Multi-target dynamic optimization control method and device for optical storage system

Abstract

The invention discloses a multi-target dynamic optimization control method and device for an optical storage system, and the method achieves the multi-target optimization through the cooperative control of dynamic weight migration triggered by a temperature gradient and high-frequency fluctuation prediction, and comprises the steps: constructing a fuzzy temperature-service life-power grid three-domain dynamic coupling decision model; the battery temperature gradient, the charge state change rate and the power grid frequency deviation are input into a three-dimensional fuzzy inference engine, and a temperature influence factor eta (T) and an economical efficiency weight lambda eco are output; and when the battery temperature T enters a critical threshold offset range, activating a dual-channel temperature response mechanism with conflict resolution capability: a prevention channel and a protection channel (T is greater than or equal to Tcrit). The invention further provides a multi-target dynamic optimization control device for the optical storage system. The multi-target dynamic optimization control device comprises an edge calculation unit, a dynamic compiling module and an instruction generation module for executing the control method. According to the invention, multi-objective optimization is realized through cooperative control of dynamic weight migration triggered by temperature gradient and high-frequency fluctuation prediction.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic energy storage technology, and more specifically to a multi-objective dynamic optimization control method and device for a photovoltaic energy storage system. Background Art

[0002] In the context of the accelerated construction of new power systems, the photovoltaic storage system is a key facility that supports a high proportion of new energy consumption and flexible grid regulation. Its multi-objective coordinated control capability directly affects the economy and safety of the system. The current mainstream control strategy has exposed several technical bottlenecks when dealing with complex working conditions such as high-temperature frequency modulation and sudden changes in cloud cover: the traditional method adjusts the optimization target weight based on fixed rules or a single temperature threshold. Although it can achieve basic multi-objective optimization, it has no effect on the battery temperature change rate (such as temperature gradient). The dynamic characteristics of the battery's temperature (unit: °C / s) are poorly understood, especially in rapid temperature rise scenarios, where delayed weight migration significantly exacerbates battery lifespan loss. Furthermore, the linear power attenuation strategy used in existing derating control differs fundamentally from the nonlinear characteristics of the battery's actual temperature rise rate, making it difficult to balance protection effectiveness with system performance during sudden temperature changes. Aging batteries face a higher risk of thermal runaway due to the rigidity of their protection threshold parameters.

[0003] At the same time, when the energy storage system enters a silent period due to temperature protection, the contradiction between output interruption and the grid's inertia support requirements becomes prominent. Secondary frequency fluctuations frequently occur in low-inertia grids, directly impacting frequency regulation quality and power supply reliability. This contradiction is particularly prominent in areas with high penetration of new energy sources, exposing a lack of coordinated design between temperature protection mechanisms and the dynamic needs of the grid.

[0004] In addition, the limitations of photovoltaic power prediction models exacerbate the pressure on system control. Existing methods are unable to capture the short-term, high-frequency fluctuations caused by cloud movement, resulting in timing deviations between energy storage charging and discharging instructions and actual power demand, indirectly causing frequent charging and discharging of batteries and temperature fluctuations, forming a vicious cycle of life degradation and inaccurate control.

[0005] The existence of the above-mentioned problems makes it difficult for photovoltaic storage systems to simultaneously take into account both service life and grid stability under extreme working conditions, restricting their large-scale application and efficiency release in the power system. Summary of the Invention

[0006] In order to solve one or more of the above technical problems, the present invention provides the following technical solutions:

[0007] A multi-objective dynamic optimization control method for a photovoltaic storage system achieves multi-objective optimization through the coordinated control of dynamic weight migration triggered by temperature gradient and high-frequency fluctuation prediction, including:

[0008] Construct a fuzzy temperature-life-grid three-domain dynamic coupling decision model to convert the battery temperature gradient The three-dimensional fuzzy inference engine inputs the temperature impact factor η(T) (dimensionless, representing the weight of the temperature's impact on battery life) and the economic weight λ_eco (dimensionless, representing the weight of the economic target):

[0009] Prevention channel (T ≥ T_crit - 5°C): uses an S-shaped power derating curve that is adaptive to the dynamic trigger temperature T_trig. Its steepness coefficient α decays linearly with the battery state of health (SOH), satisfying α = α_0 (1-0.3 (1-SOH)), where α_0 is the initial steepness coefficient of 2.0 to 5.01 / °C.

[0010] Protection channel (T ≥ T_crit): Based on the Arrhenius-rainflow coupling model, the life loss rate D_rate is calculated in real time. The charge and discharge silent period Δt is generated by the coordinated calculation of the damage rate compensation coefficient k_d and the cooling system response time t_cool. During the silent period, the grid frequency feedforward compensation Δf_comp is injected. The compensation amount satisfies Δf_comp = K_p Δf (1-e^{-t / τ_c}), where K_p is the grid inertia compensation coefficient of 0.5 to 1.2, and τ_c is the inertia time constant of 2 to 5 seconds.

[0011] Multi-objective collaborative optimization expressed by the following function:

[0012] min(η(T)·C deg +(1-η(T))·[λ eco C eco +(1-λ eco )C grid ])

[0013] Where η(T) is corrected by coupling the hyperbolic tangent function with the sigmoid function:

[0014]

[0015] Where k is the slope coefficient (unit: 1 / °C, range of 0.5 to 2.0), Tsafe is the battery's nominal temperature upper limit minus 10°C, and δ is the temperature gradient enhancement factor (unit: °C / s, range of 0.2 to 0.5). This formula dynamically adjusts the weight of temperature's impact on battery life, η(T), by coupling the sigmoid function with the hyperbolic tangent function.

[0016] And the life loss cost C_deg introduces the temperature gradient enhancement factor:

[0017]

[0018] Furthermore, the dynamic trigger temperature T_trig is determined by the battery health state (SOH) and the temperature gradient Dynamic calibration to meet:

[0019] Wherein, ΔTbase is the reference offset (unit: °C) with a value range of 3 to 5 °C, and γ is the health compensation coefficient (unit: °C / %) with a value range of 0.5 to 1.5 °C / %. is the temperature gradient (unit: ℃ / s), when When the temperature is 0.04°C, T_trig is reduced by an additional 1 to 2°C. This formula comprehensively considers the impact of the battery health state (SOH) and temperature gradient on the trigger temperature, ensuring dynamic adjustment of the protection threshold under different operating conditions.

[0020] Furthermore, the calculation of the life loss rate D_rate introduces a dynamic activation energy compensation mechanism:

[0021] When the internal pressure fluctuation rate dP / dt of the battery is detected to be greater than 10 kPa / s, the corrected activation energy E_a is:

[0022] E'a=E a ·[1+0.15·sgn(dP / dt)·ln(1+|dP / dt|)]

[0023] The duration of the silent period Δt takes into account the grid frequency regulation requirements. When the grid equivalent inertia H_sys is less than 4 seconds, the following settings are mandatory:

[0024] Δtmax=min(t cool ,10·H sys )

[0025] Furthermore, the membership function of the three-dimensional fuzzy inference engine satisfies:

[0026] Temperature gradient The membership boundary value is dynamically adjusted according to the battery chemistry type. The "high" gradient threshold of lithium iron phosphate battery is 1.3 times that of ternary lithium battery.

[0027] The warning threshold of the grid frequency deviation Δf is negatively correlated with the grid inertia H_sys, satisfying Δf_{alert}=0.15+0.02·(10-H_sys)Hz. When H_sys≤5s, the emergency membership mapping is started.

[0028] Furthermore, the photovoltaic output is separated into a low-frequency component \(P_{low}\) and a high-frequency component \(P_{high}\) through empirical mode decomposition (EMD). An ARIMA model is constructed for \(P_{low}\) to predict the trend value, and an LSTM network is used for \(P_{high}\) to generate a probability distribution band \([P_{high}^{min}, P_{high}^{max}]\), and a cloud motion compensation term \(\Delta H_k\) is injected through a Kalman filter:

[0029]

[0030] where the cloud movement speed \(v_{cloud}\) is calculated through the power gradient of adjacent photovoltaic arrays:

[0031]

[0032] Furthermore, a double-layer determination mechanism for the transition zone is set:

[0033] Early warning area (\(T_{base}\leq T < T_{crit}-5^{\circ}C\)): The power is adjusted by linear attenuation weighted by the aging state:

[0034]

[0035] Emergency area (\(T\geq T_{crit}-5^{\circ}C\)): Switch to the S-shaped derating strategy, and the derating rate is increased to 1.5 to 2 times that of the early warning area.

[0036] Furthermore, dynamic inertial filtering is used for the smoothing process of the control command, and the time constant \(\tau\) satisfies:

[0037]

[0038] When a sudden change in the grid frequency (\(\Delta f > 0.3Hz\)) is detected, \(\tau\) is forced to be reduced to 0.05 seconds to accelerate the response.

[0039] Furthermore, the cyclic identification window of the rain flow counting method is dynamically associated with the temperature gradient Dynamic association:

[0040] When occurs, the cyclic identification window is compressed from the standard 30 minutes to 10 minutes, and the damage weight of small-amplitude cycles is increased by 20% to 50%.

[0041] Furthermore, the grid support cost \(C_{grid}\) in the objective function includes a frequency deviation penalty term:

[0042]

[0043] Among them, K_f = 0.8 to 1.5 yuan / Hz^{1.5}, K_{df} = 0.2 to 0.5 yuan / (Hz / s), and when H_sys < 5 seconds, K_f is increased to 1.5 times the original value.

[0044] Furthermore, the present invention also provides a control device for a multi-objective dynamic optimization control method of a photovoltaic storage system, comprising:

[0045] The edge computing unit collects battery temperature, SOC, and grid frequency at a sampling rate of 100Hz and calculates the temperature gradient in real time and dSOC / dt;

[0046] Dynamic compilation module, automatically switches the objective function calculation mode according to the temperature gradient range: when The module activates the emergency optimization algorithm, compressing the computation cycle to 5ms. This module, implemented on an edge computing unit (such as a 1GHz processor), ensures real-time performance through task priority scheduling and hardware acceleration. Experimental verification shows that the module's average runtime on actual hardware is 4.8ms, meeting the 5ms computation cycle requirement.

[0047] An instruction generation module, executing any of the above methods to generate control instructions, and outputting the instructions to the power conversion system through a dynamic inertia filter;

[0048] The device firmware integrates a battery parameter self-learning function, which can dynamically correct the T_crit, α, and k parameters based on historical operating data, with the correction range not exceeding ±20% of the initial value.

[0049] Beneficial effects

[0050] (1) The present invention uses a fuzzy decision-making mechanism based on temperature gradient perception to deeply integrate the dynamic characteristics of battery temperature rise, aging status, and real-time grid demand, breaking through the static weight limitations of traditional strategies and significantly improving the life protection response capability in high-temperature frequency modulation scenarios while ensuring grid frequency stability.

[0051] (2) The present invention matches the nonlinear characteristics of the battery temperature rise rate through a hierarchical protection mechanism, achieving coordinated optimization of protection efficiency and system performance during the temperature mutation stage, effectively reducing the risk of thermal runaway and extending the battery life;

[0052] (3) The present invention introduces an inertia compensation mechanism when the temperature protection is triggered, and balances the energy storage protection and the grid inertia demand by dynamically adjusting the output curve, significantly alleviating the frequency secondary fluctuation problem in the low-inertia grid;

[0053] (4) The present invention integrates high-frequency fluctuation prediction and compensation technology to enhance the ability to capture short-term sudden changes in photovoltaic power, reduce the timing deviation between energy storage charging and discharging instructions and actual demand, and reduce the impact of temperature rise fluctuations on battery life from the source. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is the overall system architecture diagram, showing the multi-level modular design and data interaction relationship between the data acquisition layer, edge computing layer, decision control layer and execution layer;

[0055] Figure 2 This is a control flow chart that describes the logic judgment process of the temperature gradient triggering dual-channel protection mechanism and the multi-objective optimization instruction generation path;

[0056] Figure 3 The three-dimensional dynamic weight factor distribution surface reveals that η(T) varies with battery temperature T and temperature gradient nonlinear coupling characteristics.

[0057] Figure 4 The figure shows the comparison of S-shaped derating curves under different SOH conditions, reflecting the accelerated power attenuation characteristics of aging batteries in the critical temperature zone.

[0058] Figure 5 The comparison curve for photovoltaic high-frequency fluctuation prediction is used to verify the LSTM model's ability to capture cloud mutation events in advance and its compensation effect. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention clearer, the following will combine the technical logic and core innovations in the embodiments of the present invention to fully and detailedly explain the multi-objective dynamic optimization control method and device of the photovoltaic storage system. It should be noted that the specific embodiments described here are only used to explain the present invention, and do not limit the scope of protection of the present invention. Based on the technical essence of the present invention, the implementation methods derived by those skilled in the art without creative work are all within the scope of the claims of the present invention.

[0060] Example 1:

[0061] This embodiment is based on a fuzzy temperature-lifespan-grid three-domain dynamic coupling decision model, combined with an S-shaped weight migration mechanism and a three-dimensional fuzzy inference engine to achieve coordinated optimization of battery temperature, lifespan, and grid stability. The specific implementation process is as follows:

[0062] (1) Collaborative perception and feature fusion of multi-source data

[0063] like Figure 1 As shown, a distributed sensor network is deployed at key nodes of the optical storage system to collect the following parameters in real time:

[0064] Battery domain: Cell temperature T i (t)(i=1,2,...,n), temperature rise gradient Health status Among them Ck (t) is the measured capacity at the kth cycle;

[0065] Grid domain: frequency deviation Δf(t) = f grid (t)-50Hz, system inertia time constant J g is the inertia constant of the synchronous unit, S base is the system baseline capacity;

[0066] Photovoltaic domain: Through the joint modeling of empirical mode decomposition (EMD) and long short-term memory network (LSTM), the high-frequency fluctuation components of photovoltaic output are extracted. IMF m (t) is the mth order eigenmode function decomposed by EMD.

[0067] (2) Fuzzy temperature-life-grid three-domain dynamic coupling decision

[0068] Build a three-dimensional fuzzy inference engine and define the mapping relationship between input variables and output weights:

[0069] Input variables:

[0070] Temperature gradient membership function: where k T is the slope adjustment coefficient;

[0071] Grid frequency deviation membership function: Gaussian distribution is used to characterize the frequency urgency;

[0072] SOH attenuation membership function: μ SOH (SOH)=1-SOH.

[0073] Fuzzy rule base: defines 27 fuzzy rules, such as: HIGH AND Δf IS LOW AND SOH IS LOW, THEN w life IS is extremely high;

[0074] Defuzzification output: Calculate dynamic life weight using the center of gravity method:

[0075]

[0076] where μ r is the activation degree of the rth rule, w life,r Preset weight values ​​for the rules.

[0077] See also Figure 2 Control flow chart, 3D fuzzy inference engine receiving dSOC / dt and Δf input, through Figure 3The nonlinear mapping of the three-dimensional dynamic weight factor surface η(T) outputs the life protection weight and economic weight. When triggered Figure 2 The S-type derating strategy of the prevention channel, the parameterization process of the power adjustment curve is as follows Figure 4 shown.

[0078] (3) Implementation of hierarchical dual-channel protection mechanism

[0079] when When the prevention channel is activated, the aging adaptive S-shaped derating curve is adopted:

[0080]

[0081] Where k derate =k0·(2-SOH(t)), which increases the derating rate of aged batteries (SOH<0.8) to 1.5 times.

[0082] When any monomer temperature T i (t)>T th (SOH), the protection channel is triggered:

[0083] Dynamic temperature threshold:

[0084] T th (SOH) = T th0 -η·(1-SOH(t))

[0085] Where η = 5°C is the aging compensation coefficient, which reduces the threshold by 1.5°C when SOH = 0.7;

[0086] Inertia compensation power calculation:

[0087]

[0088] The compensation intensity is controlled by adjusting λ∈[0.1,0.5] to avoid power mutation exacerbating temperature rise.

[0089] When entering Figure 2 After the protection channel is completed, the charge and discharge silent period Δt is dynamically generated based on the life loss rate D_rate calculated by the modified Arrhenius-Rainflow model, and the grid frequency fluctuation is alleviated through the feedforward compensation mechanism. Figure 4 As shown in the comparison of S-shaped derating curves under different SOH, the trigger temperature threshold T_trig of the aged battery (SOH=0.7) is 1.8°C lower than that of the new battery, and the steepness coefficient α of the derating curve is increased to 1.3 times the initial value.

[0090] (4) Photovoltaic high-frequency fluctuation prediction-compensation linkage

[0091] First, perform EMD decomposition and convert the original photovoltaic power P pv (t) is decomposed into M intrinsic mode functions (IMFs) and residual terms:

[0092]

[0093] Then, perform LSTM prediction: input the high-frequency component IMF1(t) into the LSTM network and output the predicted fluctuation value within the next Δt The network structure contains 2 hidden layers, and the loss function is:

[0094]

[0095] The prediction results are fed forward to the energy storage control instruction generation module to correct the charge and discharge power instructions:

[0096]

[0097] Where γ = 0.8 is the compensation gain, which is dynamically adjusted by the Kalman filter to balance the impact of the prediction error.

[0098] like Figure 5 The PV high-frequency fluctuation prediction comparison curve shows that the LSTM-EMD joint model predicts cloud cover mutation events up to 18 seconds ahead (marked at t = 18s), with a prediction error of less than 8%. By feeding the compensation value ΔH_k into the energy storage control command, the temperature rise fluctuation caused by prediction lag is significantly reduced.

[0099] (5) Multi-objective dynamic optimization execution

[0100] Construct an objective function that includes life loss, grid support cost, and economic indicators, and use the following function to express multi-objective collaborative optimization:

[0101]

[0102] Constraints: T i ≤T th (SOH)

[0103] The improved particle swarm algorithm (introducing the inertia weight adaptive mechanism) is used for real-time optimization, and the particle velocity update formula is as follows:

[0104]

[0105] in K maxis the maximum number of iterations; where ω(k) is the inertia weight, c1=2.0 and c2=2.0 are acceleration constants, and r1 and r2 are uniformly distributed random numbers. The algorithm dynamically adjusts the inertia weight to balance global search and local optimization capabilities, ensuring rapid convergence under complex working conditions.

[0106] Through the above-mentioned implementation methods, the present invention systematically solves the problems of life protection, temperature rise suppression and grid stability coordination of photovoltaic storage systems in complex scenarios such as high-temperature frequency modulation and low-inertia power grids, significantly improving the economy and reliability of the system.

[0107] Example 2:

[0108] like Figure 1 As shown in the system architecture, a photovoltaic storage system in a coastal industrial park is connected to the regional power grid. The grid characteristics are as follows: renewable energy penetration rate: 62% (photovoltaic 45%, wind power 17%), system inertia time constant H sys =3.2s (5s below the safety threshold); the typical operating condition is the sudden drop in photovoltaic output at noon in summer (cloud cover) and the surge in air conditioning load, which causes frequency fluctuations (maximum deviation reaches ±0.5Hz). At the same time, the ambient temperature reaches 38°C, causing the temperature rise rate of the energy storage battery to exceed 0.8°C / min; the current battery status is a lithium iron phosphate battery pack (rated capacity 2MWh, SOH = 0.76), and the average annual capacity decay rate under the original control strategy is 8.3%.

[0109] The hardware deployment and parameter configuration implemented are as follows: an infrared temperature sensor (accuracy ±0.5°C) is deployed in each battery cluster, with a sampling period of 1s; a broadband measurement unit (PMU) is deployed at the grid connection point, with a frequency monitoring bandwidth of 0.1-100Hz; and an irradiance mutation detection module is deployed in the photovoltaic array (response time <200ms). Temperature warning threshold Critical threshold Dynamic temperature threshold T th0 =45℃, aging compensation coefficient η=6℃; inertia compensation nonlinear penalty coefficient λ=0.3, feedforward compensation gain γ=0.75.

[0110] The specific implementation process is as follows:

[0111] Step 1: Cloud layer mutation triggers compound disturbance (12:00-12:15), photovoltaic fluctuation: irradiance changes from 980W / m 2 Dropped to 320W / m 2 (lasting 5 minutes), the EMD-LSTM model predicted 18 seconds in advance The temperature rise accelerated: the battery pack increased from 0.5MW to 1.8MW due to compensation discharge power, and the maximum temperature rose from 39.2℃ to 43.1℃. The temperature rise gradient Reach 0.73℃ / min.

[0112] Step 2: Dynamic weight migration and hierarchical protection triggering

[0113] Fuzzy decision engine calculation:

[0114]

[0115] The life protection weight is increased by 70% compared to the traditional fixed weight (40%);

[0116] Prevent channel activation:

[0117]

[0118] The derating rate is 19%, and the temperature rise rate is suppressed to 0.51℃ / min.

[0119] Step 3: Inertia compensation and predictive feedforward linkage

[0120] Inertia Compensation Power Generation:

[0121]

[0122] That is, while derating, -0.12MW compensation power is provided to narrow the frequency deviation from -0.35Hz to -0.18Hz;

[0123] Photovoltaic fluctuation feedforward compensation:

[0124]

[0125] After correction, the actual discharge power is reduced by 0.9MW, and the temperature rise contribution is reduced by about 0.2℃ / min.

[0126] Step 4: Multi-objective optimization convergence

[0127] The improved particle swarm optimization algorithm converges after 35 iterations, and the optimal solution satisfies:

[0128] Temperature constraint: T max =44.2℃<T th =45-6×(1-0.76)=43.56°C (triggering protection channel boundary condition);

[0129] Frequency constraint: |Δf| max =0.21Hz<0.25Hz;

[0130] Economic loss: Deviation of charge and discharge plan is less than 12%.

[0131] Comparison of implementation effects:

[0132] Compared with the original control method of the industrial park's photovoltaic storage system, which was based on fixed rules or a single temperature threshold to adjust the optimization target weight, adopt a linear derating control strategy, and optimize a single grid frequency support target, the implementation effect achieved by the control of the present invention is as follows (this is only an example data, for reference only):

[0133] index Original control strategy Control strategy of the present invention Improvement Maximum temperature rise rate 0.85℃ / min 0.51℃ / min 40% reduction Maximum frequency deviation -0.52Hz -0.21Hz 60% smaller Battery capacity decay Single event accelerated decay 0.08% Single event accelerated decay 0.03% 62% reduction Photovoltaic fluctuation compensation delay 8-12 seconds Feedforward compensation advance 18 seconds 225% increase

[0134] This example verifies the technical advantages of the present invention in extreme complex disturbance scenarios. The coupled model automatically increases the life weight to 68% during the temperature rise acceleration phase, avoiding the fixed weight rigidity problem of traditional strategies; the S-shaped derating curve is used to match the power attenuation rate with the temperature rise gradient, and the derating process is smooth without power mutation; the inertia compensation module responds to frequency changes within 200ms; the improved particle swarm algorithm refreshes the global optimal solution every 5 minutes; the dynamic threshold T th The SOH (State of Health) is continuously corrected as it ages. While suppressing temperature rise, the frequency deviation is strictly controlled within ±0.25Hz, solving the mutually exclusive dilemma of "lifespan preservation" and "grid stability." The above case demonstrates that this invention is particularly suitable for complex scenarios such as high-proportion renewable energy and low-inertia grids, providing reliable technical support for the participation of photovoltaic storage systems in grid frequency regulation.

[0135] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0136] To better understand the design logic and implementation details of the technical solution, the following parameter characteristic table of the present invention is given as follows:

[0137]

[0138]

[0139] The actual values ​​of the above parameters depend on the actual working conditions and theoretical calculations to ensure the feasibility and robustness of the method. Most parameters (such as T_trig, α, γ, etc.) can be adjusted according to the battery health state (SOH), temperature gradient, and grid characteristics (such as H_sys) to adapt to different working conditions.

Claims

1. A multi-objective dynamic optimization control method for a photovoltaic storage system, characterized in that: Multi-objective optimization is achieved through the collaborative control of dynamic weight migration triggered by temperature gradient and high-frequency fluctuation prediction, including: Construct a fuzzy temperature-life-grid three-domain dynamic coupling decision model to convert the battery temperature gradient The state-of-charge change rate dSOC / dt and the grid frequency deviation Δf are input into a three-dimensional fuzzy inference engine, which outputs a temperature impact factor η(T) and an economic weight λ_eco. When the battery temperature T enters the critical threshold deviation range, a dual-channel temperature response mechanism with conflict resolution capability is activated: Preventing the channel T≥T_crit - 5°C: Adopting a sigmoidal power derating curve with the dynamic trigger temperature T_trig being adaptive, and its steepness coefficient α decays linearly with the state of health (SOH) of the battery; Protecting the channel T≥T_crit: Based on the Arrhenius-Rainflow coupling model, the life loss rate D_rate is calculated in real time, and the charge-discharge silent period duration Δt is generated through the collaborative calculation of the damage rate compensation coefficient k_d and the cooling system response time t_cool, and a grid frequency feedforward compensation amount Δf_comp is injected during the silent period; Multi-objective collaborative optimization expressed by the following function: min(η(T)·C deg +(1-η(T))·[λ eco C eco +(1-l eco )C grid ]) Where η(T) is the temperature T and the temperature gradient The function is dynamically corrected, and the life loss cost C_deg includes the temperature gradient enhancement factor.

2. The method according to claim 1, characterized in that The dynamic trigger temperature T_trig is determined by the battery health state (SOH) and temperature gradient Dynamic calibration to meet: Where ΔT_base is the reference offset of 3 to 5°C, γ is the health compensation coefficient of 0.5 to 1.5°C / %, and when When the temperature is increased, T_trig decreases by an additional 1 to 2°C.

3. The method according to claim 2, characterized in that The calculation of the life loss rate D_rate introduces a dynamic activation energy compensation mechanism: When it is detected that the internal pressure volatility dP / dt of the battery exceeds the threshold, the activation energy E_a is corrected; and the silent period duration Δt synchronously considers the dynamic constraints of the grid frequency regulation demand.

4. The method according to claim 2 or 3, characterized in that The membership function of the three-dimensional fuzzy inference engine satisfies: Temperature gradient The membership boundary value of is dynamically adjusted according to the battery chemistry type; The warning threshold of the grid frequency deviation Δf is negatively correlated with the grid inertia H_sys.

5. The method according to claim 1, wherein The photovoltaic output is separated into a low-frequency component P_low and a high-frequency component P_high through empirical mode decomposition (EMD). An ARIMA model is constructed for P_low to predict the trend value, and an LSTM network is used for P_high to generate a probability distribution band [P_{high}^{min}, P_{high}^{max}], and a cloud movement compensation term ΔH_k is injected through a Kalman filter.

6. The method according to claim 5, characterized in that Setting a double-layer determination mechanism for the transition zone: Early warning zone T_base≤T<T_crit - 5°C: The power is adjusted by linearly decaying according to the aging state; Emergency zone T≥T_crit - 5°C: Switch to the sigmoidal derating strategy, and the derating rate is increased to 1.5 to 2 times that of the early warning zone.

7. The method according to claim 6, characterized in that The smoothing of control instructions adopts dynamic inertia filtering, and the time constant τ varies with the and dSOC / dt changes dynamically, and responds faster when the grid frequency changes suddenly.

8. The method according to any one of claims 5 to 7, characterized in that: Cycle identification window and temperature gradient of the rainflow counting method Dynamic association: When The cycle recognition window is compressed from the standard 30 minutes to 10 minutes, and the damage weight of small-amplitude cycles is increased by 20% to 50%.

9. The method according to claim 1, characterized in that The grid support cost C_grid in the objective function, where C_grid includes a non-linear penalty term for the frequency deviation and its rate of change, and the penalty coefficient is dynamically adjusted according to the grid inertia.

10. A multi-objective dynamic optimization control device for a photovoltaic storage system, characterized in that: Including: The edge computing unit collects battery temperature, SOC, and grid frequency at a sampling rate of 100Hz and calculates the temperature gradient in real time and dSOC / dt; Dynamic compilation module, automatically switches the objective function calculation mode according to the temperature gradient range: The emergency optimization algorithm is started at this time, and the calculation cycle is compressed to 5ms; An instruction generation module that executes any of the methods in claims 1-9 to generate control instructions and outputs them to the power conversion system through a dynamic inertia filter; The device firmware integrates a battery parameter self-learning function, which can dynamically correct the parameters of T_crit, α, and k according to historical operation data, and the correction amplitude does not exceed ±20% of the initial value.

Images