A distributed light storage grid-connected power optimization control method and system

Abstract

The application discloses a kind of distributed light storage grid-connected power optimization control method and system, belong to distribution network operation and control technical field.The application realizes the optimization solution of energy storage charge-discharge power by heuristic improved particle swarm algorithm, specifically includes: heuristic initialization, energy storage dynamic model and transformer temperature rise and aging factor quantization model, system economic cost model containing transformer life loss, system total fitness function, asymmetric nonlinear adaptive particle swarm iteration optimization based on population state feedback, optimal solution analysis and result verification.The application forces energy storage to charge preferentially and discharge power to be zero when photovoltaic reverse overload period, effectively reduces transformer load rate;Quantify transformer life loss into cost model, realize the collaborative optimization of equipment safety and economic operation;Improved particle swarm algorithm solves the problem that traditional algorithm is easy to fall into local optimum and slow convergence speed.The application has the advantages of strong engineering practicability and significant optimization effect.

Description

Technical Field

[0001] This invention belongs to the field of distributed photovoltaic and energy storage grid connection technology, and particularly relates to a method and system for power optimization control of distributed photovoltaic and energy storage grid connection. Background Technology

[0002] Currently, distributed photovoltaic (PV) power has become the core form of clean energy utilization on the distribution network side. However, the strong intermittency and volatility of its output lead to a series of problems in the distribution network, such as power backfeeding, node voltage exceeding limits, and transformer overload and aging, which seriously affect the safe and stable operation of the distribution network. Distributed energy storage systems, with their flexible charging and discharging characteristics, can smooth out PV power output fluctuations and absorb excess PV power, making them a key technical means to solve the above problems.

[0003] Existing distributed photovoltaic-storage grid-connected power optimization methods mostly employ traditional intelligent algorithms, but these methods have several shortcomings: First, the algorithm initialization phase lacks targeted heuristic guidance, leading to an imbalance between population diversity and optimization directionality, resulting in slow convergence and a tendency to get trapped in local optima. Second, they fail to consider transformer protection during photovoltaic reverse overload periods as a core constraint, and fail to implement strategic control for priority charging of energy storage. Long-term transformer overload can easily lead to excessive temperature rise and accelerated lifespan loss. Third, the cost model only considers basic costs such as electricity purchase and sale and charging and discharging losses, without quantifying transformer lifespan loss into the economic cost system, making it difficult to achieve the dual objectives of economic operation and equipment safety optimization. Fourth, the constraint design is relatively simple, lacking targeted soft constraint penalties for photovoltaic reverse feed and extreme transformer load rates, resulting in insufficient engineering applicability of the optimization results.

[0004] Particle swarm optimization (PSO), a classic swarm intelligence optimization algorithm, boasts advantages such as simple structure, fast convergence speed, and ease of integration with engineering scenarios, making it widely applicable to power system optimization problems. However, the fixed inertia weights and random initialization of traditional PSO algorithms fail to meet the specific requirements of distributed photovoltaic (PV) and energy storage (ESS) grid-connected systems. Therefore, it is crucial to improve the PSO algorithm by designing heuristic initialization strategies that incorporate PV output, load fluctuations, and electricity price characteristics. Simultaneously, a comprehensive economic cost model including transformer lifetime losses should be established, along with targeted control rules for PV reverse overload, enabling refined and scenario-specific optimization control of distributed PV and ESS grid-connected power. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to address the issues of poor algorithm convergence, failure to consider transformer life loss, and lack of targeted control strategies during photovoltaic reverse overload in traditional optimization methods. The core objectives are to achieve priority charging of energy storage and effective reduction of transformer load rate during photovoltaic reverse overload, while minimizing the total daily economic cost of the system and ensuring the safe operation of energy storage, transformers, and other equipment, thereby improving the engineering practicality of the optimization results.

[0006] The present invention solves the above-mentioned technical problems through the following technical means: A method for optimizing and controlling the grid-connected power of distributed photovoltaic energy storage includes the following steps: Pre-calculate the initial power of the grid when there is no energy storage charging and discharging, extract the reverse overload period when the photovoltaic output is greater than the load, and perform heuristic initialization of the particle swarm based on the prior knowledge of time-of-use pricing, photovoltaic output and load fluctuations. A quantitative model for transformer temperature rise and aging factor considering the thermal inertia effect of dual time constants is established, and the transformer life loss is monetized and included in the total daily economic cost of the system, which includes equipment investment depreciation, charging and discharging losses, and net cost of purchasing and selling electricity from the grid. Construct a fitness function based on the total daily economic cost of the system, superimposed with a soft penalty term for transformer load rate, a closed-loop penalty term for SOC, and a hard constraint penalty term for equipment; An improved particle swarm optimization is performed, which uses personalized inertia weights based on individual fitness feedback and asymmetric trigonometric function dynamic learning factors to update particle states. During the reverse overload period, energy storage is forced to charge and discharge power is zero, and the optimal energy storage charging and discharging power scheduling scheme is output.

[0007] Furthermore, the heuristic initialization specifically includes: For 80% of the particles, perform directional initialization according to the following rules: During off-peak electricity price periods, initialize the particle charging power dimension to 0.6 to 0.8 times the maximum charging power; during reverse overload periods, initialize the particle charging power dimension to 0.8 to 1.0 times the maximum charging power, and forcibly initialize the discharge power dimension to 0; during peak electricity price periods, initialize the particle discharge power dimension to 0.5 to 0.7 times the maximum discharge power. The remaining 20% ​​of the particles are kept globally randomized to maintain population diversity.

[0008] Furthermore, the quantitative model for transformer temperature rise and aging factors includes the following: The temperature rise of the top oil is divided into steady-state value and dynamic value. The steady-state temperature rise of the top oil is determined by the reference top oil temperature rise, transformer load rate, impedance ratio and load loss ratio. The dynamic temperature rise of the top oil is based on the top oil time constant and the thermal inertia hysteresis effect is introduced for correction. Hot spot temperature rise is divided into steady-state value and dynamic value. The steady-state hot spot temperature rise is determined by the reference hot spot temperature rise, transformer load rate and hot spot load coefficient. The dynamic hot spot temperature rise is based on the hot spot time constant and the thermal inertia hysteresis effect is introduced for correction. The transformer hot spot temperature is obtained by superimposing the ambient temperature and the dynamic hot spot temperature rise. The aging factor is calculated exponentially based on hot spot temperature to quantify the life loss of insulation materials; the transformer life loss cost is calculated in monetary terms by combining the aging factor with time step, expected transformer life, and life loss amplification factor, and is included in the system economic cost.

[0009] Furthermore, the total daily economic cost of the system is the sum of the following items: The average daily investment depreciation cost is calculated by multiplying the rated capacity and rated power of the energy storage system and the rated capacity of the transformer by the corresponding unit cost, and then dividing by the equipment lifespan. The cost of charging and discharging losses is the sum of the charging and discharging power and unit loss cost at each time period and the time step. The net cost of electricity purchase and sale by the power grid is the sum of the product of the time-of-use electricity purchase price and the purchased power, minus the sum of the product of the fixed electricity sales price and the sold power. The cost of transformer life loss is determined by the aging factor, transformer life loss coefficient, the ratio of time step to expected transformer life, and life loss amplification factor, and is quantified in monetary terms.

[0010] Furthermore, the method for determining the personalized inertia weight based on individual fitness feedback is as follows: The global baseline inertia weight is based on the Sigmoid mapping and decays nonlinearly with the iteration process. The parameters controlling the steepness of the curve and the position of the inflection point are configurable. In each iteration, the current best fitness and average fitness of the population are statistically analyzed, and the normalized relative fitness of each particle is calculated. Particles with fitness better than the average level of the population are identified as dominant particles, and are given a lower inertial weight based on their target difference on the basis of the benchmark weight to perform fine local search. Particles with fitness worse than the average level of the population are identified as inferior particles, and are given an inertial weight higher than the benchmark to encourage them to jump out of the local extreme value region. In the asymmetric trigonometric function dynamic learning factor, the cognitive factor decreases non-linearly from the initial value to the final value according to the cosine function as the iteration process progresses, and the social factor increases non-linearly from the initial value to the final value according to the sine function as the iteration process progresses; after the particle velocity is updated, it is limited to a set range; after the position is updated, the discharge power position corresponding to the reverse overload period is forcibly set to zero, and all positions are limited to the normalized interval.

[0011] The present invention also provides a distributed photovoltaic-storage grid-connected power optimization control system, comprising: The initialization module is used to obtain basic system parameters, pre-calculate the initial power of the grid when there is no energy storage charging and discharging, extract the reverse overload period when the photovoltaic output is greater than the load, and perform heuristic initialization of the particle swarm based on the prior knowledge of time-of-use pricing, photovoltaic output and load fluctuations. The dynamic modeling module is used to establish a dynamic model of the energy storage state of charge and a power interaction model with the power grid, as well as a quantitative model of transformer temperature rise and aging factor that takes into account the thermal inertia effect of dual time constants. The economic cost model module is used to establish a system daily total economic cost model that includes equipment investment depreciation costs, charging and discharging loss costs, net electricity purchase and sale costs from the grid, and transformer life loss costs. The fitness function construction module is used to construct the system fitness function by adding a transformer load rate soft penalty term, a SOC closed-loop penalty term, and a device hard constraint penalty term to the system's daily total economic cost. The particle swarm optimization module is improved by using personalized inertia weights based on Sigmoid baseline weights and relative fitness feedback, along with asymmetric trigonometric function dynamic learning factors to update particle states. During the reverse overload period, energy storage is forced to charge while the discharge power is zero, and the optimal energy storage charging and discharging power scheduling scheme is output.

[0012] Furthermore, the initialization module performs directional initialization on 80% of the particles according to the following rules: during off-peak electricity price periods, the particle charging power dimension is initialized to 0.6 to 0.8 times the maximum charging power; during reverse overload periods, the particle charging power dimension is initialized to 0.8 to 1.0 times the maximum charging power and the discharge power is forcibly initialized to 0; during peak electricity price periods, the particle discharge power dimension is initialized to 0.5 to 0.7 times the maximum discharge power; the remaining 20% ​​of the particles are kept globally randomized to maintain population diversity.

[0013] Furthermore, in the dynamic modeling module: The dynamic model of energy storage state of charge takes into account charging and discharging efficiency and time step, and applies hard constraints on the upper and lower limits of state of charge and the sustainability constraints of electricity within a 24-hour operating cycle. The initial state of charge is set as a fixed proportion of the rated capacity. In the quantitative model of transformer temperature rise and aging factor, the steady-state value and dynamic thermal inertia correction value based on the corresponding time constant are calculated for both top oil temperature rise and hot spot temperature rise. The aging factor is calculated exponentially according to the hot spot temperature. The transformer life loss cost is incorporated into the economic cost after being calculated in monetary terms by combining the aging factor, time step, expected transformer life and life loss amplification factor.

[0014] Furthermore, in the fitness function construction module: The transformer load rate soft penalty term applies a cubic penalty to load rates exceeding a set threshold, selectively suppressing extreme transformer overloads without applying indiscriminate penalties; The SOC closed-loop penalty term applies a high-weight penalty to the absolute value of the deviation of the state of charge from the initial value at the end of the operating cycle, ensuring that the system achieves complete power closed-loop within the 24-hour scheduling cycle; The hard constraint penalty term applies high penalties for exceeding the limits of state of charge, charging power, and discharging power, respectively, to ensure that the physical boundaries of the energy storage device are not breached.

[0015] Furthermore, in the improved particle swarm optimization module: The global baseline inertia weight is based on the Sigmoid mapping and decays nonlinearly with the iteration process. In each iteration, the optimal fitness and average fitness of the population are statistically analyzed, and the normalized relative fitness of each particle is calculated. The dominant particles are given an inertia weight lower than the baseline for fine local search, and the inferior particles are given an inertia weight higher than the baseline to escape local extrema. Cognitive factors decrease non-linearly according to a cosine function, while social factors increase non-linearly according to a sine function. The guiding algorithm evolves from individual independent exploration in the early stage to collaborative convergence towards the global optimum in the later stage. After the particle velocity is updated, it is limited to the range of [-0.15, 0.15], and after the particle position is updated, it is limited to the normalized interval [0, 1]. The positions of all discharge power corresponding to the reverse overload period are forcibly set to zero.

[0016] The advantages of this invention are: This invention, through the improvement of the particle swarm optimization algorithm and its deep integration with distributed photovoltaic-storage grid-connected scenarios, achieves targeted power control during photovoltaic reverse overload periods, offering the following significant advantages compared to existing technologies: 1. Improved algorithm convergence and optimization performance: A heuristic initialization strategy based on electricity price, photovoltaic output, and load fluctuations is designed. At the same time, an inertial weight decay strategy is adopted to solve the problems of random initialization and poor local search ability in the later stage of traditional particle swarm optimization algorithm. The algorithm converges faster, is less likely to get trapped in local optima, and the optimization results are more in line with engineering practice.

[0017] 2. Significant safety protection effect of transformer: The core control strategy of prioritizing charging of energy storage and forcing discharge to zero during photovoltaic reverse overload periods is formulated, which effectively absorbs excess photovoltaic power and reduces the reverse load rate of transformer. At the same time, transformer temperature rise and aging factors are incorporated into the cost model to achieve quantitative control of transformer life loss and significantly reduce the risk of transformer overload aging.

[0018] 3. Achieve multi-objective collaborative optimization: Establish a comprehensive economic cost model that includes equipment investment, charging and discharging losses, electricity purchase and sale, and transformer life loss. This model takes into account both the economic operation of the system and the safety of energy storage / transformer equipment. The optimization results can not only reduce the total daily economic cost, but also ensure that the energy storage SOC is within a safe range and that the transformer load rate and temperature are at a reasonable level.

[0019] 4. Scientific and reasonable constraint system design: A multi-dimensional constraint system of hard constraints and soft constraints is adopted. Hard constraints ensure that the physical boundaries of the equipment are not broken, while soft constraints suppress photovoltaic reverse feed and extreme load rate of transformers through targeted penalties, avoiding over-optimization caused by indiscriminate penalties and improving the engineering practicality of the optimization results.

[0020] 5. Strong scenario adaptability: The algorithm and control strategy are designed based on actual 24-hour distribution network load and photovoltaic output data with a time resolution of 15 minutes. It can be directly adapted to the distributed photovoltaic-storage grid-connected scenario of low-voltage distribution network. Moreover, the parameters can be flexibly adjusted and adapted according to the electricity price and equipment parameters in different regions, which has broad application value. Attached Figure Description

[0021] Figure 1 This embodiment presents a distributed photovoltaic-storage grid-connected power optimization control method based on an improved particle swarm optimization algorithm. Figure 2 A comparison of the improved particle swarm optimization algorithm and the traditional particle swarm optimization iterative convergence curves in this embodiment; Figure 3 This embodiment compares the improved particle swarm optimization algorithm with the traditional particle swarm optimization transformer state. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0023] This invention provides a distributed photovoltaic-storage grid-connected power optimization control method based on an improved particle swarm optimization algorithm. For distributed photovoltaic-storage grid-connected scenarios on a 24-hour timescale, the method divides the time period into 15-minute sub-periods and uses a heuristic improved particle swarm optimization algorithm to optimize the energy storage charging and discharging power. Figure 1 As shown, the specific steps include: S1: Determine system basic parameters and perform heuristic initialization S2: Establish a dynamic model for energy storage and a quantitative model for transformer temperature rise and aging factors. S3: Establish a system economic cost model that includes transformer lifespan losses. S4: Construct the overall fitness function of the system S5: Implement improved particle swarm master iterative optimization by introducing a nonlinear feedback mechanism. S6: Testing Process, Comparison Scaling, and Result Analysis Furthermore, in step S1, the heuristic initialization of the improved particle swarm optimization algorithm specifically includes: pre-calculating the initial power of the power grid when there is no charging or discharging. ,in It is the initial power of the power grid. It is the load power. This refers to photovoltaic power; the index of the reverse overload period when photovoltaic output exceeds load is extracted. To accelerate algorithm convergence and protect the transformer, a heuristic guidance ratio is set. The following directional guidance rule is applied to this portion of particles (80% of particles), while the remaining particles are kept randomly initialized to maintain population diversity: Off-peak period inspiration: During off-peak electricity price periods, force the charging power of particles to be initialized to a higher value to encourage exploration of charging using low-priced electricity; Reverse overload period inspiration: During the reverse overload period, the charging power of the particles is forced to be initialized to a value close to the maximum power, and the discharge power is forced to be 0, so as to achieve priority absorption; Peak period inspiration: During periods of highest electricity prices, initial particles tend to discharge at higher power to achieve peak shaving and valley filling.

[0024] Furthermore, the modeling of the photovoltaic system and the distributed energy storage system in step S2 includes the following specific steps: S21: Dynamic model of energy storage state of charge (SOC): The state of charge (SBC) of energy storage reflects the remaining amount of energy stored. Its dynamic changes are related to charging and discharging power, charging and discharging efficiency, and time step, and must also meet certain requirements. The hard constraint is that the initial charge state SOC(1) = 0.1E. b To ensure the sustainability of the energy storage system's power supply over multiple days of continuous operation, the state of charge of the energy storage system is measured within a 24-hour operating cycle. The constraints are as follows:

[0025] in, Indexed by time period; The energy storage charging power during time period t; The energy storage discharge power during time period t; =0.25h is the time step; For charging efficiency, This refers to the discharge efficiency.

[0026] S22: Power Grid Interaction Model With the participation of energy storage charging and discharging power, the interaction power between the distribution network and the upstream grid is jointly determined by the load power, photovoltaic output, and energy storage charging and discharging power, as shown in the formula:

[0027] in, This indicates that the distribution network purchases electricity from the upstream power grid. This indicates that the distribution network is sending power in the reverse direction to the grid (photovoltaic reverse overload).

[0028] S23: Calculation Model for Transformer Temperature Rise and Aging Factor Taking thermal inertia into account, the temperature rise of the transformer top oil and the temperature rise of the hot spot are calculated as follows:

[0029] in, It is a steady-state top oil temperature rise. For reference, the temperature rise of the top oil layer; The transformer load rate during time period t; This refers to the transformer impedance ratio; n This represents the load loss ratio. It is a dynamic top-layer oil temperature rise (considering thermal inertia). The time constant of the top oil layer; Let e ​​be the time step, and e be the natural constant. It is a steady-state hotspot temperature rise. For reference, the temperature rises in hot spots; m This is the hotspot load factor. It is a dynamic hot spot temperature rise (considering thermal inertia). This refers to the hot spot temperature of the transformer. For hotspot time constant, The ambient temperature.

[0030] The aging factor reflects the impact of hot spot temperature on the insulation life of a transformer. The higher the temperature, the greater the aging factor. The formula is as follows: It is an aging factor, and exp is the natural exponential function. This refers to the hot spot temperature of the transformer.

[0031]

[0032] Where 16158 is the activation energy constant, 371 is the reference temperature of the transformer's rated hot spot under absolute temperature, and 273 is the conversion constant for converting Celsius to absolute temperature.

[0033] Furthermore, the system's true economic cost model in step S3 includes the following specific steps: System daily total real economic cost The sum of daily average investment cost, charging and discharging loss cost, net cost of electricity purchase and sale from the grid, and transformer life loss cost is the core optimization objective of this invention, and the formula is as follows:

[0034] in, It is the system's total daily real economic cost. This is the average daily investment cost. These are energy storage capacity, energy storage power, and transformer unit cost, respectively. These are energy storage capacity, energy storage power, and transformer rated capacity, respectively. N This refers to the equipment's lifespan in years. The cost of charging and discharging losses is determined by the charging and discharging power, the unit cost of loss, and the time step. Cost per unit of charge / discharge loss. It is the net cost of electricity purchased and sold by the power grid. It refers to the power purchased by the power grid. The electricity purchase price for period t; It refers to the power output of the power grid. The electricity price is fixed. The cost of transformer lifespan loss is determined by the aging factor, lifespan loss coefficient, and time percentage, considering a lifespan loss amplification factor of 1.2. This is the transformer life loss factor. This is the time step, where 1440 represents the number of minutes in a day. The expected lifespan of the transformer is taken as 20.5 years.

[0035] Furthermore, the fitness function model in step S4 is as follows: S41: Algorithm Fitness Function Model The fitness function is the objective function of the particle swarm optimization algorithm. Based on the actual economic cost, soft and hard constraint penalty terms are introduced to guide the algorithm to converge to an engineering-feasible optimal solution. The formula is:

[0036] It's about adaptability. It is the system's total daily real economic cost. This is a soft penalty term for transformer load rate, which applies a cubic penalty to extreme load rates exceeding a threshold to selectively suppress transformer overload. This is the transformer load rate threshold.

[0037] The intraday SOC cycle penalty term applies a very large penalty weight to the absolute value of the final SOC deviating from the initial SOC at the end of the day, ensuring that the system achieves a complete power closed loop within a 24-hour cycle. It is the initial value of SOC during the day. It is the final value of SOC in a day.

[0038] It is a hard constraint penalty item: a high penalty is imposed for exceeding the SOC limit or the charging and discharging power limit to ensure that the physical boundaries of the equipment are not breached. It is the minimum SOC value of the day. It is the minimum SOC value of the day. It refers to the battery charging power. It is the battery discharge power.

[0039] Furthermore, in step S5, in response to the problem that the traditional particle swarm optimization algorithm's linear decay strategy is prone to getting trapped in local optima and lacks the ability to find high-dimensional optimization, this invention proposes an asymmetric nonlinear adaptive strategy based on population state feedback to reconstruct the dynamic parameter update mechanism.

[0040] S51: Particle Swarm Optimization Heuristic Initialization Heuristic initialization is performed to generate the initial population for the improved particle swarm optimization algorithm. The heuristic initialization includes: pre-calculating the initial grid power when there is no energy storage charging and discharging, and extracting the index of reverse overload periods when photovoltaic output is greater than the load. The directional guidance rules include: during off-peak electricity price periods, initializing the charging power dimension of the particles to 0.6 to 0.8 times the maximum charging power; during reverse overload periods, initializing the charging power dimension of the particles to 0.8 to 1.0 times the maximum charging power, and forcibly initializing the discharging power to 0; during periods with the highest electricity price, initializing the discharging power of the particles to 0.5 to 0.7 times the maximum discharging power. S52: Personalized Inertia Weights Based on Individual Fitness Feedback Instead of the traditional particle swarm optimization algorithm where all particles share a uniform weight, a dynamic feedback weight system is constructed that varies depending on the individual particle's state, as shown in the following formula:

[0041] The global baseline inertia weight at the k-th iteration is determined by the current iteration number k. It is the given minimum inertia weight. Given the maximum inertia weight, exp is the natural exponential function. It is the given maximum number of iterations. a , b It is a given constant ，a It is the attenuation coefficient, which controls the steepness of the curve. a The larger the value, the more drastic the weight decay during the descent phase; b is the center offset, which controls the inflection point position.

[0042] For the first i The normalized relative fitness of each particle. For the first i The current fitness value of each particle. It is the optimal (minimum) fitness value of the entire population in the current iteration. It is the average fitness value of the entire population in the current iteration. To prevent extremely small constants with a denominator of zero.

[0043] For the k-th iteration, the inertial weight actually assigned to the i-th particle is given by e, which is a natural constant.

[0044] To address the shortcomings of traditional decreasing strategies in clearly defining the boundaries between global exploration and local development, an innovative S-shaped nonlinear inertial weight based on the Sigmoid map is introduced. This mechanism exhibits a significant "smoothing locking" effect in both the early and later stages of iteration, avoiding premature convergence in the early stages and greatly enhancing the micro-level fine-grained search capability in the later stages. Subsequently, the optimal fitness of the current individual in the population is calculated in each iteration. With average fitness .

[0045] When the particle performs better than the average ( )hour, The calculation result will be between [0, 1]. The closer the particle is to the current global optimum ( The closer ), The closer it gets to 0.

[0046] According to particles Current fitness Dynamically allocate personalized inertia weights : like If a particle is identified as dominant, it is subjected to local fine mining based on its target gap; otherwise, it is identified as inferior and given a higher baseline inertia to enable it to leap out of the current local minimum region in a large step.

[0047] Among them, S53: Dynamic learning factor for asymmetric trigonometric functions: Cosine and sine functions are used to control cognitive factors respectively. Social factors The algorithm undergoes nonlinear cross-evolution. This mechanism enables the algorithm to possess strong independent exploration capabilities in the early stages of iteration, and to quickly and smoothly converge towards the global optimum in the later stages. The formula is as follows:

[0048] in, k This represents the current iteration number. This represents the maximum number of iterations. The first k Cognitive and social factors in the next iteration These are the given initial and final values ​​of the cognitive factors. These are the initial and final values ​​of the given social factors, respectively.

[0049] Particle velocity update: Introducing individual and global learning factors, combined with random numbers, to achieve velocity updates, while limiting the velocity range to [-0.15, 0.15]. The formula is as follows:

[0050] in, Let be the velocity of the i-th particle in the k-th iteration; This represents the position of the i-th particle in the (k-1)-th iteration. A random number within the range [0,1]. Let i be the optimal position for the i-th particle. This is the globally optimal position.

[0051] Particle position update: Adjust particle positions according to the updated velocity, while limiting the position range to [0,1] (after normalization), and force the discharge power position to 0 during the photovoltaic reverse overload period. The formula is:

[0052] in, Let i be the position of the i-th particle after the k-th iteration. Let i be the position of the i-th particle after the previous iteration. Let be the velocity of the i-th particle after the k-th iteration. T=96 represents the charging power dimension. T+ overload This is an index of the discharge power location during the photovoltaic reverse overload period.

[0053] Furthermore, the system's true economic cost model in step S6 includes the following specific steps: S61: Test Scenarios and Parameter Settings A typical low-voltage distribution network area containing distributed photovoltaic and energy storage was selected as the test object. The time resolution was set to 15 minutes, and the optimization cycle was 24 hours (i.e., 96 time periods). The basic system parameters are as follows: Equipment parameters: The rated capacity of the energy storage system is 500 kWh, the rated charge and discharge power is 100 kW, and the initial SOC is set to 0.1; the rated capacity of the transformer is 100 kVA. Electricity price parameters: A time-of-use pricing mechanism is adopted, with off-peak hours from 00:00 to 03:00 and 20:00 to 24:00 (price 0.4 yuan / kWh), flat hours from 03:00 to 14:00 (price 0.8 yuan / kWh), and peak hours from 14:00 to 20:00 (price 1.2 yuan / kWh). The grid-connected electricity price is uniformly 0.25 yuan / kWh. Algorithm parameter settings: The population size of the improved particle swarm algorithm in this invention is set to 300, and the maximum number of iterations is set to 5000.

[0054] S62: Scale settings To highlight the superiority of this invention, the following two methods are used for simulation comparison under the same conditions: Comparative Example (Traditional Method): The standard particle swarm optimization (PSO) algorithm is used. Its inertia weights employ a fixed linear decreasing strategy, and the learning factor is constant. Global random generation is used during the initialization phase. The fitness function only considers basic electricity purchase and sale costs and charging / discharging losses, without including transformer lifespan losses in the quantification cost, and does not include a mandatory control strategy for photovoltaic reverse overload periods.

[0055] The method of this invention (proposed method) employs the improved particle swarm optimization algorithm described in this invention, which considers transformer aging and multiple constraints. It includes heuristic initialization, personalized nonlinear inertia weights, and dynamic learning factors; it comprehensively quantifies the transformer lifetime loss cost in the fitness function and introduces soft and hard constraint penalty mechanisms; during reverse overload periods when photovoltaic output exceeds load, it enforces a strategy of prioritizing energy storage charging and setting the discharge power to 0.

[0056] S63: Algorithm Performance Comparison Analysis The iterative convergence curves of the two algorithms are as follows: Figure 2 As shown, the comprehensive economic cost includes four parts: daily investment depreciation, net cost of electricity purchase and sale from the grid, energy storage charging and discharging losses, and transformer life loss.

[0057] With the same population size (300) and maximum number of iterations (5000), the global optimal cost of the traditional standard PSO stops improving after about 41 iterations, and then stagnates completely for nearly 5000 iterations, with a final daily comprehensive cost of 783.80 yuan. The method of this invention shortens the convergence generation to about 3 generations, with a final daily comprehensive cost of 403.24 yuan, representing a cost reduction of 48.55%.

[0058] The superiority of this invention lies in its fast convergence speed and strong global optimization capability. Traditional PSO's fitness function does not include transformer aging costs, leading the algorithm to blindly pursue low electricity prices by subjecting transformers to prolonged severe overload (load rate as high as 183.70%) to reduce electricity purchase costs. Its search direction is blind and even erroneous in terms of real economic efficiency. In contrast, the method of this invention fully quantifies transformer lifetime losses and embeds them into the fitness function, ensuring that the search direction is completely consistent with the actual comprehensive cost, fundamentally guaranteeing the accuracy of convergence. This is due to: first, heuristic initialization utilizes prior knowledge of photovoltaic power consumption and time-of-use pricing, providing the algorithm with a high-quality initial population; second, personalized inertial weights based on individual fitness feedback and asymmetric trigonometric function dynamic learning factors successfully achieve a balance between broad-area exploration in the early stage and fine-tuning in the later stage.

[0059] S64: Comparison of Transformer Operating Status and Lifespan Loss Comparison of transformer temperature using two algorithms Figure 3 As shown in the diagram, under traditional methods, the combined effect of photovoltaic reverse overload and peak-hour heavy load periods resulted in a transformer maximum average load rate of 155.56%, with the hot spot temperature reaching a peak of 167.59 ℃, far exceeding the transformer's insulation heat resistance limit. The quantified cost of transformer life loss per day reached as high as 249.34 yuan. Using the method of this invention, the transformer's maximum average load rate was reduced to below 125.00%, and the peak hot spot temperature significantly decreased by 31.99 ℃ (from 167.59 ℃ to 135.60 ℃), drastically reducing the daily transformer life loss cost to 78.19 yuan.

[0060] The superiority of this invention lies in its effective protection of the operational safety of underlying transformer equipment. This is because the invention constructs a quantitative model of transformer temperature rise and aging factors based on the dual-time-constant thermal inertia effect, accurately monetizing lifespan loss and introducing it into economic cost and fitness functions. This allows the optimizer to automatically perceive the economic losses caused by high temperatures during the search process. Simultaneously, combined with soft and hard constraint penalties for exceeding transformer load rate limits, the algorithm is forced to actively avoid extreme operating conditions that would cause excessive transformer temperature rise during optimization, thereby significantly delaying equipment aging and extending the actual service life of the transformer.

[0061] S65: Comparison of Comprehensive Economic Benefits Taking into account daily investment depreciation (90.6 yuan), energy storage charging and discharging losses, net electricity purchase and sale costs from the grid, and transformer lifespan losses, the total daily real economic cost of the traditional method is 783.80 yuan. The total daily real economic cost of the method of this invention is 403.24 yuan, representing a 48.55% reduction in overall cost.

[0062] In the traditional method, transformer life loss (249.34 yuan) is the largest cost item, which exposes the serious economic flaws of the dispatch strategy that does not take into account transformer aging. Although the method of this invention compromises on the power purchase strategy in local periods to protect the transformer and maintain the SOC closed loop (the power purchase cost is about 189 yuan, which is basically the same as the traditional method), it achieves true global comprehensive economic optimization from the perspective of the whole life cycle system by reducing the transformer life loss cost from 249.34 yuan to 78.19 yuan.

[0063] The superiority of this invention lies in achieving global economic optimization under multi-objective coordination. It does not unilaterally pursue the minimization of a single cost item, but rather places energy storage cycle, grid interaction, and transformer aging within a unified monetization framework for coordinated optimization, fully embodying the idea of ​​source-grid-load-storage coordinated scheduling that takes into account equipment lifespan constraints.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distributed photovoltaic-storage grid-connected power optimization control method, characterized in that, Includes the following steps: Pre-calculate the initial power of the grid when there is no energy storage charging and discharging, extract the reverse overload period when the photovoltaic output is greater than the load, and perform heuristic initialization of the particle swarm based on the prior knowledge of time-of-use pricing, photovoltaic output and load fluctuations. A quantitative model for transformer temperature rise and aging factor considering the thermal inertia effect of dual time constants is established, and the transformer life loss is monetized and included in the total daily economic cost of the system, which includes equipment investment depreciation, charging and discharging losses, and net cost of purchasing and selling electricity from the grid. Construct a fitness function based on the total daily economic cost of the system, superimposed with a soft penalty term for transformer load rate, a closed-loop penalty term for SOC, and a hard constraint penalty term for equipment; An improved particle swarm optimization is performed, which uses personalized inertia weights based on Sigmoid baseline weights and relative fitness feedback, along with asymmetric trigonometric function dynamic learning factors, to update the particle state. During the reverse overload period, energy storage is forced to charge and the discharge power is zero, and the optimal energy storage charging and discharging power scheduling scheme is output.

2. The method according to claim 1, characterized in that, The heuristic initialization specifically refers to: For 80% of the particles, perform directional initialization according to the following rules: During off-peak electricity price periods, initialize the particle charging power dimension to 0.6 to 0.8 times the maximum charging power; during reverse overload periods, initialize the particle charging power dimension to 0.8 to 1.0 times the maximum charging power, and forcibly initialize the discharge power dimension to 0; during peak electricity price periods, initialize the particle discharge power dimension to 0.5 to 0.7 times the maximum discharge power. The remaining 20% ​​of the particles are kept globally randomized to maintain population diversity.

3. The method according to claim 1, characterized in that, The quantitative model for transformer temperature rise and aging factors includes the following: The temperature rise of the top oil is divided into steady-state value and dynamic value. The steady-state temperature rise of the top oil is determined by the reference top oil temperature rise, transformer load rate, impedance ratio and load loss ratio. The dynamic temperature rise of the top oil is based on the top oil time constant and the thermal inertia hysteresis effect is introduced for correction. Hot spot temperature rise is divided into steady-state value and dynamic value. The steady-state hot spot temperature rise is determined by the reference hot spot temperature rise, transformer load rate and hot spot load coefficient. The dynamic hot spot temperature rise is based on the hot spot time constant and the thermal inertia hysteresis effect is introduced for correction. The transformer hot spot temperature is obtained by superimposing the ambient temperature and the dynamic hot spot temperature rise. The aging factor is calculated exponentially based on hot spot temperature to quantify the life loss of insulation materials; the transformer life loss cost is calculated in monetary terms by combining the aging factor with time step, expected transformer life, and life loss amplification factor, and is included in the system economic cost.

4. The method according to claim 1, characterized in that, The total daily economic cost of the system is the sum of the following items: The average daily investment depreciation cost is calculated by multiplying the rated capacity and rated power of the energy storage system and the rated capacity of the transformer by the corresponding unit cost, and then dividing by the equipment lifespan. The cost of charging and discharging losses is the sum of the charging and discharging power and unit loss cost at each time period and the time step. The net cost of electricity purchase and sale by the power grid is the sum of the product of the time-of-use electricity purchase price and the purchased power, minus the sum of the product of the fixed electricity sales price and the sold power. The cost of transformer life loss is determined by the aging factor, transformer life loss coefficient, the ratio of time step to expected transformer life, and life loss amplification factor, and is quantified in monetary terms.

5. The method according to any one of claims 1 to 4, characterized in that, The specific steps for updating the particle state are as follows: The global baseline inertia weight is based on the Sigmoid mapping and decays nonlinearly with the iteration process. The parameters controlling the steepness of the curve and the position of the inflection point are configurable. In each iteration, the current optimal fitness and average fitness of the population are statistically analyzed, and the normalized relative fitness of each particle is calculated. Particles with fitness better than the average level of the population are identified as dominant particles, and are given an inertial weight lower than the benchmark weight based on their target difference to perform fine local search. Particles with fitness worse than the average level of the population are identified as suboptimal particles, and are given an inertial weight higher than the benchmark weight to encourage them to escape local extreme regions. In the asymmetric trigonometric function dynamic learning factor, the cognitive factor decreases non-linearly from the initial value to the final value according to the cosine function as the iteration process progresses, and the social factor increases non-linearly from the initial value to the final value according to the sine function as the iteration process progresses; after the particle velocity is updated, it is limited to a set range; after the position is updated, the discharge power position corresponding to the reverse overload period is forcibly set to zero, and all positions are limited to the normalized interval.

6. A distributed photovoltaic-storage grid-connected power optimization control system, characterized in that, include: The initialization module is used to obtain basic system parameters, pre-calculate the initial power of the grid when there is no energy storage charging and discharging, extract the reverse overload period when the photovoltaic output is greater than the load, and perform heuristic initialization of the particle swarm based on the prior knowledge of time-of-use pricing, photovoltaic output and load fluctuations. The dynamic modeling module is used to establish a dynamic model of the energy storage state of charge and a power interaction model with the power grid, as well as a quantitative model of transformer temperature rise and aging factor that takes into account the thermal inertia effect of dual time constants. The economic cost model module is used to establish a system daily total economic cost model that includes equipment investment depreciation costs, charging and discharging loss costs, net electricity purchase and sale costs from the grid, and transformer life loss costs. The fitness function construction module is used to construct the system fitness function by adding a transformer load rate soft penalty term, a SOC closed-loop penalty term, and a device hard constraint penalty term to the system's daily total economic cost. The particle swarm optimization module is improved by using personalized inertia weights based on Sigmoid baseline weights and relative fitness feedback, along with asymmetric trigonometric function dynamic learning factors to update particle states. During the reverse overload period, energy storage is forced to charge while the discharge power is zero, and the optimal energy storage charging and discharging power scheduling scheme is output.

7. The system according to claim 6, characterized in that, The initialization module performs targeted initialization on 80% of the particles according to the following rules: during off-peak electricity price periods, the particle charging power dimension is initialized to 0.6 to 0.8 times the maximum charging power; during reverse overload periods, the particle charging power dimension is initialized to 0.8 to 1.0 times the maximum charging power and the discharge power is forcibly initialized to 0; during peak electricity price periods, the particle discharge power dimension is initialized to 0.5 to 0.7 times the maximum discharge power; the remaining 20% ​​of the particles are kept globally randomized to maintain population diversity.

8. The system according to claim 6, characterized in that, In the dynamic modeling module: The dynamic model of energy storage state of charge takes into account charging and discharging efficiency and time step, and applies hard constraints on the upper and lower limits of state of charge and the sustainability constraints of electricity within a 24-hour operating cycle. The initial state of charge is set as a fixed proportion of the rated capacity. In the quantitative model of transformer temperature rise and aging factor, the steady-state value and dynamic thermal inertia correction value based on the corresponding time constant are calculated for both top oil temperature rise and hot spot temperature rise. The aging factor is calculated exponentially according to the hot spot temperature. The transformer life loss cost is incorporated into the economic cost after being calculated in monetary terms by combining the aging factor, time step, expected transformer life and life loss amplification factor.

9. The system according to claim 6, characterized in that, In the fitness function construction module: The transformer load rate soft penalty term applies a cubic penalty to load rates exceeding a set threshold, selectively suppressing extreme transformer overloads without applying indiscriminate penalties; The SOC closed-loop penalty term applies a high-weight penalty to the absolute value of the deviation of the state of charge from the initial value at the end of the operating cycle, ensuring that the system achieves complete power closed-loop within the 24-hour scheduling cycle; The hard constraint penalty term applies high penalties for exceeding the limits of state of charge, charging power, and discharging power, respectively, to ensure that the physical boundaries of the energy storage device are not breached.

10. The system according to claim 6, characterized in that, In the improved particle swarm optimization module: The global baseline inertia weight is based on the Sigmoid mapping and decays nonlinearly with the iteration process. In each iteration, the optimal fitness and average fitness of the population are statistically analyzed, and the normalized relative fitness of each particle is calculated. The dominant particles are given an inertia weight lower than the baseline for fine local search, and the inferior particles are given an inertia weight higher than the baseline to escape local extrema. Cognitive factors decrease non-linearly according to a cosine function, while social factors increase non-linearly according to a sine function. The guiding algorithm evolves from individual independent exploration in the early stage to collaborative convergence towards the global optimum in the later stage. After the particle velocity is updated, it is limited to the range of [-0.15, 0.15], and after the particle position is updated, it is limited to the normalized interval [0,1]. The positions of all discharge power corresponding to the reverse overload period are forcibly set to zero.

Images