energy harvesting control input data, the series connection, parallel connection or group connection of the thermoelectric generator unit is selected, and the series-parallel switching command is obtained.

[0214] When multiple thermoelectric generator units predict low voltage but have stable output current, the system selects a series connection method to increase the output voltage.

[0215] When the predicted voltages of multiple thermoelectric generators are close and their output currents are low, the system selects a parallel connection method to increase the output current.

[0216] When the output of thermoelectric generator units in different regions varies greatly, the system selects a group connection method to combine units with similar output characteristics into the same group to reduce mismatch losses.

[0217] Based on the energy storage voltage, remaining capacity, and allowable charging current in the energy harvesting control input data, the system selects the energy storage unit access state, energy storage unit disconnection state, or energy storage bypass state to obtain the energy storage access command.

[0218] When the energy storage voltage is lower than the preset charging upper limit and the predicted output meets the charging conditions, the system generates an energy storage unit access command.

[0219] When the energy storage voltage approaches the upper limit or the allowable charging current is insufficient, the system generates an energy storage unit disconnection command.

[0220] When there is an immediate power demand on the load side and the energy storage side is not suitable for connection at the moment, the system generates an energy storage bypass command, so that the output of the thermoelectric generator unit is connected to the load side through the load matching circuit.

[0221] Based on the load status, predicted energy value, and energy harvesting channel status in the energy harvesting control input data, the system adjusts the load matching impedance and converter duty cycle to obtain load matching parameters. These parameters include impedance adjustment level, converter duty cycle, channel switching delay, and low-power sustaining threshold. The system selects an appropriate impedance matching level based on the predicted output status of the thermoelectric generator (TEG) and adjusts the converter duty cycle to match the TEG output with the receiving conditions on the energy storage side or load side. S65 sends the series-parallel switching command, energy storage access command, and load matching parameters to the energy harvesting control circuit to obtain the energy harvesting control results.

[0222] refer to Figure 1 and Figure 3 In this embodiment, S7 is the feedback correction step.

[0223] The actual collected electrical energy, actual output voltage, actual output current, energy storage voltage change, and load connection status are collected after the energy harvesting control results are executed to obtain actual energy harvesting data.

[0224] Actual energy harvesting data is collected jointly by the energy metering module, the thermoelectric generator output acquisition interface, the energy storage controller, and the load detection interface, and stored according to the area number, unit number, channel number, and sampling time.

[0225] The actual energy harvesting data is matched with the predicted energy harvestable by region according to the region number, thermoelectric power unit number, channel number and sampling time to obtain the corresponding predicted data.

[0226] The system maps the predicted energy, voltage, and current values ​​of the same thermoelectric generator unit within the same energy harvesting cycle to the actual harvested energy, actual output voltage, and actual output current, forming the predicted-actual correspondence data.

[0227] Based on the actual corresponding data, the predicted energy deviation, predicted voltage deviation, and predicted current deviation are calculated to obtain energy harvesting deviation data. Energy harvesting deviation data reflects the difference between the model's predicted results and the actual energy harvesting results.

[0228] Based on the energy harvesting deviation data and the thermoelectric power generation characteristic data at the corresponding sampling time, the correction amounts of the thermal inertia weight parameter, the thermoelectric conversion gating coefficient, and the partitioned energy mapping layer parameter are calculated to obtain the model correction parameters.

[0229] For regions where the predicted energy is higher than the actual collected energy for multiple consecutive cycles, the system reduces the thermal inertia weighting parameter or thermoelectric conversion gating coefficient for that region.

[0230] For regions where the predicted energy is lower than the actual collected energy for several consecutive cycles, the system increases the corresponding partition energy mapping weight for that region; for regions with large voltage prediction deviations but small energy prediction deviations, the system prioritizes correcting the electrical matching related gating coefficients.

[0231] The model correction parameters, energy harvesting bias data, and control parameter corrections for the next energy harvesting cycle are combined to obtain feedback correction data. This feedback correction data is then incorporated into the next energy harvesting cycle for model prediction and control parameter generation.

[0232] In this embodiment, the photovoltaic thermoelectric power generation energy harvesting system can be deployed on the photovoltaic module side, the string combiner side, or the edge controller side.

[0233] Temperature sampling points and thermoelectric power generation units are arranged on the photovoltaic module side. The thermoelectric power generation units are connected to the energy harvesting channel through a switch matrix.

[0234] The string busbar side is equipped with energy storage interfaces, load interfaces, and energy metering modules.

[0235] The edge controller operates a data standardization processing program, a partition mapping program, an improved thermal inertia liquid time constant (LTC) energy harvesting model, and a control command generation program.

[0236] After receiving multi-source thermoelectric power generation data, the edge controller sequentially generates standardized thermoelectric power generation data, thermoelectric power generation characteristic data, zoned thermoelectric power generation characteristic sequence, zoned collectable energy prediction results, zoned energy collection strategy, and energy collection control results, and writes the feedback correction data into the next control cycle.

[0237] Compared with the thermoelectric energy harvesting method that uses a fixed threshold for judgment, this implementation method no longer determines whether the thermoelectric power generation unit should be connected based solely on the current temperature difference. Instead, it makes a comprehensive judgment by combining regional temperature difference, temperature difference change rate, regional temperature difference gradient, output voltage, output current, energy storage status, and load status.

[0238] Compared with the conventional recurrent neural network prediction method, the improved thermal inertia liquid time constant (LTC) energy harvesting model introduces thermal inertia time constant generation units during the state evolution process, enabling thermoelectric power generation units in different regions to form different evolution rates according to the heating, steady state, and cooling states.

[0239] Compared with the single maximum power point tracking control method, this implementation method can simultaneously generate parameters for partition access, series-parallel switching, energy storage access, and load matching, so that the prediction results can be transformed into actual energy harvesting control results.

[0240] Through the above treatments, this embodiment can improve the stability, continuity, and adjustability of the thermoelectric power generation energy harvesting process under conditions of photovoltaic module temperature fluctuations, environmental disturbances, and changes in energy storage load status.

[0241] Example 1: To verify the feasibility of this invention in practice, it was applied to the thermoelectric power generation energy harvesting process of a distributed photovoltaic system. The system comprises 12 photovoltaic modules, with 6 thermoelectric power generation units arranged on the back panel of each module, forming a total of 72 thermoelectric power generation units. Each photovoltaic module has 4 temperature sampling points on its front side and 4 on its back panel, and is equipped with interfaces for acquiring ambient temperature, irradiance, wind speed, energy storage voltage, load status, output voltage, and output current. The system continuously collects data on the temperature difference changes of the photovoltaic modules and the output status of the thermoelectric power generation units at 60-second sampling intervals and 15-minute energy harvesting control cycles.

[0242] In this embodiment, the raw data collected by the system includes the front temperature of the component, the back panel temperature, the ambient temperature, the irradiance, the wind speed, the output voltage and current of the thermoelectric generator unit, the energy storage voltage, and the load connection status. Due to short-term packet loss in some temperature sensors and spikes in some output current data, the system first aggregates the raw data according to the collection time and region number, and then performs timestamp correction and region number matching based on a unified sampling time. For data exceeding preset temperature, voltage, and current ranges, the system performs anomaly identification and rejection, and fills in missing data using adjacent sampling points and historical sampling records from the same region to obtain standardized thermoelectric generator data.

[0243] The system calculates the temperature difference between the front and back panels based on standardized thermoelectric power generation data, obtaining the module area temperature difference data. Statistical analysis shows that during the stable irradiation phase, the average front temperature of the photovoltaic module is 58.4℃, the average back panel temperature is 43.7℃, and the average regional temperature difference is 14.7℃. During the irradiation reduction phase, the average regional temperature difference decreases to 8.9℃, with some peripheral areas experiencing rapid temperature difference decay due to wind speed disturbances. The system further calculates the rate of temperature difference change and the regional temperature difference gradient, and jointly encodes the output voltage, output current, energy storage voltage, and load status of the thermoelectric power generation unit to generate thermoelectric power generation characteristic data.

[0244] Subsequently, the system partitions and maps the thermoelectric power generation characteristic data according to the row and column positions of photovoltaic modules, the installation positions of thermoelectric power generation units, and the connection relationships of energy harvesting channels. Each photovoltaic module is divided into a central region, an upper edge region, a lower edge region, a left edge region, and a right edge region. The system merges the temperature difference characteristics, thermoelectric output characteristics, energy storage status characteristics, and load status characteristics within the same region to form a partitioned thermoelectric power generation characteristic sequence. According to the partitioned statistics, the average temperature difference in the central region is 15.6℃, the average temperature difference in the edge regions is 11.8℃, the average output voltage of the thermoelectric power generation units in the central region is 1.86V, and the average output voltage in the edge regions is 1.34V.

[0245] The system inputs the regional thermoelectric power generation characteristic sequence into the improved thermal inertia liquid time constant (LTC) energy harvesting model. The thermal inertia time constant generation unit in the model generates liquid time constant parameters based on the temperature difference amplitude, temperature difference rate of change, regional temperature difference gradient, irradiance variation, and wind speed disturbance. The liquid time constant state evolution layer continuously updates the hot-end state, cold-end state, voltage output state, and current output state. The thermoelectric conversion constraint gating unit calculates the gating coefficient based on the output voltage, output current, energy storage voltage, and load state. The regional energy mapping layer outputs the predicted harvestable energy for different regions in the next energy harvesting cycle. The model predicts that the harvestable energy for a single cycle is 8.6 Wh in the central region, 5.1 Wh in the upper edge region, 5.4 Wh in the lower edge region, 4.8 Wh in the left edge region, and 4.6 Wh in the right edge region.

[0246] Based on the predicted energy harvestable by region, the system prioritizes the access of thermoelectric generators (TEGs). TEGs with high predicted energy, stable output voltage, and storage voltage that allows access are preferentially allocated to the main collection channel; TEGs with medium predicted energy and small output fluctuations are allocated to the maintenance collection channel; and TEGs with low predicted energy or unstable output are placed in the waiting access channel. The system adjusts the series-parallel switching mode of TEGs, the energy storage access status, and load matching parameters according to the regional energy harvesting strategy. In this embodiment, the central area uses a grouped series-parallel access mode, while the edge area uses a low-voltage parallel maintenance collection mode. The energy storage side connects to the main collection channel when the voltage is below 11.8V and switches to the maintenance collection channel when the voltage is above 13.2V.

[0247] In a continuous energy harvesting process, the fixed threshold harvesting scheme accumulated 128.4 Wh of energy, the conventional MPPT harvesting scheme accumulated 143.7 Wh of energy, and the initial control result of this invention accumulated 161.5 Wh of energy. After three feedback correction cycles, the system uses the actual harvested energy, actual output voltage, and actual output current to calculate the prediction deviation, and corrects the thermal inertia weight parameters, thermoelectric conversion gating coefficient, and partitioned energy mapping layer parameters. The accumulated harvested energy increased to 168.9 Wh, the number of invalid accesses due to low temperature difference decreased from 18 to 5, and the voltage fluctuation range on the energy storage side decreased from 1.42 V to 0.76 V. This indicates that this invention can improve the energy harvesting efficiency of photovoltaic thermoelectric power generation and enhance the stability of energy storage load matching.

[0248] Table 1 Comparison data of different energy harvesting schemes in the photovoltaic thermoelectric power generation scenario.

[0249] Cumulative collected electricity / Wh 128.4 143.7 161.5 168.9 Average single-cycle harvested energy / Wh 5.35 5.99 6.73 7.04 Central area temperature difference utilization rate / % 61.8 68.5 78.6 82.3 Edge area temperature difference utilization rate / % 43.2 49.7 61.4 65.8 Average output voltage / V 9.84 10.37 11.26 11.58 Average output current / A 0.54 0.61 0.69 0.72 Energy storage voltage fluctuation range / V 1.42 1.18 0.89 0.76 Invalid access attempts / times 18 13 7 5 Number of series-parallel switching operations / times 32 27 21 18 Percentage of low temperature difference unit connected / % 26.4 20.1 11.8 8.6 Average error in predicted energy / % — — 8.7 5.2 Channel matching success rate / % 71.5 78.9 88.6 92.4 Load matching stability rate / % 74.2 81.3 89.1 93.0 Energy harvesting efficiency / % 47.6 53.2 59.8 62.5

[0250] As shown in Table 1, the cumulative collected energy after feedback correction in this invention is 168.9Wh, which is significantly higher than the 128.4Wh of the fixed threshold collection scheme and the 143.7Wh of the conventional MPPT collection scheme. This indicates that this invention, through the regional thermoelectric power generation characteristic sequence and the improved thermal inertia liquid time constant (LTC) energy harvesting model, can more accurately identify the collectable energy of thermoelectric power generation units in different regions in the next energy harvesting cycle, avoiding energy loss caused by access control based solely on instantaneous temperature difference or instantaneous voltage.

[0251] In terms of temperature difference utilization, the temperature difference utilization rate in the central area after feedback correction in this invention reaches 82.3%, and the temperature difference utilization rate in the edge area reaches 65.8%, both higher than the two comparative schemes. This result mainly comes from the partition mapping and channel association processing. The system can prioritize matching the central high temperature difference area with the main collection channel, and allocate the edge medium and low temperature difference areas to the maintenance collection channel or the channel to be connected, so that different temperature difference areas participate in energy collection according to their actual output capacity, thereby improving the full utilization of the waste heat temperature difference of photovoltaic modules.

[0252] From the perspective of control stability, after feedback correction, the energy storage voltage fluctuation range of this invention is reduced to 0.76V, the number of invalid connections is reduced to 5, the number of series-parallel switching is reduced to 18, and the proportion of low temperature differential unit connections is reduced to 8.6%. These data indicate that the thermoelectric conversion constraint gating unit can combine output voltage, output current, energy storage voltage, and load status to constrain the connection value of thermoelectric power generation units, reducing the connection of inefficient units and frequent switching. The average error of predicted energy is reduced from 8.7% of the initial control result to 5.2%, indicating that the feedback correction data can effectively correct the thermal inertia weight parameter, thermoelectric conversion gating coefficient, and zoned energy mapping parameter. Overall, this invention outperforms the comparative scheme in terms of energy harvesting capacity, thermoelectric conversion utilization rate, channel matching success rate, and load matching stability rate, and can improve the continuous energy recovery capability and operational stability of photovoltaic thermoelectric power generation systems.

[0253] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A photovoltaic thermoelectric power harvesting method based on deep learning, characterized in that, Includes the following steps: S1. Obtain multi-source thermoelectric power generation data of photovoltaic modules during the energy collection cycle, and perform time alignment, anomaly removal, missing data filling and normalization on the multi-source thermoelectric power generation data to obtain standardized thermoelectric power generation data. S2. Based on standardized thermoelectric power generation data, extract component area temperature difference, temperature difference change rate, area temperature difference gradient, thermoelectric output status, energy storage status and load status to generate thermoelectric power generation characteristic data; S3. Based on the location of photovoltaic modules and the connection relationship of thermoelectric power generation units, perform partition mapping and channel association processing on the thermoelectric power generation characteristic data to obtain the partition thermoelectric power generation characteristic sequence. S4. Input the regional thermoelectric power generation characteristic sequence into the improved thermal inertia liquid time constant (LTC) energy harvesting model. After thermal inertia time constant generation, thermoelectric state evolution, thermoelectric conversion constraint gating, and regional energy mapping processing, the predicted results of regional harvestable energy are obtained. S5. Based on the predicted energy harvestable by zone and the energy storage status, prioritize the access of different thermoelectric power generation units and match the collection channels to generate a zoned energy harvesting strategy. S6. Adjust the series-parallel switching mode, energy storage access status and load matching parameters of the thermoelectric power generation unit according to the zoned energy harvesting strategy to obtain the energy harvesting control result; S7. Collect the actual collected electrical energy, actual output voltage and actual output current corresponding to the energy harvesting control results, calculate the deviation between the predicted energy harvestable data and the actual energy harvesting data of the zone, and obtain feedback correction data.

2. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S1 includes the following steps: S11. Acquire temperature data, environmental data, thermoelectric output data, energy storage data and load data of photovoltaic modules during the energy collection cycle, and collect them according to the collection time and area number to obtain the original thermoelectric power generation dataset; S12. Based on the sampling time of the energy harvesting cycle, perform timestamp correction, sampling interval unification, and area number matching on the original thermoelectric power generation dataset to obtain time-aligned thermoelectric power generation data. S13. Based on the preset temperature range, voltage range, current range and load access records, perform anomaly identification, anomaly removal and missing data filling on the time-aligned thermoelectric power generation data to obtain complete thermoelectric power generation data. S14. Perform dimension unification, normalization, and time-series processing on the temperature data, environmental data, electrical output data, energy storage data, and load data in the complete thermoelectric power generation data to obtain standardized thermoelectric power generation data.

3. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S2 includes the following steps: S21. Based on the front temperature data and back sheet temperature data in the standardized thermoelectric power generation data, calculate the front and back temperature difference according to the area number to obtain the module area temperature difference data. S22. Based on the changes in temperature difference data of the component area between continuous sampling times, calculate the temperature difference increase, temperature difference decrease, and temperature difference hold-up to obtain the temperature difference change rate data. S23. Based on the spatial adjacency relationship between different regions of the photovoltaic module, perform horizontal and vertical differential processing on the temperature difference data of the module region to obtain the regional temperature difference gradient data. S24. Based on the output voltage, output current, energy storage voltage, energy storage current and load access records in the standardized thermoelectric power generation data, perform joint encoding to obtain thermoelectric output status data, energy storage operation status data and load operation status data. S25. The component area temperature difference data, temperature difference change rate data, area temperature difference gradient data, thermoelectric output status, energy storage status and load status are spliced ​​together according to the sampling time to generate thermoelectric power generation characteristic data.

4. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S3 includes the following steps: S31. Based on the row and column positions of photovoltaic modules, the installation positions of thermoelectric power generation units, and the positions of temperature sampling points, establish a mapping relationship between module area numbers and thermoelectric power generation unit numbers to obtain an area unit mapping table. S32. Based on the series connection relationship, parallel connection relationship and switching switch connection relationship between the thermoelectric power generation units, establish the channel association relationship between the thermoelectric power generation units and the energy harvesting channel to obtain the unit channel association table. S33. Based on the regional unit mapping table, the thermoelectric power generation characteristic data is divided into regions. The temperature difference characteristics, thermoelectric output characteristics, energy storage status characteristics and load status characteristics in the same region are merged to obtain the regional thermoelectric power generation characteristics. S34. Based on the unit channel association table, channel marking and connection status marking are performed on the partitioned thermoelectric power generation characteristics to obtain a partitioned thermoelectric power generation characteristic sequence with region number, unit number and channel number.

5. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S4 includes the following steps: S41. Input the temperature difference amplitude, temperature difference rate of change, regional temperature difference gradient, irradiance change and wind speed disturbance in the regional thermoelectric power generation characteristic sequence into the thermal inertia time constant generation unit to generate liquid time constant parameters corresponding to different thermoelectric power generation units. S42. Input the partitioned thermoelectric power generation characteristic sequence and liquid time constant parameter into the liquid time constant state evolution layer, and continuously update the hot end state, cold end state, voltage output state and current output state to obtain the dynamic characteristics of thermoelectric power generation. S43. Input the dynamic characteristics of thermoelectric power generation, output voltage data, output current data, energy storage voltage data and load state data into the thermoelectric conversion constraint gating unit, perform gating coefficient calculation and feature fusion, and obtain the energy harvesting state characteristics. S44. Input the energy harvesting status characteristics into the partitioned energy mapping layer, and perform mapping processing on the energy output data of different regions and different thermoelectric power generation units to obtain the partitioned harvestable energy prediction results.

6. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 5, characterized in that, S41 includes the following steps: S411. Extract the temperature difference amplitude sequence, temperature difference rate of change sequence, and regional temperature difference gradient sequence of the same region at continuous sampling time from the regional thermoelectric power generation characteristic sequence to form the thermoelectric thermal inertia input sequence. S412. Extract the changes in irradiance intensity, wind speed, and ambient temperature from the regional thermoelectric power generation characteristic sequence to form an environmental disturbance input sequence; S413. Input the temperature difference thermal inertia input sequence and the environmental disturbance input sequence into the thermal inertia weight mapping layer, and assign weights to the heating response, steady state maintenance and cooling decay of different regions to obtain thermal inertia weight parameters. S414. Associate the thermal inertia weight parameter with the thermoelectric power generation unit number, region number, and channel number to generate liquid time constant parameters for each region, unit, and channel. S415. Configure the liquid time constant parameter to the state update term of the liquid time constant state evolution layer to obtain the state evolution input of the improved thermal inertial liquid time constant LTC energy harvesting model.

7. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 5, characterized in that, S43 includes the following steps: S431. Generate thermoelectric state fusion characteristics based on the hot end state characteristics, cold end state characteristics, voltage output state characteristics and current output state characteristics in the dynamic characteristics of thermoelectric power generation. S432. Based on the output voltage data, output current data, energy storage voltage data, and load status data, calculate the electrical matching characteristics between the thermoelectric generator unit and the energy storage side and the load side. S433. Input the thermoelectric state fusion characteristics and electrical matching characteristics into the gating mapping layer to calculate the thermoelectric conversion gating coefficient of each thermoelectric power generation unit under the corresponding energy harvesting channel. S434. The dynamic characteristics of thermoelectric power generation are weighted according to the thermoelectric conversion gate coefficient to obtain the gated dynamic characteristics of thermoelectric power generation. S435. The dynamic characteristics of the temperature difference power generation after gating are spliced ​​and mapped with the energy storage state characteristics and the load state characteristics to obtain the energy harvesting state characteristics.

8. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S5 includes the following steps: S51. Based on the predicted energy value, predicted voltage value and predicted current value of different regions in the next energy collection cycle, extract the predicted energy value, predicted voltage value and predicted current value of different regions according to the predicted energy of the region, and obtain the region prediction output data. S52. Based on the energy storage voltage, remaining capacity and allowable charging current in the energy storage state, perform access condition matching on the partition prediction output data to obtain a set of candidate thermoelectric generator units that meet the energy storage access conditions. S53. Based on the predicted energy value, predicted voltage value, predicted current value and channel connection status in the candidate thermoelectric power generation unit set, prioritize the candidate thermoelectric power generation units to obtain the access order of the thermoelectric power generation units. S54. Based on the access sequence of the thermoelectric power generation units and the current occupancy status of the energy collection channels, the energy collection channels are divided into main collection channels, maintenance collection channels and channels to be accessed, and corresponding channels are matched for different thermoelectric power generation units to obtain the collection channel matching results. S55. Combine the connection sequence of the thermoelectric generator units and the matching results of the collection channels to generate a zoned energy collection strategy.

9. The photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S6 includes the following steps: S61. Read the access sequence of thermoelectric power generation units, the matching result of collection channels and the access conditions of energy storage in the partitioned energy harvesting strategy to form energy harvesting control input data; S62. Based on the predicted voltage value, predicted current value and channel connection status in the energy harvesting control input data, select the series connection method, parallel connection method or group connection method of the thermoelectric power generation unit, and obtain the series-parallel switching command. S63. Based on the energy storage voltage, remaining capacity and allowable charging current in the energy harvesting control input data, select the energy storage unit access state, energy storage unit disconnect state or energy storage bypass state, and obtain the energy storage access command. S64. Based on the load status, predicted energy value, and energy harvesting channel status in the energy harvesting control input data, adjust the load matching impedance and converter duty cycle to obtain the load matching parameters. S65. Send the series-parallel switching command, energy storage access command and load matching parameters to the energy harvesting control circuit to obtain the energy harvesting control result.

10. A photovoltaic thermoelectric power harvesting method based on deep learning according to claim 1, characterized in that, S7 includes the following steps: S71. Collect the actual collected electrical energy, actual output voltage, actual output current, energy storage voltage change and load connection status after the energy harvesting control results are executed, and obtain the actual energy harvesting data; S72. Match the actual energy collection data with the predicted energy collection results of the zone according to the zone number, thermoelectric power generation unit number, channel number and sampling time to obtain the predicted actual corresponding data; S73. Calculate the predicted energy deviation, predicted voltage deviation, and predicted current deviation based on the actual predicted data to obtain energy harvesting deviation data; S74. Based on the energy harvesting deviation data and the temperature difference power generation characteristic data at the corresponding sampling time, calculate the correction amount of the thermal inertia weight parameter, the correction amount of the thermoelectric conversion gate coefficient, and the correction amount of the partitioned energy mapping layer parameter to obtain the model correction parameters. S75. Combine the model correction parameters, energy harvesting deviation data, and control parameter correction amounts for the next energy harvesting cycle to obtain feedback correction data.

Images