Energy management system for hydroelectric coupling system and method for use therein

Abstract

The invention provides an energy management system for a hydrogen-electricity coupling system and a method applied to the energy management system. The energy management system provided by the invention is set to be a cloud edge-end architecture, the end-side calculation layer is used for collecting the operation data of the hydrogen-electricity coupling system, the meteorological data influencing photovoltaic power generation, the timestamp data, the cloud layer image data and the like, and the collected data is reported to the edge side and the end side. And generating a scheduling strategy based on the operation data reported by the end side by using the side central coordination control layer, and issuing the scheduling strategy to the end side, so that the end side schedules each device in the hydrogen-electricity coupling system based on the scheduling strategy. A cloud side energy management layer is utilized to train and generate a power prediction model based on data reported by an end side, and the power generation power of the distributed photovoltaic system is predicted, so that a side can optimize a scheduling strategy according to a prediction result of the power generation power, and more efficient energy management of the hydrogen-electricity coupling system is realized.

Description

Technical Field

[0001] The present application relates to the technical field of coordinated control of new energy, and in particular to an energy management system for a hydrogen-electric coupling system and a scheduling optimization method and a model training method applied therein. Background Art

[0002] The hydrogen-electric coupling system can uniformly manage and dispatch distributed photovoltaics, new energy storage and various energy loads, thereby reducing the impact of photovoltaic fluctuations on the system voltage frequency. It is a core component of the future energy and power system. In order to ensure the economy and stability of the future energy and power system, it is of great significance to ensure the efficient use of hydrogen energy in the hydrogen-electric coupling system and the stable consumption of renewable energy. Based on this, how to efficiently manage the energy of the hydrogen-electric coupling system, ensure the efficient use of hydrogen energy, and the stable consumption of renewable energy has become a technical problem that needs to be solved in this field. Summary of the invention

[0003] In order to solve the problem of how to efficiently manage the energy of the hydrogen-electricity coupling system, ensure the efficient utilization of hydrogen energy and the stable consumption of renewable energy, the present application provides an energy management system for the hydrogen-electricity coupling system and a scheduling optimization method and a model training method applied therein.

[0004] In a first aspect, an embodiment of the present application provides an energy management system for a hydrogen-electricity coupling system, the energy management system comprising: an edge-side central coordination control layer and an end-side computing layer; the end-side computing layer is used to obtain the actual power generation power of the distributed photovoltaic system, the actual pressure of the hydrogen storage tank, the actual energy storage power and the load power of the energy storage battery in the hydrogen-electricity coupling system, and upload the collected actual power generation power, actual pressure, actual energy storage power and load power to the edge-side central coordination control layer; the edge-side central coordination control layer is used to: determine the net power between the actual power generation power and the load power; when the net power is greater than zero, based on the relationship between the net power and the minimum operating power and the maximum operating power of the hydrogen production electrolyzer in the hydrogen-electricity coupling system, the actual pressure The method comprises the following steps: generating a first scheduling strategy based on the relationship between the net power and the upper limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the upper limit of the energy storage power of the energy storage battery, so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen production electrolyzer and the energy storage battery in the hydrogen-electric coupling system based on the first scheduling strategy to absorb the net power; and generating a second scheduling strategy based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure and the lower limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the lower limit of the energy storage power of the energy storage battery, so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen production electrolyzer and the energy storage battery based on the second scheduling strategy to compensate for the net power when the net power is less than zero.

[0005] In the second aspect, an embodiment of the present application also provides a scheduling optimization method, which includes: obtaining the actual power generation power of the distributed photovoltaic system in the hydrogen-electricity coupling system, the actual pressure of the hydrogen storage tank, the actual energy storage power and the load power of the energy storage battery through the end-side computing layer of the energy management system, and uploading the actual power generation power, actual pressure, actual energy storage power and load power to the edge-side central coordination control layer of the energy management system; determining the net power between the actual power generation power and the load power through the edge-side central coordination control layer; and, when the net power is greater than zero, based on the relationship between the net power and the minimum operating power and the maximum operating power of the hydrogen production electrolyzer in the hydrogen-electricity coupling system, the actual pressure and the hydrogen storage tank. The relationship between the pressure upper limit and the pressure lower limit, and the relationship between the actual energy storage power and the energy storage power upper limit of the energy storage battery, generate a first scheduling strategy so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen production electrolyzer and the energy storage battery in the hydrogen-electric coupling system based on the first scheduling strategy to absorb the net power; when the net power is less than zero, based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure and the pressure lower limit of the hydrogen storage tank, and the relationship between the actual energy storage power and the energy storage power lower limit of the energy storage battery, generate a second scheduling strategy so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen production electrolyzer and the energy storage battery based on the second scheduling strategy to compensate for the net power.

[0006] In a third aspect, an embodiment of the present application also provides a model training method, which includes: obtaining a historical data set; the historical data set includes multiple historical samples, each of the historical samples includes a first historical text data sequence and a first historical cloud image data sequence; the first historical text data sequence includes a first meteorological data sequence of a first meteorological parameter related to photovoltaic power generation power, a first power generation data sequence of photovoltaic power generation power, and a first timestamp data sequence; constructing a training data set based on the historical data set; the training data set includes multiple training samples, each of the training samples includes a second historical text data sequence and a second historical cloud image data sequence; the second historical text data sequence includes a second meteorological data sequence of a second meteorological parameter, A second power generation data sequence and a second timestamp data sequence; the second meteorological parameter includes the first meteorological parameter in the first meteorological parameter whose correlation with photovoltaic power generation is greater than or equal to a correlation threshold; the second historical cloud image data sequence is generated based on the first historical cloud image data sequence; the second power generation data sequence is generated based on the first power generation data sequence; the second timestamp data sequence is generated based on the first timestamp data sequence; the training data set is used to train a pre-set prediction model to be trained to obtain a trained power prediction model; the prediction model to be trained includes a temporal convolutional network model and a Transformer model; the power prediction model is used to predict the power generation of the distributed photovoltaic system.

[0007] The embodiment of the present application provides an energy management system for a hydrogen-electric coupling system and a scheduling optimization method and a model training method used therein. The energy management system provided by the present application is set as a cloud-edge-end architecture, and uses the end-side computing layer to collect the operating data of the hydrogen-electric coupling system, meteorological data affecting photovoltaic power generation, timestamp data, cloud image data, etc., and reports the collected data to the edge and end. The edge-side central coordination control layer is used to generate a scheduling strategy based on the operating data reported by the end side, and the scheduling strategy is sent down to the end side, so that the end side can schedule each device in the hydrogen-electric coupling system based on the scheduling strategy. The cloud-side energy management layer is used to train and generate a power prediction model based on the data reported by the end side, and the power generation power of the distributed photovoltaic system is predicted, so that the edge side can optimize the scheduling strategy according to the prediction results of the power generation power, and realize more efficient energy management of the hydrogen-electric coupling system.

[0008] Among them, when training and generating the power prediction model, the cloud side uses time features and cloud image features that have a strong correlation with the power generation as the features of the training samples, and the trained power prediction model has higher accuracy. In addition, the pre-set prediction model to be trained adopts a fusion architecture based on the time convolutional network model and the Transformer model, which can simultaneously learn the local power generation features and global time series dependencies in the training samples, thereby improving the prediction accuracy and generalization ability of the model, and further improving the accuracy of power supply prediction.

[0009] When generating the scheduling strategy, the edge side takes the minimum operating power of the hydrogen electrolyzer, the maximum operating power of the hydrogen electrolyzer, and the maximum operating power of the fuel cell as the conditions for scheduling optimization, fully considering the operating safety of the equipment. While managing energy, it also ensures the operating safety of the equipment to a great extent, making energy management more efficient. In addition, the edge side can also perform scheduling optimization based on the cloud side's prediction results of the power supply, further improving the efficiency of energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only 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.

[0011] Figure 1 A schematic diagram of the architecture of a hydrogen-electricity coupling system provided in an embodiment of the present application.

[0012] Figure 2 A schematic diagram of the architecture of an energy management system for a hydrogen-electricity coupling system provided in an embodiment of the present application.

[0013] Figure 3 A flowchart of a scheduling strategy generation method provided in an embodiment of the present application.

[0014] Figure 4 A schematic diagram of an application scenario provided for an embodiment of the present application.

[0015] Figure 5 A flowchart of a model training method provided in an embodiment of the present application.

[0016] Figure 6 A schematic diagram of another application scenario provided for an embodiment of the present application.

[0017] Figure 7 A flowchart of a scheduling optimization method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The present application is further described in detail below through the accompanying drawings and embodiments. Through these descriptions, the characteristics and advantages of the present application will become clearer and more specific.

[0019] The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise noted.

[0020] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0021] See also Figure 1 , Figure 1 This is a schematic diagram of the architecture of a hydrogen-electric coupling system provided in an embodiment of the present application. Figure 1 As shown, the hydrogen-electricity coupling system may include: a distributed photovoltaic system, an energy storage battery, a hydrogen production electrolyzer, a hydrogen storage tank, a fuel cell, a DC (direct current) bus, an AC (alternating current) bus, a DC / DC (direct current-direct current) converter, a DC / AC (direct current-alternating current) converter, etc.

[0022] The distributed photovoltaic system and the energy storage battery can be connected to the DC bus through the same photovoltaic and storage bidirectional DC / DC converter. The hydrogen production electrolyzer and the fuel cell can be connected to the DC bus through different DC / DC converters. It should be understood that the distributed photovoltaic system and the energy storage battery can also be connected to the DC bus through different DC / DC converters. The energy storage battery can be an electrochemical energy storage battery, for example, the energy storage battery can be a lithium energy storage battery.

[0023] The hydrogen storage tank can be connected between the hydrogen production electrolyzer and the fuel cell. The hydrogen produced by the hydrogen production electrolyzer can be stored in the hydrogen storage tank, and the fuel cell can use the hydrogen in the hydrogen storage tank to generate electricity.

[0024] The DC bus can be connected to the AC bus through a DC / AC converter. In addition, the AC bus can also connect the power grid and the load.

[0025] The role or function of each device in the hydrogen-electricity coupling system, as well as the relationship between the devices, can be referred to the contents of the subsequent embodiments and will not be described in detail here.

[0026] See also Figure 2 , Figure 2 The schematic diagram of the architecture of an energy management system for a hydrogen-electricity coupling system provided in an embodiment of the present application is shown in FIG. The energy management system may include a cloud-side energy management layer, an edge-side central coordination control layer, and a terminal-side computing layer.

[0027] Among them, the end-side computing layer can collect the operating data of each device in the hydrogen-electricity coupling system in real time, and then report the collected operating data to the edge-side central coordination layer and / or the cloud-side energy management layer. In addition, the end-side computing layer can also optimize the scheduling of the equipment in the hydrogen-electricity coupling system according to the control instructions issued by the central coordination control layer and / or the cloud-side energy management layer. For specific implementation, please refer to the contents of the subsequent embodiments.

[0028] Optionally, the end-side computing layer may include multiple end-side acquisition and computing architectures, and each end-side acquisition and computing architecture may include an end-side acquisition device and an end-side computing device. Among them, each end-side acquisition device can be used to collect the operating data of a device in the hydrogen-electricity coupling system. In other words, a separate end-side acquisition device can be configured for each device in the hydrogen-electricity coupling system to collect the operating data of the corresponding device. Each end-side acquisition device can be communicatively connected to the end-side computing device in the same architecture, and send the collected operating data to the end-side computing device. The end-side computing device can pre-process the operating data collected by the end-side acquisition device, for example, data cleaning, etc., and then report the pre-processed operating data to the edge central coordination control layer and / or the cloud-side energy management layer.

[0029] It should be understood that the end-side computing layer can also be set to other architecture modes. For example, all end-side acquisition devices in the end-side computing layer are connected to the same end-side computing device. Alternatively, some end-side acquisition devices in the end-side computing layer are connected to the same end-side computing device, and the remaining end-side acquisition devices are connected to one or more end-side computing devices, etc. This application does not limit this.

[0030] Optionally, the end-side computing layer may include multiple modules. Exemplarily, the end-side computing layer may include a power management module, an auxiliary function module, an online detection module, a control and protection module, a communication management module, etc.

[0031] Optionally, the multiple modules included in the end-side computing layer can be arranged in different devices of the end-side computing layer. Alternatively, optionally, some of the multiple modules included in the end-side computing layer can also be arranged in the same device of the end-side computing layer, which is not limited in this application.

[0032] Optionally, the power management module of the end-side computing layer can be used to maintain the power supply stability of the hydrogen-electric coupling system. Exemplarily, the power management module can uniformly control the power supply of each device such as the hydrogen production electrolyzer, energy storage battery, fuel cell, etc. in the hydrogen-electric coupling system to maintain the consistency of the voltage level between each device. For example, the power management module can control the scheduled on / off, delayed startup, safety protection, etc. of each device in the hydrogen-electric coupling system.

[0033] An interface for adjusting the operating parameter configuration can be set in the auxiliary function module of the edge computing layer, which facilitates the adjustment of the energy management strategy, optimization algorithm, objective function, etc. in the system, so that it can be debugged from the software layer in the edge central coordination control layer without adjusting the hardware equipment, thereby improving the flexibility of the energy management system.

[0034] The online detection module of the end-side computing layer is used to collect the operating data (or actual operating data, actual operating indicators, etc.) of each device in the hydrogen-electric coupling system. Exemplarily, the operating data collected by the online detection module may include: the actual power generation power of the distributed photovoltaic system, the hydrogen production power of the hydrogen production electrolyzer, the actual energy storage power of the energy storage battery, the actual pressure of the hydrogen storage tank, the operating power of the fuel cell, the load power, the interaction power between the hydrogen-electric coupling system and the power grid, etc. The end-side computing layer can calculate the charge state of the energy storage battery, the cumulative power generation of the distributed photovoltaic system, the total operating time of the hydrogen production electrolyzer, etc. according to the operating data collected by the online detection module. It should be understood that other data that need to be calculated can also be calculated, which will not be listed here one by one.

[0035] The control and protection module of the end-side computing layer can be used to detect the operating temperature of each device in the hydrogen-electricity coupling system in real time, determine the accuracy and delay of data collection, analyze the error rate, etc., to avoid system collapse due to excessive load.

[0036] The communication management module of the end-side computing layer can be used for protocol conversion during data transmission and control instruction execution. For example, the protocols may include the Modbus protocol used in distributed photovoltaic systems, the TCP / IP protocol and UDP protocol used in cloud transmission, the PLC communication protocol used in hydrogen electrolyzers, and the CAN protocol used in energy storage batteries and fuel cells.

[0037] The edge-side central coordination and control layer can receive the operating data reported by the end-side computing layer in real time, and generate scheduling strategies in real time based on the operating data reported by the end-side computing layer, and send the scheduling strategies to the end-side computing layer, so that the end-side computing layer can optimize the scheduling of each device in the hydrogen-electricity coupling system in real time based on the scheduling strategies sent by the edge-side central coordination and control layer, ensure the efficient utilization of hydrogen energy in the hydrogen-electricity coupling system, and the stable consumption of photovoltaic power generation energy, and realize efficient energy management of the hydrogen-electricity coupling system.

[0038] See also Figure 3 , Figure 3 A schematic diagram of a scheduling strategy generation method provided in an embodiment of the present application. In an optional implementation of the present application, the side central coordination control layer can be based on Figure 3 Generate a scheduling strategy in the following way:

[0039] Step S101, the actual power generation power of the distributed photovoltaic system, the actual pressure of the hydrogen storage tank, the actual energy storage power of the energy storage battery and the load power reported by the computing layer on the receiving end side.

[0040] In combination with the contents of the foregoing embodiments, it can be seen that the end-side computing layer can collect the operating data of each device in the hydrogen-electric coupling system in real time, and then report the collected operating data to the edge-side central coordination and control layer. Correspondingly, the edge-side central coordination and control layer can receive the operating data reported by the end-side computing layer in real time. Exemplarily, the operating data received by the edge-side central coordination and control layer may include the actual power generation power of the distributed photovoltaic system, the actual pressure of the hydrogen storage tank, the actual energy storage power of the energy storage battery, and the load power.

[0041] Step S102: determine the net power between the actual power generation power and the load power of the distributed photovoltaic system.

[0042] Net power refers to the difference between the actual power generation power of the distributed photovoltaic system and the load power.

[0043] Step S103: When the net power is greater than zero, a first scheduling strategy is generated based on the relationship between the net power and the minimum operating power of the hydrogen production electrolyzer, the maximum operating power of the hydrogen production electrolyzer, the relationship between the actual pressure of the hydrogen storage tank and the upper limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power of the energy storage battery and the upper limit of the energy storage power of the energy storage battery.

[0044] After generating the first scheduling strategy, the edge central coordination control layer can send the first scheduling strategy to the end-side computing layer in real time, so that the end-side computing layer can adjust the operating status of the fuel cell, hydrogen production electrolyzer and energy storage battery in the hydrogen-electric coupling system based on the first scheduling strategy, thereby absorbing the net power.

[0045] Step S104: When the net power is less than zero, a second scheduling strategy is generated based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure of the hydrogen storage tank and the lower limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery.

[0046] After generating the second scheduling strategy, the edge central coordination control layer can send the second scheduling strategy to the end-side computing layer in real time, so that the end-side computing layer adjusts the operating status of the fuel cell, hydrogen production electrolyzer and energy storage battery based on the second scheduling strategy, thereby compensating for the net power.

[0047] It should be noted that the above steps S101 to S104 are based on the actual power generation of the distributed photovoltaic system, the actual pressure of the hydrogen storage tank, the actual energy storage power of the energy storage battery and the load power reported by the computing layer of the receiving end, and are used as an example to illustrate the implementation method of the generation of the scheduling strategy by the central coordination control layer on the edge. In the actual application scenario, the central coordination control layer on the edge can Figure 3 In the manner shown, a scheduling strategy is generated in real time so that the end-side computing layer can schedule and optimize the operating status of each device in the hydrogen-electricity coupling system in real time.

[0048] It should be understood that the end-side computing layer can also periodically collect the operating data of each device in the hydrogen-electricity coupling system according to a preset period, and report the collected operating data to the edge-side central coordination control layer. Correspondingly, the edge-side central coordination control layer can also periodically receive the operating data reported by the end-side computing layer, so as to periodically schedule and optimize the operating status of each device in the hydrogen-electricity coupling system, depending on the needs of the application scenario, and this application does not limit this.

[0049] In an optional implementation manner of the present application, the first scheduling strategy may include the following strategies:

[0050] Strategy 1: If the net power is less than the minimum operating power of the hydrogen production electrolyzer, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, that is, the fuel cell and the hydrogen production electrolyzer are controlled to remain in a shutdown state.

[0051] And, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is greater than or equal to the net power, the energy storage battery is charged based on the net power. That is, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is greater than or equal to the net power, the net power can be fully charged into the energy storage battery, and the charging power of the energy storage battery is equal to the net power.

[0052] Alternatively, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power of the energy storage battery based on the net power, and then the remaining power is transmitted to the power grid. That is to say, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is less than the net power, the energy storage battery is charged first, and after charging to the upper limit of the energy storage power of the energy storage battery, the charging of the energy storage battery is stopped, and the remaining power of the net power that has not been charged into the energy storage battery is transmitted to the power grid.

[0053] Strategy 2: If the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer and less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure of the hydrogen storage tank is less than the upper pressure limit of the hydrogen storage tank, the fuel cell is controlled to stop operating and the hydrogen production electrolyzer is started to produce hydrogen.

[0054] That is to say, when the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer and less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure of the hydrogen storage tank is less than the upper pressure limit of the hydrogen storage tank, the fuel cell is controlled to remain in a shutdown state, the hydrogen production electrolyzer is started to produce hydrogen, the hydrogen production power of the hydrogen production electrolyzer is equal to the net power, and the net power is consumed by producing hydrogen through the hydrogen production electrolyzer.

[0055] Strategy 3: If the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer, less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure of the hydrogen storage tank is greater than or equal to the upper pressure limit of the hydrogen storage tank, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating.

[0056] And, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is greater than or equal to the net power, the energy storage battery is charged based on the net power, that is, all the net power is charged into the energy storage battery, and the charging power of the energy storage battery is equal to the net power.

[0057] Alternatively, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power of the energy storage battery based on the net power and then the remaining power is transmitted to the power grid, that is, the energy storage battery is charged first, and then the charging of the energy storage battery is stopped after charging to the upper limit of the energy storage power of the energy storage battery, and the remaining power in the net power that has not been charged into the energy storage battery is transmitted to the power grid.

[0058] Strategy 4: If the net power is greater than the maximum operating power of the hydrogen production electrolyzer and the actual pressure of the hydrogen storage tank is less than the upper pressure limit of the hydrogen storage tank, the fuel cell is controlled to stop running and the hydrogen production electrolyzer is started to produce hydrogen.

[0059] And, when the hydrogen production power of the hydrogen production electrolyzer reaches the upper limit of the hydrogen production power, hydrogen production is stopped. The remaining power, i.e., the first power difference, can be absorbed by the energy storage battery and the power grid. The first power difference is the difference between the net power and the upper limit of the hydrogen production power of the hydrogen production electrolyzer.

[0060] When the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is greater than or equal to the first power difference, the energy storage battery is charged based on the first power difference, that is, the first power difference is fully charged into the energy storage battery, and the charging power of the energy storage battery is equal to the first power difference.

[0061] Alternatively, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is less than the first power difference, the energy storage battery is charged to the upper limit of the energy storage power of the energy storage battery based on the first power difference, and then the remaining power is transmitted to the power grid, that is, the energy storage battery is charged first, and then the charging of the energy storage battery is stopped after charging to the upper limit of the energy storage power of the energy storage battery, and the remaining power of the first power difference that has not been charged into the energy storage battery is transmitted to the power grid.

[0062] Strategy 5: If the net power is greater than the maximum operating power of the hydrogen production electrolyzer, and the actual pressure of the hydrogen storage tank is greater than or equal to the upper pressure limit of the hydrogen storage tank, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating.

[0063] And, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is greater than or equal to the net power, the energy storage battery is charged based on the net power, that is, all the net power is charged into the energy storage battery, and the charging power of the energy storage battery is equal to the net power.

[0064] Alternatively, when the difference between the upper limit of the energy storage power of the energy storage battery and the actual energy storage power of the energy storage battery is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power of the energy storage battery based on the net power and then the remaining power is transmitted to the power grid, that is, the energy storage battery is charged first, and then the charging of the energy storage battery is stopped after charging to the upper limit of the energy storage power of the energy storage battery, and the remaining power in the net power that has not been charged into the energy storage battery is transmitted to the power grid.

[0065] In an optional implementation manner of the present application, the second scheduling strategy may include the following strategies:

[0066] Strategy 6: If the absolute value of the net power is less than the maximum operating power of the fuel cell, and the actual pressure of the hydrogen storage tank is greater than the lower pressure limit of the hydrogen storage tank, the hydrogen production electrolyzer is controlled to stop operating, and the fuel cell is started to generate electricity based on the absolute value of the net power. In other words, the net power is compensated by the fuel cell power generation, and the operating power of the fuel cell is equal to the absolute value of the net power.

[0067] Strategy 7: If the absolute value of the net power is less than the maximum operating power of the fuel cell, and the actual pressure of the hydrogen storage tank is less than or equal to the lower pressure limit of the hydrogen storage tank, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, that is, the hydrogen production electrolyzer and the fuel cell are controlled to remain in a shutdown state.

[0068] And, when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is greater than or equal to the absolute value of the net power, the energy storage battery is controlled to discharge, that is, the net power is compensated by discharging the energy storage battery, and the discharge power of the energy storage battery is equal to the absolute value of the net power.

[0069] Alternatively, when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power of the energy storage battery, and then the remaining power is absorbed from the power grid, that is, the energy storage battery is first controlled to discharge, and the energy storage battery is stopped from discharging after discharging to the lower limit of the energy storage power of the energy storage battery, and the remaining power in the net power is absorbed from the power grid.

[0070] Strategy 8: If the absolute value of the net power is greater than the maximum operating power of the fuel cell, and the actual pressure of the hydrogen storage tank is greater than the lower pressure limit of the hydrogen storage tank, the hydrogen production electrolyzer is controlled to stop operating and the fuel cell is started to generate electricity.

[0071] And, after the operating power of the fuel cell reaches the maximum operating power of the fuel cell, the fuel cell is stopped from generating electricity, and when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is greater than or equal to the second power difference, the energy storage battery is controlled to discharge based on the second power difference, that is, the second power difference is compensated by discharging the energy storage battery, and the discharge power of the energy storage battery is equal to the second power difference. The second power difference is the difference between the absolute value of the net power and the maximum operating power of the fuel cell.

[0072] Alternatively, when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is less than the second power difference, the energy storage battery is controlled to discharge to the lower limit of the energy storage power of the energy storage battery, and then the remaining power is absorbed from the power grid, that is, the energy storage battery is first controlled to discharge, and the energy storage battery is stopped from discharging after discharging to the lower limit of the energy storage power of the energy storage battery, and the remaining power in the second power difference is absorbed from the power grid.

[0073] Strategy 9: If the absolute value of the net power is greater than the maximum operating power of the fuel cell and the actual pressure of the hydrogen storage tank is less than the lower pressure limit of the hydrogen storage tank, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating.

[0074] Furthermore, when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is greater than or equal to the absolute value of the net power, the discharge of the energy storage battery is controlled based on the absolute value of the net power, that is, the net power is compensated by the discharge of the energy storage battery, and the discharge power of the energy storage battery is equal to the absolute value of the net power.

[0075] Alternatively, when the difference between the actual energy storage power of the energy storage battery and the lower limit of the energy storage power of the energy storage battery is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power of the energy storage battery, and then the remaining power is absorbed from the power grid, that is, the energy storage battery is first controlled to discharge, and the energy storage battery is stopped from discharging after discharging to the lower limit of the energy storage power of the energy storage battery, and the remaining power in the net power is absorbed from the power grid.

[0076] After simulation analysis, if the dispatch control is carried out according to the dispatch strategy generated by the central coordination control layer on the side, the energy management analysis curve of the hydrogen-electric coupling system can be referred to Figure 4 . Combined Figure 4 It can be seen that during the period of 8:00-20:00, the distributed photovoltaic system in the hydrogen-electricity coupling system is in the photovoltaic power generation stage, and photovoltaic power generation can be used for loads, hydrogen production by hydrogen production electrolyzers, and charging of energy storage batteries. Among them, the hydrogen-electricity coupling system at 8:00 operates according to the above strategy 2; 10:00-14:00 is the stage of hydrogen production by hydrogen production electrolyzers; 10:00-12:00 the hydrogen-electricity coupling system operates according to the above strategy 2; 13:00-14:00 the hydrogen-electricity coupling system operates according to the above strategy 4; 15:00-17:00 the hydrogen-electricity coupling system operates according to the above strategy 5; 18:00-20:00 the hydrogen-electricity coupling system operates according to the above strategies 6, 3, and 8. During the night when there is no photovoltaic power generation, power is supplemented by fuel cells and energy storage batteries. From 21:00 to 24:00 and from 1:00 to 2:00, the hydrogen-electricity coupling system operates according to the above strategy 8; from 3:00 to 7:00, the hydrogen-electricity coupling system operates according to the above strategy 9. By Figure 4 It can be seen that by operating according to the scheduling strategy generated by the central coordination control layer on the side, the hydrogen-electricity coupling system can achieve safe and stable operation with high controllability.

[0077] Optionally, the edge central coordination and control layer may include multiple edge central coordination and control devices. Each edge central coordination and control device may be connected to multiple end-side acquisition and computing architectures in the end-side computing layer to receive operation data reported by the end-side computing layer and issue control instructions to the end-side computing layer, such as scheduling strategies.

[0078] Optionally, the side central coordination and control layer may include multiple modules. Exemplarily, the side central coordination and control layer may include a main program initialization module, a core scheduling module, a synchronous grid-connected cabinet control module, an equipment warning information detection module, an optimization scheduling operation standby module, a debugging and timing module, and a cloud interaction module. It should be understood that the multiple modules included in the side central coordination and control layer can be separately arranged in different side central coordination and control devices, or some or all of the modules can be arranged in the same side central coordination and control device, and this application does not limit this.

[0079] Among them, the main program initialization module can be used to define global variables, receive device information, and send control instructions such as telemetry, telesignaling, remote control and remote adjustment. When initializing the main program, the main program initialization module can establish global variable interlocking in the system to ensure the consistency of global variables during the operation of the main program. During the main program initialization process, the main program initialization module can check the connection status of each device and the occupancy of the communication port when the system starts, and can establish a task scheduling plan for each subsystem.

[0080] The core scheduling module is used to generate scheduling strategies for the hydrogen-electricity coupling system, such as the aforementioned strategies 1-9, etc., and the calculation results of the optimization scheduling program (such as the aforementioned strategies 1-9) are global variables for use by various processes in the main program.

[0081] The synchronous grid-connected cabinet control module is used to control the off-grid / grid-connected mode switching of the hydrogen-electric coupling system. For example, if the interaction power P between the hydrogen-electric coupling system and the grid is grid is zero, it means that the hydrogen-electric coupling system is switched to off-grid mode. grid If P is greater than zero, it means that the hydrogen-electricity coupling system is switched to the grid-connected mode and the hydrogen-electricity coupling system transmits energy to the grid. grid If it is less than zero, it means that the hydrogen-electric coupling system is switched to the grid-connected mode, and the hydrogen-electric coupling system obtains energy from the grid. The control main process of the synchronous grid-connected cabinet is established through the TCP protocol. During the initialization of the main program, the working mode of the off-grid / grid-connected mode switching of the hydrogen-electric coupling system is initialized by initializing the global variables. The working states of the inverter corresponding to the working mode of the hydrogen-electric coupling system include grid-connected mode, constant current mode, standby mode, constant power mode, shutdown, independent inverter mode and voltage drop seamless switching mode.

[0082] The equipment warning information detection module is used to detect the operating status of equipment such as distributed photovoltaic systems, hydrogen production electrolyzers, fuel cells, energy storage batteries, DC / DC converters and inverters, generate corresponding equipment warning information, and report the corresponding fault data to the cloud energy management layer. In addition, the equipment warning information detection module can collect characteristic parameters of fault occurrence, use models such as machine learning or deep learning to achieve fault warning, and facilitate operators to prevent faults in time. In addition, the equipment warning information detection module is different from control processes such as remote signaling, telemetry and remote adjustment, and runs as a separate process. Exemplarily, the equipment warning information generated by the equipment warning information detection module may include energy storage battery (such as lithium battery) warning information, photovoltaic system warning information, fuel cell warning information, hydrogen production electrolyzer warning information, purification system warning information and other hydrogen production equipment warning information.

[0083] The optimization scheduling operation standby module is used to use the main program debugging process as a separate command line interactive process to achieve online adjustment and improve system reliability when the optimization scheduling process stops running due to program errors or other reasons.

[0084] The debugging and timing module is used to control the consistency of communication time within the system, set the timing function of the hydrogen-electric coupling system for the central coordination control layer on the side, and ensure the uniformity of data collection / transmission by adding a timestamp in each received / transmitted communication frame.

[0085] The cloud-side interaction module is used to receive the power generation prediction results and load power prediction results of the distributed photovoltaic system generated by the cloud-side energy management layer, and send the power generation prediction results and load power prediction results to the optimization scheduling operation process of the edge-side central coordination control layer through the TCP / IP protocol. In addition, the cloud-side interaction module is also used to receive the actual energy storage power of the energy storage battery, the actual pressure of the hydrogen storage tank, the alkali solution temperature, the opening state of the hydrogen production electrolyzer, the grid-connected / off-grid state of the hydrogen-electric coupling system and other information reported by the end-side computing layer, and send the actual energy storage power of the energy storage battery, the actual pressure of the hydrogen storage tank, the alkali solution temperature, the opening state of the hydrogen production electrolyzer, the grid-connected / off-grid state of the hydrogen-electric coupling system and other information reported by the end-side computing layer to the optimization scheduling operation process of the edge-side central coordination control layer.

[0086] The cloud-side energy management layer can be used to predict the power generation of the distributed photovoltaic system, and then send the prediction results of the power generation to the edge-side central coordination and control layer, so that the edge-side central coordination and control layer can generate a scheduling strategy based on the prediction results of the power generation. In other words, the edge-side central coordination and control layer can also use the prediction results of the power generation obtained by the cloud-side energy management layer as the actual power generation of the distributed photovoltaic system to generate a scheduling strategy, so that the end-side computing layer can timely schedule and optimize each device in the hydrogen-electricity coupling system to improve the management efficiency of the system.

[0087] The inventors of the present application have found in practical applications and research that photovoltaic power generation is not only related to general meteorological information, but also has a strong correlation with power generation time, cloud information, etc. Therefore, in an embodiment of the present application, when the cloud-side energy management layer predicts the power generation of a distributed photovoltaic system and constructs a power prediction model, it not only considers general meteorological information, but also considers information on power generation time and cloud images that represent cloud information.

[0088] See also Figure 5 , Figure 5 A flow chart of a model training method provided in an embodiment of the present application. Figure 5 As shown, the model training method may include:

[0089] Step S201: Obtain historical data set.

[0090] In combination with the contents of the foregoing embodiments, it can be seen that the cloud-side energy management layer can receive in real time the actual power generation power (which can be referred to as photovoltaic power generation power) of the distributed photovoltaic system reported by the end-side computing layer. In addition, the cloud-side energy management layer can also receive in real time the first meteorological data of the first meteorological parameter reported by the end-side computing layer, the timestamp data of the power generation time, and the cloud layer image data.

[0091] After the cloud-side energy management layer receives the power generation data, first meteorological data, timestamp data and cloud image data reported by the end-side computing layer, it can store the power generation data, first meteorological data, timestamp data and cloud image data as historical data according to the generation time.

[0092] Exemplarily, the first meteorological parameter is a meteorological parameter related to photovoltaic power generation power. Exemplarily, the first meteorological parameter may include parameters such as wind speed, sunlight irradiance, air pressure, temperature, and relative air humidity.

[0093] Afterwards, the cloud-side energy management layer can construct a historical data set based on the historical data stored above. The historical data set may include multiple historical samples, and each historical sample may include multiple data sequences. The sequence length of the data sequence of the historical sample can be set according to the needs of the actual application scenario. For example, if it is necessary to predict the power generation power on a certain day in the future, the sequence length of each data sequence can be set to one day. If it is necessary to predict the power generation power in a certain time period in the future, the sequence length of each data sequence can be set to the length of the time period, etc., and this application does not limit this.

[0094] Exemplarily, the historical data set may include multiple historical samples, each of which may include a first historical text data sequence and a first historical cloud layer image data sequence. The first historical text data sequence may include a first meteorological data sequence of a first meteorological parameter related to photovoltaic power generation, a first power generation data sequence of photovoltaic power generation, and a first timestamp data sequence.

[0095] The first meteorological data sequence is constructed based on the first meteorological data in the historical data. The first power generation data sequence is constructed based on the power generation data in the historical data. The first timestamp data sequence is constructed based on the timestamp data in the historical data. The first historical cloud image data sequence is constructed based on the cloud image data in the historical data.

[0096] Step S202: construct a training data set based on the historical data set.

[0097] After the historical data set is constructed, the cloud-side energy management layer can construct a training data set based on the historical data set.

[0098] The training data set includes a plurality of training samples, each of which includes a second historical text data sequence and a second historical cloud image data sequence. The second historical text data sequence includes a second meteorological data sequence of a second meteorological parameter, a second power generation data sequence, and a second timestamp data sequence. The second meteorological parameter includes a first meteorological parameter in the first meteorological parameter whose correlation with photovoltaic power generation is greater than or equal to a correlation threshold. The second historical cloud image data sequence is generated based on the first historical cloud image data sequence. The second power generation data sequence is generated based on the first power generation data sequence. The second timestamp data sequence is generated based on the first timestamp data sequence.

[0099] In an optional implementation of the present application, constructing a training data set based on a historical data set can be implemented in the following manner:

[0100] The first step is to normalize each data sequence included in the historical data set to obtain a normalized data sequence corresponding to each data sequence included in the historical data set.

[0101] Optionally, the fourth target data sequence in the historical data set may be normalized according to the following formula (1):

[0102]

[0103] In formula (1), x 1 represents any data (or element) in the fourth target data sequence. min(x 1 ) represents the minimum value in the fourth target data sequence. max(x 1 ) represents the maximum value in the fourth target data sequence. 1 Indicates x 1 The data obtained after normalization processing. The fourth target data sequence is any data sequence in the historical data set.

[0104] The second step is to perform outlier clustering processing on each normalized data sequence, remove the outliers in each normalized data sequence, and obtain the outlier-removed data sequence corresponding to each normalized data sequence.

[0105] Optionally, K-means outlier cluster analysis can be performed on each normalized data sequence. Since the similarity between data is inversely proportional to the Euclidean distance, in the K-means outlier cluster analysis, the Euclidean distance can be used to measure the similarity between data. The smaller the calculated Euclidean distance, the higher the similarity between data. Based on this, a distance threshold can be set in advance, and data with a Euclidean distance greater than the distance threshold can be treated as outliers and eliminated. After eliminating the outliers in each normalized data sequence, the mean of the data adjacent to the outlier is used as the filling of the vacant value after eliminating the outlier, thereby obtaining the outlier-eliminated data sequence corresponding to the normalized data sequence.

[0106] Optionally, the Euclidean distance between the data and the cluster center can be calculated according to the following formula (2):

[0107]

[0108] In formula (2), d 1 (x′ 1 ,C a ) represents the data x′ 1 and the ath cluster center C a The Euclidean distance between 1 Indicates x 1 The data obtained after normalization, C a represents the ath cluster center, k 1 Indicates the data dimension, x′ 1r and C ar C a The rth attribute value of .

[0109] In the third step, the outlier-removed data sequences corresponding to the first power generation data sequence, the first timestamp data sequence and the first historical cloud image data sequence are respectively determined as the second power generation data sequence, the second timestamp data sequence and the second historical cloud image data sequence.

[0110] After obtaining the outlier-removed data sequence corresponding to each normalized data sequence, the outlier-removed data sequence corresponding to the first power generation data sequence can be determined as the second power generation data sequence. The outlier-removed data sequence corresponding to the first timestamp data sequence can be determined as the second timestamp data sequence. And, the outlier-removed data sequence corresponding to the first historical cloud layer image data sequence can be determined as the second historical cloud layer image data sequence.

[0111] The fourth step is to use the Pearson correlation analysis method to select the second meteorological data sequence of the second meteorological parameter from the data sequence corresponding to the first meteorological parameter excluding outliers.

[0112] The second meteorological parameter includes the first meteorological parameter whose correlation with the photovoltaic power generation power is greater than or equal to the correlation threshold value. The data sequence corresponding to the second meteorological parameter with the exception value removed is the second meteorological data sequence.

[0113] After obtaining the outlier-removed data sequence corresponding to each normalized data sequence, the Pearson correlation analysis method can be used to screen out the outlier-removed data sequence of the second meteorological parameter with a high correlation with photovoltaic power generation from the outlier-removed data sequence corresponding to the first meteorological parameter as the second meteorological data sequence of the second meteorological parameter. In this way, meteorological factors with a high correlation with photovoltaic power generation can be screened out, and then the power prediction model obtained by training has a higher accuracy.

[0114] Optionally, a Pearson correlation analysis method may be used to calculate the Pearson correlation coefficient between each outlier-removed data sequence corresponding to the first meteorological parameter and the second power generation data sequence. Then, the first meteorological parameter whose Pearson correlation coefficient is greater than the correlation coefficient threshold is determined as the second meteorological parameter, and the outlier-removed data sequence corresponding to the second meteorological parameter is determined as the corresponding second meteorological data sequence.

[0115] Optionally, the Pearson correlation coefficient may be calculated according to the following formula (3):

[0116]

[0117] Wherein, in formula (3), PCC represents the Pearson correlation coefficient, Z represents the target outlier-removed data sequence, and the target outlier-removed data sequence is any outlier-removed data sequence in the outlier-removed data sequence corresponding to the first meteorological parameter; Z s Represents the data at the sth position in the target outlier removal sequence; represents the arithmetic mean of all data in the target sequence with outliers removed; E represents the second power generation data sequence; E s represents the data at the sth position in the second generated power data sequence; represents the arithmetic mean of all data in the second power generation data sequence; s represents the sth position in the target outlier elimination data sequence; and N represents the length of the target outlier elimination data sequence.

[0118] Step 5: construct the training data set based on all second meteorological data sequences, all second power generation data sequences, all second timestamp data sequences and all second historical cloud image data sequences.

[0119] After obtaining all the second meteorological data sequences, all the second power generation data sequences, all the second timestamp data sequences and all the second historical cloud image data sequences, each training sample can be constructed according to the corresponding generation time, and then a training data set can be constructed based on all the training samples.

[0120] Step S203: train the preset prediction model to be trained using the training data set to obtain a trained power prediction model.

[0121] Among them, the power prediction model is used to predict the power generation of distributed photovoltaic systems in hydrogen-electricity coupling systems.

[0122] Exemplarily, the preset prediction model to be trained may include a temporal convolutional network (TCN) model (e.g. Figure 6 The temporal convolutional neural network shown in Figure 1 and the Transformer model. The architecture of the prediction model to be trained can be referenced Figure 6 The model architecture is shown in Figure 2. Figure 6 As shown, the TCN model may include at least one temporal convolution block, and each temporal convolution block may include at least one causal expansion convolution structure (e.g. Figure 6 The dilated causal convolutional layer shown) and a residual connection layer (e.g. Figure 6 The upper residual module output layer shown).

[0123] Optionally, each temporal convolutional block may also include other network architecture layers, such as Figure 6 The weight regularization layer, the rectified linear unit (ReLU) layer and the Dropout layer are shown. It should be understood that each time convolution block may also include other network architecture layers, which is not limited in this application.

[0124] Optionally, the Transformer model may include at least one encoder-decoder block, wherein the encoder and the decoder in each encoder-decoder block include a multi-layer network structure, and each layer of the network structure includes a dot product attention network (e.g. Figure 6 The self-attention mechanism layer shown) and a fully connected feedforward neural network (e.g. Figure 6 The feed-forward network shown).

[0125] Optionally, the Transformer model can also include other network architecture layers, such as Figure 6As shown, the encoder may also be provided with an addition and normalization layer. The decoder may also be provided with an addition and normalization layer, and an encoding / decoding self-attention layer. It should be understood that the Transformer model may also include other network architecture layers, which are not limited in this application.

[0126] Similarly, the prediction model to be trained can also include other network architecture layers, such as Figure 6 The position encoding layer shown in FIG. 1 and the fully connected neural network for regression prediction. It should be understood that the prediction model to be trained may also include Figure 6 There are multiple network architecture layers, and this application does not impose any restrictions on this.

[0127] In an optional implementation of the present application, a pre-set prediction model to be trained is trained using a training data set to obtain a trained power prediction model, which can be implemented in the following manner:

[0128] Each training sample in the training data set is input into the prediction model to be trained in turn in an iterative manner, and the prediction model to be trained is trained until the loss converges and is minimized, thereby obtaining a trained power prediction model.

[0129] In any iteration of training, the target causal dilation convolution structure is used to perform causal dilation convolution processing on the first target data sequence input into the target causal dilation convolution structure according to the following formula (4).

[0130]

[0131] Wherein, in formula (4), F(X) represents the data with position index X in the output sequence obtained after causal expansion convolution processing is performed on the first target data sequence; f represents the convolution kernel (also called filter); m represents the length of the convolution kernel; j represents the position index of the element in the convolution kernel; f(j) represents the jth element in the convolution kernel, that is, the jth weight in the convolution kernel; l represents the expansion coefficient; ψ represents the first target data sequence, and the first target data sequence refers to the data sequence output by the previous layer of the target causal expansion convolution structure, and the target causal expansion convolution structure is any causal expansion convolution structure in any temporal convolution block included in the temporal convolution network model; Ψ X-l×j Represents the data with position index Xl×j in the first target data sequence.

[0132] Through the target residual connection layer, residual connection processing is performed according to the following formula (5).

[0133] D (h) =Activation(P(D (h-1) )+D (h-1) ) (5)

[0134] In formula (5), D (h) represents the data sequence output by the temporal convolution block of the target residual connection layer; Activation represents the residual activation function; D (h-1) Represents the data sequence output by the previous time convolution block of the time convolution block to which the target residual connection layer belongs; P(D (h-1) ) represents the data sequence obtained by processing the data sequence output by the previous time convolution block by the time convolution block to which the target residual connection layer belongs; the target residual connection layer is the residual connection layer in any time convolution block included in the time convolution network model.

[0135] According to the following formulas (6) and (7), the data sequence output by the temporal convolutional network model is position-encoded, and the position code is embedded into the data sequence to obtain a position-encoded data sequence.

[0136] PE(pos,2i)=sin(pos / 10000 2i / n ) (6)

[0137] PE(pos,2i+1)=cos(pos / 10000 2i / n ) (7)

[0138] In formulas (6) and (7), pos represents the position index of the data in the data sequence output by the temporal convolutional network model, i represents the dimension index of the position encoding, and n represents the dimension of the position encoding.

[0139] Through the target dot product attention network, a scaled dot product attention operation is performed on the second target data sequence input into the target dot product attention network according to the following formula (8).

[0140]

[0141] In formula (8), Attention represents the scaled dot product attention operation, Attention(Q, K, V) represents the data sequence output by the scaled dot product attention operation, Q represents the query vector matrix generated after query transformation of the second target data sequence, K represents the key vector matrix generated after key transformation of the second target data sequence; V represents the value vector matrix generated after value transformation of the second target data sequence; K represents the value vector matrix generated after value transformation of the second target data sequence. T represents the transposed matrix of the key vector matrix; represents a scaling factor; softmax represents an activation function; the second target data sequence is a data sequence generated based on the position-encoded data sequence; the target dot-product attention network is any dot-product attention network included in the Transformer model.

[0142] Through the target dot product attention network, the data sequence output by the scaled dot product attention operation is processed by the multi-head attention mechanism according to the following formulas (9) and (10) to generate a multi-head spliced ​​data sequence.

[0143]

[0144] Multihead(Q,K,V)=Concat(head 1 ,…,head q )W o (10)

[0145] Among them, in formulas (9) and (10), head b represents the output sequence of the attention head with index b, b = 1, 2, ..., q, q is a positive integer; Represents the query mapping matrix corresponding to the attention head with index b; Represents the key mapping matrix corresponding to the attention head with index b; represents the value mapping matrix corresponding to the attention head with index b; Multihead(Q,K,V) represents the data sequence of multi-head splicing; W o Represents a linear transformation matrix, which is used to transform the output sequence of q attention heads into the output dimension of the model.

[0146] The data at each position in the third target data sequence input to the target fully connected feedforward neural network is processed according to the following formula (11) through the target fully connected feedforward neural network.

[0147] FFN(x)=max(0,xW 1 +b 1 )W 2 +b 2 (11)

[0148] Wherein, in formula (11), FFN(x) represents the processing result of the target fully connected feedforward neural network on the data x in the third target data sequence; x represents the data at any position in the third target data sequence; max represents the activation function; W 1 and W 2 represents the weight matrix of the linear transformation, b 1 and b 2 represents the bias term; the target fully connected feedforward neural network and the target dot product attention network belong to the same layer of network structure; the third target data sequence is a data sequence generated based on the multi-head splicing data sequence.

[0149] The last decoder of the Transformer model generates a predicted power sequence (e.g. Figure 6After the prediction sequence shown in , a fully connected neural network can be used to globally focus on the sequence, for example Figure 6 The neural network regression prediction process shown generates the predicted value of the power generation of the distributed photovoltaic system.

[0150] After training the power prediction model, the power prediction model can be used to predict the power generation of the distributed photovoltaic system in the hydrogen-electric coupling system. Combined with the content of the aforementioned embodiment, it can be seen that according to the needs of the application scenario, the power prediction model can be used to predict the power generation of the distributed photovoltaic system a day or a day, or the power generation of other time periods in the future can be predicted. As long as the corresponding power prediction model is constructed, the specific implementation method can refer to the content of the aforementioned embodiment, which will not be repeated here.

[0151] Optionally, the cloud-side energy management layer is arranged with a cloud-side energy management system. Exemplarily, the cloud-side energy management system may include multiple cloud devices. It should be understood that the architecture of the cloud-side energy management system can be arranged according to the needs of actual application scenarios, and this application does not limit this.

[0152] Optionally, the cloud-side energy management layer may include multiple modules. Exemplarily, the cloud-side energy management layer may include a hydrogen-electricity coupling system operation module, a system topology analysis module, an economic and benefit analysis module, a system model prediction module, a real-time database module, a fault prevention control module, an operation parameter configuration module, etc. It should be understood that the multiple modules included in the cloud-side energy management layer may be arranged in the same cloud device, or the multiple modules included in the cloud-side energy management layer may also be arranged in different cloud devices respectively, and this application does not limit this.

[0153] Among them, the hydrogen-electricity coupling system operation module can be used for real-time data monitoring, historical data storage, graphical interface services, general chart services, authority management services and system operation alarms.

[0154] The system topology analysis module can be used for topological display of hydrogen-electricity coupling system, as well as multi-energy flow analysis between distributed photovoltaic system, hydrogen production electrolyzer and power conversion unit.

[0155] The economic and benefit analysis module can be used to calculate energy conversion efficiency, carbon emissions, cost-benefit, etc.

[0156] The system model prediction module can be used to train and generate a power prediction model, optimize the power prediction model, and predict the power generation of a distributed photovoltaic system. Figure 5 The power prediction model is trained in the manner shown.

[0157] The real-time database module can be used to meet the multi-concurrent access requirements of the edge central coordination control layer at different time scales, and to send the prediction results of power generation and other scheduling data to the edge central coordination control layer, so that the edge central coordination control layer can generate scheduling strategies based on the prediction results of power generation and other scheduling data.

[0158] Optionally, the multiple modules included in the cloud-side energy management layer can be arranged in the same cloud device, or each of the multiple modules can be arranged in a different cloud device, or some of the multiple modules can be arranged in the same cloud device, and this application does not impose any restrictions on this.

[0159] The energy management system for the hydrogen-electricity coupling system provided in the embodiment of the present application is arranged as a cloud-edge-end architecture, and uses the end-side computing layer to collect the operating data of the hydrogen-electricity coupling system, meteorological data affecting photovoltaic power generation, timestamp data, cloud image data, etc., and reports the collected data to the edge and end sides. The edge-side central coordination control layer generates a scheduling strategy based on the operating data reported by the end side, and sends the scheduling strategy to the end side, so that the end side schedules each device in the hydrogen-electricity coupling system based on the scheduling strategy. The cloud-side energy management layer is used to train and generate a power prediction model based on the data reported by the end side, and predict the power generation of the distributed photovoltaic system, so that the edge side can optimize the scheduling strategy according to the prediction result of the power generation, and realize more efficient energy management of the hydrogen-electricity coupling system.

[0160] Among them, when training and generating the power prediction model, the cloud side uses time features and cloud image features that have a strong correlation with the power generation as the features of the training samples, and the trained power prediction model has higher accuracy. In addition, the pre-set prediction model to be trained adopts a fusion architecture based on the time convolutional network model and the Transformer model, which can simultaneously learn the local power generation features and global time series dependencies in the training samples, thereby improving the prediction accuracy and generalization ability of the model, and further improving the accuracy of power supply prediction.

[0161] When generating the scheduling strategy, the edge side takes the minimum operating power of the hydrogen electrolyzer, the maximum operating power of the hydrogen electrolyzer, and the maximum operating power of the fuel cell as the conditions for scheduling optimization, fully considering the operating safety of the equipment. While managing energy, it also ensures the operating safety of the equipment to a great extent, making energy management more efficient. In addition, the edge side can also perform scheduling optimization based on the cloud side's prediction results of the power supply, further improving the efficiency of energy management.

[0162] It can be understood that the above embodiments are only examples and can be modified in actual implementation. Those skilled in the art can understand that modifications of the above embodiments without creative work fall within the protection scope of the present application and will not be described in detail in the embodiments.

[0163] Based on the same inventive concept, an embodiment of the present application also provides a scheduling optimization method applied to the above-mentioned energy management system, and a model training method applied to the above-mentioned energy management system. Since the principles of solving the problems by the scheduling optimization method and the model training method are similar to those of the aforementioned energy management system for the hydrogen-electricity coupling system, the implementation of the scheduling optimization method and the model training method can refer to the implementation of the aforementioned energy management system for the hydrogen-electricity coupling system, and the repeated parts will not be repeated.

[0164] See also Figure 7 , Figure 7 A flow chart of a scheduling optimization method provided in an embodiment of the present application. Figure 7 As shown, the scheduling optimization method may include:

[0165] Step S301: Obtain the actual power generation power of the distributed photovoltaic system in the hydrogen-electric coupling system, the actual pressure of the hydrogen storage tank, the actual energy storage power and the load power of the energy storage battery through the end-side computing layer of the energy management system, and upload the actual power generation power, actual pressure, actual energy storage power and load power to the edge-side central coordination control layer of the energy management system.

[0166] Step S302: determine the net power between the actual generated power and the load power through the edge central coordination control layer.

[0167] Step S303: When the net power is greater than zero, based on the relationship between the net power and the minimum operating power of the hydrogen-producing electrolyzer in the hydrogen-electricity coupling system, the maximum operating power of the hydrogen-producing electrolyzer, the relationship between the actual pressure and the upper limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the upper limit of the energy storage power of the energy storage battery, generate a first scheduling strategy so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen-producing electrolyzer and the energy storage battery in the hydrogen-electricity coupling system based on the first scheduling strategy to absorb the net power.

[0168] Step S304: When the net power is less than zero, based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure and the lower limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the lower limit of the energy storage power of the energy storage battery, generate a second scheduling strategy so that the end-side computing layer adjusts the operating status of the fuel cell, the hydrogen production electrolyzer and the energy storage battery based on the second scheduling strategy to compensate for the net power.

[0169] In a possible implementation, the first scheduling strategy may include:

[0170] If the net power is less than the minimum operating power of the hydrogen production electrolyzer, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power based on the net power and then the remaining power is transmitted to the power grid;

[0171] If the net power is greater than or equal to the minimum operating power of the hydrogen-producing electrolyzer and less than or equal to the maximum operating power of the hydrogen-producing electrolyzer, and the actual pressure is less than the upper pressure limit, the fuel cell is controlled to stop operating, and the hydrogen-producing electrolyzer is started to produce hydrogen;

[0172] If the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer and less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and then the remaining power is transmitted to the power grid;

[0173] If the net power is greater than the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is less than the pressure upper limit, the fuel cell is controlled to stop operating, the hydrogen production electrolyzer is started to produce hydrogen, and the hydrogen production is stopped after the hydrogen production power of the hydrogen production electrolyzer reaches the upper limit of the hydrogen production power, and when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the first power difference, the energy storage battery is charged based on the first power difference, or when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the first power difference, the energy storage battery is charged to the upper limit of the energy storage power based on the first power difference and then the remaining power is transmitted to the power grid; the first power difference is the difference between the net power and the upper limit of the hydrogen production power;

[0174] If the net power is greater than the maximum operating power of the hydrogen production electrolyzer and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and the remaining power is transmitted to the power grid.

[0175] In a possible implementation, the second scheduling strategy may include:

[0176] If the absolute value of the net power is less than the maximum operating power of the fuel cell and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, and the fuel cell is started to generate electricity based on the net power;

[0177] If the absolute value of the net power is less than the maximum operating power of the fuel cell, and the actual pressure is less than or equal to the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the net power, the energy storage battery is controlled to discharge, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid;

[0178] If the absolute value of the net power is greater than the maximum operating power of the fuel cell, and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, the fuel cell is started to generate electricity, and the fuel cell is stopped to generate electricity after the operating power of the fuel cell reaches the maximum operating power of the fuel cell, and when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to a second power difference, the energy storage battery is controlled to discharge based on the second power difference, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the second power difference, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid; the second power difference is the difference between the absolute value of the net power and the maximum operating power of the fuel cell;

[0179] If the absolute value of the net power is greater than the maximum operating power of the fuel cell and the actual pressure is less than the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the absolute value of the net power, the energy storage battery is controlled to discharge based on the absolute value of the net power, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid.

[0180] In addition, the present application also provides a model training method. The specific content of the model training method can refer to the content of the above embodiments and Figure 5 The model training method shown is not repeated here.

[0181] Although the present application provides method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the actual device or client product is executed, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel (for example, in a parallel processor or multi-threaded processing environment).

[0182] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, devices (systems) or computer program products. Therefore, the embodiments of this specification may take the form of complete hardware embodiments, complete software embodiments or embodiments combining software and hardware. Moreover, the application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0183] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0184] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0185] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0186] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. In this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to the specific circumstances. It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The present application is not limited to any single aspect, nor is it limited to any single embodiment, nor is it limited to any combination and / or permutation of these aspects and / or embodiments. Furthermore, each aspect and / or embodiment of the present application may be used alone or in combination with one or more other aspects and / or embodiments thereof.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and they should all be included in the scope of the present application.

[0188] The present application has been described above in conjunction with preferred embodiments, but these embodiments are only exemplary and serve only as an illustration. On this basis, various replacements and improvements may be made to the present application, all of which fall within the scope of protection of the present application.

Claims

1. An energy management system for a hydrogen-electricity coupling system, characterized in that: include: The edge-side central coordination control layer and the end-side computing layer; The end-side computing layer is used to obtain the actual power generation power of the distributed photovoltaic system, the actual pressure of the hydrogen storage tank, the actual energy storage power and load power of the energy storage battery in the hydrogen-electricity coupling system, and upload the collected actual power generation power, actual pressure, actual energy storage power and load power to the edge-side central coordination control layer; The side central coordination control layer is used for: Determining the net power between the actual generated power and the load power; In the case where the net power is greater than zero, based on the relationship between the net power and the minimum operating power of the hydrogen-producing electrolyzer in the hydrogen-electricity coupling system, the maximum operating power of the hydrogen-producing electrolyzer, the relationship between the actual pressure and the upper limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the upper limit of the energy storage power of the energy storage battery, a first scheduling strategy is generated, so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen-producing electrolyzer, and the energy storage battery in the hydrogen-electricity coupling system based on the first scheduling strategy to absorb the net power; When the net power is less than zero, a second scheduling strategy is generated based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure and the lower limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the lower limit of the energy storage power of the energy storage battery, so that the end-side computing layer adjusts the operating status of the fuel cell, the hydrogen production electrolyzer and the energy storage battery based on the second scheduling strategy to compensate for the net power.

2. The energy management system according to claim 1, characterized in that: The first scheduling strategy includes: If the net power is less than the minimum operating power of the hydrogen production electrolyzer, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power based on the net power and then the remaining power is transmitted to the power grid; If the net power is greater than or equal to the minimum operating power of the hydrogen-producing electrolyzer and less than or equal to the maximum operating power of the hydrogen-producing electrolyzer, and the actual pressure is less than the upper pressure limit, the fuel cell is controlled to stop operating, and the hydrogen-producing electrolyzer is started to produce hydrogen; If the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer and less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and then the remaining power is transmitted to the power grid; If the net power is greater than the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is less than the pressure upper limit, the fuel cell is controlled to stop operating, the hydrogen production electrolyzer is started to produce hydrogen, and the hydrogen production is stopped after the hydrogen production power of the hydrogen production electrolyzer reaches the upper limit of the hydrogen production power, and when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the first power difference, the energy storage battery is charged based on the first power difference, or when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the first power difference, the energy storage battery is charged to the upper limit of the energy storage power based on the first power difference and then the remaining power is transmitted to the power grid; the first power difference is the difference between the net power and the upper limit of the hydrogen production power; If the net power is greater than the maximum operating power of the hydrogen production electrolyzer and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and the remaining power is transmitted to the power grid.

3. The energy management system according to claim 1, characterized in that: The second scheduling strategy includes: If the absolute value of the net power is less than the maximum operating power of the fuel cell and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, and the fuel cell is started to generate electricity based on the net power; If the absolute value of the net power is less than the maximum operating power of the fuel cell, and the actual pressure is less than or equal to the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, in the case where the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the absolute value of the net power, the energy storage battery is controlled to discharge, or in the case where the difference between the actual energy storage power and the lower limit of the energy storage power is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid; If the absolute value of the net power is greater than the maximum operating power of the fuel cell, and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, the fuel cell is started to generate electricity, and the fuel cell is stopped to generate electricity after the operating power of the fuel cell reaches the maximum operating power of the fuel cell, and when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to a second power difference, the energy storage battery is controlled to discharge based on the second power difference, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the second power difference, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid; the second power difference is the difference between the absolute value of the net power and the maximum operating power of the fuel cell; If the absolute value of the net power is greater than the maximum operating power of the fuel cell and the actual pressure is less than the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the absolute value of the net power, the energy storage battery is controlled to discharge based on the absolute value of the net power, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid.

4. The energy management system according to claim 1, characterized in that: The energy management system also includes a cloud-side energy management layer; The cloud-side energy management layer is used to: Acquire a historical data set; the historical data set includes a plurality of historical samples, each of the historical samples includes a first historical text data sequence and a first historical cloud image data sequence; the first historical text data sequence includes a first meteorological data sequence of a first meteorological parameter related to photovoltaic power generation, a first photovoltaic power generation data sequence, and a first timestamp data sequence; Building a training data set based on the historical data set; The training data set includes a plurality of training samples, each of which includes a second historical text data sequence and a second historical cloud image data sequence; the second historical text data sequence includes a second meteorological data sequence of a second meteorological parameter, a second power generation data sequence, and a second timestamp data sequence; the second meteorological parameter includes a first meteorological parameter in the first meteorological parameter whose correlation with photovoltaic power generation is greater than or equal to a correlation threshold; the second historical cloud image data sequence is generated based on the first historical cloud image data sequence; the second power generation data sequence is generated based on the first power generation data sequence; The second timestamp data sequence is generated based on the first timestamp data sequence; The training data set is used to train a pre-set prediction model to be trained to obtain a trained power prediction model; the prediction model to be trained includes a time convolutional network model and a Transformer model; the power prediction model is used to predict the power generation power of the distributed photovoltaic system.

5. The energy management system according to claim 4, characterized in that: The first meteorological parameters include: wind speed, sunlight irradiance, air pressure, temperature, and relative air humidity.

6. The energy management system according to claim 4 or 5, characterized in that: The cloud-side energy management layer is used to construct a training data set based on the historical data set, including: the cloud-side energy management layer is used to: Normalizing each data sequence included in the historical data set to obtain a normalized data sequence corresponding to each data sequence included in the historical data set; Perform outlier clustering processing on each normalized data sequence, remove outliers in each normalized data sequence, and obtain an outlier-removed data sequence corresponding to each normalized data sequence; Determine the outlier-removed data sequences corresponding to the first power generation data sequence, the first timestamp data sequence, and the first historical cloud layer image data sequence as the second power generation data sequence, the second timestamp data sequence, and the second historical cloud layer image data sequence; Using a Pearson correlation analysis method, a second meteorological data sequence of the second meteorological parameter is selected from the outlier-removed data sequence corresponding to the first meteorological parameter; The training data set is constructed based on all second meteorological data sequences, all second power generation data sequences, all second timestamp data sequences and all second historical cloud layer image data sequences.

7. The energy management system according to claim 4 or 5, characterized in that: The temporal convolutional network model includes at least one temporal convolutional block, each of which includes at least one causal dilation convolution structure and a residual connection layer; the Transformer model includes at least one encoder-decoder block, and the encoder and decoder in each encoder-decoder block include a multi-layer network structure, and each layer of the network structure includes a dot product attention network and a fully connected feedforward neural network; The cloud-side energy management layer is used to train a preset prediction model to be trained using the training data set to obtain a trained power prediction model, including: the cloud-side energy management layer is used to: Inputting each training sample in the training data set into the prediction model to be trained in turn in an iterative manner, training the prediction model to be trained until the loss reaches convergence and is minimized, thereby obtaining a trained power prediction model; Wherein, in any iteration process, through the target causal dilation convolution structure, according to the following formula, the first target data sequence input into the target causal dilation convolution structure is subjected to causal dilation convolution processing; Wherein, F(X) represents the data with position index X in the output sequence obtained after performing causal expansion convolution processing on the first target data sequence; f represents the convolution kernel; m represents the length of the convolution kernel; j represents the position index of the element in the convolution kernel; f(j) represents the jth element in the convolution kernel; l represents the expansion coefficient; ψ represents the first target data sequence, which refers to the data sequence output by the previous layer of the target causal expansion convolution structure, and the target causal expansion convolution structure is any causal expansion convolution structure in any temporal convolution block included in the temporal convolution network model; Ψ X-l×j represents data with position index Xl×j in the first target data sequence; Through the target residual connection layer, residual connection processing is performed according to the following formula; D (h) =Activation(P(D (h-1) )+D (h-1) ); Among them, D (h) represents the data sequence output by the temporal convolution block of the target residual connection layer; Activation represents the residual activation function; D (h-1) represents the data sequence output by the previous time convolution block of the time convolution block to which the target residual connection layer belongs; P(D (h-1) ) represents a data sequence obtained by processing a data sequence output by a previous time convolution block by the time convolution block to which the target residual connection layer belongs; the target residual connection layer is a residual connection layer in any time convolution block included in the time convolution network model; According to the following formula, the data sequence output by the temporal convolutional network model is position-encoded, and the position code is embedded into the data sequence to obtain a position-encoded data sequence; PE(pos,2i)=sin(pos / 10000 2i / n ); PE(pos,2i+1)=cos(pos / 10000 2i / n ); Among them, pos represents the position index of the data in the data sequence output by the temporal convolutional network model, i represents the dimension index of the position encoding, and n represents the dimension of the position encoding; Through the target dot product attention network, a scaled dot product attention operation is performed on the second target data sequence input into the target dot product attention network according to the following formula; Wherein, Attention represents a scaled dot product attention operation, Attention(Q, K, V) represents a data sequence output by the scaled dot product attention operation, Q represents a query vector matrix generated after query transformation of the second target data sequence, K represents a key vector matrix generated after key transformation of the second target data sequence; V represents a value vector matrix generated after value transformation of the second target data sequence; K T represents the transposed matrix of the key vector matrix; represents a scaling factor; softmax represents an activation function; the second target data sequence is a data sequence generated based on the position-encoded data sequence; the target dot product attention network is any dot product attention network included in the Transformer model; Through the target dot product attention network, the data sequence output by the scaled dot product attention operation is processed by the multi-head attention mechanism according to the following formula to generate a multi-head spliced ​​data sequence; Multihead(Q,K,V)=Concat(head1,…,head q )W o ; Among them, head b represents the output sequence of the attention head with index b, b = 1, 2, ..., q, q is a positive integer; Represents the query mapping matrix corresponding to the attention head with index b; Represents the key mapping matrix corresponding to the attention head with index b; represents the value mapping matrix corresponding to the attention head with index b; Multihead(Q,K,V) represents the multi-head concatenated data sequence; W o represents the linear transformation matrix, which is used to convert the output sequence of q attention heads into the output dimension of the model; By means of a target fully connected feedforward neural network, the data at each position in the third target data sequence input into the target fully connected feedforward neural network is processed according to the following formula; FFN(x)=max(0,xW1+b1)W2+b2; Among them, FFN(x) represents the processing result of the target fully connected feedforward neural network on the data x in the third target data sequence; x represents the data at any position in the third target data sequence; max represents the activation function; W1 and W2 represent the weight matrices of the linear transformation, b1 and b2 represent the bias terms; the target fully connected feedforward neural network and the target dot product attention network belong to the same layer network structure; the third target data sequence is a data sequence generated based on the multi-head spliced ​​data sequence.

8. A scheduling optimization method, characterized in that: Applied to the energy management system according to any one of claims 1 to 7, the method comprising: The actual power generation power of the distributed photovoltaic system in the hydrogen-electricity coupling system, the actual pressure of the hydrogen storage tank, the actual energy storage power and the load power of the energy storage battery are obtained through the end-side computing layer of the energy management system, and the actual power generation power, actual pressure, actual energy storage power and load power are uploaded to the edge-side central coordination control layer of the energy management system; Determine the net power between the actual generated power and the load power through the edge central coordination control layer; and In the case where the net power is greater than zero, based on the relationship between the net power and the minimum operating power of the hydrogen-producing electrolyzer in the hydrogen-electricity coupling system, the maximum operating power of the hydrogen-producing electrolyzer, the relationship between the actual pressure and the upper limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the upper limit of the energy storage power of the energy storage battery, a first scheduling strategy is generated, so that the end-side computing layer adjusts the operating states of the fuel cell, the hydrogen-producing electrolyzer, and the energy storage battery in the hydrogen-electricity coupling system based on the first scheduling strategy to absorb the net power; When the net power is less than zero, a second scheduling strategy is generated based on the relationship between the net power and the maximum operating power of the fuel cell, the relationship between the actual pressure and the lower limit of the pressure of the hydrogen storage tank, and the relationship between the actual energy storage power and the lower limit of the energy storage power of the energy storage battery, so that the end-side computing layer adjusts the operating status of the fuel cell, the hydrogen production electrolyzer and the energy storage battery based on the second scheduling strategy to compensate for the net power.

9. The method according to claim 8, characterized in that The first scheduling strategy includes: If the net power is less than the minimum operating power of the hydrogen production electrolyzer, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the net power, the energy storage battery is charged to the upper limit of the energy storage power based on the net power and then the remaining power is transmitted to the power grid; If the net power is greater than or equal to the minimum operating power of the hydrogen-producing electrolyzer and less than or equal to the maximum operating power of the hydrogen-producing electrolyzer, and the actual pressure is less than the upper pressure limit, the fuel cell is controlled to stop operating, and the hydrogen-producing electrolyzer is started to produce hydrogen; If the net power is greater than or equal to the minimum operating power of the hydrogen production electrolyzer and less than or equal to the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or, when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and then the remaining power is transmitted to the power grid; If the net power is greater than the maximum operating power of the hydrogen production electrolyzer, and the actual pressure is less than the pressure upper limit, the fuel cell is controlled to stop operating, the hydrogen production electrolyzer is started to produce hydrogen, and the hydrogen production is stopped after the hydrogen production power of the hydrogen production electrolyzer reaches the upper limit of the hydrogen production power, and when the difference between the upper limit of the energy storage power and the actual energy storage power is greater than or equal to the first power difference, the energy storage battery is charged based on the first power difference, or when the difference between the upper limit of the energy storage power and the actual energy storage power is less than the first power difference, the energy storage battery is charged to the upper limit of the energy storage power based on the first power difference and then the remaining power is transmitted to the power grid; the first power difference is the difference between the net power and the upper limit of the hydrogen production power; If the net power is greater than the maximum operating power of the hydrogen production electrolyzer and the actual pressure is greater than or equal to the pressure upper limit, the fuel cell and the hydrogen production electrolyzer are controlled to stop operating, and, when the difference between the energy storage power upper limit and the actual energy storage power is greater than or equal to the net power, the energy storage battery is charged based on the net power, or when the difference between the energy storage power upper limit and the actual energy storage power is less than the net power, the energy storage battery is charged to the energy storage power upper limit based on the net power and the remaining power is transmitted to the power grid.

10. The method according to claim 8, characterized in that The second scheduling strategy includes: If the absolute value of the net power is less than the maximum operating power of the fuel cell and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, and the fuel cell is started to generate electricity based on the net power; If the absolute value of the net power is less than the maximum operating power of the fuel cell, and the actual pressure is less than or equal to the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the net power, the energy storage battery is controlled to discharge, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid; If the absolute value of the net power is greater than the maximum operating power of the fuel cell, and the actual pressure is greater than the lower pressure limit, the hydrogen production electrolyzer is controlled to stop operating, the fuel cell is started to generate electricity, and the fuel cell is stopped to generate electricity after the operating power of the fuel cell reaches the maximum operating power of the fuel cell, and when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to a second power difference, the energy storage battery is controlled to discharge based on the second power difference, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the second power difference, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid; the second power difference is the difference between the absolute value of the net power and the maximum operating power of the fuel cell; If the absolute value of the net power is greater than the maximum operating power of the fuel cell and the actual pressure is less than the lower pressure limit, the hydrogen production electrolyzer and the fuel cell are controlled to stop operating, and, when the difference between the actual energy storage power and the lower limit of the energy storage power is greater than or equal to the absolute value of the net power, the energy storage battery is controlled to discharge based on the absolute value of the net power, or when the difference between the actual energy storage power and the lower limit of the energy storage power is less than the absolute value of the net power, the energy storage battery is controlled to discharge to the lower limit of the energy storage power and then absorb the remaining power from the power grid.

11. A model training method, characterized in that: Applied to the energy management system according to any one of claims 1 to 7, the method comprising: Acquire a historical data set; the historical data set includes a plurality of historical samples, each of the historical samples includes a first historical text data sequence and a first historical cloud image data sequence; the first historical text data sequence includes a first meteorological data sequence of a first meteorological parameter related to photovoltaic power generation, a first photovoltaic power generation data sequence, and a first timestamp data sequence; A training data set is constructed based on the historical data set; the training data set includes multiple training samples, each of which includes a second historical text data sequence and a second historical cloud image data sequence; the second historical text data sequence includes a second meteorological data sequence of a second meteorological parameter, a second power generation data sequence, and a second timestamp data sequence; the second meteorological parameter includes a first meteorological parameter in the first meteorological parameter whose correlation with photovoltaic power generation is greater than or equal to a correlation threshold; the second historical cloud image data sequence is generated based on the first historical cloud image data sequence; the second power generation data sequence is generated based on the first power generation data sequence; the second timestamp data sequence is generated based on the first timestamp data sequence; The training data set is used to train a pre-set prediction model to be trained to obtain a trained power prediction model; the prediction model to be trained includes a time convolutional network model and a Transformer model; the power prediction model is used to predict the power generation power of the distributed photovoltaic system.

Images