Photovoltaic energy storage energy management system

Abstract

The invention relates to the technical field of energy management, in particular to a photovoltaic energy storage energy management system, which comprises a working condition acquisition module used for acquiring environment irradiance and output power data of a photovoltaic array, charge state and temperature data of an energy storage unit and power consumption data of a local load according to a preset acquisition period, packaging the data into a periodic working condition data set; and the power prediction module is used for receiving the periodic working condition data set. According to the invention, through a power prediction module, collected power consumption and photovoltaic output data are calculated as a system net load, a net load time sequence stored according to a time sequence is constructed, and an expected net load curve of a next scheduling period is extrapolated by adopting a time sequence prediction model; and the curve is provided for an energy storage response equalization module to carry out multi-objective optimization solution, so that an optimal initial charging and discharging instruction is solved in advance, and the problems of power matching misalignment and poor economical efficiency caused by control lag are solved.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, specifically a photovoltaic energy storage management system. Background Technology

[0002] In photovoltaic energy storage systems, the energy management system is the core for achieving efficient, safe, and economical operation. It requires real-time monitoring of the photovoltaic array's power generation, the energy storage unit's state of charge and temperature, as well as the real-time power consumption of the local load and the interaction with the power grid. Traditional energy management strategies are typically based on simple threshold control or instantaneous power balancing logic. For example, when the photovoltaic power generation exceeds the load power consumption, the excess electricity is used to charge the energy storage unit; when the load power consumption exceeds the photovoltaic power generation, the energy storage unit discharges to compensate. This control logic has many problems under complex operating conditions.

[0003] For example, Chinese Patent Publication No. CN119482566A discloses a control method, terminal and storage medium for a microgrid system, which includes: acquiring the energy storage capacity (such as the SOC value) of the energy storage subsystem; and setting a preset energy threshold. For example, when the energy storage capacity is less than the first preset energy threshold, the photovoltaic power generation system is controlled to start working. This method mainly focuses on the overcharge and over-discharge boundary management of the energy storage unit itself, which is a basic threshold safety control.

[0004] For example, Chinese Patent Publication No. CN107222013B discloses an energy control method for an independent photovoltaic hybrid energy storage system. This method uses hybrid energy storage (batteries and supercapacitors) and dynamically allocates power based on the state of charge (SOC) partitioning of the energy storage to smooth out power fluctuations in photovoltaics. This method is a smoothing strategy based on the current state of the system.

[0005] As illustrated in the examples above, existing technologies have achieved energy storage protection and power smoothing to some extent by managing the threshold of the energy storage state of charge or compensating for instantaneous fluctuations in grid-connected power. However, these technologies do not fully consider the coordinated optimization among multiple variables within the system, such as photovoltaic output, load power consumption, and energy storage state of charge. Their control logic is passive and lagging, lacking prediction of changes in photovoltaic output and load power consumption in the next scheduling cycle. This results in the energy storage unit's charging and discharging strategy not being globally optimal, often merely responding to events that have already occurred rather than planning energy flow in advance. Consequently, this leads to problems such as low energy scheduling efficiency, insufficient constraints on the ramp-up rate of energy storage units, and poor overall system operating economy. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: a photovoltaic energy storage management system, comprising:

[0007] The operating condition acquisition module is used to collect environmental irradiance and output power data of the photovoltaic array, state of charge and temperature data of the energy storage unit, and power consumption data of the local load according to a preset acquisition cycle, and package the data into a periodic operating condition dataset.

[0008] The power prediction module is used to receive the periodic operating condition dataset, extract power consumption data and photovoltaic output power data from it, calculate the current system net load, and generate the expected net load curve for the next scheduling cycle based on the changing trend of the net load.

[0009] The energy storage response balancing module is used to receive the expected net load curve, extract the current state of charge of the energy storage unit from the periodic operating condition dataset, and calculate the initial charge and discharge power command required to balance the expected net load.

[0010] The power smoothing constraint module is used to analyze the initial charge and discharge power command, constrain instantaneous requests in the command that exceed the preset ramp rate or the preset energy storage safety threshold, and generate a smoothed charge and discharge strategy.

[0011] The energy flow topology control module is used to decompose the smoothed charging and discharging strategy into specific control timing sequences for the photovoltaic inverter, energy storage converter and grid interface, and generate a set of coordinated operation instructions for multiple devices.

[0012] The closed-loop correction execution module is used to send the collaborative operation instruction set to each device, and obtain the actual power feedback value of each device after execution within the current scheduling cycle. Based on the deviation between the feedback value and the smoothed charging and discharging strategy, the initial parameters calculated in the next scheduling cycle are adjusted.

[0013] Preferably, the specific implementation of the operating condition acquisition module includes:

[0014] The ambient irradiance, output power data, state of charge, temperature data, and power consumption data are collected by polling from the photovoltaic inverter, battery management system, and smart meter via an industrial bus.

[0015] Add a uniform timestamp to all collected data and verify the data. If the data packet checksum is incorrect or the data value exceeds the preset physical range, mark it as invalid data and enable neighboring data interpolation.

[0016] All data that pass verification within the same acquisition cycle and have the same timestamp are combined into a complete periodic operating condition dataset and forwarded to the power prediction module.

[0017] Preferably, the power prediction module is implemented in the following ways:

[0018] After receiving the periodic operating condition dataset, the system net load is calculated, whereby the net load is the power consumption data minus the photovoltaic output power data.

[0019] The calculated current system net load value is constructed into a net load time series stored in chronological order;

[0020] A pre-trained time-series prediction model is used to extrapolate the net load time series to generate an expected net load curve covering the next scheduling cycle and containing multiple time steps.

[0021] Preferably, the method for generating the expected net load curve further includes:

[0022] When extrapolating the net load time series, the environmental irradiance data in the periodic operating condition dataset is also introduced as an additional input variable for the time series prediction model.

[0023] When a sharp drop in environmental irradiance data is detected within a short period of time, the time-series prediction model responds to the sharp drop by increasing the influence weight of the additional input variables to immediately correct the shape of the expected net load curve.

[0024] Preferably, the specific implementation of the energy storage response balancing module includes:

[0025] Receive the expected net load curve and extract the current state of charge and temperature data of the energy storage unit from the periodic operating condition dataset;

[0026] Based on the temperature data, a target range of state of charge is determined from a preset lookup table or function;

[0027] Using the expected net load curve and the target range of state of charge as input, a multi-objective optimization solution is performed. The main objective is to minimize the power grid interaction, while the constraints are that the energy storage unit does not exceed the upper and lower limits of the safe state of charge and the safe temperature threshold. The initial charge and discharge power command required at each time step within the scheduling cycle is calculated.

[0028] Preferably, the power smoothing constraint module is implemented in the following ways:

[0029] Receive the initial charge and discharge power command sequence, and obtain the current maximum available charge and discharge rate of the energy storage unit and the preset ramp rate from the battery management system;

[0030] The initial charge and discharge power command is checked point by point, compared with the command at the previous moment, and the instantaneous power change rate is calculated.

[0031] When the instantaneous power change rate exceeds the preset ramp rate, or when the absolute value of the command exceeds the maximum available charge / discharge rate, the command point is corrected and limited to the boundary of the preset ramp rate or the maximum rate, thereby generating a smooth charge / discharge strategy.

[0032] Preferably, the implementation method for correcting the instruction point further includes:

[0033] When the instantaneous power change rate is limited to the preset ramp rate boundary, the power demand exceeding the limit is allocated to one or more subsequent scheduling time steps in the future on the time axis.

[0034] Ensure that the total charge and discharge energy of the smoothed charge and discharge strategy remains consistent with the total energy required by the initial charge and discharge power command throughout the entire scheduling cycle.

[0035] Preferably, the specific implementation of the energy flow topology control module includes:

[0036] The smoothed charge / discharge strategy is received, which defines the target power of the energy storage unit at each time step;

[0037] Based on the current operating mode of the system, which includes grid-connected mode and off-grid mode, the smoothed charging and discharging strategy is decomposed into the active power derating command of the photovoltaic inverter, the constant power charging and discharging command of the energy storage converter, and the on / off logic command of the static switching switch of the grid interface.

[0038] The multiple instructions are encoded according to their execution order and security dependencies, and packaged into a collaborative execution instruction set containing timing information.

[0039] Preferably, the specific implementation of the closed-loop correction execution module includes:

[0040] The coordinated operation instruction set is sent to the corresponding photovoltaic inverter, energy storage converter and static switching switch via the control bus;

[0041] After the instruction is executed, the actual output power feedback value is read back from the power meters of each device;

[0042] Calculate the deviation between the actual power feedback value and the target value of the smoothed charge-discharge strategy at that moment;

[0043] The deviation value is used as a correction factor and input into the energy storage response balancing module in the next round to dynamically adjust the power grid interaction weight in the multi-objective optimization solution in order to compensate for system response lag.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] (1) The present invention uses a power prediction module to calculate the collected power consumption and photovoltaic output data as the system net load and constructs a net load time series stored in time order. The time series prediction model is used to extrapolate the expected net load curve for the next scheduling cycle and provides the curve to the energy storage response balancing module for multi-objective optimization solution, so as to calculate the optimal initial charging and discharging command in advance. This makes energy scheduling change from the passive response of the existing technology to active pre-planning, and fundamentally solves the problems of power matching inaccuracy and poor economy caused by control lag.

[0046] (2) The present invention uses a power smoothing constraint module to check whether the instantaneous power change rate of the command exceeds the preset ramp rate after receiving the initial charge and discharge power command. When the limit is exceeded, the excess power demand is allocated to one or more future scheduling time steps. This ensures the executability of the command and avoids the impact and lifespan decay of the energy storage unit caused by high-rate instantaneous power changes through the redistribution of energy on the time axis. This achieves the unity of power smoothing and energy conservation.

[0047] (3) The present invention obtains the actual power feedback value of each device after the instruction is issued through the closed-loop correction execution module, calculates the deviation between the feedback value and the target value of the smoothed charging and discharging strategy, and uses the deviation value as a correction factor to input to the energy storage response balancing module in the next round, so as to dynamically adjust the weight parameters in the multi-objective optimization solution, thereby forming a cross-cycle adaptive correction closed loop, enabling the system to learn and compensate for the response lag caused by device aging or model error, thereby continuously optimizing the accuracy of the system model and ensuring high-precision control under long-term operation. Attached Figure Description

[0048] Figure 1 This is a system timing diagram of a photovoltaic energy storage management system according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 The present invention provides a technical solution: a photovoltaic energy storage management system, comprising:

[0051] The operating condition acquisition module is used to collect environmental irradiance and output power data of the photovoltaic array, state of charge and temperature data of the energy storage unit, and power consumption data of the local load according to a preset acquisition cycle, and package the data into a periodic operating condition dataset.

[0052] The power prediction module is used to receive periodic operating condition datasets, extract power consumption data and photovoltaic output power data from them, calculate the current system net load, and generate the expected net load curve for the next scheduling cycle based on the changing trend of the net load.

[0053] The energy storage response balancing module is used to receive the expected net load curve, extract the current state of charge of the energy storage unit from the periodic operating condition dataset, and calculate the initial charge and discharge power command required to balance the expected net load.

[0054] The power smoothing constraint module is used to analyze the initial charge and discharge power command, constrain instantaneous requests in the command that exceed the preset ramp rate or the preset energy storage safety threshold, and generate a smoothed charge and discharge strategy.

[0055] The energy flow topology control module is used to decompose the smoothed charging and discharging strategy into specific control timing sequences for photovoltaic inverters, energy storage converters and grid interfaces, and generate a set of coordinated operation instructions for multiple devices.

[0056] The closed-loop correction execution module is used to issue a set of collaborative operation instructions to each device and obtain the actual power feedback value of each device after execution within the current scheduling cycle. Based on the deviation between the feedback value and the smoothed charging and discharging strategy, the initial parameters calculated for the next scheduling cycle are adjusted.

[0057] In this invention, the periodic operating condition dataset generated by the operating condition acquisition module is the data foundation for the entire energy flow optimization. This dataset is first processed by the power prediction module to generate the expected net load curve for future net load. This step realizes the transformation from passive response to active prediction. Subsequently, the expected net load curve is transmitted to the energy storage response balancing module to calculate the theoretically optimal initial charge and discharge power command. This command is then subject to practical constraints by the power smoothing constraint module, especially the constraint on the ramp rate, to generate the final safe and smooth charge and discharge strategy. This series of data processing and transmission constitutes the core decision chain from prediction to planning to constraint, and finally achieves accurate control and iterative optimization at the physical level through the energy flow topology control module and the closed-loop correction execution module.

[0058] It should be noted that the operating data collected in the current scenario includes, but is not limited to, photovoltaic inverters, battery management systems, and smart meters. These devices provide key data required for energy management, such as ambient irradiance, photovoltaic output power, state of charge of energy storage units, temperature, and power consumption data of local loads. The system processes these multi-dimensional data collaboratively and ultimately outputs a set of collaborative operation instructions for energy storage converters and grid interfaces to achieve efficient and economical operation of the entire photovoltaic energy storage system.

[0059] In one optional embodiment, the specific implementation of the working condition acquisition module includes:

[0060] The system uses an industrial bus to collect ambient irradiance, output power data, state of charge, temperature data, and power consumption data from the photovoltaic inverter, battery management system, and smart meter, respectively.

[0061] Specifically, the industrial bus can be a combination of one or more physical media and protocols, such as Modbus-RTU (based on RS485) or Modbus-TCP (based on Ethernet). In the energy management system of this invention, the operating condition acquisition module acts as the communication master station, while the photovoltaic inverter, battery management system and smart meter act as communication slave stations. Polling acquisition refers to the operating condition acquisition module sending read instructions to the preset register addresses of each slave station in sequence according to a preset acquisition period (e.g., set to 1 second) to obtain their latest operating data.

[0062] All collected data are annotated with a uniform timestamp and the data is verified. If the data packet checksum is incorrect or the data value exceeds the preset physical range, it is marked as invalid data and neighboring data interpolation is enabled.

[0063] Specifically, the unified timestamp does not refer to the time reported by each device (inverter, BMS, etc.), as their clocks may be out of sync. Instead, it refers to the time when the operating condition acquisition module (EMS master controller) initiates this round of polling, or after receiving all response data, when it uses the EMS's own high-precision system clock (e.g., a clock synchronized via the NTP network time protocol) as a global timestamp to mark all data (irradiance, power, SOC, temperature, power consumption) collected within this cycle. Data verification includes two levels: the first is communication layer verification, i.e., data packet checksum and error checking. For example, in Modbus-RTU communication, the cyclic redundancy check (CRC) code of the received data frame is checked for correctness. If the CRC check fails, it means that the data has been corrupted during transmission, and the data point (e.g., the current SOC value) is discarded. The second is physical layer verification, i.e., data values ​​exceeding the preset physical range. This is based on the knowledge of those skilled in the art regarding the design. The system recognizes the boundaries set by the system. For example, the SOC value reported by the BMS must theoretically be between 0% and 100% (which can be relaxed to a tolerance range of -5% to 105%), and the temperature data must be between -40℃ and 85℃. If the module receives a value of SOC=200% or temperature=500℃, it is judged as a sensor fault or data anomaly, and the data point is marked as invalid data. After the data is marked as invalid data, in order to ensure that the downstream power prediction module (usually the algorithm model) will not crash due to receiving null values ​​or extreme abnormal values, the system must enable neighbor interpolation. The simplest and most robust interpolation method is to keep the previous value, that is, to directly use the valid data of the previous acquisition cycle to fill the current invalid data point. For example, if the SOC value of T=10:00:01 is invalid due to CRC check failure, the system will automatically fill the valid SOC value of T=10:00:00 (such as 65.5%) into the BMS data field of the current cycle.

[0064] All data that pass verification within the same acquisition cycle and have the same timestamp are combined into a complete periodic operating condition dataset and forwarded to the power prediction module.

[0065] Specifically, combining data into a complete periodic operating condition dataset means constructing a structured data body or data snapshot in the system's memory from all the data that has undergone verification and interpolation and has been assigned the same global timestamp (e.g., T=10:00:01). This snapshot ensures that at the moment T=10:00:01, all states of photovoltaic, energy storage, and load are simultaneously aligned. Subsequently, this periodic operating condition dataset is forwarded, for example, pushed into a first-in-first-out data queue, waiting for the power prediction module to read and consume it from the queue as input to its prediction algorithm.

[0066] In one optional embodiment, the power prediction module is specifically implemented in the following ways:

[0067] After receiving the periodic operating condition dataset, the system net load is calculated, which is the power consumption data minus the photovoltaic output power data.

[0068] Specifically, in the system net load calculation, this calculation is performed in real time, cycle by cycle. That is, the power prediction module immediately performs a calculation every time it receives a periodic operating condition dataset from the operating condition acquisition module (e.g., every 1 second). The calculation formula is: System Net Load = Power Consumption Data of Local Load - Output Power Data of Photovoltaic Array, where:

[0069] When the net load is positive (>0), it means that the power consumption of the local load is greater than the power generation of the photovoltaic system, and the system is in a "power deficit" state, requiring energy storage discharge or grid supplementation.

[0070] When the net load is negative (<0), it means that the photovoltaic power generation is greater than the power consumption of the local load, the system is in a state of power surplus, and the excess energy can be used for energy storage charging.

[0071] The calculated current system net load value is constructed into a net load time series stored in chronological order.

[0072] Specifically, the time series does not refer to a single current net load value, but rather a historical data window used to describe the recent load characteristics of the system. The power prediction module internally maintains a fixed-length first-in-first-out data buffer (or sliding window). For example, this buffer is configured to store net load data for the past hour. When the current system net load value calculated in the system net load calculation (e.g., the value at T=10:00:01) is obtained, this value is pushed into the latest end of the buffer. At the same time, the data at the oldest end of the buffer (e.g., the value at T=09:00:01) is popped and discarded. In this way, the buffer always dynamically maintains the latest net load time series that best reflects the current operating conditions.

[0073] A pre-trained time-series prediction model is used to extrapolate the net load time series and generate an expected net load curve covering the next scheduling cycle and containing multiple time steps.

[0074] Specifically, pre-training means that the model has been fully trained offline using a large amount of historical operating data (e.g., the operating data of the site over the past year), so that it has learned the typical variation patterns of photovoltaic and load at the site under different times (e.g., morning, noon, evening) and different conditions (e.g., weekdays, weekends, seasons).

[0075] The time series prediction model itself can be, but is not limited to, the following:

[0076] Statistical models, such as the autoregressive integral moving average model or its seasonal variant, the seasonal autoregressive moving average model, are adept at capturing the inherent periodicity in the data.

[0077] Machine learning models: such as support vector regression;

[0078] Deep learning models, such as Long Short-Term Memory networks or gated recurrent units, are variants of recurrent neural networks specifically designed to process and predict time-series data. They are able to effectively learn long-term dependencies in the data and are well-suited for predicting complex fluctuations in photovoltaics and loads.

[0079] It should be noted that the specific steps for extrapolating the net load time series are as follows: the entire historical data window constructed in the previous step (e.g., 3600 data points in the past hour) is taken as input and fed into a selected pre-trained time series prediction model (e.g., a long short-term memory network model). The output of the model is the expected net load curve, which covers the next scheduling cycle (e.g., the next 15 minutes) and includes multiple time steps (e.g., the net load value predicted by the model every minute in the next 15 minutes). Finally, this curve (i.e., a time series containing 15 predicted values) is generated and immediately forwarded to the energy storage response balancing module.

[0080] In an alternative embodiment, the generation of the expected net load curve further includes:

[0081] When extrapolating the net load time series, environmental irradiance data from the periodic operating condition dataset is also introduced as an additional input variable for the time series prediction model.

[0082] Specifically, the input structure of the net load time series has changed here. In the original model, the input at each time step t may only be a single value: [NetLoad_t]. However, in this embodiment, the input at each time step t is expanded into a feature vector, which contains at least two dimensions: [NetLoad_t, Irradiance_t]. Correspondingly, the pre-training mentioned above refers to the fact that the model (e.g., a long short-term memory network or a gated recurrent unit) is trained from the beginning on this multivariate input structure (e.g., simultaneously inputting the net load sequence and irradiance sequence of the past hour).

[0083] When a sharp drop in environmental irradiance data is detected within a short period of time, the time-series forecasting model responds to the sharp drop by increasing the influence weight of additional input variables in order to immediately correct the shape of the expected net load curve.

[0084] Specifically, in the step of detecting a sharp drop in ambient irradiance data within a short period of time, a real-time anomaly detection trigger is defined. This detection is not performed by the time series prediction model itself, but by a parallel monitoring unit in the power prediction module. The sharp drop is quantified by calculating the first derivative (rate of change) of the ambient irradiance data. For example, the monitoring unit is configured with a threshold: when the average rate of decrease of ambient irradiance exceeds 50 W / m² per second within 10 consecutive seconds (or the cumulative decrease exceeds 400 W / m² within 1 minute), it is determined that a sharp drop event has occurred, and a "correction" trigger signal is immediately sent to the time series prediction model.

[0085] Specifically, this step describes the model's dynamic response mechanism after receiving a "correction" trigger signal, which in deep learning models can be implemented through an attention mechanism or gating structure:

[0086] During normal training and runtime, the model, through its internal weights, has learned how much influence the NetLoad history and Irradiance history should each have under different operating conditions. When the "correction" trigger signal is activated, the model's internal state is dynamically modulated. Specifically, the trigger signal forcibly increases (or opens) a specific gating mechanism that is specifically responsible for the attention score of the Irradiance_t (irradiance) input channel. This forces the model to break its usual behavior at this moment, and it will temporarily "distrust" the "smooth" pattern it has learned from the historical NetLoad (e.g., historical experience suggests "now"). (It's 1 PM, the net load should be very low), and then "highly trusts" the just-input, real-time Irradiance_t (e.g., "irradiance is currently only 100W / m²"). In this way, the expected net load curve to be generated is corrected in real time. Because the decline in photovoltaic (PV) power generation will lead to an increase in net load, the model will immediately (instead of lagging by a period) significantly raise the predicted value of the curve in its output expected net load curve (e.g., in the next 15 minutes), thus accurately reflecting the power deficit caused by this sharp decline. This corrected, more accurate curve is then sent to the energy storage response balancing module.

[0087] It should be noted that this embodiment is a further improvement on the aforementioned "time series prediction model". The aforementioned model only uses the single variable "net load" for prediction. Its drawback is that when the system encounters sudden, non-historical weather events (such as cloud cover), the model will produce a prediction lag due to the lack of external driving factors. It can only see this change in the next collection cycle after the photovoltaic power has decreased, resulting in an increase in net load, and cannot predict it.

[0088] In one optional embodiment, the specific implementation of the energy storage response equalization module includes:

[0089] Receive the expected net load curve and extract the current state of charge and temperature data of the energy storage unit from the periodic operating condition dataset;

[0090] Based on temperature data, a target range of states of charge is determined from a pre-defined lookup table or function.

[0091] Specifically, since the electrochemical activity and safety margins of lithium batteries differ at different temperatures, a pre-defined lookup table or function solidifies this expert experience. For example, the lookup table could be:

[0092] When the temperature data is <0℃ (low temperature), in order to prevent lithium plating, the target range of state of charge is dynamically narrowed to [20%, 80%];

[0093] When 0℃≤temperature data≤45℃ (normal), the target range is relaxed to [10%, 90%];

[0094] When the temperature data is >45℃ (high temperature), the target range is narrowed to [15%, 85%] to prevent the risk of thermal runaway;

[0095] The target range of charged states is a flexible constraint, representing the ideal range that the optimization algorithm attempts to reach at the end of the scheduling cycle.

[0096] Using the expected net load curve and the target range of state of charge as input, a multi-objective optimization solution is performed. The main objective is to minimize the power grid interaction, while the constraints are that the energy storage unit does not exceed the upper and lower limits of the safe state of charge and the safe temperature threshold. The initial charging and discharging power command required at each time step within the scheduling cycle is calculated.

[0097] Specifically, minimizing grid interaction power as the primary objective physically means maximizing photovoltaic self-consumption rate. The solver (e.g., using linear programming (LP) or dynamic programming (DP) algorithms) receives the generated expected net load curve (e.g., a prediction for the next 15 minutes) and attempts to completely fill this curve with the charging and discharging of energy storage (within the target state of charge range), so that the net grid interaction power approaches zero after 15 minutes. Furthermore, the constraints are rigid and must be satisfied at every time step. That is, the upper and lower limits of the safe state of charge and the safe temperature threshold are absolutely inviolable boundaries to prevent instantaneous overcharging, over-discharging, and over-temperature of energy storage. Therefore, the essence of the solution is: under the premise of satisfying rigid safety constraints, using the expected net load curve as input, the target state of charge range as flexible guidance, and minimizing grid interaction as the objective function, the final calculated initial charging and discharging power command is a power sequence containing multiple time steps (e.g., one power value per minute for the next 15 minutes). This sequence represents the theoretically optimal operating strategy that the system can achieve under the current operating conditions.

[0098] In an optional embodiment, the power smoothing constraint module is specifically implemented in the following ways:

[0099] Receive the initial charge and discharge power command sequence, and obtain the current maximum available charge and discharge rate and preset ramp rate of the energy storage unit from the battery management system;

[0100] Specifically, the maximum available charge / discharge rate is calculated and reported in real time by the battery management system. It is not a fixed value, but rather the instantaneous maximum power that the battery management system can withstand based on a comprehensive assessment of the current state of charge, health status, temperature, and other factors of the battery. The preset ramp rate is a system-level configuration parameter, which is usually determined by the hardware specifications of the energy storage converter or the grid connection requirements of the power grid (for example, requiring that the power change per second should not exceed 10% of the rated capacity).

[0101] Check the initial charge and discharge power command point by point, compare it with the command at the previous moment, and calculate the instantaneous power change rate;

[0102] When the instantaneous power change rate exceeds the preset ramp rate, or when the absolute value of the command exceeds the maximum available charge / discharge rate, the command point is corrected and limited to the boundary of the preset ramp rate or the maximum rate, thereby generating a smooth charge / discharge strategy.

[0103] Specifically: Assuming the preset ramp rate is 30kW / s, the system's (executed) power at time T=1 is 100kW (discharge). The initial command sequence requires a power of 150kW at time T=2. The module calculates the instantaneous power change rate as (150kW-100kW) / 1 second = 50kW / s. This change rate of 50kW / s exceeds the preset 30kW / s. Therefore, the module corrects the command point at T=2 to limit it to the boundary. The corrected command at T=2 = power at T=1 + maximum ramp rate = 100kW + 30kW = 130kW. Assuming the battery management system reports a power corresponding to the current maximum available charge / discharge rate of 200kW, the initial command sequence requires a power of 210kW at time T=3. The absolute value of this command, 210kW, exceeds the 200kW allowed by the battery management system. Therefore, the module corrects the command point at T=3 to limit it to the boundary. The corrected command at T=3 = 200kW.

[0104] It should be noted that this embodiment is intended to convert the theoretically optimal command calculated above into a physically safe command to be executed. The optimization solver may not have considered the instantaneous dynamic response limitations of the energy storage converter and battery during the calculation, which may result in the output initial charge and discharge power command containing drastic and abrupt power requests.

[0105] In an optional embodiment, the method for modifying the instruction point further includes:

[0106] When the instantaneous power change rate is limited to the preset ramp rate boundary, the power demand exceeding the limit is allocated to one or more subsequent scheduling time steps in the future on the time axis.

[0107] Ensure that the total charge and discharge energy of the smoothed charge and discharge strategy remains consistent with the total energy required by the initial charge and discharge power command throughout the entire scheduling cycle.

[0108] Specifically, this step implements a rolling compensation mechanism for power gaps. At time T=2, the initial command is 150kW, and the corrected command is 130kW. The system calculates an unmet power demand of 150kW - 130kW = 20kW. Assuming the command at time T=3 in the initial command sequence is 110kW, this module adds the unmet 20kW at time T=2 to the initial command at time T=3. At this time, the new initial command at time T=3 is 110kW + 20kW = 130kW. Subsequently, the module checks the ramp rate (130kW - 130kW = 0, satisfying the 30kW / s constraint) and maximum rate constraint between this new initial command of 130kW and T=2 (130kW). This ensures that the total charging and discharging energy of the smoothed charging and discharging strategy remains consistent with the total energy required by the initial charging and discharging power command throughout the entire scheduling cycle.

[0109] It should be noted that this embodiment is a further optimization of the correction step, which can solve the problem of a 20kW power gap that may occur during the correction process (e.g., correcting 150kW to 130kW). If this gap is not addressed, it will lead to a deviation in the total energy, making it impossible to achieve the target state of charge calculated in the solution at the end of the scheduling cycle. Through the time axis redistribution implemented above, the power smoothing constraint module only changes the instantaneous form of charging and discharging power, but does not change the total energy that needs to be exchanged in the entire scheduling cycle calculated by the energy storage response balancing module. This ensures that the energy management of the system satisfies the hardware safety constraints without sacrificing the upper-level optimization algorithm.

[0110] In one optional embodiment, the energy flow topology control module is specifically implemented in the following ways:

[0111] A smoothed charge / discharge strategy is received, which defines the target power of the energy storage unit at each time step;

[0112] Based on the current operating mode of the system, which includes grid-connected mode and off-grid mode, the smoothed charging and discharging strategy is decomposed into the active power derating command of the photovoltaic inverter, the constant power charging and discharging command of the energy storage converter, and the on / off logic command of the static switching switch of the grid interface.

[0113] Specifically, the single strategy for energy storage output from the upstream module is first translated and decomposed into specific execution instructions that are sent to multiple different physical devices (inverters, energy storage converters, switches). Then, the module must decompose the target energy storage power to different actuators according to different operating modes:

[0114] Under normal grid-connected operating conditions, the smoothed strategy is P_storage = -100kW (charging). The decomposed commands are: photovoltaic inverter command = maximum power point tracking to maximize power generation; energy storage converter command = constant power mode, P_target = -100kW; grid interface command is closed (maintaining grid connection); in terms of energy flow distribution, photovoltaic power is prioritized to supply the load, and excess power is used to charge the energy storage (-100kW). If there is still excess power, it is fed into the grid. If photovoltaic power is insufficient, it is supplemented by the grid.

[0115] Under the anti-reverse current condition in grid-connected mode, the smoothed strategy = P_storage = 0kW (BMS reports full), and the system detects that photovoltaic power generation > local load (there is a surplus of 50kW); in the decomposition command, the photovoltaic inverter command = active power derating, P_limit = local load (for example, commanding the inverter to reduce from 150kW to 100kW to eliminate the 50kW surplus); the energy storage converter command = constant power mode, P_target = 0kW; the grid interface command = closed (but the power meter reading is 0).

[0116] In off-grid mode, the smoothed strategy = P_storage = +50kW (discharge to meet load); decomposed instructions: photovoltaic inverter instruction = maximum power point tracking; energy storage converter instruction = V / F (voltage frequency) control mode (i.e., "grid creation"), P_target is used as a reference rather than a mandatory instruction at this time, and the energy storage converter is responsible for maintaining the voltage and frequency stability of the off-grid system; grid interface instruction = disconnect.

[0117] Multiple instructions are encoded according to their execution order and security dependencies, and packaged into a collaborative execution instruction set containing timing information.

[0118] Specifically, security dependencies mean that when switching from grid-connected mode to off-grid mode (e.g., during planned power outages or grid disconnections), instructions must never be issued simultaneously; the correct timing information is:

[0119] T=0: Send a command to the power grid interface to disconnect it;

[0120] T=1: Confirms that the power grid interface has been disconnected;

[0121] T=2: A command is sent to the energy storage converter to start the V / F control mode;

[0122] It should be noted that if the sequence is incorrect (starting V / F first and then disconnecting the grid interface), it will cause the energy storage converter to fail to connect to the grid or short-circuit. That is, this module generates a time-safe set of cooperative operation instructions by encoding this dependency relationship to ensure that the switching of energy flow topology is smooth and safe.

[0123] In an optional embodiment, the closed-loop correction execution module is specifically implemented in the following ways:

[0124] The system sends a set of collaborative operation instructions to the corresponding photovoltaic inverters, energy storage converters, and static switching switches via the control bus.

[0125] After the command is executed, the actual output power feedback value is read back from the power meters of each device;

[0126] Calculate the deviation between the actual power feedback value and the target value of the smoothed charge-discharge strategy at that moment;

[0127] Specifically, the process of reading back and calculating the deviation is actually a real-time execution monitoring. For example, the smoothed charge and discharge strategy (target value) at time T=5 is: P_target=+150kW (discharge); the energy flow topology module has issued this instruction to the energy storage converter; at time T=5 (or T=5.1, considering bus delay), the actual power feedback value read back from the outlet power meter of the energy storage converter by this module is: P_actual=+148.5kW; deviation = P_target-P_actual=150kW-148.5kW=+1.5kW; this 1.5kW deviation is the comprehensive response lag or error of the system (including model, communication, and hardware); then this deviation value is used as a correction factor and input to the next round of energy storage response balancing module to dynamically adjust the power weight of grid interaction in the multi-objective optimization solution to compensate for the system response lag.

[0128] It should be noted that the module will not catch up with the 1.5kW deviation at the current time T=5 (this is what the inner loop control of the energy storage converter does). This module stores this 1.5kW deviation (or its average deviation over the current scheduling cycle). When the next round of energy storage response balancing module is about to begin its multi-objective optimization solution, this module will use this deviation value (1.5kW) as the correction factor input. During the dynamic adjustment process, when the system detects a continuous deviation of +1.5kW (i.e., the system discharge capacity is 1.5kW weaker than expected), the correction factor will be dynamically and slightly increased. The penalty cost of the power weighting of the grid interaction forces the next round of optimization solver to automatically plan a more aggressive initial charge and discharge power command during calculation in order to avoid this expensive grid interaction. For example, it may calculate P_initial = +151.5kW. When this new command of +151.5kW is smoothed and finally executed by the calculation deviation, its actual output will be very close to the real target of +150kW. This can compensate for the systematic response lag caused by model inaccuracy or equipment aging, and ensure that the control accuracy adaptively converges in long-term operation.

[0129] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A photovoltaic energy storage management system, characterized in that, include: The operating condition acquisition module is used to collect environmental irradiance and output power data of the photovoltaic array, state of charge and temperature data of the energy storage unit, and power consumption data of the local load according to a preset acquisition cycle, and package the data into a periodic operating condition dataset. The power prediction module is used to receive the periodic operating condition dataset, extract power consumption data and photovoltaic output power data from it, calculate the current system net load, and generate the expected net load curve for the next scheduling cycle based on the changing trend of the net load. The energy storage response balancing module is used to receive the expected net load curve, extract the current state of charge of the energy storage unit from the periodic operating condition dataset, and calculate the initial charge and discharge power command required to balance the expected net load. The power smoothing constraint module is used to analyze the initial charge and discharge power command, constrain instantaneous requests in the command that exceed the preset ramp rate or the preset energy storage safety threshold, and generate a smoothed charge and discharge strategy. The energy flow topology control module is used to decompose the smoothed charging and discharging strategy into specific control timing sequences for the photovoltaic inverter, energy storage converter and grid interface, and generate a set of coordinated operation instructions for multiple devices. The closed-loop correction execution module is used to send the collaborative operation instruction set to each device, and obtain the actual power feedback value of each device after execution within the current scheduling cycle. Based on the deviation between the feedback value and the smoothed charging and discharging strategy, the initial parameters calculated in the next scheduling cycle are adjusted.

2. The photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the operating condition acquisition module include: The ambient irradiance, output power data, state of charge, temperature data, and power consumption data are collected by polling from the photovoltaic inverter, battery management system, and smart meter via an industrial bus. Add a uniform timestamp to all collected data and verify the data. If the data packet checksum is incorrect or the data value exceeds the preset physical range, mark it as invalid data and enable neighboring data interpolation. All data that pass verification within the same acquisition cycle and have the same timestamp are combined into a complete periodic operating condition dataset and forwarded to the power prediction module.

3. The photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the power prediction module include: After receiving the periodic operating condition dataset, the system net load is calculated, whereby the net load is the power consumption data minus the photovoltaic output power data. The calculated current system net load value is constructed into a net load time series stored in chronological order; A pre-trained time-series prediction model is used to extrapolate the net load time series to generate an expected net load curve covering the next scheduling cycle and containing multiple time steps.

4. A photovoltaic energy storage management system according to claim 3, characterized in that, The method for generating the expected net load curve also includes: When extrapolating the net load time series, the environmental irradiance data in the periodic operating condition dataset is also introduced as an additional input variable for the time series prediction model. When a sharp drop in environmental irradiance data is detected within a short period of time, the time-series prediction model responds to the sharp drop by increasing the influence weight of the additional input variables to immediately correct the shape of the expected net load curve.

5. A photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the energy storage response balancing module include: Receive the expected net load curve and extract the current state of charge and temperature data of the energy storage unit from the periodic operating condition dataset; Based on the temperature data, a target range of state of charge is determined from a preset lookup table or function; Using the expected net load curve and the target range of state of charge as input, a multi-objective optimization solution is performed. The main objective is to minimize the power grid interaction, while the constraints are that the energy storage unit does not exceed the upper and lower limits of the safe state of charge and the safe temperature threshold. The initial charge and discharge power command required at each time step within the scheduling cycle is calculated.

6. A photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the power smoothing constraint module include: Receive the initial charge and discharge power command sequence, and obtain the current maximum available charge and discharge rate of the energy storage unit and the preset ramp rate from the battery management system; The initial charge and discharge power command is checked point by point, compared with the command at the previous moment, and the instantaneous power change rate is calculated. When the instantaneous power change rate exceeds the preset ramp rate, or when the absolute value of the command exceeds the maximum available charge / discharge rate, the command point is corrected and limited to the boundary of the preset ramp rate or the maximum rate, thereby generating a smooth charge / discharge strategy.

7. A photovoltaic energy storage management system according to claim 6, characterized in that, The implementation method for correcting the instruction point also includes: When the instantaneous power change rate is limited to the preset ramp rate boundary, the power demand exceeding the limit is allocated to one or more subsequent scheduling time steps in the future on the time axis. Ensure that the total charge and discharge energy of the smoothed charge and discharge strategy remains consistent with the total energy required by the initial charge and discharge power command throughout the entire scheduling cycle.

8. A photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the energy flow topology control module include: The smoothed charge / discharge strategy is received, which defines the target power of the energy storage unit at each time step; Based on the current operating mode of the system, which includes grid-connected mode and off-grid mode, the smoothed charging and discharging strategy is decomposed into the active power derating command of the photovoltaic inverter, the constant power charging and discharging command of the energy storage converter, and the on / off logic command of the static switching switch of the grid interface. The multiple instructions are encoded according to their execution order and security dependencies, and packaged into a collaborative execution instruction set containing timing information.

9. A photovoltaic energy storage management system according to claim 1, characterized in that, The specific implementation methods of the closed-loop correction execution module include: The coordinated operation instruction set is sent to the corresponding photovoltaic inverter, energy storage converter and static switching switch via the control bus; After the instruction is executed, the actual output power feedback value is read back from the power meters of each device; Calculate the deviation between the actual power feedback value and the target value of the smoothed charge-discharge strategy at that moment; The deviation value is used as a correction factor and input into the energy storage response balancing module in the next round to dynamically adjust the power grid interaction weight in the multi-objective optimization solution in order to compensate for system response lag.

Images