A distributed integrated photovoltaic, energy storage and utilization energy dispatching method
By analyzing cross-user load complementarity and separating battery aging characteristics, a precise distributed photovoltaic-storage-utilization integrated scheduling scheme is generated, which solves the problems of photovoltaic curtailment and shortened energy storage life, and achieves efficient photovoltaic absorption and extended battery life.
Patent Information
- Application Number
- CN202611303141.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
AI Technical Summary
In a distributed photovoltaic-storage-utilization integrated system at the district level, the independent scheduling of a single household cannot perceive the energy complementarity of users, resulting in curtailment of photovoltaic power during periods of surplus and the purchase of electricity during periods of shortage. Furthermore, existing scheduling methods fail to distinguish the differentiated damage responses of energy storage battery aging mechanisms, leading to a shortened energy storage lifespan.
Load complementarity is obtained through cross-user time-shift cross-correlation analysis, battery aging characteristic parameters are separated, a multi-user joint Pareto optimization model is used to generate a scheduling scheme, and time-series verification and correction are performed to generate the final scheduling instruction. Combined with frequency domain separation and diagnostic network, the severity of aging and evolution rate are accurately assessed.
It enables on-site photovoltaic consumption under cross-user energy complementarity, extends the life of energy storage batteries, and optimizes dispatch efficiency and lifespan assurance.
Smart Images

Figure CN123121752A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of distributed energy dispatching technology, and specifically to a distributed integrated photovoltaic, energy storage and utilization energy dispatching method. Background Technology
[0002] In a distributed photovoltaic-storage integrated system at the substation level, multiple users each configure their own photovoltaic modules and adjustable load devices, and share the substation-level energy storage battery. The existing scheduling method independently schedules the adjustable load on a per-user basis, increasing the load of a user when the photovoltaic system has a surplus and decreasing the load when the photovoltaic system is insufficient.
[0003] However, the energy consumption behaviors of different users are naturally complementary in the time dimension. Individual household dispatch cannot perceive the complementary relationship between users, resulting in a situation where there is a large surplus of photovoltaic power during certain periods, but no adjustable load can be started by the household, resulting in curtailment of photovoltaic power. At the same time, neighboring users purchase electricity from the grid due to insufficient photovoltaic power in their own households, causing unnecessary power loss.
[0004] While treating multiple adjustable loads as a whole for cross-user coordinated dispatch can improve photovoltaic (PV) absorption rates, shared energy storage batteries will be subject to high-frequency charge-discharge switching and irregular power pulses triggered alternately by multiple complementary users. The three microscopic aging mechanisms within the battery—SEI film thickening, lithium dendrite growth, and cathode material phase transition—exhibit distinctly different accelerating effects at different temperature ranges and charge-discharge rates. Existing dispatch methods for aging assessment conflate multiple aging mechanisms into a single decay curve, failing to distinguish the differentiated damage responses of each mechanism under coordinated dispatch mode. This means that the optimal short-term absorption scheme may significantly shorten the long-term energy storage lifespan. Summary of the Invention
[0005] This invention provides a distributed photovoltaic-storage-utilization integrated energy dispatching method, which solves the technical problems in related technologies such as the difficulty in accurately sensing the aging status of shared energy storage batteries, the lack of effective exploration of the complementary potential of multi-user adjustable loads, and the lack of aging damage constraints in dispatching schemes, which leads to excessively rapid loss of energy storage life.
[0006] This invention discloses a distributed photovoltaic-storage-utilization integrated energy dispatching method, including: acquiring the power data and photovoltaic output data of each user in the distribution area, decomposing the electricity consumption data of each user into a load-level electricity consumption behavior timeline, and simultaneously acquiring the operation data of the shared energy storage battery; Cross-user time-shift cross-correlation analysis is performed on the adjustable load available capacity curves of each user to generate a cross-user behavior complementarity matrix; The voltage and current curves of the shared energy storage battery are separated in the frequency domain, and three sets of frequency domain feature parameters (high frequency, mid frequency and low frequency) are extracted. These parameters are then input into the corresponding diagnostic network, and the severity index and evolution rate prediction values of each aging mechanism are output. Based on the severity index and evolution rate prediction, the marginal damage increment matrix is calculated and transformed into a set of constraints. Based on the cross-user behavior complementarity matrix and the constraint set, the scheduling scheme is solved by a multi-user joint Pareto optimization model. The scheduling scheme is subjected to timing expansion verification and correction, and the final scheduling instruction set is generated and distributed for execution.
[0007] Furthermore, the cross-user time-shift cross-correlation analysis includes: For any two households' adjustable load available capacity time series, calculate the normalized cross-correlation coefficients for each integer time shift within the preset maximum time shift window; The maximum value of the absolute value of the normalized cross-correlation coefficient under each time shift is taken as the complementarity index between the two households. The complementarity index reflects the degree of negative correlation between the adjustable load available capacity of the two households on the time axis. The cross-user behavior complementarity matrix is generated based on the complementarity index of all user pairs.
[0008] Furthermore, the diagnostic networks corresponding to the three sets of frequency domain characteristic parameters are the SEI film thickening diagnostic network, the lithium dendrite risk assessment network, and the cathode phase transition diagnostic network, respectively. The SEI membrane thickening diagnostic network takes as input a time series of high-frequency SEI membrane impedance parameters and outputs as the current severity index and evolution rate prediction of SEI membrane thickening. The input to the lithium dendrite risk assessment network is the time series of mid-frequency charge transfer impedance parameters, and the output is the current severity index and evolution rate prediction of lithium dendrite growth. The input to the cathode phase transition diagnostic network is the time series of low-frequency diffusion impedance parameters, and the output is the current severity index and evolution rate prediction of the cathode material phase transition. The severity index and evolution rate output by the three diagnostic networks are dimensionless values normalized to the interval between zero and one.
[0009] Furthermore, the SEI film thickening diagnostic network includes an impedance sequence encoding unit, a temperature modulation unit, and a severity inference unit; The impedance sequence encoding unit uses a recurrent neural network to encode the time series of high-frequency SEI membrane impedance parameters and outputs a fixed-dimensional impedance feature vector. The temperature modulation unit generates a fused feature vector by fusing the impedance feature vector and the surface temperature sequence encoding vector through element-wise gating. The gating coefficient is generated by the surface temperature sequence encoding vector through a learnable weight matrix linear transformation and then through a sigmoid activation function, which is used to control the mixing ratio of the impedance feature vector and the temperature encoding vector in each dimension. The severity inference unit outputs the current severity index and evolution rate prediction value of the SEI film thickening through the fully connected output layer from the fused feature vector.
[0010] Furthermore, the lithium dendrite risk assessment network includes a charge transfer feature encoding unit, a current pulse correlation unit, and a dendrite risk inference unit; The charge transfer feature encoding unit uses a recurrent neural network to encode the time series of the mid-frequency charge transfer impedance parameters and outputs a charge transfer feature vector. The current pulse association unit concatenates the charge transfer feature vector with the amplitude and interval sequence encoding vector of the charge and discharge current pulses, performs linear transformation by the learnable weight matrix, calculates the attention weight vector through softmax, and then uses the attention weight vector to perform a weighted sum of the charge transfer feature vector and the pulse encoding vector to output the pulse association feature vector. The dendrite risk inference unit outputs the current severity index and evolution rate prediction value of lithium dendrites through the pulse correlation feature vector via a fully connected output layer.
[0011] Furthermore, each element in the marginal damage increment matrix is the weighted sum of the three aging mechanisms caused by performing unit charge operation under a specific temperature range, specific charge / discharge rate and specific switching frequency conditions. The weighting coefficients for each aging mechanism are obtained by normalizing the current severity index output by the corresponding diagnostic network among the three aging mechanisms. When the severity index of a certain aging mechanism is high, the weighting coefficient corresponding to that mechanism increases. The unit electrical damage contribution value of each aging mechanism is obtained by multiplying the evolution rate prediction value output by the corresponding diagnostic network by the conditional amplification factor. The conditional amplification factor reflects the joint modulation factor of temperature, rate and switching frequency on the evolution rate of the mechanism, and is determined by a piecewise linear lookup table fitted from historical experimental data.
[0012] Furthermore, the set of constraints includes an upper limit constraint on the charging / discharging switching frequency and an upper limit constraint on the power rate in each temperature range. The upper limit constraint for the charging and discharging switching frequency is: within each scheduling cycle, the number of charging and discharging direction switching times shall not exceed the upper limit determined by the difference value of the marginal damage increment matrix along the switching frequency dimension; The upper limit constraint of the power rate in each temperature range is: within each temperature range, the rate of change of charging and discharging power shall not exceed the upper limit determined by the difference value of the marginal damage increment matrix along the rate dimension. When the difference value of a certain dimension exceeds the preset gradient threshold, the upper limit of the operation parameter corresponding to that dimension is taken as the current level value.
[0013] Furthermore, the multi-household joint Pareto optimization model takes maximizing the overall photovoltaic absorption capacity of the distribution area, minimizing the weighted sum of the disturbances caused by the behavior of each household, and minimizing the incremental damage during the energy storage cycle as three optimization objectives. The interference level of each household's behavior is the weighted cumulative amount of the deviation between the actual dispatched power and the power of the typical power consumption mode of each household, and the weight of each household is determined by the power consumption priority declared in advance by each household. The output of the multi-household joint Pareto optimization model is a set of non-dominated solutions on the Pareto front. The solution with the highest comprehensive score is selected as the scheduling scheme.
[0014] Furthermore, the timing decomposition verification and correction includes: The number of charge-discharge switching times and power rates of the shared energy storage battery in each time period are simulated under the execution of the scheduling scheme. The simulation results are then verified against the damage thresholds of the corresponding condition combinations in the marginal damage increment matrix on a time-by-time basis. When the simulated damage value for a certain period exceeds the corresponding damage threshold, the inter-household power exchange volume for that period is multiplied by a reduction factor to reduce it. The reduction factor is the ratio of the damage threshold for that period to the simulated damage value. The scheduling task of the corresponding adjustable load is then shifted to the nearest period with a damage value lower than the damage threshold and with available adjustable load capacity. After the execution of the final scheduling instruction set is completed, the changes in the three sets of frequency domain characteristic parameters during the actual charging and discharging process of the shared energy storage battery are recorded. The deviation between the actual changes and the predicted values of each diagnostic network is calculated, and the weight parameters of each diagnostic network and the weighting coefficients in the marginal damage increment matrix are updated online using the predicted deviation values.
[0015] This invention provides a distributed photovoltaic-storage-utilization integrated energy dispatching system, comprising: The data acquisition module is used to acquire power data and photovoltaic output data of each user in the distribution area, decompose the electricity consumption data of each user to generate a timeline of equipment-level electricity consumption behavior, and simultaneously acquire the operation data of the shared energy storage battery. The complementarity analysis module is used to perform cross-user time-shift cross-correlation analysis on the adjustable load available capacity curves of each user, and generate a cross-user behavior complementarity matrix; The aging feature separation module is used to perform frequency domain separation on the voltage and current curves of the shared energy storage battery, extract three sets of frequency domain feature parameters (high frequency, mid frequency and low frequency), input them into the corresponding diagnostic network, and output the severity index and evolution rate prediction value of each aging mechanism. The damage matrix calculation module is used to calculate the marginal damage increment matrix based on the severity index and the evolution rate prediction value and convert it into a set of constraint conditions. The collaborative optimization solution module is used to solve the scheduling scheme based on the cross-user behavior complementarity matrix and the constraint set through a multi-user joint Pareto optimization model. The verification and correction module is used to perform timing expansion verification and correction on the scheduling scheme, generate the final scheduling instruction set, and distribute it for execution.
[0016] This invention quantifies the temporal complementarity of different users' energy consumption behaviors at the distribution area level through a cross-user behavior complementarity matrix. In dispatching decisions, the excess power of photovoltaic surplus households is locally consumed by increasing the adjustable load of complementary households, overcoming the problem of curtailment and unnecessary power purchase caused by the inability of single-household independent dispatch to perceive cross-user complementarity.
[0017] This invention applies frequency domain separation to the battery operation dataset to independently extract the characteristic parameters of three aging mechanisms: SEI film thickening, lithium dendrite growth, and cathode material phase transition. The invention then outputs differentiated severity indices and evolution rate predictions through their respective diagnostic networks. This overcomes the problem in existing methods that mix multiple aging mechanisms into a single decay curve and fail to distinguish the differentiated damage responses of each mechanism under the cooperative scheduling mode.
[0018] This invention quantifies the interaction effects of temperature range, charge / discharge rate, and switching frequency on various aging mechanisms into a marginal damage increment matrix, and transforms it into a set of constraints for a multi-household joint Pareto optimization model, so that the scheduling scheme is constrained by the aging damage boundary while pursuing photovoltaic absorption rate.
[0019] This invention achieves a joint optimization effect between multi-user collaborative scheduling efficiency and energy storage life guarantee by verifying the timing of the scheduling scheme and correcting the cross-period shift of flexible loads. This is done by ensuring that the final scheduling instruction set does not trigger the acceleration threshold of the aging mechanism in each temperature range. Attached Figure Description
[0020] Figure 1 This is a flowchart of the distributed photovoltaic-storage-utilization integrated energy dispatching method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram showing the comparison of net load and photovoltaic output for each household at 09:00, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the cross-user behavior complementarity index matrix provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the severity index and evolution rate of three aging mechanisms (11:00 time period) provided in the embodiments of the present invention; Figure 5 This is a heat map of the marginal damage increment matrix in the high-temperature range provided in this embodiment of the invention. A schematic diagram of the interval; Figure 6 This is a schematic diagram comparing the photovoltaic output and adjustable load available capacity of each household at 11:00 AM, provided in an embodiment of the present invention. Figure 7 This is a schematic diagram of the damage value verification and correction results for the period from 13:00 to 16:00 provided in an embodiment of the present invention; Figure 8 This is a schematic diagram comparing the predicted values and actual observed values of the SEI membrane thickening diagnostic network provided in this embodiment of the invention; Figure 9 This is a schematic diagram showing the comparison of the complementarity index between user A and user C before and after time-shift update, provided in an embodiment of the present invention. Detailed Implementation
[0021] In a distributed photovoltaic-storage-utilization integrated system, multiple users within the same distribution area each have their own photovoltaic modules and adjustable load devices, sharing the distribution area-level energy storage battery. Existing scheduling methods independently schedule adjustable loads on a per-user basis, increasing a user's load when there is a photovoltaic surplus and decreasing it when there is a photovoltaic shortage. However, different users' energy consumption behaviors are naturally complementary over time; for example, user A's electricity consumption is concentrated at noon while user B's is concentrated in the evening. Independent scheduling for each user cannot utilize this complementarity, leading to situations where there is a large photovoltaic surplus during certain periods, but no adjustable load can be started, resulting in curtailment of photovoltaic power, while neighboring users purchase electricity from the grid due to insufficient photovoltaic power in their own areas. While treating the adjustable loads of multiple users as a whole for cross-user coordinated scheduling can improve photovoltaic absorption, the shared energy storage battery will be subjected to high-frequency charge-discharge switching and irregular power pulses triggered alternately by multiple complementary users. The three microscopic aging mechanisms within the battery—SEI film thickening, lithium dendrite growth, and cathode material phase transition—exhibit drastically different acceleration effects at different temperature ranges and charge-discharge rates. Existing scheduling methods for aging assessment mix multiple mechanisms into a single decay curve, failing to distinguish the differentiated damage responses of each mechanism under coordinated scheduling modes. This means that the optimal short-term absorption scheme may significantly shorten the long-term energy storage life.
[0022] In this embodiment, each user side within the distribution area is equipped with a smart meter and a photovoltaic inverter. The smart meter records the power data of each circuit at a sampling frequency of seconds. Adjustable load devices for each household include electric water heaters, air conditioners, and electric vehicle charging piles, which can accept dispatch commands to adjust their power. The distribution area-level shared energy storage battery is equipped with a battery management system, which collects terminal voltage, current, and surface temperature data at a frequency of seconds. The distribution area energy management terminal, as the dispatch execution entity, is connected to each household's smart meter, photovoltaic inverter, adjustable load device, and energy storage controller via a communication network.
[0023] According to an embodiment of the present invention, a distributed photovoltaic-storage-utilization integrated energy dispatching method is as follows: Figure 1 As shown, it includes the following steps: Step 1: Collect behavioral data and battery operation data from multiple households in the transformer area; Acquire power data and photovoltaic output data for each user and circuit within the distribution area. Utilize a non-intrusive load identification algorithm to decompose the electricity consumption data for each household. The algorithm takes the total power time series of each household as input and outputs the independent power time series of each device within each household. Based on the decomposition results, generate a device-level electricity consumption behavior timeline for each household. Simultaneously acquire the second-by-second time series of terminal voltage, current, surface temperature, and ambient temperature of the shared energy storage battery, forming a multi-household behavior dataset and a battery operation dataset.
[0024] Before proceeding to the next steps, the multi-household behavior dataset and the battery operation dataset are preprocessed. Z-score standardization is applied to the power time series, photovoltaic output data, terminal voltage, current, surface temperature, and ambient temperature data for each household to eliminate the influence of differences in the dimensions of different physical quantities on subsequent feature extraction and model input.
[0025] Step 2: Extract complementary features across user behaviors; Sequence pattern mining was performed on the timeline of each household's equipment-level electricity consumption behavior to extract typical electricity consumption patterns for each household under different day types and time periods. Day types included weekdays, rest days, and public holidays. Time periods were divided according to fixed time intervals, and the available adjustable capacity of each household's adjustable load equipment was calculated within each time period.
[0026] Cross-user time-shift cross-correlation analysis is performed on the adjustable load capacity curves of any two users. The input to the cross-user time-shift cross-correlation analysis is the time series of the adjustable load capacity of the two users, and the output is the complementarity index between the two users at each time period. The complementarity index reflects the degree of negative correlation between the adjustable load capacity of the two users on the time axis; a higher complementarity index indicates a greater degree of temporal peak-shaving in the energy consumption behavior of the two users. Based on the complementarity index of all user pairs, a cross-user behavior complementarity matrix is generated.
[0027] It should be noted that the above complementarity index is a scalar quantity obtained by taking the maximum absolute value of the normalized cross-correlation coefficients after calculating the available capacity curves of the two households under multiple time shifts. Let the time series of the available capacity of the two households be as follows: and ,in For sequence length, and Two households respectively in the Available adjustable capacity per time step If it is a time step index, then in the time shift... Normalized cross-correlation coefficients for:
[0028] in, and Sequences and The mean, This is a time shift value, and its range covers all integer steps within the preset maximum time shift window. Complementarity index. Defined as:
[0029] When the available capacity curves for the two households show opposite trends... The value approaches 1; when the fluctuation trends of the available capacity curves for two households are similar, The value approaches 0.
[0030] Step 3: Separate the multi-mechanism aging characteristic parameters of the battery; A Fast Fourier Transform (FFT) based on short-time impulse response is applied to the voltage and current curves in the battery operation dataset. The input to the FFT is a second-by-second time series of terminal voltage and current, and the output is the impedance spectrum. Three sets of frequency domain characteristic parameters are separated from the impedance spectrum: high-frequency SEI film impedance parameters, mid-frequency charge transfer impedance parameters, and low-frequency diffusion impedance parameters.
[0031] Three sets of frequency domain characteristic parameters were input into the SEI film thickening diagnostic network, the lithium dendrite risk assessment network, and the cathode phase transition diagnostic network, respectively. The current severity index and predicted evolution rate of each of the three microscopic aging mechanisms were output. The severity index and evolution rate output by the SEI film thickening diagnostic network, the lithium dendrite risk assessment network, and the cathode phase transition diagnostic network were all normalized to [value missing]. Dimensionless values within an interval.
[0032] It should be noted that the SEI membrane thickening diagnostic network includes an impedance sequence encoding unit, a temperature modulation unit, and a severity inference unit. The impedance sequence encoding unit takes as input a time series of high-frequency SEI membrane impedance parameters, encodes the time series using a recurrent neural network, and outputs a fixed-dimensional feature vector. The input to the temperature modulation unit is a feature vector. and surface temperature sequence encoding vector The modulation effect of temperature on the SEI film growth rate is embedded into the feature representation through element-wise gated fusion. The calculation method of gated fusion is as follows:
[0033] in, This is the fused feature vector after temperature modulation. It is the sigmoid activation function. The learnable gated weight matrix, For bias vectors, This represents element-wise multiplication. The input to the severity inference unit is the fused feature vector. The network outputs the current severity index and evolution rate prediction of SEI membrane thickening via a fully connected output layer. The output layer contains two neurons, corresponding to the severity index and evolution rate, respectively. The SEI membrane thickening diagnostic network uses historically labeled SEI membrane severity index and evolution rate as supervision labels, employs a mean squared error loss function, and is trained using the Adam optimization algorithm.
[0034] It should be noted that the lithium dendrite risk assessment network includes a charge transfer feature encoding unit, a current pulse correlation unit, and a dendrite risk inference unit. The input to the charge transfer feature encoding unit is the time series of mid-frequency charge transfer impedance parameters, which is encoded using a recurrent neural network, and the output is a feature vector. The input to the current pulse correlation unit is the feature vector. And the amplitude and interval sequence encoding vector of the charging and discharging current pulses. The current pulse characteristics and impedance characteristics are fused using attention weighting. The attention weighting fusion first calculates the attention weight vector using softmax. ,in The input is a concatenated vector. Learnable weight matrix and bias vector The result after linear transformation; then calculate the pulse correlation feature vector. ,in and These are attention weight vectors. Chinese correspondence and The amount, This represents vector concatenation. The input to the dendrite risk inference unit is the impulse correlation feature vector. The network outputs the current severity index and predicted evolution rate of lithium dendrites via a fully connected output layer. The output layer contains two neurons, corresponding to the severity index and evolution rate, respectively. The lithium dendrite risk assessment network uses historically labeled lithium dendrite severity index and evolution rate as supervision labels, employs a mean squared error loss function, and is trained using the Adam optimization algorithm.
[0035] It should be noted that the positive electrode phase transition diagnostic network includes a diffusion impedance mode extraction unit, a charge state correlation unit, and a phase transition state inference unit. The input to the diffusion impedance mode extraction unit is a time series of low-frequency diffusion impedance parameters. A convolutional neural network is used to extract periodic variation patterns from the time series, and the output is a feature vector of these periodic variation patterns. The input to the state-of-charge (SOC) correlation unit is a periodically changing mode feature vector. and the historical trajectory encoding vector of the state of charge The state-of-charge interval residence distribution and diffusion characteristics are concatenated and fused to output a state-related feature vector. ,in This represents a vector concatenation operation. The input to the phase transition state inference unit is the state-related feature vector. The fully connected output layer outputs the current severity index and evolution rate prediction of the cathode material phase transition. The output layer contains two neurons, corresponding to the severity index and evolution rate, respectively. The cathode phase transition diagnostic network uses historically labeled phase transition severity index and evolution rate as supervision labels, employs the mean squared error loss function, and is trained using the Adam optimization algorithm.
[0036] Step 4: Calculate the marginal damage increment matrix; Based on the severity index and evolution rate predictions of three aging mechanisms, and combined with the environmental temperature prediction curve, a marginal damage increment matrix is calculated. Each element in the marginal damage increment matrix is defined as: within a specific temperature range... Below, at a specific charge / discharge rate and switching frequency The sum of the three mechanistic weighted damages caused by performing a unit of electrical charge operation. The formula for calculating this element is as follows:
[0037] in, Temperature range Charge and discharge rate and switching frequency Marginal damage increment value under combination of conditions; Number the aging mechanism. Corresponding to SEI film thickening, Corresponding to lithium dendrite growth, For the corresponding phase transition of the cathode material, summation upper bound This represents the total number of the three aging mechanisms; For the first Aging mechanisms within a certain temperature range The weighting coefficients below; For the first Aging mechanisms within a certain temperature range Charge and discharge rate and switching frequency The unit charge damage contribution value. Due to and All are dimensionless values. It is also a dimensionless scalar, and the damage contributions of each mechanism can be directly weighted and summed.
[0038] It should be noted that the above weighting coefficients It is obtained by normalizing the current severity index output by the corresponding diagnostic network. Let the... Aging mechanisms within a certain temperature range The current severity index is ,but:
[0039] in, To sum and iterate through the aging mechanism numbers, For the first Aging mechanisms within a certain temperature range The current severity index is summed to the upper bound. This represents the total number of the three aging mechanisms. When the current severity index of a certain aging mechanism is high, the weighting coefficient corresponding to that mechanism increases, making the marginal damage increment matrix more sensitive to the damage contribution of that mechanism.
[0040] It should be noted that the above-mentioned unit power consumption damage contribution value The value is determined by extrapolating the evolution rate prediction output by the corresponding diagnostic network under this combination of conditions. Let the evolution rate prediction output by the corresponding diagnostic network in the current scheduling period be... ,but:
[0041] in, For the first Predicted evolution rate of various aging mechanisms For the first Aging mechanisms under various conditions The conditional amplification factor, which reflects the combined modulation factor of temperature, rate, and switching frequency on the rate of evolution of this mechanism, is determined by a piecewise linear lookup table fitted from historical experimental data. and Both are dimensionless values, and the product of the two is the result. Maintain dimensionless consistency.
[0042] Step 5: Solve for the multi-household collaborative scheduling scheme with aging perception. The marginal damage increment matrix is transformed into a set of constraints. This set includes an upper limit constraint on the charge / discharge switching frequency and an upper limit constraint on the power rate for each temperature range. The upper limit constraint on the charge / discharge switching frequency is: within each scheduling cycle, the number of charge / discharge direction switches does not exceed the upper limit determined by the damage gradient in the corresponding switching frequency dimension of the marginal damage increment matrix. The upper limit constraint on the power rate for each temperature range is: within each temperature range, the rate of change of charge / discharge power does not exceed the upper limit determined by the damage gradient in the corresponding temperature range and rate dimensions.
[0043] It should be noted that the damage gradient mentioned above refers to the marginal damage increment matrix. The difference along the corresponding dimension. The damage gradient in the upper limit constraint of charge / discharge switching frequency is... along The difference in dimensionality refers to the difference in damage increment between adjacent switching frequency levels; the damage gradient in the power rate upper limit constraint for each temperature range is... along The difference in dimensions refers to the difference in damage increment between adjacent rate levels. When the damage gradient in a certain dimension exceeds a preset gradient threshold, the upper limit of the operating parameter corresponding to that dimension is set to the value of the current rate level to prevent the damage increment from increasing sharply as the parameter increases.
[0044] In each scheduling decision cycle, the user combination with the highest complementarity index for the current time period is extracted from the cross-user behavior complementarity matrix. The predicted photovoltaic output, current state of charge of energy storage, available adjustable load capacity, and predicted electricity consumption intention vector for each user are obtained for the current time period. This data is then input into a multi-user joint Pareto optimization model for solution. The multi-user joint Pareto optimization model has three optimization objectives: maximizing the overall photovoltaic absorption capacity of the distribution area, minimizing the weighted sum of disturbances caused by user behavior, and minimizing the incremental damage during the energy storage cycle. A set of constraints is superimposed as the boundary of the feasible region. This is applied in each scheduling cycle. The above methods jointly solve for the adjustable load dispatch power and energy storage charging and discharging power of each household, where For scheduling cycle index, This represents the total number of time periods within the scheduling cycle. The output of the multi-user joint Pareto optimization model is a set of non-dominated solutions on the Pareto front. The solution with the highest comprehensive score is selected as the distributed photovoltaic-storage-utilization integrated energy scheduling scheme. The distributed photovoltaic-storage-utilization integrated energy scheduling scheme is output in the form of scheduling power commands and energy storage charging and discharging power commands for each user's adjustable load in each time period. These commands are directly used for timing expansion verification and command distribution execution in subsequent steps.
[0045] It should be noted that the above-mentioned household behavior interference levels are indicators used to measure the degree to which the dispatching plan deviates from the normal electricity usage habits of each household. Let the first household... During the time period The actual scheduling power is The corresponding typical power consumption mode is The priority weight for this household's electricity use is: Then the first User behavior interference for:
[0046] in, Number the user. For time period index, The total number of time periods within the scheduling period. The electricity usage priority is determined by each household's pre-declared priority; the higher the priority, the higher the priority. The larger the value, the greater the penalty for the user's deviation in the objective function.
[0047] It should be noted that the above-mentioned electricity consumption intention prediction vector is a probability distribution expression of each household's electricity consumption intention for each time period within the future scheduling cycle. The electricity consumption intention prediction vector is generated based on the historical device-level electricity consumption behavior timeline of each household, utilizing the statistical distribution of each household's typical electricity consumption patterns under corresponding day types and time periods. Let the... During the time period The electricity consumption intention prediction vector is ,in Number the user. For time period index, The number of adjustable load devices. For the first The first household The equipment during the time period Activated historical frequencies The device index is obtained by normalizing the statistical counts of historical device-level power consumption behavior timelines under the corresponding day type and time period.
[0048] In this embodiment, to avoid excessively concentrated inter-household power exchange in extreme complementary scenarios using a single Pareto optimization solution, a single-household power exchange ratio constraint is applied to the Pareto front solution set in step 5. This constraint limits the amount of inter-household power received by any user within a single scheduling cycle to no more than a preset proportion of that user's photovoltaic installed capacity, thereby suppressing excessive occupation of the power channel by individual users while utilizing complementarity.
[0049] Step 6: Perform timing expansion verification and correction on the scheduling scheme; A time-series expansion is performed on the cross-user power exchange sequence in a distributed photovoltaic-storage-utilization integrated energy dispatch scheme. The number of charge-discharge switching operations and the power rate of the shared energy storage battery in each time period are simulated under the execution of the distributed photovoltaic-storage-utilization integrated energy dispatch scheme. The simulation results are then verified time-by-time against the damage thresholds corresponding to the condition combinations in the marginal damage increment matrix.
[0050] It should be noted that the above damage thresholds are marginal damage increment matrices. In the corresponding combination of conditions The preset upper limit value is determined by the critical damage amount that triggers accelerated aging under the corresponding temperature range, charge / discharge rate and switching frequency conditions of each aging mechanism in historical experimental data, and is stored in the power management terminal of the distribution area in the form of a piecewise linear lookup table.
[0051] When the simulated damage value for a certain period does not exceed the corresponding damage threshold, the distributed photovoltaic-storage integrated energy dispatch scheme for that period remains unchanged. When the simulated damage value for a certain period exceeds the corresponding damage threshold, the inter-household power exchange volume for that period is reduced proportionally, and the dispatching task for the corresponding adjustable load is shifted to the adjacent low-damage period. The selection principle for the target period for shifting is: the closest period in the marginal damage increment matrix where the damage value is lower than the damage threshold and there is available capacity margin for adjustable loads. After period-by-period verification and shifting, a corrected final dispatch instruction set is generated. The final dispatch instruction set is output in the form of specific power instruction values and energy storage charging and discharging power instruction values for each household's adjustable load equipment in each period, for direct distribution and execution in step 7.
[0052] It should be noted that the aforementioned proportional reduction refers to multiplying the inter-household power exchange volume during periods exceeding the damage threshold by a reduction factor. The reduction factor is the ratio of the damage threshold to the simulated damage value for that period. When the simulated damage value is twice the damage threshold, the reduction factor is 0.5, meaning the inter-household power exchange volume is reduced to half of the original distributed photovoltaic-storage-utilization integrated energy dispatch scheme.
[0053] In this embodiment, to address the extreme case where multiple consecutive time periods exceed the damage threshold, a rolling backtracking adjustment is introduced based on step 6. When the number of consecutive time periods exceeding the damage threshold exceeds a preset consecutive threshold, backtracking is performed sequentially from the last time period exceeding the damage threshold. The remaining power exchange amount that cannot be digested in adjacent low-damage time periods after reduction in each time period exceeding the damage threshold is gradually allocated to earlier low-damage time periods until the simulated damage value of all time periods does not exceed the corresponding damage threshold.
[0054] Step 7: Execute the scheduling instructions and update the diagnostic parameters online; The final set of dispatch instructions is distributed to each household's adjustable load equipment and energy storage controller for execution.
[0055] The changes in three sets of frequency domain characteristic parameters during the actual charging and discharging process of the shared energy storage battery are recorded. The actual changes in these three sets of frequency domain characteristic parameters are then compared with the predicted values of the SEI film thickening diagnostic network, the lithium dendrite risk assessment network, and the cathode phase transition diagnostic network within the corresponding scheduling period. The prediction deviation values of each diagnostic network are then used to update the weight parameters of each diagnostic network online, ensuring that the outputs of each diagnostic network converge to the actual observed values. The online update employs a stochastic gradient descent optimization algorithm.
[0056] Simultaneously, the weighting coefficients in the marginal damage increment matrix are updated based on the prediction bias values of each diagnostic network. .in For the first Aging mechanisms within a certain temperature range The weighting coefficients below, Number the aging mechanism. The temperature range is defined. When the prediction bias of a diagnostic network remains consistently high, the weighting coefficient corresponding to that mechanism is increased; when the prediction bias remains consistently low, the weighting coefficient corresponding to that mechanism is decreased.
[0057] The actual user response data is fed back to the cross-user behavior complementarity matrix for incremental updates. The actual response data includes the actual execution power and execution time of each user's adjustable load equipment. Incremental updates are achieved by weighted fusion of the actual response data with historical typical electricity consumption patterns. The fused typical electricity consumption pattern replaces the original typical electricity consumption pattern in the complementarity calculation of subsequent scheduling cycles.
[0058] This implementation method quantifies the temporal complementarity of different users' energy consumption behaviors at the distribution area level through a cross-user behavior complementarity matrix. In dispatching decisions, excess power from photovoltaic surplus users is locally absorbed by increasing the adjustable load of complementary users. This overcomes the problems of curtailment and unnecessary power purchases caused by independent dispatching of single users due to their inability to perceive cross-user complementarity. This implementation method also extracts the characteristic parameters of three aging mechanisms—SEI film thickening, lithium dendrite growth, and cathode material phase transition—by applying frequency domain separation to the battery operation dataset. Differentiated severity indices and evolution rate predictions are output through SEI film thickening diagnostic networks, lithium dendrite risk assessment networks, and cathode phase transition diagnostic networks, respectively. This overcomes the problem of existing methods mixing multiple aging mechanisms into a single decay curve, making it impossible to distinguish the differentiated damage responses of each mechanism under the collaborative dispatching mode. This implementation quantifies the interaction effects of temperature range, charge / discharge rate, and switching frequency on various aging mechanisms into a marginal damage increment matrix, and transforms it into a constraint set for a multi-user joint Pareto optimization model. This ensures that the distributed photovoltaic-storage-utilization integrated energy dispatch scheme is constrained by the aging damage boundary while pursuing photovoltaic absorption rate. Through time-series expansion verification and elastic load cross-period shift correction of the distributed photovoltaic-storage-utilization integrated energy dispatch scheme, this implementation ensures that the final dispatch command set does not trigger the acceleration threshold of aging mechanisms in any temperature range, thereby achieving a joint optimization effect between multi-user collaborative dispatch efficiency and energy storage lifetime assurance.
[0059] The following is an example of an application of the present invention, such as... Figure 2-9 As shown, the implementation process is as follows: There are four households (Users A, B, C, and D) in a certain transformer substation area. Each household has photovoltaic modules installed on its roof and shares a substation-level energy storage battery (rated capacity 100kWh). During a scheduling cycle from 07:00 to 19:00 on a certain working day in 20XX, the substation energy management terminal implemented distributed integrated photovoltaic-storage-utilization energy dispatch for each household's adjustable loads (electric water heater, air conditioner, and electric vehicle charging station) and the shared energy storage battery. The ambient temperature on that day varied between 28℃ and 36℃, and the initial state of charge of the energy storage battery was 52%.
[0060] Implementation example of step 1 The district energy management terminal obtains the total power time series from each household's smart meter at a second-level sampling frequency, and obtains the photovoltaic output data of each household from the photovoltaic inverter. Taking the period from 09:00 to 09:01 as an example, the non-intrusive load identification algorithm decomposes the total power time series of user A, separating it into independent power time series for electric water heaters, air conditioners, and electric vehicle charging piles, and generates equipment-level electricity consumption behavior timelines. Simultaneously, it collects second-level terminal voltage, current, and surface temperature data of the energy storage battery. Z-score normalization is applied to all collected data to eliminate dimensional differences between power (kW), voltage (V), current (A), and temperature (°C).
[0061] Table 1. Timeline of device-level electricity consumption behavior for each household (excerpt from the 09:00 time period)
[0062] The negative net load values of users C and D indicate that both users have a photovoltaic surplus during the 09:00 period. Meanwhile, user B has a higher net load due to the operation of electric vehicle charging stations, thus forming the basic conditions for cross-user complementarity.
[0063] Implementation example of step 2 Sequence pattern mining was performed on the timeline of each household's equipment-level electricity consumption behavior to statistically analyze the adjustable load and available capacity for each time period on weekdays. Taking users A and C as examples, the complementarity index of their available capacity curves was calculated. The available capacity time series for two households in terms of time shift. (Time period) Normalized cross-correlation coefficient To reach the maximum absolute value, take This indicates a significant temporal shift in the energy consumption behavior of User A and User C. Complementarity indices were calculated for each of the six user pairs, generating a cross-user behavior complementarity matrix.
[0064] Table 2 Cross-user behavior complementarity matrix
[0065] The matrix shows that the complementarity index of the three user pairs, namely A and C, A and D, and B and C, all exceed 0.75. Subsequent scheduling decisions will prioritize extracting coordinated scheduling combinations from these three pairs.
[0066] Implementation example of step 3 A fast Fourier transform based on short-time impulse response (FFT) is applied to the second-by-second terminal voltage and current time series of the energy storage battery to obtain the impedance spectrum. Three sets of frequency domain feature parameters are then extracted from the impedance spectrum and input into three diagnostic networks. Taking the time period of 11:00 AM (surface temperature 34℃, corresponding to the high-temperature range) as an example, in the SEI film thickening diagnostic network, the impedance sequence encoding unit outputs a feature vector. In the temperature modulation unit, the high-temperature acceleration effect corresponding to 34℃ is embedded into the feature expression through gated fusion. The severity inference unit outputs a SEI film severity index of 0.61 and an evolution rate of 0.48. In the lithium dendrite risk assessment network, the current pulse correlation unit performs attention-weighted fusion on the high-frequency pulse sequence caused by multi-household collaborative scheduling on the same day. The dendrite risk inference unit outputs a severity index of 0.29 and an evolution rate of 0.37. In the cathode phase transition diagnosis network, the state of charge correlation unit splices and fuses the historical trajectory of the state of charge (the state of charge fluctuates frequently between 38% and 71% on the same day) with the diffusion impedance features. The phase transition state inference unit outputs a severity index of 0.44 and an evolution rate of 0.52.
[0067] Table 3. Diagnostic network outputs for three aging mechanisms (11:00 time period, surface temperature 34℃)
[0068] Implementation example of step 4 Based on the diagnostic output in Table 3, the high temperature range at 11:00 is calculated ( Weighting coefficients at temperatures between 30℃ and 35℃. The sum of the severity indices for the three mechanisms is... Thus we obtain , , The sum of the three is 1.00.
[0069] At medium charge / discharge rate ( ) and medium switching frequency level ( Taking the conditional combination of 4 times / hour as an example, the conditional amplification factor of the three mechanisms. The values were determined by a piecewise linear lookup table fitted from historical experimental data, and were 1.32, 1.68, and 1.21, respectively. The contribution of each mechanism to damage per unit charge. They are respectively , , The marginal damage increment value is .
[0070] Table 4 Marginal Damage Increment Matrix (High Temperature Range) (Partial combination of conditions)
[0071] The matrix shows that when the rate increases to And the switching frequency has increased to At that time, the damage increment value reached 1.02, which is much higher than 0.41 under low rate and low frequency conditions, providing a quantitative basis for the constraint generation in step 5.
[0072] Implementation example of step 5 Extract the damage gradient from the marginal damage increment matrix. Along... Dimension From gear down arrive The difference is This exceeds the preset gradient threshold of 0.13, therefore the upper limit constraint on the charging / discharging switching frequency will be... The maximum frequency of gear switching is set to (4 times / hour). Along Dimension From gear down arrive The difference is The value exceeds the preset gradient threshold of 0.15, therefore the upper limit of the power rate in the high-temperature range is set to the upper limit of the charge and discharge power rate. (0.5C).
[0073] The user pair with the highest complementarity index at 11:00 was extracted from the cross-user behavior complementarity matrix and identified as A and C. After obtaining the current predicted photovoltaic output values for two households (User A 5.2kW, User C 6.8kW), energy storage state of charge (58%), adjustable load capacity for each household, and electricity consumption intention prediction vectors, all data are input into a multi-household joint Pareto optimization model. Taking User A as an example, its electricity consumption intention prediction vector at 11:00 is... The values correspond to the historical frequency of the use of electric water heaters, air conditioners, and charging piles during this period, respectively, indicating that Household A frequently uses electric water heaters and air conditioners during the 11:00 time period but rarely uses charging piles.
[0074] Table 5 Input data for the multi-household joint Pareto optimization model (11:00 time period)
[0075] Under the constraints of the Pareto optimization model (switching frequency not exceeding 4 times / hour, power rate not exceeding 0.5C), a set of non-dominated solutions on the Pareto front is output. The solution with the highest comprehensive score allocates the 2.3kW surplus photovoltaic power from user C to user B's charging pile through energy storage, and locally increases the air conditioning load of user A's 1.4kW surplus photovoltaic power. The energy storage charging power command is set to 14kW (0.14C, below the 0.5C upper limit), the switching frequency is controlled within 3 times / hour, and the single-household power exchange ratio constraint is met (the 2.3kW cross-household power received by user B is less than 80% of its 4kW photovoltaic installed capacity upper limit).
[0076] Implementation example of step 6 A time-series simulation was performed on the period from 13:00 to 15:00 in the optimized scheme to simulate the number of charge-discharge switching times and power rates of the energy storage battery during each period. The simulation results for the 13:00 period showed a switching frequency of 5 times / hour, corresponding to a simulated damage increment of 0.81, which exceeds the high-temperature range. Gear down The corresponding damage threshold is 0.72. The proportional reduction factor is... The inter-household power exchange volume during that period was reduced from 2.6kW to 2.3kW, and the remaining 0.3kW scheduling task was moved to the 14:00 time period (damage increment value 0.54, lower than the threshold of 0.72 and with 1.8kW of available capacity margin). After the 14:00 time period took over the relocation task, the simulated damage value rose to 0.61, still lower than the threshold of 0.72, and the verification passed.
[0077] If the simulated damage values for three consecutive time periods from 15:00 to 17:00 exceed the threshold, exceeding the preset consecutive threshold of 2, a rolling backtracking adjustment is triggered. Starting from 17:00, backtracking is performed hour by hour, and the remaining power exchange that cannot be digested in adjacent time periods after reducing the time periods exceeding the threshold is gradually allocated to the 14:00 time period (damage margin 0.11) and the 12:00 time period (damage margin 0.18), until the simulated damage values for all time periods do not exceed the threshold, and the corrected final scheduling instruction set is generated.
[0078] Table 6. Results of time series expansion verification and correction (13:00 to 17:00)
[0079] Implementation example of step 7 After the final scheduling instruction set is distributed to each user's adjustable load equipment and energy storage controller for execution, the changes in three sets of frequency domain characteristic parameters during the actual charging and discharging process of the energy storage battery are recorded. Taking the SEI film thickening diagnostic network as an example, the actual change in the high-frequency SEI film impedance parameter during the period from 11:00 to 13:00 is 0.053, while the predicted value of the diagnostic network is 0.047, with a prediction deviation of 0.006. This deviation value is used to update the network weight parameters online using the stochastic gradient descent algorithm. Because the prediction deviation of the SEI film thickening diagnostic network remains high for three consecutive scheduling cycles, the weighting coefficients are adjusted accordingly. The value was increased from 0.46 to 0.51, which makes the marginal damage increment matrix more sensitive to the damage contribution of the SEI film thickening mechanism.
[0080] The actual power and execution time data of each household's adjustable load equipment are fed back to the cross-user behavior complementarity matrix. Taking users A and C as examples, the actual execution data for the day shows that the peak-shifting time shift for the two households is 2.5 time periods, which deviates from the 3 time periods in the historical typical electricity consumption pattern. After updating through weighted fusion (the actual data for the day has a weight of 0.3, and the historical typical pattern has a weight of 0.7), the typical electricity consumption pattern time shift for users A and C is updated to... Complementary indicators for each time period The value was subsequently revised from 0.87 to 0.84, replacing the original typical power consumption pattern in the complementary calculation of subsequent scheduling cycles.
[0081] Throughout the implementation process, data begins with the second-level power sequence and battery operation data collected in step 1. Step 2 involves sequence pattern mining to form a cross-user behavior complementarity matrix. Step 3 uses frequency domain separation to generate severity indices and evolution rates for three mechanisms. Step 4 uses weighted calculation to form a marginal damage increment matrix. Step 5 transforms this into a constraint set and drives Pareto optimization. Step 6 performs time-by-time verification and shift correction to generate the final scheduling instruction set. Finally, step 7 executes the instruction and updates the diagnostic network weights and complementarity matrix based on actual observations. The output data from each step serves as direct input for subsequent steps, forming a complete closed-loop data flow chain.
[0082] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.
Claims
1. A distributed energy storage integrated energy scheduling method, characterized in that, include: The system acquires power data and photovoltaic output data of each user in the distribution area, performs load decomposition on the electricity consumption data of each user to generate a timeline of equipment-level electricity consumption behavior, and extracts the time series of adjustable load available capacity of each user based on the timeline of equipment-level electricity consumption behavior. The voltage and current time series of the shared energy storage battery are acquired synchronously as operating data, and the overall photovoltaic output data of the distribution area is used as the boundary condition input for optimized scheduling. Cross-user time-shift cross-correlation analysis was performed on the time series of adjustable load available capacity for each user to generate a cross-user behavior complementarity matrix; The voltage and current curves of the shared energy storage battery are separated in the frequency domain, and three sets of frequency domain feature parameters (high frequency, mid frequency and low frequency) are extracted. These parameters are then input into the corresponding diagnostic network, and the severity index and evolution rate prediction values of each aging mechanism are output. Based on the severity index and evolution rate prediction, the marginal damage increment matrix is calculated and transformed into a set of constraints. Based on the cross-user behavior complementarity matrix and the constraint set, the scheduling scheme is solved by a multi-user joint Pareto optimization model. The scheduling scheme is subjected to timing expansion verification and correction, and the final scheduling instruction set is generated and distributed for execution.
2. The integrated energy scheduling method for distributed photovoltaic storage according to claim 1, wherein, The cross-user time-shift cross-correlation analysis includes: For any two households' adjustable load available capacity time series, calculate the normalized cross-correlation coefficients for each integer time shift within the preset maximum time shift window; The maximum value of the absolute value of the normalized cross-correlation coefficient under each time shift is taken as the complementarity index between the two households. The complementarity index reflects the degree of negative correlation between the adjustable load available capacity of the two households on the time axis. The cross-user behavior complementarity matrix is generated based on the complementarity index of all user pairs.
3. The method of claim 1, wherein the method is integrated with energy scheduling of distributed photovoltaic storage. The diagnostic networks corresponding to the three sets of frequency domain characteristic parameters are the SEI film thickening diagnostic network, the lithium dendrite risk assessment network, and the cathode phase transition diagnostic network, respectively. The SEI membrane thickening diagnostic network takes as input a time series of high-frequency SEI membrane impedance parameters and outputs as the current severity index and evolution rate prediction of SEI membrane thickening. The lithium dendrite risk assessment network takes as input a time series of mid-frequency charge transfer impedance parameters and outputs as the current severity index and evolution rate prediction of lithium dendrite growth. The input to the cathode phase transition diagnostic network is the time series of low-frequency diffusion impedance parameters, and the output is the current severity index and evolution rate prediction of the cathode material phase transition. The severity index and evolution rate output by the three diagnostic networks are dimensionless values normalized to the range of zero to one. 4.The integrated energy scheduling method for distributed photovoltaic storage system according to claim 3, wherein, The SEI film thickening diagnostic network includes an impedance sequence encoding unit, a temperature modulation unit, and a severity inference unit; The impedance sequence encoding unit uses a recurrent neural network to encode the time series of high-frequency SEI membrane impedance parameters and outputs a fixed-dimensional impedance feature vector. The temperature modulation unit first transforms the surface temperature sequence encoding vector linearly through a learnable weight matrix and then generates a gate coefficient vector through a sigmoid activation function. The gate coefficient vector has the same component dimension as the impedance feature vector, and each component value represents the fusion weight of temperature encoding information and impedance feature information in the corresponding feature component. Then, the corresponding components of the impedance feature vector and the surface temperature sequence encoding vector are weighted and summed using each component of the gate coefficient vector to generate a fused feature vector. The severity inference unit outputs the current severity index and evolution rate prediction value of the SEI film thickening through the fully connected output layer from the fused feature vector.
5. The method of claim 3, wherein the method further comprises: The lithium dendrite risk assessment network includes a charge transfer feature encoding unit, a current pulse correlation unit, and a dendrite risk inference unit. The charge transfer feature encoding unit uses a recurrent neural network to encode the time series of the mid-frequency charge transfer impedance parameters and outputs a charge transfer feature vector. The current pulse correlation unit first uses a recurrent neural network to encode the amplitude and interval time sequence of the charging and discharging current pulses, and outputs a pulse coding vector; then it concatenates the charge transfer feature vector with the pulse coding vector, performs a linear transformation of the learnable weight matrix, calculates the attention weight vector through softmax, and then uses the attention weight vector to perform a weighted sum of the charge transfer feature vector and the pulse coding vector to output a pulse correlation feature vector. The dendrite risk inference unit outputs the current severity index and evolution rate prediction value of lithium dendrites through the pulse correlation feature vector via a fully connected output layer.
6. The method of claim 1, wherein the method is integrated with energy scheduling of distributed photovoltaic storage. Each element in the marginal damage increment matrix is the weighted sum of damage caused by three aging mechanisms—SEI film thickening, lithium dendrite growth, and cathode material phase transition—under a specific combination of conditions including a specific temperature range, a specific charge / discharge rate, and a specific switching frequency, when performing a unit charge operation. The damage contribution per unit of electrical charge for each aging mechanism is obtained by multiplying the evolution rate prediction value output by the corresponding diagnostic network by the conditional amplification factor, which represents the amount of damage accumulated by the mechanism for each unit of electrical charge operation under specific operating conditions; the conditional amplification factor reflects the joint modulation factor of temperature, rate and switching frequency on the evolution rate of the mechanism, and is determined by a piecewise linear lookup table fitted from historical experimental data. The weighting coefficient of each aging mechanism is determined by the proportion of the current severity index output by the corresponding diagnostic network in the sum of the severity indices of the three aging mechanisms. It is used to weight and sum the unit power damage contribution value of the three aging mechanisms to obtain the weighted damage sum. When the severity index of a certain aging mechanism is higher than the severity index of the other two mechanisms, the weighting coefficient of that mechanism is larger, so that the damage assessment is biased towards the most severe aging mechanism at present.
7. The method of claim 1, wherein the method is integrated with energy scheduling. The set of constraints includes upper limit constraints on charge / discharge switching frequency and upper limit constraints on power rate at each temperature level. The upper limit constraint on the charging and discharging switching frequency is: within each scheduling cycle, the number of charging and discharging direction switching times shall not exceed the upper limit determined by the difference value of the marginal damage increment matrix along the switching frequency dimension; The upper limit constraint of the power rate at each temperature level is as follows: for each temperature level interval on the temperature dimension of the marginal damage increment matrix, the rate of change of charging and discharging power shall not exceed the upper limit determined by the difference value of the marginal damage increment matrix along the rate dimension. When the difference value of a certain dimension exceeds the preset gradient threshold, the upper limit of the operation parameter corresponding to that dimension is taken as the current level value.
8. The distributed photovoltaic-storage-utilization integrated energy dispatching method according to claim 1, characterized in that, The multi-household joint Pareto optimization model has three optimization objectives: maximizing the overall photovoltaic absorption capacity of the distribution area, minimizing the weighted sum of the disturbances caused by the behavior of each household, and minimizing the incremental damage during the energy storage cycle. The interference level of each household's behavior is the weighted cumulative amount of the deviation between the actual dispatched power and the power of the typical power consumption mode of each household, and the weight of each household is determined by the power consumption priority declared in advance by each household. The output of the multi-household joint Pareto optimization model is a set of non-dominated solutions on the Pareto front, from which the solution with the highest comprehensive score is selected as the scheduling scheme.
9. The distributed photovoltaic-storage-utilization integrated energy dispatching method according to claim 1, characterized in that, The timing decomposition verification and correction include: The number of charge-discharge switching times and power rates of the shared energy storage battery in each time period under the execution of the scheduling scheme are simulated, and the simulation results are verified against the damage threshold of the corresponding condition combination in the marginal damage increment matrix in each time period. When the simulated damage value for a certain period exceeds the corresponding damage threshold, the inter-household power exchange volume for that period is multiplied by a reduction factor to reduce it. The reduction factor is the ratio of the damage threshold for that period to the simulated damage value. The scheduling task of the corresponding adjustable load is then shifted to the nearest period with a damage value lower than the damage threshold and with available adjustable load capacity. After the execution of the final scheduling instruction set is completed, the changes in the three sets of frequency domain characteristic parameters during the actual charging and discharging process of the shared energy storage battery are recorded. The deviation between the actual changes and the predicted values of each diagnostic network is calculated, and the weight parameters of each diagnostic network and the weighting coefficients in the marginal damage increment matrix are updated online using the predicted deviation values.
10. A distributed photovoltaic-storage-utilization integrated energy dispatching system, used to execute the distributed photovoltaic-storage-utilization integrated energy dispatching method according to any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire power data and photovoltaic output data of each user in the distribution area, decompose the electricity consumption data of each user to generate a timeline of equipment-level electricity consumption behavior, and simultaneously acquire the operation data of the shared energy storage battery. The complementarity analysis module is used to perform cross-user time-shift cross-correlation analysis on the time series of adjustable load available capacity of each user, and generate a cross-user behavior complementarity matrix; The aging feature separation module is used to perform frequency domain separation on the voltage and current curves of the shared energy storage battery, extract three sets of frequency domain feature parameters (high frequency, mid frequency and low frequency), input them into the corresponding diagnostic network, and output the severity index and evolution rate prediction value of each aging mechanism. The damage matrix calculation module is used to calculate the marginal damage increment matrix based on the severity index and the evolution rate prediction value and convert it into a set of constraint conditions. The collaborative optimization solution module is used to solve the scheduling scheme based on the cross-user behavior complementarity matrix and the constraint set through a multi-user joint Pareto optimization model. The verification and correction module is used to perform timing expansion verification and correction on the scheduling scheme, generate the final scheduling instruction set, and distribute it for execution.
