New energy power station operation and maintenance method and system based on artificial intelligence

Abstract

The invention discloses a new energy power station operation and maintenance method and system based on artificial intelligence, and relates to the field of photovoltaic operation and maintenance management, and the method comprises the steps: collecting the output power and module temperature of each group string at a combiner box side, calculating the mean value and variance of power, generating node features through combining with the temperature, and obtaining the environment vector of a whole station; constructing a graph structure by taking the group string as a node and the electrical connection as an edge, and inputting the graph structure and the environment vector into a graph attention network to obtain a mismatch score of each group string; and constructing a decision model by taking the mismatch score and the number of optimizers as input, solving an optimal compensation scheme and issuing a compensation instruction to a corresponding DC / DC optimizer to realize dynamic mismatch compensation. Therefore, the self-adaptive capability of the photovoltaic power station in a complex environment is improved, and the operation and maintenance intelligent level is improved.

Description

Technical Field

[0001] This application relates to the technical field of photovoltaic operation and maintenance management, and particularly to an operation and maintenance method and system for new energy power stations based on artificial intelligence. Background Art

[0002] With the large-scale development of photovoltaic new energy power stations, the booster station plays a key role in collecting the DC power from each string, boosting the voltage and connecting to the grid. However, due to complex factors such as different component aging rates, local shading, surface contamination, and non-uniform light distribution among the strings, a mismatch effect will occur on the DC side, which will further lead to a decrease in the power generation efficiency of the entire power station. At present, passive bypass diode technology or centralized DC / DC optimizers are mainly used to solve the string mismatch problem.

[0003] In the passive bypass diode technology, a bypass diode is connected in parallel to each photovoltaic component. When the output power of a component is significantly lower than that of other components in the same string, the bypass diode conducts, cutting off the influence of this component on the entire series current. However, bypass conduction is usually accompanied by additional power loss, and it only isolates the faulty component without reconstructing the string IV characteristic curve, and cannot eliminate the influence of mismatch on the system-level MPPT (Maximum Power Point Tracking).

[0004] In the booster station, an independent DC / DC converter is configured for each string to independently adjust the voltage-current characteristics of each string for string-level MPPT optimization to achieve an output close to the maximum power point. Although each string-level DC / DC optimizer can independently adjust the voltage-current curve according to the electrical characteristics, its compensation decision is still only based on the information in the dimension of electrical measurement, ignoring the on-site environment and operation and maintenance status. For example, in the case of sudden changes in light intensity, component surface contamination or shading, simply relying on the current-voltage curve cannot accurately identify the root cause of the mismatch, resulting in problems such as target deviation or response lag in the compensation action. Summary of the Invention

[0005] This application provides an operation and maintenance method, system, storage medium, computer program product and electronic device for new energy power stations based on artificial intelligence, so as to at least solve the problems of inaccurate detection, response lag and unreasonable resource allocation of photovoltaic string mismatch in the current related technologies.

[0006] In the first aspect, the embodiment of the present application provides an artificial intelligence-based new energy power station operation and maintenance method, including: collecting the output power and module temperature values of each photovoltaic string connected to the combiner box on the combiner box side, extracting the output power mean and output power variance of each photovoltaic string and generating a corresponding initial feature vector in combination with the module temperature value, and generating a global environment vector according to the environmental irradiance of the entire station and the environmental temperature of the entire station; constructing an undirected graph structure with each photovoltaic string as a node and the electrical connection relationship between each photovoltaic string as an edge; the node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string; the undirected graph structure and the global environment vector are input into a graph attention network, and L After the iterative fusion of the multi-head attention layers, L The output features of the layer are linearly mapped and ReLU activated to obtain the string mismatch score of each photovoltaic string; L represents the total number of graph attention layers in the graph attention network; takes the string mismatch scores of each photovoltaic string and the number of DC / DC optimizers as input, constructs a compensation decision model; solves the compensation decision model to obtain the optimal compensation solution ;in, and Respectively represent The optimal target power and activation flag decision value corresponding to each PV string; according to the optimal compensation scheme Generate compensation instructions, and send the generated compensation instructions to the DC / DC optimizer corresponding to each photovoltaic string to implement dynamic compensation for string mismatch.

[0007] In the second aspect, the embodiment of the present application provides an artificial intelligence-based new energy power station operation and maintenance system, including: a multi-source feature acquisition unit, which is used to collect the output power and module temperature values of each photovoltaic string connected to the combiner box on the combiner box side, extract the output power mean and output power variance of each photovoltaic string and generate a corresponding initial feature vector in combination with the module temperature value, and generate a global environment vector according to the environmental irradiance of the whole station and the environmental temperature of the whole station; a graph structure construction unit, which is used to construct an undirected graph structure with each photovoltaic string as a node and the electrical connection relationship between each photovoltaic string as an edge; the node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string; a node feature update unit, which is used to input the undirected graph structure and the global environment vector into a graph attention network, and then generate a global environment vector according to the environmental irradiance of the whole station and the environmental temperature of the whole station; L After the iterative fusion of the multi-head attention layers, L The output features of the layer are linearly mapped and ReLU activated to obtain the string mismatch score of each photovoltaic string: Lrepresents the total number of graph attention layers in the graph attention network; a compensation decision modeling unit, used to construct a compensation decision model with the string mismatch score of each photovoltaic string and the number of DC / DC optimizers as input; a compensation decision solving unit, used to solve the compensation decision model to obtain the optimal compensation solution ;in, and Respectively represent The optimal target power and activation flag decision value corresponding to each photovoltaic string; the string mismatch dynamic compensation unit is used to calculate the optimal compensation scheme according to the optimal target power and activation flag decision value corresponding to each photovoltaic string; the string mismatch dynamic compensation unit is used to calculate ... Generate compensation instructions, and send the generated compensation instructions to the DC / DC optimizer corresponding to each photovoltaic string to implement dynamic compensation for string mismatch.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the steps of the artificial intelligence-based new energy power station operation and maintenance method of any embodiment of the present application.

[0009] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the artificial intelligence-based new energy power station operation and maintenance method of any embodiment of the present application are implemented.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the artificial intelligence-based new energy power station operation and maintenance method of any embodiment of the present application.

[0011] The artificial intelligence-based new energy power station operation and maintenance method and system provided by this application can produce at least the following technical effects: (1) Through multi-dimensional information fusion and intelligent optimization decision-making, a complete dynamic closed loop from mismatch signal capture to compensation execution is established. The power statistical characteristics and temperature data of each connected photovoltaic string are collected on the combiner box side, and combined with the whole station environmental monitoring, the graph structure constructed by electrical connections and the multi-head attention network are used to deeply mine the relationship between strings. Then, through optimal scheduling under resource constraints, real-time and accurate identification and rapid dynamic compensation of string mismatch are achieved, which improves the system-level maximum power tracking efficiency and operation and maintenance automation level as a whole.

[0012] (2) By collecting the output power mean, variance and module temperature information of each photovoltaic string on the junction box side, and combining the environmental irradiance and ambient temperature of the whole station to construct a global environmental vector, the connection relationship of each string is characterized by an undirected graph structure, and the above features are input into the graph attention network for multi-level information fusion processing, thereby generating a robust mismatch degree score, realizing global perception and root cause location of the mismatch degree of the string. Compared with the current related technology that is only based on the voltage-current characteristics of the detection method, this technical solution can effectively distinguish different types of mismatch problems such as component aging, shading, pollution, and uneven illumination, significantly improving the accuracy and pertinence of mismatch detection.

[0013] (3) A compensation decision model is constructed based on the string mismatch score and DC / DC optimizer resource constraints. The optimal compensation strategy is constructed by minimizing the system power loss and optimizing the resource activation cost. The mismatch state is quickly identified and the optimal compensation strategy is generated. The string characteristics are dynamically adjusted by issuing compensation instructions, achieving the global optimal allocation of optimizer resources. This effectively optimizes the resource waste or unreasonable resource allocation of uncompensated mismatched strings that exist in the current related technologies, significantly improves the utilization efficiency of compensation resources, and ensures the maximization of overall power generation efficiency.

[0014] Through this technical solution, the string mismatch of the photovoltaic strings on the junction box side is accurately detected and dynamically compensated, effectively reducing the power generation loss caused by string mismatch and improving the overall power generation efficiency of the photovoltaic power station. At the same time, through the reasonable allocation and control of the resources of each optimizer, the energy waste caused by redundant compensation or the situation where the mismatched strings are not compensated is avoided, the mismatch state can be quickly identified and the optimal compensation strategy can be generated. By issuing compensation instructions, the dynamic adjustment of the string characteristics can be achieved, which significantly improves the adaptive ability of the photovoltaic power station in complex environments, reduces the frequency of manual intervention, and improves the intelligent level of operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A flowchart showing an example of an artificial intelligence-based new energy power station operation and maintenance method according to an embodiment of the present application is shown; Figure 2 A schematic diagram of a topological structure showing an example of a layout of power generation equipment of a new energy power station; Figure 3Schematic diagram showing an example of the current-voltage characteristic curve of a photovoltaic string; Figure 4 Schematic diagram of an operation flow showing an example of constructing an undirected graph structure according to an embodiment of the present application; Figure 5 Schematic diagram of an operation flow showing an example of iteratively solving a compensation decision model by the three-stage alternating direction method of multipliers according to an embodiment of the present application; Figure 6 Schematic block diagram showing an example of a new energy power station operation and maintenance system based on artificial intelligence according to an embodiment of the present application; Figure 7 Schematic diagram of the hardware structure of an electronic device for executing a new energy power station operation and maintenance method based on artificial intelligence provided in another embodiment of the present application. Detailed implementation manners

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0018] Figure 1 Schematic flow chart showing an example of a new energy power station operation and maintenance method based on artificial intelligence according to an embodiment of the present application.

[0019] Regarding the execution subject of the method in the embodiments of the present application, it can be any controller or processor with computing or processing capabilities. Specifically, it can be implemented by a power station operation and maintenance platform or edge ECUs (Energy Control Units) distributed in each busbar box. Through multi-dimensional intelligent feature extraction, graph network correlation analysis, and resource optimization scheduling, not only high-precision identification and rapid compensation of photovoltaic string mismatch are achieved, but also the maximum power tracking efficiency is significantly improved, the manual operation and maintenance cost is reduced, and the reliability and intelligent level of the booster station operation are enhanced.

[0020] Figure 2 Schematic diagram of the topological structure showing an example of the layout of the power generation equipment in a new energy power station.

[0021] Such as Figure 2As shown, each photovoltaic string is first connected in parallel to the corresponding combiner box, and then the inverter is centrally inverted and boosted before being connected to the mains power grid. Specifically, photovoltaic strings 211, 213, and 215 are connected in parallel to combiner box 221 through their respective DC / DC optimizers 21a, 21c, and 21e; photovoltaic strings 212, 214, and 216 are connected in parallel to combiner box 222 through DC / DC optimizers 21b, 21d, and 21f. Combiner boxes 221 and 222 will collect the DC currents of their respective parallel strings and send them to inverter 240, which will be connected to the grid for power generation after being inverted and boosted.

[0022] Here, in order to monitor the power generation mismatch status between each string in real time and perform dynamic compensation when necessary, edge ECUs are deployed on each combiner box side, namely edge ECU 231 and edge ECU 232. Specifically, each edge ECU will continuously collect multi-source data of the multiple photovoltaic strings it is connected to, and realize the identification of the string mismatch status and power compensation decision between different photovoltaic strings in the combiner box through edge computing, and realize dynamic compensation by sending power compensation instructions to the corresponding DC / DC optimizer through precise adjustment of the power of each string.

[0023] In some examples, it may be integrated and configured in an electronic device or terminal by means of software, hardware, or a combination of software and hardware, and the type of the terminal or electronic device may be diverse, such as a mobile phone, a tablet computer, or a desktop computer, etc.

[0024] like Figure 1 As shown, in step S110, multi-source sensor data of each photovoltaic string connected to the combiner box is collected on the combiner box side, and the initial feature vector of each photovoltaic string is extracted, and a global environment vector is generated according to the global environmental irradiance and the global environmental temperature.

[0025] Specifically, the output power and module temperature values of each photovoltaic string connected to the combiner box are collected on the combiner box side, the output power mean and output power variance of each photovoltaic string are extracted, and the corresponding initial feature vector is generated in combination with the module temperature value.

[0026] It should be noted that the fundamental reason for the mismatch of photovoltaic strings is that different photovoltaic strings under the same junction box produce asynchronous fluctuations in output power due to factors such as shadow obstruction, component aging, and thermal attenuation differences. Therefore, relying solely on a single source of electrical parameters (such as voltage or current) cannot accurately reflect the true degree of mismatch, and it is also difficult to distinguish between environmental fluctuations and component abnormalities. In an embodiment of the present application, by collecting multi-source sensor data, synchronously collecting and preprocessing power, module temperature, and environmental meteorological indicators, the "external disturbance" and "internal state" are quantified separately, so that the extracted structure can not only reflect the performance of the string itself but also eliminate the initial characteristics of environmental interference.

[0027] In some embodiments, by installing power sensors (for measuring output power) and module surface temperature sensors for each connected PV string on each combiner box side, and configuring corresponding power sampling frequencies and temperature sampling frequencies respectively. Exemplarily, by connecting a set of shunt resistors in series on the DC bus connecting the combiner box and each PV string to measure the real-time output power of each PV string. Additionally, thermocouples (such as PT100) are fixedly mounted on the back or solder strip area of the representative module in each string to measure the surface temperature in real time. Furthermore, at the central position of the power station or deployed by partition, a two-axis PV-specific irradiance meter is installed to measure the irradiance of the whole station and the ambient temperature of the whole station is measured by the installed ambient temperature and humidity sensors .

[0028] All channels uniformly divide the time window in steps of 100 ms. During the sampling process, the latest data in each time slot is marked with the same timestamp and stored in a circular queue in the edge ECU to ensure that the data of each subsequent channel can correspond one by one

[0029] In terms of initial feature calculation, for the output power mean , it can be calculated within each 30-second time window as follows: , Equation (1) For the time window length (such as corresponding to 5 minutes), the arithmetic mean is calculated as: , Equation (2) The output power variance is calculated within the same time window as: , Equation (3) The power fluctuation amplitude is directly reflected by this variance value, providing quantitative support for mismatch fluctuations

[0030] The module temperature value can take the latest smoothed value of the temperature signal within this time window as the representative temperature input without further statistics, ensuring sensitivity to hot spot mutations

[0031] Furthermore, the three are concatenated to obtain the initial feature vector .

[0032] In addition, the irradiance of the whole station and the ambient temperature of the whole station are subjected to time series exponential smoothing, and the latest smoothed values can be used as the irradiance of the whole station environment and the ambient temperature of the whole station, and the two are fused to serve as the global environment vector

[0033] In this embodiment, a time window sampling and filtering strategy at the 30 - second level is adopted, and through smoothing and synchronization processing, power and temperature jumps caused by sudden weather changes (such as cloud shadows and dust) are quickly filtered, ensuring that subsequent fusion focuses on real mismatch signals.

[0034] Figure 3 The effect schematic diagram shows an example of the current - voltage characteristic curve of a photovoltaic string.

[0035] As Figure 3 shown, the two light yellow and orange straight lines respectively represent the current - voltage characteristic curves (i.e., I - V curves) of the same photovoltaic string at 25°C and 45°C. It can be seen that as the module temperature increases, the open - circuit voltage decreases significantly (the right - hand end of the curve moves down), while the short - circuit current increases slightly (the left - hand end of the curve moves up). However, the voltage - current product at the maximum power point (at the dot) is lower than the performance at low temperature overall. This characteristic shift is a direct reflection of the increased carrier recombination and voltage attenuation of photovoltaic materials at high temperatures.

[0036] In the embodiment of the present application, considering the sensitive influence of temperature on the I - V curve and the position of the maximum power point, the module temperature is taken as an important dimension for mismatch evaluation. Compared with only relying on output power or electrical quantities, fusing temperature characteristics can effectively improve the accuracy of mismatch identification. In addition, when a section of a string has hot spots due to reasons such as shading, local damage, or poor solder tape contact, the local module temperature will be significantly higher than that of the surrounding modules, resulting in limited current output of this string, while adjacent strings may maintain normal output due to lower temperatures. Therefore, through the temperature difference, not only can internal or surface abnormalities of the components be quickly reflected, but also a current "restraint" effect will be caused within the same junction box, thus exacerbating the mismatch between strings. Incorporating the module temperature into the features helps the model to prioritize focusing on those nodes with mismatches caused by temperature abnormalities during fusion, supplementing internal fault signals that are difficult to capture only by electrical quantities from a physical level.

[0037] In step S120, an undirected graph structure is constructed with each photovoltaic string as a node and the electrical connection relationship between each photovoltaic string as an edge. The node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string.

[0038] It should be noted that the electrical connection relationship between strings determines the coupling mode of their output power, that is, for strings connected in parallel to the same junction box, their power changes often have a correlation. Specifically, taking the wiring relationship between each string and its junction box as the physical connection basis, if the i th string and the j th string are in the same junction box, then an undirected edge is drawn between them.

[0039] It should be understood that the initialization method of the edge weights of each edge in the undirected graph structure can be diversified, for example, initialized according to a preset value. In addition, the edge weights can also be set according to the correlation degree of the statistical results of the data trend.

[0040] Figure 4 FIG. shows an operation flowchart of an example constructed according to the undirected graph structure of an embodiment of the present application.

[0041] As Figure 4 shown, in step S410, in the node set corresponding to each photovoltaic string, an undirected edge is constructed for any pair of nodes to form a complete graph between the complete node sets. In step S420, for each undirected edge in the complete graph, the Pearson correlation coefficient is calculated based on the output power time series data of the two photovoltaic strings corresponding to the undirected edge within a preset time window, and the absolute value of the Pearson correlation coefficient is used as the initial weight of the undirected edge.

[0042] In some embodiments, at the edge ECU or the busbar box side controller, the output power of each string is synchronously collected at a fixed rate , and the time window length is set (for example, 5 minutes). Before each compensation scheduling, the power time series data within the most recent is collected.

[0043] For each edge in the complete graph , the power sequences of the two strings at the same time point are respectively taken , , and according to the Pearson correlation coefficient calculation formula: , formula (4) calculate the Pearson correlation coefficient , and take the absolute value: , formula (5) The initial weight of each edge is stored in the corresponding position of the adjacency matrix.

[0044] It should be noted that the Pearson correlation can reflect the synchronous fluctuation degree of the output power of the two strings under the same meteorological and load conditions. A high correlation indicates that they may be jointly affected by local shading or irradiation gradient, and the absolute value can effectively reflect the corresponding correlation strength, so as to dynamically capture short-term dynamic effects such as shadow movement and cloud mass change.

[0045] In step S430, for each node in the complete graph, the initial weights of all the connected edges of the node are normalized row by row so that the sum of the normalized weights of all the outgoing edges of each node is 1 to update the edge weights of the connected edges.

[0046] In some embodiments, for each node , take the set of initial weights of all its connected edges , and calculate their sum , Equation (6) Set the final weight of each edge to: , Equation (7) Furthermore, save the normalized weighted adjacency matrix in the edge ECU. Through the normalization operation, it is avoided that the power amplitude directly affects the attention, ensuring that all nodes fairly distribute attention within their neighborhoods.

[0047] It should be noted that if fixed initial edge weights are used (such as uniformly set to 0 or other constants), the contributions of different neighbors to the mismatch evaluation cannot be distinguished, resulting in the neglect of neighborhood differences and making the aggregated information have a large noise content. In contrast, in the embodiments of the present application, the initial edge weights are assigned based on the absolute value of the Pearson correlation coefficient of the power time series. Through dynamic correlation weighting, multi-source information of the environment and component states is supplemented at the physical level, and at the same time, a prior distribution in mathematical trend statistics is provided, which helps to automatically focus on those strings with highly synchronized or synchronously reversed operation fluctuations during the node aggregation stage, so as to truly reflect the collaborative effects of disturbances such as dynamic electrical coupling and shadow occlusion.

[0048] In step S130, the undirected graph structure and the global environment vector are input into the graph attention network for node feature aggregation, so as to obtain the string mismatch scores of each path of photovoltaic strings.

[0049] It should be noted that the whole-station environmental parameters can have an overall trend impact on all photovoltaic strings. Using the graph structure to model and using the electrical topology to associate each string, and then aggregating the node features through the graph neural network, so that each string can "see" its own state with adjacent strings and also "perceive" the global environment synchronously, thereby obtaining a mismatch score that comprehensively considers local and global factors.

[0050] More specifically, within the entire power station, irradiance and ambient temperature are the most critical external factors determining the output power of photovoltaic modules, and their impacts on the performance of each path of strings are unified and synchronous. Using the whole-station environmental irradiance and the whole-station environmental temperature to form the global environment vector can distinguish the overall power fluctuations caused by weather changes (such as rapid cloud movement or sunlight intensity fluctuations) from the local mismatch behaviors of single-path strings. Therefore, injecting this global vector as context information into the graph attention network enables the model to automatically "deduct" the equilibrium changes caused by environmental fluctuations when calculating the attention weights between nodes, thereby focusing on the true component differences and electrical coupling anomalies.

[0051] Specifically, after iterative fusion of multi-head attention through layers, linear mapping and ReLU activation are performed on the output features of the L -th layer to obtain the string mismatch score of each photovoltaic string: L , Equation (8) In the formula, represents the string mismatch score of the -th photovoltaic string, represents the trainable weight vector of linear mapping, represents the vector transpose operation, represents the bias term of linear mapping, L represents the total number of graph attention layers in the graph attention network, represents the -th photovoltaic string after passing through L layers of graph attention layer fusion and output feature vector.

[0052] During the attention fusion process, each node inputs its initial feature vector and the global environmental vector not included in the node features into the attention scoring function in the same frame. The multi-head attention mechanism is used for branched parallel calculation, which enables the model to learn multiple association patterns at different "focus points". Through the feature iteration of multiple attention layers, the weighted features of neighbor nodes are fused with its own features layer by layer, gradually expanding the receptive field. Through linear mapping and ReLU activation, the fused features are uniformly mapped to a non-negative scalar score to measure the current mismatch degree of the string with the whole.

[0053] Through the learnable attention mechanism of the Graph Attention Network (GAT), the edge weights of each edge are dynamically adjusted, enabling the model to adaptively focus on the neighbor nodes that contribute the most to the string mismatch score. At the same time, the attention intensity is adjusted in combination with the global environmental context, so that the differential associations of different strings in terms of mismatch degree can be effectively reflected in the node aggregation process.

[0054] Specifically, in the string mismatch scenario, if the output fluctuations of a certain string are significantly different from those of some neighbors (i.e., high difference), GAT will automatically amplify the edge weights of these significantly different edges through the attention mechanism. Conversely, it will weaken the irrelevant or highly similar edge weights, so that each node finally aggregates the information that best represents its "local mismatch degree" and "neighborhood contrast relationship". After multiple iterative fusions, the node forms a high-order representation that combines local features and neighborhood differences. Finally, through linear mapping and non-linear activation, it is compressed into a non-negative scalar, that is, the string mismatch score, which is used to quantify the mismatch score of "the output of this string is different from the expectation / neighbors under the current environment and neighborhood contrast". ​

[0055] In step S140, a compensation decision model is constructed with the string mismatch score of each photovoltaic string and the number of DC / DC optimizers as inputs.

[0056] It should be noted that the string mismatch score represents the degree of deviation of each string from the maximum output capacity compared to neighboring strings. For strings with severe mismatch or high contribution, power optimization references need to be provided for the corresponding DC / DC optimizers for mismatch compensation. In addition, over-compensation should be avoided to prevent system oscillations and maintain low resource consumption for the optimizers.

[0057] It should be understood that the compensation decision model can be diverse, such as a priority greedy allocation model, a particle swarm meta-heuristic model, etc.

[0058] Exemplarily, in the priority greedy allocation model, instead of solving the global optimal binary mixed-integer programming, the marginal benefit of unit compensation for each string is directly calculated using the mismatch score and the compensation strength range .

[0059] , Equation (9) In other words, when enabling an optimizer for the -th string and fully compensating, the maximum benefit that can be obtained is . Through the prior sorting of the marginal benefit , the DC / DC optimizer corresponding to the string with the highest support benefit can be preferentially enabled, thereby minimizing the overall mismatch to the greatest extent. However, the greedy strategy is an approximate approach and cannot guarantee global optimality.

[0060] Preferably, in some examples of the embodiments of the present application, by constructing an optimization model based on the mismatch score and compensation cost, and under the premise of satisfying the power constraint and the available number constraint of the optimizer, a set of optimal power settings and compensator enabling schemes are solved to achieve dynamic compensation decision-making.

[0061] , Equation (10) where , .

[0062] In the formula, N represents the total number of photovoltaic strings, represents the number of DC / DC optimizers, is the rated power of the -th photovoltaic string, is the target power of the -th photovoltaic string after compensation, is the optimizer enabling penalty coefficient; is the enabling flag of the DC / DC optimizer corresponding to the th string of the PV array, indicating that the DC / DC optimizer is enabled for the th string of the PV array, indicating that the DC / DC optimizer is not enabled for the th string of the PV array.

[0063] In the compensation decision model, the compensation power of each string and its mismatch score form a product term , which intuitively reflects "the mismatch loss that can be reduced by compensating 1 watt"; the additional penalty term quantifies the resource cost of enabling each DC / DC optimizer as a unified "cost" for constraint. On the premise of ensuring and the total number of enabled ones does not exceed the number of available optimizers , this model aims to minimize the sum of the total loss and the total resource cost, and directly derives the optimal trade-off of "which strings should be compensated and how many watts should be compensated". Thus, from the physical level, it takes into account the dual constraints of the power loss distribution of each string and the DC / DC resource consumption of the junction box, no longer relying on manual rules or greedy algorithms, but automatically outputting the global optimal "start-stop + power" decision by the model, ensuring that each compensation can achieve the maximum mismatch reduction effect with the least amount of hardware activation, significantly enhancing the reliability of the operation and maintenance strategy.

[0064] In step S150, solve the compensation decision model to obtain the optimal compensation plan.

[0065] Here, the optimal compensation plan can be expressed as , and respectively represent the optimal target power and the enabling flag decision value corresponding to the th string of the PV array.

[0066] The solution method for the compensation decision model can be diversified. For example, a greedy heuristic algorithm, a genetic algorithm, or an ADMM (Alternating Direction Method of Multipliers) method can be used, etc., which will not be restricted here.

[0067] For example, in the greedy heuristic algorithm, based on the ratio of "unit compensation benefit" to the enabling cost, all strings are sorted from high to low, and the optimizers are allocated in turn, and then for the enabled strings Perform a unified projection or local optimization. Although this strategy cannot guarantee global optimality, it is simple to implement and has extremely low computational overhead, making it suitable for situations with extremely high real-time requirements or a large number of strings.

[0068] In step S160, according to the optimal compensation scheme Generate compensation instructions and send the generated compensation instructions to the DC / DC optimizers corresponding to each photovoltaic string to implement dynamic compensation for string mismatch.

[0069] Exemplarily, the edge ECU sends the compensation instructions to the DC / DC optimizers corresponding to each string through an industrial bus, so that the DC / DC optimizers know to Adjust the output power and regulate it through the internal boost / buck switch to complete the compensation execution.

[0070] Through this embodiment, by introducing multi-source information fusion and graph attention network, first filter out environmental interference at the time series level, organically integrate power, module temperature and meteorological parameters, and accurately extract the component mismatch signal physically; then on the fully connected graph constructed by the electrical topology, adaptively allocate the contribution of each neighbor to the mismatch score, and accurately quantify the mismatch degree of each string; finally, combine the optimization model of the score and the optimizer resource constraint to achieve the global optimal scheduling of the power compensation demand and the optimizer activation cost, and send it through a low-latency bus. Thus, accurate dynamic mismatch compensation is achieved, promoting the intelligence and efficiency of photovoltaic power station operation and maintenance.

[0071] Regarding the description of the graph attention network in the embodiment of the present application, it can be fine-tuned on the framework of the general GAT to fit the actual needs of photovoltaic string mismatch assessment.

[0072] Specifically, in the offline training stage of the graph attention network, a large-scale time series graph snapshot dataset can be constructed. For each sampling moment, use the current irradiance, ambient temperature of the whole station, and the mean, variance and module temperature of the output power of each string as node features, and calculate the normalized weighted adjacency matrix according to the complete graph structure to form a complete graph structure. At the same time, extract the real mismatch index at the corresponding moment from the historical SCADA (Supervisory Control And Data Acquisition) records - for example, the string mismatch magnitude calculated based on the bypass diode trigger log or the unbalanced current on the inverter side, as the label for supervised learning. To ensure the robustness of the model to different weather, seasons and fault types, it is necessary to ensure that the training set covers various typical mismatch scenarios such as sunny day shadows, local hot spots, and component degradation, and perform balanced sampling or weighting on the samples, so that the model neither overemphasizes common situations nor fails to recognize rare but critical extreme mismatches.

[0073] In the construction of the specific graph attention model, a multi-layer GAT architecture is selected, and 2-3 layers can be configured, with 4-8 attention heads in each layer, and 0.2-0.5 Dropout (random inactivation) is added after each layer to prevent overfitting, and LayerNorm (layer normalization) is used to ensure training stability. The loss function is mainly based on MSE (mean square error), and high mismatch errors can be additionally weighted to highlight key nodes - that is, a higher gradient ratio is given when the label is large.

[0074] Finally, the verified model is exported into a lightweight inference format and deployed on the edge ECU for low-latency inference. It is also regularly (e.g. monthly) retrained offline using the latest collected data to cope with distribution drift caused by component aging or site topology changes. Combined with the electrical and environmental characteristics of the PV strings, the final mismatch scoring model is ensured to be superior to traditional static MPPT or bypass isolation solutions in terms of accuracy, real-time performance, and maintainability.

[0075] Regarding the details of the graph neural network calculating the string mismatch score of each photovoltaic string, in some examples of the embodiments of the present application, the node initial feature matrix is represented as: , Formula (11) in, Indicates The initial eigenvector corresponding to the photovoltaic string, and Respectively represent the first The arithmetic mean and standard deviation of the output power of the photovoltaic strings. Indicates The surface module temperature of the PV modules used in the PV strings.

[0076] The global environment vector is represented as: , Formula (12) In the formula, represents the global environment vector, represents the ambient irradiance of the entire station, and Indicates the ambient temperature of the entire station.

[0077] Concatenate the node initial feature matrix and the global environment vector in the feature dimension into the initial fusion matrix : , Formula (13) It should be noted that string mismatch, that is, the deviation of the actual output from the expected or adjacent string output, not only depends on the electrical state of the string itself (such as output power fluctuation, module temperature difference), but is also deeply affected by the current / voltage coupling of adjacent strings in the same box and the overall meteorological conditions (irradiance and ambient temperature). Therefore, initially integrating the node's own characteristics with the global environmental context, where the node characteristics reflect both the local mismatch-related signals and the overall trend brought about by environmental changes, helps to distinguish environmental disturbances from component anomalies.

[0078] For the th layer and the th attention head, represent the projection matrix as , and represent the attention parameter vector as , then calculate the original attention score of the neighbor node to the node.

[0079] The multi-head attention mechanism learns multiple neighbor information fusion modes in parallel for the same node in each layer. Each head calculates the original attention score in different projection subspaces based on the interaction between the node's own and neighbor features and the prior edge weights.

[0080] , Equation (14) In the formula, represents the original attention score, indicating the original influence magnitude of the th neighbor string corresponding to the neighbor node on the th attention head during the feature aggregation of the th photovoltaic string to the th photovoltaic string; and respectively represent the node feature vectors of the th photovoltaic string and the th photovoltaic string after the th layer iteration.

[0081] Then, update the original attention score by fusing the normalized edge weight matrix of the node: , Equation (15) In the formula, represents the attention energy value with physical prior weighting from the node of the th layer and the th attention head from the node of the th photovoltaic string to the node of the th photovoltaic string; is the matrix element in the normalized edge weight matrix, representing the normalized edge weight from the node of the th photovoltaic string to its neighbor node of the th photovoltaic string.

[0082] It should be noted that there may be differences in the current restraint effect of parallel strings in the same combiner box. By using the absolute value of the Pearson correlation coefficient based on power time series to assign initial edge weights, mapping the synchronous / antisynchronous information in actual operation to the initial weights of the edges, distinguishing the contributions of different neighbors to mismatch evaluation, and providing differential edge weight priors for different real-time correlations in the graph structure. By mapping the real-time correlation degrees of different strings to the edge weights, when the graph attention network calculates the attention scores of neighbor nodes for the target node, it can automatically identify and prioritize attention to those neighbors with strong physical coupling or the most obvious synchronization mismatch.

[0083] For example, when the temperature of a certain string suddenly rises due to local hot spots or poor solder tape contact, the fluctuation characteristics of its power curve will be highly correlated with those of its neighbors that fluctuate due to environmental occlusion and light gradient. At this time, the correlation coefficient is large and the initial attention score also increases accordingly; conversely, when the differences in the environment and component states between two strings are large, the correlation coefficient is low and the attention score is also suppressed. This usually means that the power fluctuations of the two are not caused by environmental factors (such as shadows or temperature gradients) that can be reduced by adjusting the compensator power, but are more likely to reflect component defects or electrical faults in one of the strings. Relying solely on power compensation cannot effectively restore its state, and waste of compensation resources should be avoided. Therefore, after updating the original attention scores with the normalized edge weight matrix of the fusion node, more attention can be allocated to those nodes that truly need to be compensated first.

[0084] The attention weights are obtained by performing softmax normalization on all neighbor nodes of the node: , Equation (16) In the formula, represents the th layer and the th head's normalized attention weight, which is used to measure the relative importance of the th string to the th string in this channel;

[0085] Here, the Softmax function is used to normalize all neighbors of the same node to obtain the final attention weights , ensuring that for a certain string the sum of all

[0086] Perform neighbor node feature aggregation on nodes according to attention weights: , Equation (17) where represents the aggregated vector of the -th layer, the -th head, for the -th path string node, representing the high-order features it absorbs from all neighbors and transforms with weights.

[0087] Concatenate and activate the features output by all attention heads in the -th layer to obtain the node features of the -th layer: , Equation (18) where represents the updated feature vector of the -th layer, the -th path of PV strings, represents the total number of attention heads in the -th layer.

[0088] Here, after concatenating the outputs of all attention heads in the -th layer and performing ReLU non-linear activation, different neighborhood patterns captured by each attention sub-channel can be weighted and integrated to form the high-order feature vector of the nodes in the -th layer. Through multi-head concatenation, not only is the expression ability of node representation greatly enriched, but also a wider feature subspace is provided for the next layer, enabling the network to concurrently focus on multiple different mismatch trigger patterns (such as hot spots, occlusions, temperature mutations, etc.), so that more abstract and comprehensive mismatch representations can be gradually refined during continuous iteration. Through ReLU activation, the non-linear mapping ability is further introduced, which can not only retain the positive response but also avoid negative value interference.

[0089] Perform linear mapping and ReLU activation on the node features output by the L -th layer of the last layer to obtain the string mismatch scores of each path of PV strings: , Equation (19) Here, after multi-head multi-layer attention weighted fusion, each node obtains a high-dimensional feature vector, which contains multi-scale mismatch information of the node in the overall topology and environmental context. Mapping this high-dimensional vector to one dimension and retaining the non-negative part through ReLU completes the mapping from high-order representation to mismatch score.

[0090] Through the embodiments of the present application, a mismatch scoring method based on multi-source information fusion and weighted graph attention mechanism is proposed: by taking the power statistics, component temperature of each string group, and the whole station environmental variables as node features, and constructing a complete graph with the power time series correlation as the edge weight prior, the graph attention network can adaptively measure and aggregate multi-dimensional mismatch signals from different neighbors. The model is no longer limited to the traditional data that only relies on voltage or current, but can simultaneously capture the dynamic coupling effects caused by temperature difference hot spots and local shading, thereby achieving high-precision quantification of the string mismatch degree; at the same time, the multi-head parallel attention mechanism ensures that the network can simultaneously focus on multiple mismatch trigger modes, avoiding misjudgment or information loss caused by a single feature dominance, and improving the recognition accuracy of string mismatch.

[0091] Regarding the solution details of the compensation decision model, in some examples of the embodiments of the present application, the compensation decision model can be iteratively solved by adopting the three-stage alternating direction method of multipliers to obtain the optimal compensation scheme. 。

[0092] Figure 5 The operation flowchart showing an example of the three-stage alternating direction method of multipliers iteratively solving the compensation decision model according to the embodiments of the present application is shown.

[0093] As Figure 5 shown, in step S510, in the first stage, by independently optimizing and solving each path ([ P , x ) sub-problem.

[0094] During the implementation process of the first stage, for the power compensation and the binary enable under the global constraint , by introducing a continuous auxiliary variable , the originally coupled mixed integer programming is transformed into an ADMM alternating optimization framework. Specifically, the auxiliary variable is responsible for carrying the constraint of "the upper limit M of the total number of available optimizers", while the original binary variable completely retains its {0, 1} value during the separate optimization in the first stage. In this way, the originally complex coupled problem that is simultaneously restricted by both power demand and total resources is decomposed into two major sub-problems of "independently performing for each string group" and "performing projection for all ", thus realizing the decoupling and parallelization of the problem structure.

[0095] Specifically, in the th round of iteration, based on the auxiliary variable and the Lagrange multiplier obtained in the previous round of iteration, each photovoltaic string group is separately solved: , Equation (20) where represents the augmented Lagrangian penalty coefficient represents the th iteration, the Lagrange multiplier corresponding to the th PV string represents the th iteration, the auxiliary variable corresponding to the th PV string

[0096] Thus, through the decomposition strategy, it is ensured that: on the one hand, efficient solution can still be achieved while retaining the binary decision, without the need to perform relaxation or approximation; on the other hand, the global resource constraint is only placed in the step of " Projection", significantly reducing the global coupling degree, enabling the sub-problems to run in parallel, and greatly enhancing the real-time compensation decision-making ability of the edge device for hundreds of strings

[0097] For the update of , if , the loss term decreases linearly with the increase of , and the optimal value is ; if , then is forced to be set represents the enable flag variable corresponding to the th PV string at the beginning of the th iteration

[0098] It should be noted that in the P-x sub-problem of the first stage, for each string, if the enable flag , then the loss term decreases linearly with . The model directly takes the maximum value of according to this property ; if , then is forced to avoid ineffective compensation. That is to say, the power update can be divided into two extreme cases: full compensation when enabled, and zero power output when not enabled

[0099] Thus, the index significance of the mismatch score in terms of power compensation value is fully utilized: for the enabled strings, raising the power directly to the maximum available level can maximize the reduction of mismatch losses; for the non-enabled strings, setting the power to 0 avoids compensating unnecessary nodes, saving the resource consumption of the optimizer. Thus, it ensures the maximum single-channel compensation strength and makes the overall compensation more targeted and efficient

[0100] For the update of , calculate respectively and the complete costs in both cases, where the cost function is defined by .

[0101] At , the corresponding complete cost value is .

[0102] At , the corresponding complete cost value is .

[0103] Select the one with the smaller corresponding complete cost value as .

[0104] Here, for the enable flag of each string group, the model enumerates the complete cost values in the two cases of 0 and 1, and according to the magnitude of these two values, selects the one that minimizes the cost and assigns it to . Thus, both the benefit of mismatch reduction brought by enabling and the consistency penalty with the auxiliary variable and the multiplier are considered.

[0105] Through direct enumeration and cost comparison, the system does not need to relax and then backtrack to the binary solution, but can select the most appropriate start-stop decision in one go in this round of iteration, avoiding the search tree and backtracking calculations common in mixed integer programming. This enables the algorithm to maintain a linear time complexity of each round of update when the string scale is large, thereby significantly improving the edge real-time scheduling performance and ensuring the binary executability of the final compensation action.

[0106] In step S520, in the second stage, solve the auxiliary variable sub-problem through global resource constraint projection.

[0107] Fix the solved in the first stage, and project the temporary auxiliary variable onto the convex set that satisfies : Clip the temporary auxiliary variable to [0, 1]: ; denotes the intermediate value after forcing to be constrained within the interval [0, 1].[[]]

[0108] If , then , otherwise, use the bisection method to obtain the offset , and then let .

[0109] Here, the global constraint "the total number of enabled optimizers does not exceed " and the interval constraints of each node auxiliary variable are organically processed in two steps: First, a "clipping" operation is performed on each temporary calculated value, and then, only when the sum of each after clipping exceeds does a binary search trigger to accurately calculate an offset so that all simultaneously satisfy the single-node interval constraint and the global sum is strictly equal to . Thus, when the sum after clipping does not exceed the limit, is directly used as the final output, saving computational overhead; when the sum after clipping exceeds the limit and needs to be adjusted, the offset obtained based on the binary method can not only ensure that all linear or non-linear constraint conditions are not violated, but also quickly complete the projection, greatly improving the solution efficiency of the edge ECU and ensuring stable convergence.

[0110] In step S530, in the third stage, the Lagrange multiplier is updated.

[0111] , Equation (21) In the formula, represents the Lagrange multiplier after the round of update, so as to continue to promote the reduction of in the and gap in the next round of the

[0112] round of iteration. By cumulatively updating the gap between the binary variable and the auxiliary variable and , this multiplier term is equivalent to a "pulling force" in the next round of iteration, making the subsequent

[0113] more inclined to take the same value and gradually pushing them towards consistency round by round. Thus, through multiplier update, ADMM iteration can ensure that the global resource constraint and local decision are perfectly matched after several iterations, and there will be no oscillation or non-convergence problem of no solution. and Note that, compared with the "dual saddle point" dilemma that may occur in the ordinary Lagrange relaxation method when lacking the augmented term, the augmented multiplier ensures the convergence and stability of the algorithm, and at the same time realizes the termination of iteration when the gap between

[0114] In the embodiment of the present application, when the three-stage alternating direction multiplier method (ADMM) is applied to solve the photovoltaic string mismatch compensation problem, the main difference from the traditional ADMM algorithm is that: first, the enable flag is always retained. The binary constraint of , in the sub-problem, the direct enumeration and comparison of the two values avoids The continuous relaxation and complex integerization steps of In the update of , due to its linear structure, the endpoint extreme value can be directly taken, thus decomposing the originally coupled two-dimensional optimization into two-step parallel updates for each string, greatly reducing the amount of calculation of the sub-problem. Third, when dealing with the global compensator quantity constraint, by adjusting the auxiliary variable The truncation and sorting projection algorithm can quickly complete the The global constraint projection of , which eliminates the complex quadratic programming solution that may appear in the general ADMM. Therefore, through these improvements, the three-stage alternating direction multiplier method can achieve higher decision-making execution efficiency.

[0115] Furthermore, the above three-stage operation is repeated until and The difference is less than the preset convergence threshold , then the iteration is judged to converge and a set of convergent variables is obtained and , and obtain the optimal compensation solution by setting the corresponding string parameters: , Formula (22) In the formula, It means that after the iteration converges The target power value corresponding to each PV string. It means that after the iteration converges The enable flag value corresponding to the PV string. It means that after the iteration converges The auxiliary variable value corresponding to the PV string.

[0116] Here, there is no need to perform additional integerization after convergence, which not only ensures the binary executableness of the solution, but also avoids the time overhead and error caused by backtracking after relaxation. It can be efficiently sent to each road optimizer through the industrial bus, accurately reducing the string mismatch and significantly improving the real-time, reliability and economy of the system compensation decision.

[0117] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0118] Figure 6 A structural block diagram of an example of an artificial intelligence-based new energy power station operation and maintenance system according to an embodiment of the present application is shown.

[0119] like Figure 6 As shown, the artificial intelligence-based new energy power station operation and maintenance system 600 includes a multi-source feature acquisition unit 610, a graph structure construction unit 620, a node feature update unit 630, a compensation decision modeling unit 640, a compensation decision solving unit 650 and a string mismatch dynamic compensation unit 660.

[0120] The multi-source feature acquisition unit 610 is used to collect the output power and module temperature values of each photovoltaic string connected to the combiner box on the combiner box side, extract the output power mean and output power variance of each photovoltaic string and generate the corresponding initial feature vector in combination with the module temperature value, and generate a global environment vector according to the environmental irradiance and the environmental temperature of the entire station.

[0121] The graph structure construction unit 620 is used to construct an undirected graph structure with each photovoltaic string as a node and the electrical connection relationship between the photovoltaic strings as an edge; the node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string.

[0122] The node feature updating unit 630 is used to input the undirected graph structure and the global environment vector into the graph attention network. L After the iterative fusion of the multi-head attention layers, L The layer output features are linearly mapped and ReLU activated to obtain the string mismatch score of each photovoltaic string: L Represents the total number of graph attention layers in the graph attention network.

[0123] The compensation decision modeling unit 640 is used to construct a compensation decision model by taking the string mismatch score of each photovoltaic string and the number of DC / DC optimizers as input.

[0124] The compensation decision solving unit 650 is used to solve the compensation decision model to obtain the optimal compensation solution. ; wherein, and respectively represent the optimal target power and the enable flag decision value corresponding to the th photovoltaic string.

[0125] The string mismatch dynamic compensation unit 660 is used to generate a compensation instruction according to the optimal compensation scheme and send the generated compensation instruction to the DC / DC optimizer corresponding to each photovoltaic string to implement string mismatch dynamic compensation.

[0126] In some embodiments, the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any one of the above-mentioned new energy power station operation and maintenance methods based on artificial intelligence of the present application.

[0127] In some embodiments, the present application also provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer is caused to execute the steps of any one of the above-mentioned new energy power station operation and maintenance methods based on artificial intelligence.

[0128] In some embodiments, the present application also provides an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the new energy power station operation and maintenance method based on artificial intelligence.

[0129] Figure 7 is a schematic hardware structure diagram of an electronic device for executing the new energy power station operation and maintenance method based on artificial intelligence provided by another embodiment of the present application. As Figure 7 shown, the device includes: one or more processors 710 and a memory 720, Figure 7 taking one processor 710 as an example.

[0130] The device for executing the new energy power station operation and maintenance method based on artificial intelligence may further include: an input device 730 and an output device 740.

[0131] The processor 710, the memory 720, the input device 730, and the output device 740 may be connected by a bus or other means, Figure 7 taking connection by a bus as an example.

[0132] The memory 720, being a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for operation and maintenance of a new energy power station based on artificial intelligence in the embodiments of the present application. The processor 710 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 720, that is, implements the method for operation and maintenance of a new energy power station based on artificial intelligence in the above method embodiments.

[0133] The memory 720 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 720 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 720 may optionally include a memory remotely disposed relative to the processor 710, and these remote memories can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0134] The input device 730 can receive input digital or character information, and generate signals related to the user settings and function control of the electronic device. The output device 740 may include a display device such as a display screen.

[0135] The one or more modules are stored in the memory 720 and, when executed by the one or more processors 710, execute the method for operation and maintenance of a new energy power station based on artificial intelligence in any of the above method embodiments.

[0136] The above product can execute the method provided in the embodiments of the present application, and has the corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference can be made to the method provided in the embodiments of the present application.

[0137] The electronic device in the embodiments of the present application exists in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by having mobile communication functions and mainly aiming to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0138] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, etc.

[0139] (3) Portable entertainment devices: Such devices can display and play multimedia content. This type of device includes: audio and video players, handheld game consoles, e-books, as well as smart toys and portable in-vehicle navigation devices.

[0140] (4) Other airborne electronic devices with data interaction functions, such as in-vehicle device installed on a vehicle.

[0141] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0142] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or rather the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A new energy power station operation and maintenance method based on artificial intelligence, characterized in that, The method comprises: The output power and module temperature values of each photovoltaic string connected to the combiner box are collected on the combiner box side, the output power mean and output power variance of each photovoltaic string are extracted and combined with the module temperature value to generate the corresponding initial feature vector, and the global environment vector is generated according to the environmental irradiance and the ambient temperature of the whole station; An undirected graph structure is constructed with each photovoltaic string as a node and the electrical connection relationship between the photovoltaic strings as an edge; the node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string; Input the undirected graph structure and the global environment vector into the graph attention network. After L layers of multi-head attention iterative fusion, perform linear mapping and ReLU activation on the output features of the L th layer to obtain the string mismatch scores of each photovoltaic string; L represents the total number of graph attention layers in the graph attention network; The compensation decision model is constructed with the string mismatch score of each PV string and the number of DC / DC optimizers as input; Solve the compensation decision model to obtain the optimal compensation plan ; where and respectively represent the optimal target power and the enabled flag decision value corresponding to the th photovoltaic string According to the optimal compensation scheme Generate a compensation instruction and send the generated compensation instruction to the DC / DC optimizer corresponding to each photovoltaic string to implement dynamic compensation for string mismatch.

2. The method according to claim 1, wherein The string mismatch score is expressed by the following formula: ， In the formula, represents the string mismatch score of the th string of the photovoltaic string, represents the linearly mapped trainable weight vector, represents the vector transpose operation, represents the linearly mapped bias term, represents the th photovoltaic string passing through L the feature vector output after fusion by the layer graph attention layer; The compensation decision model is: ， Among them, , ; In the formula, N represents the total number of paths of the PV string, represents the number of DC / DC optimizers, is the rated power of the th path of the PV string, is the target power of the th path of the PV string after compensation, is the penalty coefficient for optimizer enabling; is the enabling flag of the DC / DC optimizer corresponding to the th path of the PV string, represents enabling the DC / DC optimizer for the th path of the PV string, represents not enabling the DC / DC optimizer for the th path of the PV string.

3. The method according to claim 2, characterized in that, The undirected graph structure is constructed by taking each photovoltaic string as a node and taking the electrical connection relationship between the photovoltaic strings as an edge, including: In the node set corresponding to each photovoltaic string, an undirected edge is constructed for any pair of nodes to form a complete graph between the entire set of nodes; For each undirected edge in the complete graph, a Pearson correlation coefficient is calculated based on the output power time series data of two photovoltaic strings corresponding to the undirected edge within a preset time window, and the absolute value of the Pearson correlation coefficient is used as the initial weight of the undirected edge; For each node in the complete graph, the initial weights of all connected undirected edges of the node are normalized by row so that the sum of the normalized weights of all connected undirected edges of each node is 1, so as to update the edge weights of each connected undirected edge.

4. The method according to claim 3, wherein Inputting the undirected graph structure and the global environment vector into the graph attention network, after L iterative fusion of L layers of multi-head attention, performing linear mapping and ReLU activation on the output features of the L -th layer to obtain the string mismatch scores of each photovoltaic string, including: The node initial feature matrix is expressed as: ， Among them, represents the initial eigenvector corresponding to the th photovoltaic string, and respectively represent the arithmetic mean and standard deviation of the output power of the th photovoltaic string within a preset time window, represents the surface module temperature of the photovoltaic modules used in the th photovoltaic string; The global environment vector is represented as: ， In the formula, represents the global environmental vector, represents the total station environmental irradiance, and represents the total station environmental temperature; Concatenate the initial feature matrix of the node and the global environment vector in the feature dimension to form the initial fusion matrix : ， For the th layer and the th attention head, represent the projection matrix as , represent the attention parameter vector as , then calculate the raw attention score of the neighbor nodes for the node: ， In the formula, is expressed as the original attention score, indicating the th attention head in the th layer, the original influence magnitude of the th photovoltaic string pair on the th photovoltaic string during feature aggregation; and respectively represent the node feature vectors of the th layer after iteration for the th photovoltaic string and the node feature vector of the th photovoltaic string; The normalized edge weight matrix of the re-fused node is used to update the original attention score: ， In the formula, represents the th attention head in the th path of the PV string nodes, the attention energy value with physical prior weighting from the is the matrix element in the normalized edge weight matrix, representing the th path of the PV string nodes to its neighbor path of the PV string nodes, the normalized edge weight; The attention weights are obtained by performing softmax normalization on all neighboring nodes of the node: ， Wherein, represents the normalized attention weight of the head of the th layer, which is used to measure the relative importance of the th string pair to the th string in this channel; represents the set of all edges connected to the th photovoltaic string node, Perform neighbor node feature aggregation on the node according to the attention weight: ， In the formula, represents the summary vector of the head pairs of the road string nodes in the layer, representing the high-order features absorbed from all its neighbors and weighted and transformed; Concatenate and activate the features output by all attention heads in the th layer to obtain the node features of the th layer: ， In the formula, represents the updated eigenvector of the th road photovoltaic string in the th layer, represents the total number of attention heads in the Perform a linear mapping and ReLU activation on the node features output by the L layer of the last layer to obtain the string mismatch scores of each photovoltaic string: 。 5. The method according to claim 4, characterized in that, Solving the compensation decision model to obtain an optimal compensation plan , including: The three-stage alternating direction method of multipliers is used to iteratively solve the compensation decision model to obtain the optimal compensation scheme ; In the first stage, by solving the sub-problems through independent optimization for each path ( P , x ), the sub-problems are as follows: In the round of iteration, based on the auxiliary variables obtained in the previous round of iteration and Lagrange multipliers , each photovoltaic string is solved separately: ， In the formula, represents the augmented Lagrangian penalty coefficient, represents the th Lagrange multiplier corresponding to the th photovoltaic string at the th iteration; For update, if , the loss term decreases linearly with increasing, and the optimal value is ; if , then force to set ; represents the enable flag variable corresponding to the th road PV string at the beginning of the th round of iteration; For updates, calculate the full cost for each of and cases, where the cost function is defined by ; At the corresponding complete cost value is ; At the corresponding complete cost value is ; Select the one with the smaller corresponding complete cost value as ; In the second stage, the auxiliary variables are solved by global resource constraint projection Sub-problem: Fix what is solved in the first stage and project the temporary auxiliary variable onto the convex set that satisfies : Limit the temporary auxiliary variable to the range [0, 1]: ; It means the intermediate value after being forced to be within the range of [0, 1]; If , then , Otherwise, use the bisection method to obtain the offset , and then let ; In the third phase, the Lagrange multipliers are updated: ， In the formula, represents the Lagrange multiplier after the -th round of update, so as to continue to promote the reduction of the gap between and in the iteration of the next Repeat the above three-stage operation iteration until the difference between and is less than the preset convergence threshold and are obtained, and the optimal compensation scheme is obtained by setting the parameters of the corresponding string: ， In the formula, represents the target power value corresponding to the th photovoltaic string after iterative convergence, represents the enable flag value corresponding to the th photovoltaic string after iterative convergence, represents the auxiliary variable value corresponding to the th photovoltaic string after iterative convergence.

6. A new energy power station operation and maintenance system based on artificial intelligence, characterized in that, The system comprises: The multi-source feature acquisition unit is used to collect the output power and module temperature values of each photovoltaic string connected to the combiner box on the combiner box side, extract the output power mean and output power variance of each photovoltaic string and generate the corresponding initial feature vector in combination with the module temperature value, and generate the global environment vector according to the environmental irradiance and the environmental temperature of the whole station; A graph structure construction unit, for constructing an undirected graph structure with each photovoltaic string as a node and with the electrical connection relationship between the photovoltaic strings as an edge; the node feature of each node in the undirected graph structure is defined according to the initial feature vector of the corresponding photovoltaic string; A node feature update unit, configured to input the undirected graph structure and the global environment vector into a graph attention network. After L iterative fusion by L layers of multi-head attention, perform linear mapping and ReLU activation on the output features of the L layer to obtain the string mismatch scores of each photovoltaic string: A compensation decision modeling unit, used to construct a compensation decision model by taking the string mismatch score of each photovoltaic string and the number of DC / DC optimizers as input; The compensation decision solving unit is used to solve the compensation decision model to obtain an optimal compensation plan ; where and respectively represent the optimal target power and the enabled flag decision value corresponding to the th PV string String mismatch dynamic compensation unit, used to generate a compensation instruction according to the optimal compensation scheme and send the generated compensation instruction to the DC / DC optimizer corresponding to each photovoltaic string to implement string mismatch dynamic compensation.

Images