An AI prediction-based energy scheduling method, device and medium

Abstract

The application discloses an AI prediction-based energy scheduling method and device and medium. The method comprises the following steps: a host obtains multi-dimensional prediction data and real-time operation data of each slave, and fuses the data to obtain a fused data sequence; a multi-objective optimization function is constructed and solved to obtain an optimized scheduling coefficient at a current time; each slave calculates expected grid-connected power in combination with its own operation state and predicted short-time data according to the obtained optimized scheduling coefficient; and the host corrects the optimized scheduling coefficient and reallocates actual grid-connected execution power of each slave according to power deviation of actual grid-connected execution power of each slave from expected grid-connected power. The device and the medium are used to implement the AI prediction-based energy scheduling method. The application has the beneficial effect that through accurate scheduling of the multi-objective optimization function and in combination with electricity price prediction, a "low storage and high generation" arbitrage strategy is realized.

Description

Technical Field

[0001] This application relates to the field of new energy power generation technology, and in particular to an energy dispatching method, equipment and medium based on AI prediction. Background Technology

[0002] In multi-unit parallel photovoltaic energy storage systems, a centralized energy management strategy is typically employed. Existing energy management methods mainly include the average allocation method and the allocation method based on battery SOC. The average allocation method simply distributes the total system power demand evenly among the slave units, without considering the actual operating status and equipment differences of each slave unit. The battery SOC-based allocation method, on the other hand, allocates charging and discharging power proportionally based on the state of charge (SOC) of each energy storage unit, with higher SOC units discharging more and lower SOC units charging more.

[0003] The existing technology described above has the following technical defects when used:

[0004] (1) Insufficient forecasting ability: The lack of forecasting ability for information such as system PV output, load demand, weather conditions and dynamic electricity prices leads to short-sighted energy management and scheduling decisions.

[0005] (2) Weak anti-interference ability: There is a lack of effective management of the uncertainty of photovoltaic power generation and load demand, and can only respond passively to sudden situations such as weather changes.

[0006] (3) Limited economic optimization: It is difficult to achieve optimal economic efficiency while meeting technical constraints.

[0007] (4) Affects equipment lifespan: It is impossible to predict and optimize charging and discharging strategies based on equipment status. Summary of the Invention

[0008] One objective of this application is to provide an AI-based predictive energy scheduling method that can address at least one of the deficiencies in the aforementioned background technology.

[0009] Another objective of this application is to provide an electronic device capable of implementing an AI-based predictive energy dispatching method that addresses at least one of the deficiencies in the aforementioned background technology.

[0010] Another object of this application is to provide a computer-readable storage medium capable of implementing an AI-predictive energy scheduling method that addresses at least one of the deficiencies in the aforementioned background art.

[0011] To achieve at least one of the above objectives, the technical solution adopted in this application is as follows: an energy scheduling method based on AI prediction, applied to an energy storage system using a master-slave control architecture, comprising the following steps: the master acquires multi-dimensional prediction data predicted by the AI ​​prediction model and real-time operating data of each slave, and fuses the acquired multi-dimensional prediction data and real-time operating data to generate a fused data sequence; a multi-objective optimization function is constructed, and the multi-objective optimization function is solved according to the set constraints and the obtained fused data sequence to obtain the optimized scheduling coefficient at the current moment; based on the obtained optimized scheduling coefficient, each slave calculates the expected grid-connected power by combining its own operating status and predicted short-term data, and corrects the expected grid-connected power based on local constraints to obtain the actual grid-connected execution power of each slave; the master corrects the optimized scheduling coefficient and redistributes the actual grid-connected execution power of each slave according to the power deviation between the actual grid-connected execution power and the expected grid-connected power of each slave.

[0012] Preferably, the acquisition of the optimal scheduling coefficient includes the following process: constructing a multi-objective optimization function that comprehensively considers economy, equipment lifespan and system stability; setting constraints for the optimization process, including power balance constraints, energy storage battery operation constraints and grid interaction constraints; based on the obtained fused data series, under the set constraints, solving for the optimal charging and discharging power demand sequence of the system in a future optimization time period that minimizes the multi-objective optimization function, and converting the obtained optimal charging and discharging power demand into the required optimal scheduling coefficient according to the current system state.

[0013] Preferably, the expression for the multi-objective loss function J is:

[0014] J=ω1J eco +ω2J aging +ω3J stab ;

[0015] ;

[0016] ;

[0017] ;

[0018] In the formula, ω1, ω2, and ω3 all represent weighting coefficients, and J eco J represents the economic sub-objective function. aging J represents the sub-objective function of equipment lifespan. stab Let t0 represent the initial time, T represent minimizing the scheduling period, and C represent the sub-objective function for runtime stability. grid (t) represents the predicted grid electricity price at time t, P grid (t) represents the power exchanged between the system and the grid at time t. The time step is represented by n, the number of battery packs in the energy storage system is represented by n, and kaging(·) represents the state of health (SOH) of the i-th battery pack. i Real-time State of Charge (SOC) i (t), charging and discharging current I i (t) and temperature T i Aging rate model of (t); and These represent the actual and predicted values ​​of the system's total power generation, respectively. and These represent the actual and predicted values ​​of the load power, respectively.

[0019] Preferably, the operating modes of the energy storage system are divided into charging mode and discharging mode based on the operating state of the energy storage battery; the optimal charging and discharging power demand is negative when the energy storage system is operating in charging mode and positive when the energy storage system is operating in discharging mode; in the charging mode of the energy storage system, the optimized scheduling coefficient is equal to the ratio of the absolute value of the current optimal charging and discharging power demand to the sum of the remaining rechargeable capacity of each slave device; in the discharging mode of the energy storage system, the optimized scheduling coefficient is equal to the ratio of the current optimal charging and discharging power demand to the sum of the discharging capacity of each slave device.

[0020] Preferably, the process of obtaining the expected grid-connected power of the slave device includes the following steps: calculating the basic expected power based on the current operating status of the slave device and in combination with the optimized scheduling coefficient; calculating the prediction feedforward adjustment based on the deviation between the predicted values ​​and actual measured values ​​of the key variables of the energy storage system in the short term by the AI ​​prediction model; and adding the obtained basic expected power and the prediction feedforward adjustment to obtain the required expected grid-connected power.

[0021] Preferably, in charging mode, the basic expected power of each slave device is equal to its power generation minus the product of the optimized scheduling coefficient and the remaining rechargeable capacity; in discharging mode, the basic expected power of each slave device is equal to its power generation minus the product of the optimized scheduling coefficient and the discharging capacity.

[0022] Preferably, the predicted feedforward adjustment amount The calculation formula is:

[0023] ;

[0024] Where α represents the feedforward coefficient. and These represent the predicted and actual measured values ​​of the short-term power generation of the slave unit, respectively. and These represent the predicted and actual measured values ​​of the short-term load power demand of the slave device, respectively.

[0025] Preferably, the predicted values ​​of key variables are continuously compared with the actual measured values; when the deviation between the two continues to exceed the set threshold, or when the operating mode of the energy storage system changes, the AI ​​prediction model is retrained or its parameters are updated.

[0026] An electronic device includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described AI-based prediction-based energy scheduling method.

[0027] A computer-readable storage medium storing a computer program; when the computer program is executed by a processor, it implements the above-described AI-based prediction-based energy scheduling method.

[0028] Compared with the prior art, the beneficial effects of this application are as follows:

[0029] (1) By accurately scheduling the multi-objective optimization function and combining it with electricity price forecasting, the arbitrage strategy of "low storage and high generation" can be realized.

[0030] (2) By introducing the equipment life sub-objective, the battery cycle life can be extended by actively avoiding harmful operating conditions during scheduling.

[0031] (3) By introducing the predictive feedforward adjustment and centralized deviation compensation mechanism, the system’s ability to smooth out fluctuations of key variables and resist interference has been significantly improved, realizing the transformation from passive response to active optimization. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall working steps of this application.

[0033] Figure 2 This is a schematic diagram of the basic architecture of the energy storage system in this application. Detailed Implementation

[0034] The present application will now be further described in conjunction with specific embodiments. It should be noted that, in the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicates that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0035] In the description of this application, it should be noted that the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., which indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of this application.

[0036] It should be noted that the terms "first," "second," etc., in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0037] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0038] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0039] The terms “comprising” and “having”, and any variations thereof, in the specification and claims of this application are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0040] One aspect of this application provides an AI-predictive energy dispatch method applied to an energy storage system employing a master-slave control architecture. The energy storage system includes multiple parallel units; one parallel unit acts as the master, and the remaining parallel units act as slaves. Figure 1 As shown, one preferred embodiment includes the following steps: The host acquires multi-dimensional prediction data from the AI ​​prediction model and real-time operating data from each slave device, and fuses the acquired multi-dimensional prediction data and real-time operating data to generate a fused data sequence. A multi-objective optimization function is constructed, and the multi-objective optimization function is solved according to the set constraints and the obtained fused data sequence to obtain the optimized scheduling coefficient at the current time. Based on the obtained optimized scheduling coefficient, each slave device calculates the expected grid-connected power by combining its own operating status and predicted short-term data, and corrects the expected grid-connected power based on local constraints to obtain the actual grid-connected execution power of each slave device. The host device corrects the optimized scheduling coefficient and redistributes the actual grid-connected execution power of each slave device based on the power deviation between the actual grid-connected execution power and the expected grid-connected power.

[0041] Understandably, the core concept of this application lies in constructing a two-layer decision-making architecture that closely coordinates a "predictive optimization layer" and a "real-time control layer," and in achieving a fundamental shift from passive response to proactive optimization by integrating predictive information from multiple time scales.

[0042] Specifically, the prediction and optimization layer, as the core of forward-looking decision-making in this application's technical solution, suggests that the AI ​​prediction model conduct medium- to long-term predictive analysis on diverse information such as the power output of the energy storage system's power generation units, load demand, and dynamic electricity prices. Over a relatively long future period, with the optimization objectives of minimizing the total electricity cost of the energy storage system, minimizing equipment lifespan loss, and maximizing system stability, a globally optimal reference charging and discharging power plan and its corresponding optimized scheduling coefficients are solved.

[0043] The real-time control layer, serving as the dynamic execution and compensation unit of this application's technical solution, receives optimized scheduling coefficients from the prediction and optimization layer as baseline instructions. It is responsible for performing precise local power allocation and execution within extremely short time periods (seconds or minutes), based on the real-time status of each energy storage battery and combined with feedforward adjustments based on short-term / ultra-short-term predictions. Simultaneously, this layer collects power deviations during actual execution through a closed-loop feedback mechanism and performs real-time fine-tuning compensation on the optimized scheduling coefficients to eliminate the impact of prediction errors and unknown disturbances, ensuring the stability and robustness of system operation.

[0044] Understandably, based on medium- to long-term scale predictions at the predictive optimization layer, multi-objective optimization functions can be constructed and accurately scheduled, thereby dominating the charging and discharging timing planning of the energy storage system and realizing the "low storage, high generation" arbitrage strategy. Simultaneously, based on short- to ultra-short-term scale predictions at the real-time control layer, predictive feedforward adjustments are generated, enabling the system to smooth power fluctuations caused by sudden changes in illumination and load switching in advance, significantly improving dynamic response quality. Through the organic combination of the above two-layer architecture and multi-scale predictions, this application achieves a balance between long-term economic optimization and short-term operational stability, allowing the energy storage system to not only "see far ahead" to make optimal economic plans but also "respond quickly" to smooth real-time fluctuations, thereby comprehensively improving system performance, economic benefits, and equipment lifespan. Furthermore, through a centralized deviation compensation mechanism, the system's ability to smooth fluctuations in key variables and its anti-interference capabilities are further enhanced, realizing a shift from passive response to proactive optimization.

[0045] It should be understood that there are various practical application scenarios for the energy storage system that can implement the AI-based predictive energy dispatch method of this application. For example, it can be an energy storage system applied to photovoltaic power generation, hydropower generation, or wind power generation. For ease of understanding, this embodiment will take a photovoltaic power generation scenario as an example to describe the technical solution of this application in detail; for further understanding of the technical solution of this application, a photovoltaic energy storage system will be used as an example to describe the technical solution of this application in detail below.

[0046] The specific architecture of the photovoltaic energy storage system is as follows: Figure 2 As shown, the photovoltaic energy storage system comprises N parallel photovoltaic-energy storage units, which can be labeled as photovoltaic-energy storage parallel unit #1 to #N. The N parallel photovoltaic-energy storage units adopt a master-slave control architecture, where one unit acts as the master and the others as slaves; for example, photovoltaic-energy storage parallel unit #1 acts as the master, and the remaining units #2 to #N act as slaves. Each parallel photovoltaic-energy storage unit includes a photovoltaic module (PV), an energy storage battery, an inverter (INV), and a three-phase load (LOAD). The energy storage battery can be a single battery pack or n battery packs connected in parallel, which can be labeled as battery packs BAT#1 to BAT#n. The photovoltaic module (PV) and energy storage battery are connected in parallel to the input side of the inverter (INV), and the three-phase load (LOAD) is connected to the output side of the inverter (INV). Each parallel photovoltaic-energy storage unit is connected to the grid GRID via AC coupling, and the master information is determined through a higher-level designation or a contention mechanism. The energy storage batteries of each photovoltaic-storage parallel unit are connected in parallel via communication cables, and all inverters INV are also connected in parallel via communication cables.

[0047] In this embodiment, based on the specific architecture of the photovoltaic energy storage system, the host can obtain multi-dimensional prediction data from an external AI prediction model via a communication interface, or it can obtain multi-dimensional prediction data from a built-in AI prediction model. The specific architecture and working principle of the AI ​​prediction model are common knowledge to those skilled in the art, and the specific configuration location of the AI ​​prediction model can be selected according to the actual needs of those skilled in the art.

[0048] It's important to understand that the multi-dimensional prediction data output by the AI ​​prediction model includes photovoltaic power generation prediction curves, load demand power prediction curves, dynamic electricity price prediction curves, and meteorological condition prediction information. Specifically, the photovoltaic power generation prediction curve can cover a specified future time period, such as a 24-hour photovoltaic processing prediction sequence; the load demand power prediction curve outputs a prediction sequence of the total system load demand for the same time period; the dynamic electricity price prediction curve reflects a prediction sequence of the grid's time-of-use electricity price or real-time electricity price signal; and the meteorological condition prediction information includes predictions of key meteorological factors affecting photovoltaic output, such as sunlight, temperature, and cloud cover.

[0049] Meanwhile, the real-time operating data of each slave device synchronously collected by the host includes the real-time output power P on the PV side of each slave system. pv The real-time SOC and rated capacity Q of each battery pack corresponding to the energy storage battery rated Health status indicator (SOH), real-time load power (P) at the output of each inverter. load And the cumulative cycle count, temperature, and historical stress data of the energy storage battery.

[0050] In this embodiment, after the host obtains the multi-dimensional prediction data output by the AI ​​prediction model and the real-time running data of each slave, the two types of data can be spatiotemporally aligned to ensure that they are consistent in the time dimension; then, the two types of data are fused by a confidence weighting method to obtain the fused data sequence required for subsequent optimization calculations.

[0051] In this embodiment, obtaining the optimization scheduling coefficient mainly includes three steps: constructing the optimization objective, modeling the system constraints, and solving the optimization scheduling coefficient. For ease of understanding, the entire acquisition process will be described in detail below.

[0052] (1) Regarding the construction of the optimization objective, multiple aspects such as economy, equipment life and system stability can be comprehensively considered to construct a multi-objective optimization function J, which can be expressed by the following expression:

[0053] J=ω1J eco +ω2J aging +ω3J stab .

[0054] In the formula, ω1, ω2, and ω3 all represent weighting coefficients, and J eco J represents the economic sub-objective function. aging J represents the sub-objective function of equipment lifespan. stab This represents the operational stability sub-objective function.

[0055] It is understandable that the economic sub-objective function J eco To minimize the total electricity cost within the scheduling period T; through the economic sub-objective function J eco Accurate scheduling, combined with electricity price forecasting, enables an arbitrage strategy of "low storage and high generation." The equipment lifespan sub-objective function J... aging This is used to minimize the equivalent aging loss of energy storage batteries; by actively avoiding harmful operating conditions (such as deep discharge and high-rate charge / discharge) during scheduling, the cycle life of the batteries is extended. The operational stability sub-objective function J... stab This is used to minimize the risk of power imbalance caused by prediction errors; by compensating for the deviation between predicted and actual values ​​during scheduling, it enhances the energy storage system's resilience to disturbances. The specific calculation expression is as follows:

[0056] .

[0057] .

[0058] .

[0059] In the formula, t0 represents the initial time, T represents the minimum scheduling period, and C grid (t) represents the predicted grid electricity price at time t, P grid (t) represents the power exchanged between the system and the grid at time t (positive for power intake and negative for power supply). This represents the time step, and kaging(·) represents the state of health (SOH) of the i-th battery pack. i Real-time State of Charge (SOC) i (t), charging and discharging current I i (t) and temperature T i Aging rate model of (t); and These represent the actual and predicted values ​​of the total photovoltaic power generation of the system, respectively. and These represent the actual and predicted values ​​of the load power, respectively.

[0060] (2) For system constraint modeling, physical and operational constraints that must be followed during the optimization process can be defined, specifically including power balance constraints, energy storage battery operation constraints, and grid interaction constraints. Among them, the power balance constraint is used to limit the output power of the energy storage system at any time; the energy storage battery operation constraint is used to limit the SOC and charging / discharging power of each battery pack; and the grid interaction constraint is used to limit the inverter's grid-connected power and power extraction power. For ease of understanding, each constraint is expressed by an expression below.

[0061] Understandably, for power balance constraints, at any time t, the following needs to be satisfied:

[0062] .

[0063] In the formula, This indicates the charging and discharging power of the energy storage battery. This indicates the power of the power grid.

[0064] For the operational constraints of energy storage batteries, during operation, any i-th battery pack must satisfy the following:

[0065] ; .

[0066] In the formula, This represents the SOC value of the i-th battery pack at any time t. Let represent the upper and lower limits of the SOC safety limit for the i-th battery pack, respectively; This represents the charging and discharging power of the i-th battery pack at any time t. These represent the maximum charging / discharging power of i battery packs, respectively.

[0067] For grid interaction constraints, at any time t, the grid power must satisfy:

[0068] .

[0069] In the formula, These are the upper and lower limits for power intake and power supply from the grid side, respectively.

[0070] (3) To solve for the optimal scheduling coefficient, the future scheduling period is discretized based on the obtained fused data series; with the goal of minimizing the multi-objective optimization function J, the optimal charging and discharging power demand sequence of the system in the next optimization time period is solved iteratively by the optimization algorithm under the set constraints. Based on the current system state at time t, the obtained optimal charging and discharging power demand is converted into the required optimized scheduling coefficient coeff.

[0071] In this embodiment, the energy storage system operates in two modes: charging mode and discharging mode, based on the operating state of the energy storage battery. For ease of understanding, the process of determining the operating mode of the energy storage system will be described in detail below. The total power generation of the energy storage system is the sum of the output power of all photovoltaic modules (PV), and the total power demand of the energy storage system is the sum of the power demand of all three-phase loads (LOAD). Therefore, based on the obtained total power generation and total power demand, the total charging and discharging power demand can be calculated by subtracting them. The operating mode of the energy storage system can then be determined based on the sign of the total charging and discharging power demand; that is, when the total charging and discharging power demand is greater than zero, the energy storage system operates in charging mode; when the total charging and discharging power demand is less than zero, the energy storage system operates in discharging mode.

[0072] It is important to note that when the total charging and discharging power demand is zero, it means that the total power generation of the energy storage system is exactly equal to the total power demand, that is, the photovoltaic power generation can just meet the load demand, and the energy storage battery does not need to work.

[0073] In this embodiment, based on the above, it is known that when the energy storage system operates in charging mode, the optimal charging and discharging power demand is positive, indicating that the energy storage battery is charging; when the energy storage system operates in charging mode, the optimal charging and discharging power demand is negative, indicating that the energy storage battery is discharging. Therefore, in the charging mode of the energy storage system, the optimal scheduling coefficient is equal to the ratio of the absolute value of the current optimal charging and discharging power demand to the sum of the remaining rechargeable capacities of each slave device; in the discharging mode of the energy storage system, the optimal scheduling coefficient is equal to the ratio of the current optimal charging and discharging power demand to the sum of the discharging capacities of each slave device. For ease of understanding, the calculation expressions for the optimal scheduling coefficient of the energy storage system in different operating modes will be described below.

[0074] In charging mode, the formula for calculating the optimized scheduling coefficient coeff is:

[0075] .

[0076] .

[0077] In the formula, This represents the optimal charging and discharging power demand at time t. This represents the sum of the remaining rechargeable capacity of each slave device. This represents the battery capacity of the i-th battery pack.

[0078] In discharge mode, the formula for calculating the optimal scheduling coefficient coeff is:

[0079] .

[0080] .

[0081] In the formula, This represents the sum of the remaining dischargeable capacity of each slave device.

[0082] In this embodiment, after solving for the optimized scheduling coefficient, the host can send the calculated optimized scheduling coefficient (coeff) and the operating mode of the energy storage system to each slave, thereby realizing the grid-connected power allocation and related limiting control for each slave. Considering that the optimized scheduling coefficient, or the optimal charging and discharging demand power, is calculated using prediction data from an AI prediction model, there may be a difference between the actual and predicted values ​​of key variables in the energy storage system, leading to power fluctuations and affecting the accuracy of the slave's grid-connected power allocation. Therefore, when allocating the grid-connected power to each slave, it is necessary to suppress power fluctuations caused by deviations in key variables. For ease of understanding, the specific process of obtaining the expected grid-connected power for each slave will be described in detail below.

[0083] Specifically, obtaining the expected grid-connected power of the slave device includes the following process: Calculate the basic expected power based on the current operating status of the slave device and the optimized scheduling coefficient; calculate the prediction feedforward adjustment based on the deviation between the predicted values ​​and actual measured values ​​of key variables of the energy storage system in the short term by the AI ​​prediction model; and add the obtained basic expected power to the prediction feedforward adjustment to obtain the required expected grid-connected power.

[0084] Understandably, the key variables affecting energy dispatch in energy storage systems mainly include the power generation of the generating units and the power demand of the load. Therefore, when calculating the expected grid-connected power of each slave unit, the forecast feedforward adjustment is a compensation amount based on the deviation between the slave unit's predicted short-term photovoltaic output and load and the current measured values, used to smooth out power fluctuations in advance. Regarding the forecast feedforward adjustment... The calculation formula is as follows:

[0085] .

[0086] In the formula, α represents the feedforward coefficient, α∈[0,1]; and These represent the predicted and actual measured values ​​of the short-term power generation of the slave unit, respectively. and These represent the predicted and actual measured values ​​of the short-term load power demand of the slave device, respectively.

[0087] Specifically, in charging mode, the basic expected power of each slave device is equal to its power generation minus the product of the optimal scheduling coefficient and the remaining rechargeable capacity; therefore, the expected grid-connected power P of each slave device in charging mode is... desired The calculation formula is as follows:

[0088] .

[0089] In discharge mode, the basic expected power of each slave unit is equal to its generating power minus the product of the optimal dispatch coefficient and the discharge capacity; therefore, the expected grid-connected power P of each slave unit in discharge mode is... desired The calculation formula is as follows:

[0090] .

[0091] Those skilled in the art will understand that, based on safety design, each parallel unit requires local constraints during operation; for example, the inverter INV may have power limiting measures, and the energy storage battery may have charge / discharge capacity limits and SOC boundary protection measures. This results in the expected grid-connected power P of the slave unit being limited. desired This may exceed the actual operating capacity of the parallel unit, therefore the expected grid-connected power P calculated by the slave unit needs to be considered. desired Restrictions should be imposed.

[0092] Specifically, if the expected grid-connected power P calculated by the slave machine... desired It did not exceed the upper limit of the actual operating capacity P max Therefore, the expected grid-connected power is taken as the actual grid-connected execution power P. actual If the expected grid-connected power P calculated by the slave machine desired Exceeding the upper limit of actual operational capacity P max Then the upper limit of the actual operating capacity P max As the actual grid-connected power P actual This makes the actual grid-connected power P of the slave device... actual With expected grid-connected power P desired There may be a difference, which is the power deviation of the slave device, e=P. desired -P actual .

[0093] In this embodiment, since the actual energy dispatch does not meet expectations due to the power deviation of the slave device, the optimized dispatch coefficient coeff can be compensated and corrected according to the operating mode and power deviation characteristics of the energy storage system. For ease of understanding, the correction process of the optimized dispatch coefficient coeff of the energy storage system in charging mode and discharging mode will be described in detail below.

[0094] When the energy storage system is in charging mode, some slave devices may fail to complete the expected charging command due to reasons such as the energy storage battery's SOC approaching its upper limit or excessive temperature. These slave devices will generate a positive power deviation e, which is the expected grid-connected power P. desired >P actualThe power deviations e of all slave devices are summed to obtain a positive total power deviation E. Based on the obtained total power deviation E, the compensation amount coeff is calculated by the PI controller. cmp (Positive value); Optimize scheduling coefficient coeff and compensation amount coeff cmp Adding them together, we get the corrected optimized scheduling coefficient, coeff. final =coeff+coeff cmp By increasing the value of the optimization scheduling coefficient, the unrestricted slave devices can undertake more charging tasks.

[0095] When the energy storage system is in discharge mode, some slave devices may fail to complete the expected discharge command due to reasons such as the energy storage battery's SOC being close to the lower limit or power limitation. These slave devices will generate a negative power deviation e, which is the expected grid-connected power P. desired <P actual The power deviations e of all slave devices are summed to obtain a negative total power deviation E. Based on the obtained total power deviation E, the compensation amount coeff is calculated by the PI controller. cmp (Positive value); Optimize scheduling coefficient coeff and compensation amount coeff cmp Subtracting the two, we obtain the corrected optimal scheduling coefficient, coeff. final =coeff-coeff cmp By reducing the value of the optimization scheduling coefficient, the discharge capability of the unrestricted slave device is enhanced.

[0096] In this embodiment, during the energy dispatching process of the energy storage system, key variables of the energy storage system can be continuously monitored, and the predicted values ​​of the continuously monitored key variables are compared with the actual measured values. If the deviation between the two continuously exceeds a set threshold, it indicates that the prediction accuracy of the AI ​​prediction model does not meet expectations, and at this time, retraining or parameter updating of the AI ​​prediction model will be triggered. At the same time, when the operating mode of the energy storage system changes, such as changing from grid-connected operation to off-grid operation, retraining or parameter updating of the AI ​​prediction model will also be triggered, so that the AI ​​prediction model can continuously improve the subsequent prediction accuracy through online self-learning.

[0097] Another aspect of this application provides an electronic device, in one preferred embodiment of which includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described AI-based prediction-based energy scheduling method.

[0098] Another aspect of this application provides a computer-readable storage medium, in a preferred embodiment of which a computer program is stored on the storage medium; when the computer program is executed by a processor, it implements the above-described AI-based prediction-based energy scheduling method.

[0099] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. An AI-based predictive energy dispatching method, applied to an energy storage system employing a master-slave control architecture, characterized in that, Includes the following steps: The host acquires multi-dimensional prediction data from the AI ​​prediction model and real-time operating data from each slave device, and merges the acquired multi-dimensional prediction data and real-time operating data to generate a fused data sequence. The multi-dimensional prediction data includes photovoltaic power generation prediction curves, load demand power prediction curves, dynamic electricity price prediction curves, and meteorological condition prediction information. Construct a multi-objective optimization function that comprehensively considers economic efficiency, equipment lifespan, and system stability; Set constraints for the optimization process, including power balance constraints, energy storage battery operation constraints, and grid interaction constraints; Based on the obtained fused data series, under the set constraints, the optimal charging and discharging power demand sequence of the system in the future optimization time period that minimizes the multi-objective optimization function is solved, and the obtained optimal charging and discharging power demand is converted into the required optimization scheduling coefficient according to the system state at the current moment. Based on the obtained optimized scheduling coefficients, each slave machine calculates its expected grid-connected power by combining its own operating status and predicted short-term data, and corrects the expected grid-connected power based on local constraints to obtain the actual grid-connected execution power of each slave machine. The host machine adjusts the optimization scheduling coefficient and redistributes the actual grid-connected power of each slave machine based on the power deviation between the actual grid-connected power and the expected grid-connected power of each slave machine.

2. The energy scheduling method based on AI prediction as described in claim 1, characterized in that, The expression for the multi-objective loss function J is: J=ω1J eco +ω2J aging +ω3J stab ; ； ； ； In the formula, ω1, ω2, and ω3 all represent weighting coefficients, and J eco J represents the economic sub-objective function. aging J represents the sub-objective function of equipment lifespan. stab Let t0 represent the initial time, T represent minimizing the scheduling period, and C represent the sub-objective function for runtime stability. grid (t) represents the predicted grid electricity price at time t, P grid (t) represents the power exchanged between the system and the grid at time t. The time step is represented by n, the number of battery packs in the energy storage system is represented by n, and kaging(·) represents the state of health (SOH) of the i-th battery pack. i Real-time State of Charge (SOC) i (t), charging and discharging current I i (t) and temperature T i Aging rate model of (t); and These represent the actual and predicted values ​​of the system's total power generation, respectively. and These represent the actual and predicted values ​​of the load power, respectively.

3. The energy scheduling method based on AI prediction as described in claim 1 or 2, characterized in that, The working modes of an energy storage system are divided into charging mode and discharging mode based on the working state of the energy storage battery. The optimal charging and discharging power demand is negative when the energy storage system is working in charging mode and positive when the energy storage system is working in discharging mode. In the charging mode of the energy storage system, the optimization scheduling coefficient is equal to the ratio of the absolute value of the optimal charging and discharging power demand at the current moment to the sum of the remaining rechargeable capacity of each slave device. In the discharge mode of the energy storage system, the optimal scheduling coefficient is equal to the ratio of the current optimal charging and discharging power demand to the sum of the discharge capacities of each slave device.

4. The energy scheduling method based on AI prediction as described in claim 3, characterized in that, The process of obtaining the expected grid-connected power of the slave device includes the following steps: Based on the current operating status of the slave device and the optimized scheduling coefficient, the basic expected power is calculated. Based on the deviation between the predicted and actual measured values ​​of key variables of the energy storage system in the short term by the AI ​​prediction model, the prediction feedforward adjustment is calculated. The required expected grid-connected power is obtained by adding the base expected power to the predicted feedforward adjustment.

5. The energy scheduling method based on AI prediction as described in claim 4, characterized in that, In charging mode, the basic expected power of each slave device is equal to its power generation minus the product of the optimal scheduling coefficient and the remaining rechargeable capacity. In discharge mode, the basic expected power of each slave device is equal to its power generation minus the product of the optimal dispatch coefficient and the discharge capacity.

6. The energy scheduling method based on AI prediction as described in claim 4, characterized in that, Predicted feedforward adjustment The calculation formula is: ； Where α represents the feedforward coefficient. and These represent the predicted and actual measured values ​​of the short-term power generation of the slave unit, respectively. and These represent the predicted and actual measured values ​​of the short-term load power demand of the slave device, respectively.

7. The energy scheduling method based on AI prediction as described in claim 4, characterized in that, The system continuously compares the predicted values ​​of key variables with the actual measured values; when the deviation between the two exceeds the set threshold or the operating mode of the energy storage system changes, it triggers the retraining or parameter update of the AI ​​prediction model.

8. An electronic device, characterized in that, It includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the AI-based prediction-based energy scheduling method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program; when the computer program is executed by a processor, it implements the AI-based prediction-based energy scheduling method as described in any one of claims 1-7.

Images